![CHAPTER 1
Introduction
1.1 Overview of rail transit robots
Railway transit is vital to the national economy. Research in Japan and China indicates
that development of the railway can lead to economic growth in the railway field [1,2].
To stimulate economic development, governments are actively developing the railway
transport industry. In such a large industrial system, the development of automation can
improve efficiency and reduce costs, so the level of automation in railways should be
increased. The increase in automation requirements in the rail transit system generates a
pursuit for rail transit robot systems.
In worldwide, many countries have proposed similar strategic plans to encourage the
development of robots in the railway transit system. Taking China as an example, “A
Country With a Strong Transportation Network” and “Smart Railway” are two important
plans that encourage the development of automation of the rail transit system. Driven by
these plans, many Chinese rail equipment manufacturers and operators have carried out
extensive research in robotics. Zhuzhou CRRC Times Electric Co., Ltd. applied an
automatic generating line to produce high-speed train converters [3]. CRRC Qishuyan
Institute Co., Ltd. developed an intelligent manufacturing workshop for gear transmission
systems for high-speed trains. CRRC Zhuzhou Institute Co., Ltd. combined automatic
guided vehicle (AGV) technology with urban transportation equipment and designed
autonomous rail rapid transit (ART) [4].
A variety of robot systems are employed. The use of robot systems in the rail transit
system can be divided into three aspects: manufacturing, dispatch, and maintenance. In
these three parts of rail transit, different kinds of robots have completely different roles, as
shown in Fig. 1.1.
1.1.1 Rail transit robots in manufacturing
“A Country With a Strong Transportation Network” points out that intelligent manufacturing
is required for the rail transit system. In the modern production line, robots can greatly
improve processing efficiency. Taking the production of a high-speed train gearbox as an
Robot Systems for Rail Transit Applications. https://doi.org/10.1016/B978-0-12-822968-2.00001-2
Copyright © 2020 Elsevier Inc. All rights reserved.
1](https://image.slidesharecdn.com/34088-250403165553-e555ad51/85/Robot-Systems-for-Rail-Transit-Applications-1st-Edition-Hui-Liu-20-320.jpg)
![example, in the process flow of a transmission gear, robot arms can result in the efficient
transmission of gears between different machine tools; when welding the gearbox, the
welding robots can improve machining efficiency; when assembling the gearbox, assembly
robots can work with high-precision; when the train is assembled, AGV enables the gearbox
to be transported quickly between workshops. Applying robots in manufacturing not only
saves processing time and labor costs, it improves manufacturing quality.
1.1.1.1 Assembly robots
As for assembly robots, the primary mission is to achieve high-precision positioning of the
workpiece. According to previous research, assembly costs account for 50% of total
manufacturing costs [5]. Assembly robot systems can also be divided into rigid assembly
and flexible assembly robots. Rigid assembly robots are customized processing systems for
specific workpieces in the traditional industrial environment.
Rigid assembly robots have poor generalization. If the production line is replaced with
processed parts, the equipment needs to be customized. Replacement of equipment will
cause a great economic burden. Compared with rigid assembly robots, flexible assembly
robots can design customized processing programs according to the workpiece. Flexible
assembly robots are programmable, which can result in different assembly schemes for
different workpieces. Flexible assembly robots are significant for a flexible assembly system.
In current industrial development, flexible assembly robots are the focus of development [6].
In the following discussion, assembly robots refer to flexible assembly robots.
Figure 1.1
Robots in the manufacturing, dispatch, and maintenance of rail transit.
2 Chapter 1](https://image.slidesharecdn.com/34088-250403165553-e555ad51/85/Robot-Systems-for-Rail-Transit-Applications-1st-Edition-Hui-Liu-21-320.jpg)
![An assembly robot consists of four components: machinery components, sensors,
controllers, and actuators. To bring about a complex workpiece track in the real assembly
environment, assembly robots usually have more than four degrees of freedom (DOFs).
Mainstream assembly robots can be divided into two types: selective compliant assembly
robot arms (SCARAs) and six-DOF robots.
SCARAs have four DOFs, which are commonly used in electronic assembly, screw
assembly, and so on [7]. SCARAs are specially designed for assembly applications by
Yamanashi University. SCARAs contain two parallel joints, which can assemble a
workpiece in a specified plane. Compared with six-DOF robots, advantages of SCARAs
are a higher assembly speed and precision; disadvantages are limited workspace.
Commonly used control strategies for SCARAs contain adaptive control, force control,
robust control, and so forth [8]. In state-of-the-art research on robot control, intelligent
algorithms are employed to improve control performance [9]. Dulger et al. applied a
neural network to control the SCARA [10]. The neural network was optimized by particle
swarm optimization to improve performance. Son et al. adopted an optimized inverse
neural network for feedback control [11]. To deal with disturbances in running, the
parameters of the inverse neural network are updated by a back propagation algorithm.
Luan et al. used the radial basis function (RBF) neural network to achieve dynamic
control of the SCARA [12].
Six-DOF robots can locate the workpiece at almost any point. Thus, six-DOF robots can
handle the assembly task of complex three-dimensional (3D) workpieces. The dynamics
of six-DOF robots are basic for operating the robots. Zhang et al. considered the friction
of the robots and used a hybrid optimization method to model the dynamics of the
six-DOF robot [13]. After optimization, dynamic accuracy increased significantly. Yang
et al. proposed a simulator for the dynamics of the six-DOF robot [14]. Robots with a
large degree of freedom have large feasibility. However, too much freedom is
uneconomical. To handle the trade-off between economy and feasibility, the DOF can be
optimized for specific tasks. Yang et al. proposed an optimization method to minimize
the DOF [15]. This optimization method can reduce the DOF and improve the use of the
DOF.
Assembly robots should cooperate with the ancillary equipment. The fixtures are vital
equipment to ensure cooperation in performance. The fixtures can fix the relative position
between the workpiece and the robot under load. If the precision of the location of the
fixtures is low, no matter how accurate the positioning precision of the robot is, it cannot
achieve high-precision assembly. Currently, the flexible fixture is a future development
[16]. Lowth et al. proposed a unique fixture that can adjust the radial and angular
adaptively [17]. Although auxiliary devices are applied for assembly robots, the results of
assembly robots may still be unsuccessful. Avoiding unsuccessful assembly is particularly
Introduction 3](https://image.slidesharecdn.com/34088-250403165553-e555ad51/85/Robot-Systems-for-Rail-Transit-Applications-1st-Edition-Hui-Liu-22-320.jpg)
![important in electric connector assembly, because the electric connector is not a rigid
component. To detect the unsuccessful assembly of the electric connector assembly, Di
et al. proposed a hybrid detection system with a force sensor and camera [18].
The fault diagnosis and prognosis system of assembly robots guarantees assembly
accuracy. There are many studies about fault diagnosis and prognosis systems. Huang
et al. designed a classifier for the wiring harness robot [19]. That study modeled the
manufacturing process was and calculated the fault with a fuzzy model. Baydar et al.
introduced a diagnosis model with error prediction [20]. The proposed model integrated
the Monte Carlo simulation, genetic algorithm, and so forth. The functions of the fault
diagnosis and prognosis system for assembly robots should contain the main aspects as
given in Choo et al. [21]:
(a) The health states of the assembly robots are monitored in real time. The monitored
data are logged into the dataset. The health features are extracted from the health
states of the assembly robots. The faults and remaining useful life can be calculated
according to the features.
(b) According to the fault diagnosis and prognosis results, the assembly tasks are reas-
signed to make sure the failed assembly robots are replaced by the fully functioning
robots. The maintenance plans can be made to repair the failed robots.
1.1.1.2 Collaborative robots
When applying the resulting classical industrial robot systems, interaction between robots
and humans is limited. There are three reasons for this phenomenon:
(a) Traditional industrial robots do not consider moving humans. If a collision occurs along
a certain trajectory, it may cause great damage to humans. Therefore, the working area
of the industrial robot is mostly separated from the working area of the human.
(b) The weight and volume of traditional industrial robots are large, and it is difficult for
humans to operate robots.
(c) Reprogramming of robots is difficult and requires special programming tools for
tuning.
However, humanerobot collaboration can combine human creativity with the efficiency of
robots and can amplify the flexibility of robots and further improve work efficiency. Under
this demand, collaborative robots are born. Compared with traditional robots, collaborative
robots have three main advantages: safety, ease of operation, and ease of teaching. Some
robot manufacturers have launched collaborative robotics products. The robot company
Universal Robot launched the UR3 collaborative robot [22]. This robot is the first truly
collaborative robot. The UR3 collaborative robot is based on a six-DOF robotic arm and is
flexible enough to achieve complex motion trajectories. In terms of safety, the UR3
4 Chapter 1](https://image.slidesharecdn.com/34088-250403165553-e555ad51/85/Robot-Systems-for-Rail-Transit-Applications-1st-Edition-Hui-Liu-23-320.jpg)
![collaborative robot has a collision monitoring system that protects human safety by
monitoring the joint position, speed, and power of the robot.
KUKA Robotics has launched a collaborative robot, the LBR iiwa [22]. The robot’s seven-
DOF design provides greater flexibility than traditional six-DOF robots, enabling more
complex trajectories to cope with complex environments that work with humans. The
shape of the robot is designed to be ergonomic and easy for humans to operate. The outer
casing is made of aluminum alloy, which can reduce weight and improve operability. The
robot is equipped with torque sensors at each joint to monitor collisions in real time. The
teaching method of the robot is dragging, which reduces the technical threshold of the
robot operator.
ABB launched the robot YuMi [22]. The robot is highly safe and can achieve
humanerobot interaction in a small space. To improve the performance of collaborative
robots, AAB acquired Gomtec Robotics, which launched the collaborative robot Roberta
[22]. The Roberta can handle higher load applications compared with YuMi.
Franka Emika launched the Franka Collaboration [23]. Like the LBR iiwa, the robot has
seven DOFs. It is also equipped with a torque sensor on each joint to enable collision
monitoring.
Rethink Robotics launched the two-arm collaborative robot Baxter and the one-arm
collaborative robot Sawyer [24]. The two robots are exquisitely designed with high
positioning accuracy and can be assembled with high precision.
Safety in humanerobot interaction is essential for collaborative robots. Threats to the
safety of collaborative robots can be divided into two aspects [25]:
(a) The first kind of threat is from robots. During operation, the robot may collide with
workers and cause injuries. To ensure the safety of employees, the robot needs to
detect the location of workers in real time and determine whether the location of
workers is in a safe position. If workers intrude on the safe area, the robot should
immediately stop to avoid a collision. In case a collision between a person and a robot
occurs, the robot needs to detect the collision in time and change torque to minimize
damage to workers. Mohammed et al. proposed a collision avoidance system for
collaborative robots [26]. This system used depth vision sensors to detect the position
of the worker. Considering the virtual model of the collaborative robot, a collision
could be detected. The collaborative robot could take measures to avoid collisions. In
addition to collision detection, it is necessary to ensure the integrity of the collabora-
tive robot control system during operation. Failure of any part of the sensors, control-
lers, or actuators will lead to the failure of humanecomputer interaction, thus
threatening the safety of workers. In addition, interaction with robots may cause
mental stress to workers [27], which will increase the risk for a collision.
Introduction 5](https://image.slidesharecdn.com/34088-250403165553-e555ad51/85/Robot-Systems-for-Rail-Transit-Applications-1st-Edition-Hui-Liu-24-320.jpg)
![(b) The second kind of threat is from the industrial process. In the process of humane
robot interaction, workers need close contact with the manufacturing process. The
temperature of the manufactured workpieces can cause damage to workers. Fallen
workpieces can also endanger workers. Therefore, it is necessary to fully consider the
impact of machining parts on workers in the design process of collaborative robots. In
addition, the unreasonable ergonomic design of the collaborative robot during mainte-
nance will have an impact on safety.
1.1.1.3 Automatic guided vehicles
According to the level of automation, manufacturing systems can be divided into three
levels [28]. Manufacturing systems in the first level are manual. Those in the second level
have small-scale automated manufacturing in which transportation is carried out manually.
Manufacturing systems in the third level have large-scale automated manufacturing and
use automated transportation. Manufacturing systems in the third level are also called
flexible manufacturing systems (FMSs). According to the Material Handling Industry of
America, only 20% of the time is spent on processing and manufacturing; the remaining
80% is used for storage, handling, waiting for processing, and transportation [29]. As
factory automation increases, transportation efficiency between workstations needs to
improve. In the FMS, the AGV can improve the use of space in the factory and the
efficiency of transportation in the material handling system. Therefore, transportation costs
can be reduced.
The rail transit manufacturing system is a typical FMS. Transportation is an important part
of the rail transit manufacturing system. In the rail transit processing environment, not
only the transfer of workpieces between processes within the plant but also the free flow
of workpieces between plants is required. In the traditional rail transit manufacturing
environment, transportation inside the factory is carried out by gantry cranes and the
transportation between factories is carried out by trucks. These modes of transportation
have some disadvantages. There is a safety hazard when using gantry cranes to lift. If the
lifting workpiece falls, it may cause serious safety accidents. The use of trucks carries a
higher cost and is less efficient to transport. Improving the safety and economy of products
in the transportation process is important for the development of the rail transit industry. In
the current manufacturing environment, the AGV is an effective mode of transportation.
The AGV is safer compared with gantry cranes and is more automated and more efficient
for transportation than trucks.
The typical structure of an AGV consists of sensors, chassis, a control unit, and so on
[30,31]. During work, the sensor can determine the position of the AGV and transmit the
current position to the control decision system, and the control decision system plans an
6 Chapter 1](https://image.slidesharecdn.com/34088-250403165553-e555ad51/85/Robot-Systems-for-Rail-Transit-Applications-1st-Edition-Hui-Liu-25-320.jpg)
![optimal path. The chassis is driven by the controller’s control command to transport the
workpiece to the designated position.
To improve transportation performance, the model predictive control of the AGVs should
be achieved. The remaining power and the working state of the AGV should be predicted
to calculate the remaining life of the AGV, and the dispatch plan of the AGV can be
optimized with consideration of these factors to improve the operational safety of the
AGV. Popular data-driven forecasting methods contain statistical methods, intelligent
methods, and hybrid methods. The statistical methods can discover the statistical rule of
the data and generate an explicit equation for prediction. Commonly used statistical
methods contain autoregressive moving average, Winner process, Gaussian process, and so
on. Intelligent methods can generate better forecasting performance than statistical
methods with the help of the strongly fitting capacity of the neural network. The Elman
neural network, multilayer perceptron, and extreme learning machine (ELM) are the three
most popular intelligent methods. However, the training process of these neural networks
depends on the initial values in some way. If the initial value is unsuitable, training of the
neural network may stop at the locally optimal solution. To improve the performance of
intelligent prediction methods, the initial values can be optimized by optimization
algorithms [32]. Hybrid prediction methods combine data processing algorithms with
statistical or intelligent methods. The decomposition algorithms are proved to be effective
[33]. The decomposition algorithms can divide the raw series into several more stationary
subseries. Each group of subseries has a simpler fluctuation mode than the raw series, so it
is more predictable.
After optimization, the control command can be assigned to the AGVs in two different
ways: static control and dynamic control [34]:
(a) Static control. The control commands are assigned before the task. Once the AGVs
receive the control command, the transportation path will not be changed until the
AGVs receive another control command. This control scheme is simple and easy to
operate. However, flexibility is weak.
(b) Dynamic control. This control method can adjust the control commands according to
the real-time state of the AGVs, so the task scheduling strategy is complex.
The multi-AGV system is being studied worldwide. Compared with the single AGV, the
advantages of the multi-AGV system are:
(a) The multi-AGV system can cover a large area. The single AGV can achieve trans-
portation only between points. In the modern manufacturing environment, the
transportation task is far more complex than the point-to-point transportation. The
multi-AGV system can build a transportation network and improve transportation
efficiency.
Introduction 7](https://image.slidesharecdn.com/34088-250403165553-e555ad51/85/Robot-Systems-for-Rail-Transit-Applications-1st-Edition-Hui-Liu-26-320.jpg)
![(b) The multi-AGV system can execute the transportation task in parallel. A multi-AGV
system can perform tasks simultaneously for a complex task. Thus, the multi-AGV
system can greatly improve the efficiency of transportation.
The dispatch and routing of the multi-AGV are important. The conflict-free function of the
multiple AGVs is the bottleneck for the multi-AGV system. Draganjac et al. proposed a
control algorithm for the multi-AGV system [35]. The proposed algorithm can detect the
conflict between the AGVs and guarantee the safe operation of the multi-AGV by the
priority mechanism. Miyamoto et al. proposed a conflict-free routing algorithm for the
multi-AGV system [36]. Because of the limited memory space of each AGV, a heuristic
algorithm was adopted for routing. Małopolski et al. considered the transportation system
in the factory as a combination of squares [37]. Based on the square topology, a novel
conflict-free routing algorithm for the multi-AGV system was proposed.
The dispatch plan of the multi-AGV should be calculated to optimize the task waiting
time, collision, load use, and so forth [38]. The optimization methods can be divided into
single- and multiple-objective optimization. Single-objective optimization can optimize
only one objective function. If the objective function consists of several objective
subfunctions, these subfunctions should be combined as the weighted sum [39]. However,
it is difficult to design these weights. Therefore, the generated optimization results might
not be the global optimal solution. The multiple-objective optimization can balance the
trade-off between different subfunctions and generate a Pareto front [40]. The Pareto front
contains many solutions, each of which has both advantages and disadvantages. The final
optimization results should be selected according to the expert. In this manner, the
intelligent decision-making ability of humans can be used to improve the dispatch
performance of the AGV.
The environment of the factory is dynamic. The AGVs should be flexible enough to cope
with the dynamic manufacturing system. There are many studies on the dynamic
transportation system. Brito et al. proposed a dynamic obstacle avoidance algorithm for
the dynamic unstructured environment [41]. In this algorithm, model predictive control is
applied to improve control performance. This algorithm was verified in the environment
with walking humans. Li et al. proposed an integrated algorithm for obstacle avoidance
[42]. This algorithm can generate a path for the AGV by a model-predictive algorithm.
The AGV can track the generated path reliably.
1.1.1.4 Manufacturing robots
Manufacturing robots contain many types including welding robots, drilling robots,
grinding robots, milling robots, and so on. Manufacturing robots can produce better
machining performance than a classical computerized numerical control (CNC) machine.
8 Chapter 1](https://image.slidesharecdn.com/34088-250403165553-e555ad51/85/Robot-Systems-for-Rail-Transit-Applications-1st-Edition-Hui-Liu-27-320.jpg)
![For example, it is proven that a workpiece polished by a manufacturing robot has a better
surface quality than any CNC machine [43]. The better performance of the manufacturing
robots is because the robots have more flexibility to make sure the tool is in the right
position.
Welding robots are one of the most widely used manufacturing robots. Difficulties of
welding robots are [44]: (1) it is hard to observe the welding seam in a complex
manufacturing task, (b) it is hard to obtain the absolute and relative locations of the
workpieces, and (c) the trajectory is hard to track. Current commonly used location
methods for the welding seam are based on optical sensors such as a depth camera and
laser sensor [45,46]. Jia et al. proposed a welding seam location method and trajectory
tracking algorithm [44]. The proposed location method was achieved using a laser scanner,
which could obtain the location and direction of the pipe. The cubic spline was applied to
fit the obtained welding seam. The velocity control was used to track the welding
trajectory. Liu et al. proposed a trajectory planning algorithm for welding robots to cope
with a single Y-groove welding task [47]. This study provided two different velocity
planning algorithms.
A key concern with drilling robots is positioning accuracy. An inaccurate hole can reduce
the mechanical performance of the equipment. The positioning compensation method can
be divided into two types: model-based and model-free [48]. Model-based methods can
guide the robot to move according to the measured positioning error. The essentials of
model-based methods are the dynamics and kinematics of the drilling robots. Model-based
methods are time-consuming. Model-free methods can solve this drawback. Model-free
methods build a model to describe the relationship between the positioning error and the
robot’s joints’ parameters, which do not consider the dynamics and kinematics of the
robots. Commonly used model-free methods contain the interpolation method [49],
cokriging method [50], and so on. Neural networks are applied to compensate intelligently
for positioning . With the help of the strongly fitting capacity of neural networks, these
intelligent positioning compensation methods obtain good performance. Yuan et al. used
ELM to predict positioning error and guide the robot to compensate for it [51]. Chen et al.
used the RBF neural network to estimate positioning error. The bandwidth of the adopted
RBF neural was fine-tuned. Positioning accuracy can be improved by more than 80% with
these intelligent positioning compensation methods [52].
The grinding manufacturing is highly precise, so the positioning accuracy of the grinding
robots requires attention. In real applications, the grinding robots may deviate from the
preset track because of the disturbance and wear to the tools. Thus, grinding robots should
have the ability to self-adjust, to ensure manufacturing quality. Control of grinding robots
is more difficult than for CNCs because grinding robots have more DOFs. Huang et al.
proposed an intelligent gear grinding system [53] in which the robot arm can detect the
Introduction 9](https://image.slidesharecdn.com/34088-250403165553-e555ad51/85/Robot-Systems-for-Rail-Transit-Applications-1st-Edition-Hui-Liu-28-320.jpg)
![actual trajectory by a vision sensor and adjust the trajectory adaptively. Two cameras are
adopted for tool center point calibration. When coping with the complex grinding task,
multiple grinding robots are necessary, because multiple grinding robots can cope with the
complex manufacturing task more easily than a single grinding robot. Han et al. [54]
proposed a multiple grinding robot system and introduced the trajectory planning method
for the multiple-robot system. Experimental studies indicated that the multiple grinding
robot system can generate a steadier manufacturing trajectory.
The robot only needs five DOFs for the milling task; the reserved one is the DOF of the
spindle of the milling cutter [55]. In the real application, six-DOF robots are applied for
milling to improve flexibility. The stiffness of milling robots is a challenge in
manufacturing. Milling robots usually have low stiffness. Robots may deform or vibrate
when milling workpieces. There have been many studies on improving milling
performance. Peng et al. optimized the stiffness of the robots in the feed direction with a
seven-DOF robot [56].
Commonly used manufacturing robots are developed from six-DOF robots. Trajectory
planning is a common technology for manufacturing robots. The trajectory planning
algorithm can generate the optimal trajectory and the robot control algorithm can drive the
robot to follow a predetermined trajectory [57]. The optimization targets of the trajectory
are the position, velocity, accelerated velocity of each joint. The trajectory planning
algorithm is the basis of the industrial robot control system. The optimization functions of
the trajectory contain many aspects, including reduce execution time, consumed energy,
and impact.
1.1.1.5 Loadingeunloading robots
In the existing transportation manufacturing environment, processing each workpiece
requires different procedures. The connection between processes requires the workpiece to
be moved from one machine to another [58]. This repetitive work can be replaced by
loadingeunloading robots. The application of loadingeunloading robots has many
benefits. On the one hand, loading and unloading of robots is repetitive and consistent,
which can avoid a decrease in work efficiency caused by worker fatigue. On the other
hand, using the robots can parameterize the loading and unloading operation and the fault
source can be quickly located when troubleshooting problems, which is convenient for
improving the processing quality.
Loadingeunloading robots have a wide range of applications. Liu et al. adopted a
loadingeunloading robot for the CNC and designed a control system based on a
programmable logic controller [59]. Zhang et al. developed a robot system for
loadingeunloading [60]. The designed robot was able to separate good and bad
workpieces. Fan et al. used the loadingeunloading robot for electric equipment detection
10 Chapter 1](https://image.slidesharecdn.com/34088-250403165553-e555ad51/85/Robot-Systems-for-Rail-Transit-Applications-1st-Edition-Hui-Liu-29-320.jpg)
![[61]. With the help of the loadingeunloading robot, automatic and high-precision
detection is achieved.
The development of loadingeunloading robots shows a trend in intelligence. Computer
vision and neural network technology have been widely applied [62]. The adoption of
computer vision can increase the flexibility of loadingeunloading robots.
Loadingeunloading robots can observe the posture of the workpiece and adjust the robots
to grasp the workpiece from the appropriate position. The neural network is widely used in
controller design with its strong generalization ability. Gu et al. proposed a visual servo
loadingeunloading robot [63]. The camera on the manipulator can detect the location of
the workpiece and help the robot capture the workpiece precisely. The design contains two
steps: (1) extract the features of the observed image and target image and calculate the
error between these features; and (2) input the error into the neural network and output the
control command for the robot. In the proposed robot control structure, the fuzzy neural
network was adopted as the controller.
1.1.2 Rail transit robots in dispatch
The running stage of the rail transit system takes up the longest length and generates the
highest cost in the whole life cycle of the rail transit system. With the help of robot
technology, the rail transit system can achieve automatic operation, thus improving
operational safety and reducing transportation costs. Autonomous driving is the most
important form of automation in rail transit. Autonomous driving is the application of
robot control technology to the railway train. The automatic train control of the railway
train contains automatic train supervision (ATS), automatic train protection (ATP), and
automatic train operation (ATO) [64]. The ATS can supervise the running states of the
railway train. The ATP system can monitor the train’s running position and obtain its
speed limit to ensure its running interval and safety. The ATO can accept output
information of the ATS and ATP and generate control instructions [65].
The international standard International Electrotechnical Commission 62,290 defines four
grades of automation (GOA) [66]:
(a) GOA1. The train is able to monitor the train’s operating status continuously. Operation
needs to be carried out by drivers.
(b) GOA2. This level is self-driving with driver duty. The train is able to drive automati-
cally through the signal system, but the driver is required to close the door and issue
the command.
(c) GOA3. There is no need to equip drivers on this level of the train; the train can auto-
matically complete the whole process of operation, including outbound, pit stop,
Introduction 11](https://image.slidesharecdn.com/34088-250403165553-e555ad51/85/Robot-Systems-for-Rail-Transit-Applications-1st-Edition-Hui-Liu-30-320.jpg)
![switch door, and so on. However, there is still a need for onboard personnel to deal
with emergencies on this level of train.
(d) GOA4. No operator is required on this class of trains, and the train’s control system
can automatically respond to unexpected situations.
ART is an important example of the automation of rail transit operation; it is developed by
CRRC Zhuzhou Institute Co., Ltd. The core technology of this train is virtual orbit
tracking control technology. That is to say, the ART train does not need a true track, but
rather, the marks on the road serve as tracks. Because the intelligent rail train does not
have a physical track, existing positioning devices used in the subway train and high-speed
train cannot be used. The ART adopts the global satellite positioning system as the main
positioning method [4]. During the operation, inertial sensors and angle sensors installed
in intelligent trains are used to monitor the movement posture and position of the train,
thereby enabling a comprehensive perception of the state of the ART train. The monitored
train status information is fed back to the controller of the intelligent rail train to control
the intelligent rail train to run along the virtual track. Using this control technology, all
wheels of the ART train can be driven along a virtual track [4]. This position accuracy can
ensure the passing performance of the ART rail train in urban traffic. To improve
operational efficiency, the ART train is also equipped with automatic driving technology
[4]. The ART train can intelligently sense the environment and control the automatic
operation of the train.
1.1.3 Rail transit robots in maintenance
Robots have become the focus of modern rail transit equipment manufacturing and operation.
With the increase in railway operating mileage and operational density, the workload of
railway maintenance support is also increasing, and higher requirements are put forward for
the maintenance and repair of railway infrastructure equipment. Currently, the maintenance of
railway infrastructure equipment is mainly carried out by workers [67]. The main problems
Figure 1.2
Role of robots in rail transit maintenance.
12 Chapter 1](https://image.slidesharecdn.com/34088-250403165553-e555ad51/85/Robot-Systems-for-Rail-Transit-Applications-1st-Edition-Hui-Liu-31-320.jpg)
![are that (1) diversified test operations have higher requirements for operators, (2) it is
inconvenient to have operators carry measurement equipment, and (3) manual inspections
maintain labor intensity and quality is uncontrollable. As shown in Fig. 1.2, robots can have
an important role in the maintenance stage of rail transit in this case.
1.1.3.1 Inspection robots
The objectives of railway maintenance include tracks, instruments, and equipment next to
railways, traction substations, pantographs, temperature of instruments, bogies, and foreign
matter intrusion. According to the work scenario, the inspection area of rail transit
inspection robots is mainly divided into traction substations and railway.
1.1.3.1.1 Inspection in traction substation
The traction substation converts electric energy from the regional power system into
electric energy suitable for electric traction according to the different requirements of
electric traction on current and voltage. The resulting electricity is then sent to catenary
lines set up along the railway line to power electric locomotives, or to the urban rail transit
system to power electric subway vehicles. The inspection work is the top priority for the
daily maintenance of the traction substation. However, because of insufficient staffing and
other reasons, the inspection cannot meet the needs. In addition, with the increase in
substations, operation, and maintenance, personnel are struggling to comply. It is also
difficult to keep satisfactory working conditions and a strong sense of responsibility.
Currently, the technology of intelligent robots is increasingly developing and the use of
robots to replace manual inspections of equipment partially or completely has become the
trend of future substations [68]. Substation intelligent inspection robots are equipped with
advanced equipment such as infrared thermometers, high-definition cameras, and audio
signal receivers. According to the preset inspection tasks, the entire station equipment and
partitions are inspected in time. All task modules of meter reading, temperature
measurement, sound collection, and functional inspection are processed together to
identify and alert regarding discovered equipment abnormalities, defects, and hidden
dangers in a timely manner. In addition, robots intuitively form charts and report weekly
on operational data indicators of devices stored in the inspection so that maintenance
personnel can perform equipment maintenance and fault analysis better.
By using the inspection robots in the substation, the equipment in the substation can be
inspected more reliably and the data analysis returned by the system can give the
maintenance personnel a better understanding of the working state of the equipment.
Recording and analyzing historical data can also significantly improve the predictability of
equipment in the substation, providing more effective data support for the condition of
maintenance and a status assessment of the equipment, which will directly reduce the
possibility of equipment accidents [69]. Using inspection robots to inspect substation
Introduction 13](https://image.slidesharecdn.com/34088-250403165553-e555ad51/85/Robot-Systems-for-Rail-Transit-Applications-1st-Edition-Hui-Liu-32-320.jpg)
![equipment can not only keep field maintenance personnel away from the field equipment,
which improves the safety and reliability of the inspection work, it provides a new method
for spot inspection for the popularization of an unattended substation and truly improves
the safety and reliability of substation equipment.
1.1.3.1.2 Inspection along the railway
With the rapid development of rail transit and the continuous improvement in passenger
high-speed processes, railway transport has been gradually heading toward function
integration, information sharing, command concentration, and highly automated transition
and transformation. The urgent demand for railway transportation safety poses new
challenges to the railway traffic safety assurance system. To adapt to the new needs of
railway development, it has become the focus of the construction of the railway traffic
safety assurance system to build a fully covered and highly reliable railway intelligent
monitoring system based on high information sharing [70]. The higher speed brings
convenience to people, but it also increases the difficulty of preventing railway disasters.
Although a high-speed train is equipped with a more advanced safety system, the large
braking distance caused by high-speed driving reduces the flexibility of the train to
respond to emergencies. In the event of an accident, a high-speed train often has more
serious consequences than an ordinary train. Therefore, various factors such as the
intrusion of foreign matter, natural disasters, and line failures may cause major railway
accidents. Casualties and property loss caused by the intrusion are enormous and
frequently occur. Therefore, research and development into railway foreign matter
intrusion detection are of great significance to ensure the safe operation of trains [71].
The intrusion of foreign matter has the characteristics of irregularity, suddenness, and
unpredictability. The fixed transportation mode of the operation route makes emergency
measures that can be adopted limited when the train detects an intrusion on the road. If
real-time detection of invasive foreign matter can be achieved and necessary measures are
taken in time, property loss can be reduced as much as possible. The traditional foreign
matter intrusion approach is to place sensors at key positions along the railway to monitor
intrusion conditions in real-time. This method has many drawbacks, such as an
unreasonable arrangement of the sensor and fixed monitoring position. With the rapid
development of robotics, robots have entered the field of railway foreign matter invasion
detection. The railway is inspected by an infrared sensor, a laser sensor, a smart camera,
and an ultrasonic device.
1.1.3.2 Channel robots
High-speed trains may experience poor performance or even failure of key components
during long-term high-speed service. If a safety accident occurs, it will cause severe
economic loss and even casualties. Therefore, how to monitor and diagnose its operating
14 Chapter 1](https://image.slidesharecdn.com/34088-250403165553-e555ad51/85/Robot-Systems-for-Rail-Transit-Applications-1st-Edition-Hui-Liu-33-320.jpg)
![status has become an important research topic [72]. Among them, the bogie is a main
components of the train. The basic structure of the bogie mainly includes the wheel set,
suspension system, frame, auxiliary suspension system, and so on. Therefore, it is very
important to research fault diagnosis for the high-speed train, especially the diagnosis of
bogie faults [73].
With the complex diversification of high-speed rail operating conditions and the increase
in operating mileage, mechanical wear and aging of vehicles are gradually accelerating,
which causes safety hazards for train operation. Among them, the train axle is an
important component to support the running of the train. During the driving process,
almost all of the weight of the train and the impact caused by vibration are assumed.
Therefore, the axle is one of the most vulnerable parts of the train. As a result of improper
assembly of the axles and excessive operation of the axles during maintenance, the weight
of axles will be aggravated. When the axle temperature rises abnormally, movement
between the axle and bearings will be worse, friction and wear will be aggravated, and
smoothness will be reduced. In this case, if the vehicle continues to run without
emergency treatment, hidden safety hazards will occur and the train will be delayed or
even derailed. Therefore, the health status of the bogie is directly related to the safety of
the vehicle operation. The fault warning and diagnosis of the axle of the high-speed train
are an urgent problem to be solved.
With the rapid development of sensor technology, a large amount of real-time monitoring
data generated during the operation of high-speed trains has been collected. Through the
analysis and mining of historical monitoring data, the potential correlation and periodicity
can be explored to provide a basis for train fault diagnosis. By analyzing the monitoring
data collected during the train’s operation, it is important to identify the health status and
the potential safety hazards that may exist at the time of the early warning, which is
greatly significant for ensuring the safe operation of the train and improving the reliability
of the train’s operation.
Bogie fault diagnosis includes fault vibration signal processing and fault identification.
Among them, fault vibration signal processing includes time-domain analysis, frequency-
domain analysis, and time-frequency analysis. Fault identification uses the expert system,
pattern recognition, neural networks, deep learning, and so forth [74,75]. Because the
high-speed train bogie has nonlinear characteristics, its vibration signal is a nonlinear,
nonstationary signal. Therefore, applying effective fault vibration signal processing
methods to extract state information and constructing a fast and effective bogie fault
diagnosis method is critical for accurately evaluating the safety state of the train.
With the rapid development of robotics, more and more trains are on the line and rail
transit channel robots have emerged. Compared with industrial robots and inspection
robots, rail transit channel robots have been in development for a short time, but they form
Introduction 15](https://image.slidesharecdn.com/34088-250403165553-e555ad51/85/Robot-Systems-for-Rail-Transit-Applications-1st-Edition-Hui-Liu-34-320.jpg)
![a certain scale. Channel robots are mainly based on intelligent technology and integrate
various sensing equipment. It is dedicated to the collection and diagnosis of fault
information. The biggest contribution of rail transit channel robots is to replace on-site
workers to check the train’s status. The staff of the inspection station can carry out remote
control operations through network technology and can reliably monitor the site in real
time, which can objectively monitor the equipment and meet the requirements of
flexibility, high precision, and antiinterference. In general, rail transit channel robots need
to meet the following requirements:
(a) The channel robot belongs to the mobile robot. Unlike the fixed industrial robot, it
needs to move according to the actual task. Therefore, the navigation control system
needs to have a control module to perform robot motion control. Of course, the move-
ment of the channel robot can be along the guide rail or autonomous navigation.
(b) The robot needs to carry a variety of sensors to obtain information about the detected
target, laying the foundation for intelligent fault diagnosis.
(c) The navigation system needs to be modularized to make it easy to replace modules
and easy to maintain. Communication between modules must be completed.
(d) The channel robot needs to upload collected information and the corresponding fault
type, fault location, and so on to the prebuilt historical fault database. With the contin-
uous increase in fault information, the database has gradually grown. Data mining can
be used to predict the fault and life of train equipment and to identify potential
dangers in advance.
Rail transit channel robots mainly perform a fault diagnosis of trains. Generally speaking,
the workflow is as follows: (1) the maintenance task is carried out along the ground track
to the checkpoint; (2) the vehicle maintenance information is collected by the onboard
sensor including the visible light camera, the laser sensor, and the infrared camera, and
some sensors are installed on the robot arm; (3) the collected image for fault diagnosis
analysis is uploaded; and (4) the fault diagnosis results are reported and stored, including
the fault types, fault problems, fault degrees, and so on. In addition to the growing
database, the big data analysis module will perform data mining on train bogie equipment
failure information to predict potential equipment failures. In addition, the robot
continuously detects its own power and performs a recharging task once the minimum
battery threshold is reached.
Channel robots with a trouble of moving electric multiple units (EMU) detection system
(TEDS) intelligent sensing system have been used for EMU fault diagnoses, aiming to use
the rail transit channel robots to complete train status detection and fault diagnosis [76].
The hardware structure of a channel robot mainly includes a ground rail, dual-arm robot,
laser sensor, visible light camera, infrared camera, server, and recharging device. Among
them, the function of the robot guide is to drive the robot and make it move along the
16 Chapter 1](https://image.slidesharecdn.com/34088-250403165553-e555ad51/85/Robot-Systems-for-Rail-Transit-Applications-1st-Edition-Hui-Liu-35-320.jpg)
![preplanned route to expand the working radius of the industrial robot. The arm of a
two-armed robot is similar to that of an industrial robot. The difference is that a two-
armed robot can perform more complex tasks and control is more difficult. Laser sensors,
visible light cameras, infrared cameras, and other in-vehicle sensors belonging to a
channel robot are used to collect different types of data. The background server is mainly
used for the channel robot control, data storage, fault diagnosis, and big data processing.
The self-recharging technology of the recharging device is an important means to ensure
the normal operation of the channel robot. Currently, the more advanced one is noncontact
recharging. Compared with contact recharging, noncontact recharging has the advantages
of a small footprint and low cost.
The software of a rail transit channel robot mainly includes intelligent visible light image
analysis, infrared thermal imaging temperature analysis, location detection, and big data
analysis. Among them, visible image intelligent analysis and infrared thermal image
analysis are used for fault diagnosis of the train. Visible light image analysis mainly
includes image registration and change detection algorithms, and the train fault is found
by comparing images. Infrared thermal imaging temperature analysis mainly includes fault
location and image segmentation, fault location image recognition, and classification. The
former uses image processing technology and the latter relies on machine learning
methods. Location detection mainly uses laser ranging technology to determine the
location of train faults. Big data analysis is mainly based on the train bogie equipment
fault information database; it uses data mining and machine learning and predicts the
potential fault of the equipment to avoid accidents.
1.1.3.3 Unmanned aerial vehicles for inspection
Railway accidents are sudden and random, so regular inspections of railways must be
rigorous and critical. With the continuous development of unmanned aerial vehicle (UAV)
technology, the demand for civilian UAVs in many countries is growing and UAVs have an
increasingly important role in many fields. Research scholars have a major interest in
developing new types of UAVs that can fly autonomously in different environments and
locations and perform a variety of tasks [77]. Therefore, various types of the UAVs with
different sizes and weights have been invented and designed for different situations and
scenes. For example, UAVs slowly began to replace some manual work of railway
operations and maintenance and have been successfully applied in the field of railway
safety. Combining existing technical skills of UAVs with the daily safety requirements of
railway tracks, UAVs can inspect for daily monitoring of railway tracks. Rail transit
inspection UAV inspection refers to the use of the UAVs equipped with visible light
cameras and infrared sensors to detect the operation status of railway equipment and real-
time weather and discover the hidden dangers of railway train operation. Equipment along
the railway includes pantographs, power lines, and so on.
Introduction 17](https://image.slidesharecdn.com/34088-250403165553-e555ad51/85/Robot-Systems-for-Rail-Transit-Applications-1st-Edition-Hui-Liu-36-320.jpg)
![1.1.3.3.1 Pantograph
The pantograph and catenary technology can connect electrical energy from a contact
network to an electric locomotive [78]. As the speed of the train increases, the vibration
characteristics of the catenary and the pantograph are particularly important for the
stability and safety of the train. To obtain a stable current, the contact pressure should be
maintained between the contact slide of the pantograph and the contact line. Receiving
current from the traction net by the pantograph is a process in which several mechanical
movements occur simultaneously during the high-speed running of a high-speed train. The
pantograph and the locomotive are closely connected. When the locomotive is driving fast,
the pantograph is also rubbing quickly and moving forward relative to the contact line.
Because of the structural characteristics of the pantograph, it maintains tension, so it
generates vibration in the vertical direction. To ensure the pressure value, the pantograph
is also in contact with the catenary, and high-speed lateral vibration occurs. The catenary
has an inherent vibration frequency that creates a traveling wave propagating along the
catenary. These kinds of movements occur at the same time and are likely to cause
separation between the nets [78]. In the case of high-speed driving, these types of motion
are superimposed, and the accuracy of construction of each line is different. This makes it
difficult to maintain stable contact pressure between the pantograph and the contact line.
When the contact pressure is zero, it is likely that mechanical disengagement will occur,
which is the pantograph-catenary offline. When the electric locomotive runs at high speed,
contact between the pantograph and the high-voltage catenary is good, which is equivalent
to a closed circuit with an inductive load, and the load operates normally. When the net is
separated, it is equivalent to disconnecting the inductive load circuit.
Because of its circuit characteristics or overvoltage, overvoltage between networks is high
enough to rupture the air around the gap, which will cause a spark discharge and then an
arc discharge. Whether it is spark discharge or arc discharge, the locomotive current at this
stage will be greatly affected. At this stage, not only will overvoltage be generated,
electromagnetic noise will be disturbed. In addition, high heat released by the arc will
melt the metal and cause wear on the pantograph and catenary. When there is an offline
situation between the pantograph and catenary, safe and stable operation of the high-speed
railway will face great hazards: unstable operation of the electric locomotive, severe wear
of the pantograph slide, deterioration of the traction motor rectification conditions, and
generated overvoltage in the main circuit.
According to statistics, railway accidents caused by failure of the traction power supply
system accounted for 40% and 30% of accidents in 2009 and 2010, respectively [79].
The quality of the pantograph-catenary system has become an important prerequisite for
speeding and improving operational reliability [79]. Currently, the real-time dynamic
monitoring system of the pantograph-catenary system has been installed on newly built
trains, and the test system has been installed in the operation EMUs. The system is
18 Chapter 1](https://image.slidesharecdn.com/34088-250403165553-e555ad51/85/Robot-Systems-for-Rail-Transit-Applications-1st-Edition-Hui-Liu-37-320.jpg)
![installed on top of the train and can detect the working state of the pantograph in real
time. The main method is to collect image information through the camera for analysis
and processing, and the pan/tilt can adjust the shooting angle of the camera. Nevertheless,
the monitoring effect of the bow monitoring system will be limited by the fixed
installation position of the camera and the range of camera angles, so that the pantograph
can be monitored from only one direction and the pantograph cannot be fully fault-
monitored. With the gradual rise of UAV technology, a train pantograph monitoring
system based on the UAV inspection has emerged. Depending on the flexibility of the
UAVs, it is theoretically possible to achieve all-around detection of the state of the
pantograph and avoid blind spots in the pantograph failure. Moreover, considering
simultaneous work in different areas, multiple sets of parallel UAVs can be used for
detection to improve efficiency.
1.1.3.3.2 Power lines
The overhead transmission line is a kind of steel frame structure that erects the wire and
keeps it away from the ground. The safety of overhead transmission lines affects people’s
normal life, is related to economic production in the country, and even poses a threat to
national security. Long-distance transmission lines are characterized by long line distances
and high voltage levels. Transmission lines are also called overhead transmission lines.
The transmission capacity of the lines is large, the transmission distance is long, and they
are in the open air for a long time. They are often affected by the surrounding
environment and natural changes such as snow, lightning, ice, external forces, birds, and
so forth. An ultrahigh voltage (UHV) transmission line has high voltage, a long line, a
high pole tower, a large conductor, and a complex geographical environment. It puts high
requirements on operation and maintenance [80]. Traditional line maintenance and
overhaul operations have complicated processes, high risks, a poor monitoring work
environment, and low labor efficiency. If there is an emergency trip, power failure, and
nature force majeure, only low-end inspection equipment such as a telescope and night
vision device can be used. Therefore, the manual operation and maintenance mode are
unable to cope with the safe operation and maintenance of the UHV over a long distance.
1.1.3.3.3 Foreign matter invasion detection
Compared with the land inspection robots introduced earlier, the UAVs can detect foreign
matter intrusion with the obvious advantages of wide detection range and emergency
response [81]. When the railway is hit by a major natural disaster, the advantages of
inspection UAVs are highlighted. For example, serious mudslide disasters may occur along
the railway line, posing a serious hazard to the railway subgrade. If a mudslide is not
found and taken emergency measures in time, it will lead to serious traffic accidents. In
general, the occurrence of mudslide disasters is accompanied by strong rainfall. In this
case, the train driver’s line of sight is seriously degraded and it is difficult to detect the
danger in time during heavy rain. Historically, there have been train crashes in which the
Introduction 19](https://image.slidesharecdn.com/34088-250403165553-e555ad51/85/Robot-Systems-for-Rail-Transit-Applications-1st-Edition-Hui-Liu-38-320.jpg)
![accident occurred because heavy rains caused mudslides to ruin railway bridges [82].
Railway inspection UAVs can take pictures of the railway along with the high altitude and
determine whether the railway is normal along the way through wireless transmission
technology and image processing technology.
1.2 Fundamental key problems of rail transit robot systems
Key problems of rail transit robots are positioning navigation and path planning,
humanerobot interaction, and power management, (Fig. 1.3).
1.2.1 Navigation
1.2.1.1 Development progress
The development of robotics has gradually expanded from the industrial field to rail
transit, services, medical, entertainment, fire protection, and other fields. The application
of robot technology has brought tremendous productivity and economic benefits to society.
How to make robots more efficient and serve humans more intelligently is the focus of
scientific research. The positioning and navigation of robots have been the core emphases
of content since the development of robotics.
The positioning of mobile robots is a prerequisite for all of their actions. The accuracy of
positioning determines the accuracy of the robots’ feasible operation. It is the basis for
ensuring the robot navigation system. The positioning of rail transit robots is to use
sensors to sense the surrounding environment and then determine its position and posture
in the workspace. In the case of positioning, robots start the navigation operation and
move to the target autonomously through a certain guiding method. Path planning means
that the autonomous mobile robots can intelligently plan the relatively shortest and least
Figure 1.3
Key problems of rail transit robot systems.
20 Chapter 1](https://image.slidesharecdn.com/34088-250403165553-e555ad51/85/Robot-Systems-for-Rail-Transit-Applications-1st-Edition-Hui-Liu-39-320.jpg)
![time-consuming path for the target task. Path planning can independently interact with the
intelligent environment through onboard equipment such as an automatic elevator ride,
intelligent access control, and automatic obstacle avoidance function. Under the premise of
ensuring the safety of robots, people, and the work environment, the target task is
completed.
1.2.1.2 Methodologies
1.2.1.2.1 Positioning
The key to the positioning and path planning of mobile robots lies in two points: “Where
am I?” and “Where am I going?‘”. The former is positioning and the latter is path
planning. Mobile robot positioning technology solves the first problem: that is, it achieves
the precise positioning of robots. If a robot can self-navigate, the first condition is that the
robot knows where it is. There are several basic tasks for navigation: (1) global
positioning, sense the position, posture, and environment information of the robot; (2) path
planning, plan an accessible path to the target point from the environmental model; and
(3) motion control, select the motion control method, and move to the destination along
the planned safe path.
Many scholars have made great progress in the field of positioning [83]. The main
research directions of mobile robot positioning technology include simultaneous
positioning and map construction, path planning, and motion control. In the process of
robot navigation, positioning is the first problem to be solved; the posture information and
current state of robots need to be known. Existing positioning methods of intelligent
mobile robots mainly include relative and absolute positioning [84]. Absolute positioning
can obtain the posture and position information of a robot directly through the sensor
carried by the mobile robot itself. Relative positioning requires the determined original
position. Sensors for a robot can obtain the concrete position of the distance and direction
to the target point. According to the current point moving distance and the direction of
rotation angle, the robot can be aware of its posture and direction. Absolute positioning
methods include a magnetic compass and global positioning system (GPS). Relative
positioning can be divided into the dead reckoning method based on the inertial sensor and
the dead reckoning method based on the odometer. According to the different types of
sensors used to perceive the environment, navigation of mobile robots mainly includes
GPS, gyroscope, inertial, photoelectric coding, magnetic compass, laser, and ultrasonic.
1.2.1.2.2 Path planning
The path planning problem of mobile robots is a hot spot in the field of mobile robot
navigation research [85]: mobile robots can find an optimal or near-optimal path from the
starting state to the target state that avoids obstacles based on one or some performance
indicators (such as the lowest working cost, the shortest walking route, the shortest
Introduction 21](https://image.slidesharecdn.com/34088-250403165553-e555ad51/85/Robot-Systems-for-Rail-Transit-Applications-1st-Edition-Hui-Liu-40-320.jpg)
![walking time, etc.) in the motion space. Generally speaking, the path planning selects the
shortest distance of the path: that is, the shortest length of the path from the starting point
to the target point as a performance indicator. The mobile robot path planning method can
be divided into two types according to the known degree of environmental information:
path planning based on global map information or local map information [86]. These two
path planning methods are referred to as global path planning and local path planning. The
global path planning method can generate the path under the completely known
environment (the position and shape of the obstacle are predetermined). With the global
map model of the environment where the mobile robots are located, the search is
performed on the established global map model. The optimal algorithm can obtain the
optimal path. Therefore, global path planning involves two parts: establishment of the
environmental model and the path planning strategy.
Path planning requires building an environmental map. Environmental map construction
refers to the establishment of an accurate spatial location description of various objects in
the environment in which the robot is located, including obstacles, road signs, and so on:
that is, the establishment of a spatial model or map. The purpose of constructing the
environmental map is to help the mobile robot plan an optimal path from the starting point
to the target point in the established environment model with obstacles.
There are many mature methods for establishing an environment model for mobile robot
path planning. Existing basic environment models mainly include a grid decomposition
map, quad split graph, visibility graph, and Voronoi diagram. After the environmental map
is built, global path planning is carried out. Algorithms of global path planning are mainly
divided into two types: heuristic search methods and intelligent algorithms. The initial
representation of the heuristic search is the A* algorithm developed by the Dijkstra
algorithm. The A* algorithm is the most commonly used heuristic graph search algorithm
for state space. In addition to solving problems based on state space, it is often used for
the path planning of robots. Many scholars have improved the A* algorithm and obtained
other heuristic search methods [87,88]. Intelligent algorithms have lots of studies,
including ant colony [89], particle swarm [90], genetic [91], bat [92], simulated annealing
[93], and so forth.
Different from the global path planning method, the local path planning method assumes
that the position of the obstacle in the environment is unknown, and the mobile robot
perceives its surrounding environment and its state only through the sensor. Because the
global information of the environment cannot be obtained, the local path planning focuses
on the current local environment information of the mobile robot and uses the local
environment information obtained by the sensor to find an optimal path from the starting
point to the target point that does not touch the obstacle in the environment. The path
planning strategy needs to be adjusted in real time. Commonly used methods for local
22 Chapter 1](https://image.slidesharecdn.com/34088-250403165553-e555ad51/85/Robot-Systems-for-Rail-Transit-Applications-1st-Edition-Hui-Liu-41-320.jpg)
![path planning include the rolling window [94], artificial potential field [95], and various
intelligent algorithms [96].
1.2.1.3 Applications
Positioning and path planning are applied to almost all mobile robots, the most widely
used of which are the AGV robots and rail transit inspection robots.
1.2.1.3.1 Carrying automatic guided vehicle robots
To reduce labor costs and storage costs, logistics sorting has gradually shifted from manual
to automated. Intelligent unmanned storage has gradually become a hot spot for many
scholars [97]. Unmanned factories and unmanned warehousing have become a trend in
development. The intelligent logistics system began to appear, which aims to reduce
handling, management, communication, and labor costs. All links in picking and unloading
are carried out in an orderly manner without manual intervention. AGV robots have become
the main method of automated warehousing. Performing collision-free path planning for
multiple AGVs and performing tasks is a difficult problem for the AGV group.
1.2.1.3.2 Rail transit inspection robots
The main method of inspection in rail transit is to use human power to conduct
inspections. However, the work environment of rail transit equipment is complex and
conditions are harsh. In case of bad weather, there is a personal safety risk for personnel
conducting a manual inspection. At the same time, the efficiency of manual inspections is
low, depending on business proficiency and the subjective initiative of the inspection
personnel. The long-term labor of mechanical repetition easily causes negligence, so rail
transit inspection robots have emerged as the times require, and will be used in the rail
transit line and station equipment. New methods of inspection are presented and manual
inspections will be phased out. Inspection robots can use various sensors equipped to
detect the rail transit and rail transit vehicle equipment through various advanced
technologies autonomously or with assisting inspectors. The use of inspection robots can
detect the work conditions of rail transit equipment under severe conditions and ensure the
normal and orderly operation of rail transit. This is important for the future realization of
the full automation of rail transit. [98].
1.2.2 Humanerobot interaction
1.2.2.1 Development progress
Humanerobot interaction is a technology that studies people, computers, and their
interactions [99]. It is a focus of competition in the information industry. Its extensive
application will profoundly change the manufacturing, construction, and transportation
industries. So far, humanerobot interaction technology has evolved to the interface of
Introduction 23](https://image.slidesharecdn.com/34088-250403165553-e555ad51/85/Robot-Systems-for-Rail-Transit-Applications-1st-Edition-Hui-Liu-42-320.jpg)
![gestures and language, combining the powerful processing power of computers. The
design of the current humanerobot interaction system has achieved a lot of results, such
as the handwritten Chinese character recognition system, Chinese speech recognition
system, and gesture recognition. In terms of handwritten Chinese character recognition,
the most advanced system has been able to identify 28,000 Chinese characters after years
of research and development [100]. If written at a speed of 12 Chinese characters per
minute, the recognition rate of the system is nearly 100% [100]. In the aspect of sign
language recognition and synthesis, the Institute of Computing Technology, Chinese
Academy of Sciences has successfully developed a Chinese sign language recognition and
synthesis system based on multifunctional perception [101].
1.2.2.2 Methodologies
1.2.2.2.1 Humanerobot interaction based on gestures
Gestures are a natural and intuitive mode of communication. Vision-based gesture
recognition is a key technology that is indispensable for the realization of a new
generation of humanerobot interaction. Gestures are a variety of actions produced by a
human hand or arm. They include static and dynamic gestures. Because of the diversity
and ambiguity of gestures and the complex deformation of human hands, vision-based
gesture recognition is a multidisciplinary and challenging research topic. To find a
breakthrough, the use of gestures in interpersonal communication is required. In the
vision-based gesture recognition system: (1) the video data stream is acquired by one or
more cameras; (2) the system detects whether there is a gesture in the data stream
according to the interaction model; and (3) if so, the gesture is segmented from the video
signal, and the gesture model is selected for gesture analysis. The analysis process
includes feature detection and model parameter estimation. In the recognition process,
gestures are classified according to model parameters and gesture descriptions are
generated as needed. Finally, the system drives specific applications according to the
generated descriptions.
The gesture model is important for the gesture recognition system, especially for
determining the recognition range. The model selection depends on the specific
application. A simple and rough model may be sufficient for a given application. However,
to achieve natural humanecomputer interaction, a complex gesture model must be
established. In this way, the recognition system can respond correctly to most gestures
made by users. From the current literature, most gesture modeling methods can be
summarized as gesture modeling based on the apparent model and gesture modeling based
on the 3D model.
The task of the gesture analysis is to estimate the parameters of the selected gesture
model. The analysis generally consists of feature detection and parameter estimation. In
the feature detection process, the subject that makes the gesture must be located.
24 Chapter 1](https://image.slidesharecdn.com/34088-250403165553-e555ad51/85/Robot-Systems-for-Rail-Transit-Applications-1st-Edition-Hui-Liu-43-320.jpg)
![Depending on the conditions used, positioning techniques can be classified as color-based,
motion-based, and multimode. Most color-based positioning techniques rely on histogram
matching or the use of skin training data to create a lookup table. Color-based positioning
techniques have significant drawbacks (i.e., skin color changes under different lighting
conditions). Gesture recognition is the process of classifying a trajectory (or point) in a
model parameter space into a subset of that space.
The static gesture corresponds to a point in the model parameter space, and the dynamic
gesture corresponds to a trajectory in the model parameter space. Therefore, their
identification methods are different. Static gesture recognition algorithms include
recognition based on classical parameter clustering techniques and recognition based on
nonlinear clustering techniques. Unlike static gestures, dynamic gestures involve time and
space. Most dynamic gestures are modeled as a single trajectory in the parameter space.
Because of the different habits of different users, the difference in speed and proficiency in
gestures will cause nonlinear fluctuations on the time axis of the trajectory. Eliminating
these nonlinear fluctuations is an important problem that must be overcome by dynamic
gesture recognition technology. Considering the different processing of the time axis,
existing dynamic gesture recognition technology can be divided recognition based on
hidden Markov model [102], recognition based on dynamic time warping [103], and so
forth.
1.2.2.2.2 Humanerobot interaction based on human skeleton
The development of human pose estimation has become increasingly practical. In the
fields of gait analysis, humanerobot interaction, and video surveillance, human pose
estimation have broad application prospects. The mainstream human body pose estimation
algorithm can be divided into traditional and deep learning-based methods.
Traditional methods are generally based on the graph structure; they design a 2D body part
detector, use the graph model to establish the connectivity of each component, and
continuously optimize the graph structure model to estimate the human body posture
combined with the relevant constraints of human kinematics. Although traditional methods
have high time efficiency, the extracted features are mainly histogram of oriented gradient
[104] and scale-invariant feature transform features [105]. These features cannot fully use
image information; thus, the algorithm is subject to different viewing angles, occlusions,
and inherent geometric ambiguities of the image.
At the same time, because of the single structures of the traditional models, when the
posture of the human body changes greatly, the traditional models cannot be accurately
characterized and expressed. There are multiple feasible solutions to the same data; that is,
the results of attitude estimation are not unique, which limits the scope of application of
traditional methods. On the other hand, most traditional methods are based on digital
Introduction 25](https://image.slidesharecdn.com/34088-250403165553-e555ad51/85/Robot-Systems-for-Rail-Transit-Applications-1st-Edition-Hui-Liu-44-320.jpg)
![images such as depth images for feature extraction. However, because the acquisition of
depth images requires professional acquisition equipment, the cost is high, so it is difficult
to apply to all application scenarios. In addition, the acquisition process needs to
synchronize the depth cameras of multiple perspectives, which reduces the impact of the
occlusion problem. These factors will make the acquisition process of human posture data
complicated and difficult. In contrast, monocular cameras are more common, although the
color images they collect are susceptible to environmental factors such as illumination.
Neural networks can be used to extract more accurate convolution features than artificial
features and improve the performance of the forecast. Because of this, human body pose
estimation methods based on deep learning have been deeply studied [106e108]. The
human body pose estimation method based on deep learning mainly uses a convolutional
neural network (CNN) to extract the human body pose features from the image. Compared
with artificial design features of the traditional method, the CNN can obtain more rich
features of semantic information as well as obtain different types of multiscale and
multiple-type human joint point feature vectors.
1.2.2.2.3 Humanerobot interaction based on speech recognition
Speech recognition is the process of converting human sound signals into words or
instructions. Speech recognition is based on speech. It is an important research direction of
speech signal processing and a branch of pattern recognition. The research of speech
recognition involves many subject areas such as computer technology, artificial
intelligence, digital signal processing, pattern recognition, acoustics, linguistics, and
cognitive science. It is a multidisciplinary comprehensive research field. Different research
areas have emerged based on research tasks under different constraints. According to the
requirements of the speaker’s way of speaking, these areas can be divided into isolated
words, connected words, and continuous speech recognition systems. According to the
degree of dependence on the speaker, these areas can be divided into speech recognition
systems for the specific person and nonspecific person. According to the size of
vocabulary, they can be divided into small vocabulary, medium vocabulary, large
vocabulary, and infinite vocabulary speech recognition systems.
From the perspective of the speech recognition model, the theory of the speech recognition
system is based on pattern recognition. The goal of speech recognition is to transform the
input speech feature vector sequence into a sequence of words using phonetic and
linguistic information. According to the structure of the speech recognition system, a
complete speech recognition system includes a feature extraction algorithm, acoustic
model, and language model and search algorithm. The speech recognition system is
essentially a multidimensional pattern recognition system. For different speech recognition
systems, the specific recognition methods and techniques used by people are different, but
the basic principles are the same. The collected speech signals are sent to feature
26 Chapter 1](https://image.slidesharecdn.com/34088-250403165553-e555ad51/85/Robot-Systems-for-Rail-Transit-Applications-1st-Edition-Hui-Liu-45-320.jpg)
![extraction. The module processes and the obtained speech features are sent to the model
library module. The speech pattern matching module identifies the segment speech
according to the model library, and finally obtains the recognition result.
1.2.2.3 Applications
In the field of rail transit, rail transit equipment manufacturing and rail transit operation
and maintenance are the main components that determine whether the rail transit system
can operate safely, efficiently, and economically for a long time. Introducing humanerobot
interaction technology into rail transit equipment manufacturing will change the traditional
equipment manufacturing production mode and transform the original staff to work
together with production tools, which will greatly improve the quality and efficiency of
equipment manufacturing. In the field of operation and maintenance, the friendly interface
achieved by high-level humanerobot interaction will enhance the passenger’s ride
experience from the inbound to the outbound service cycle and will greatly improve the
management level of the rail transit system [109].
1.2.2.3.1 Intelligent rail transit equipment manufacturing
Intelligent industrial equipment has become a trend in the development of the global
manufacturing industry. The robotic group used for welding replaces a separate welder to
perform collaborative and high-precision welding and assembly operations. The automated
guided carrier platform connects manufacturing modules in an intelligent manufacturing
environment, based on unmanned and intelligent obstacle-avoidance delivery platforms. It
can be shuttled within a complex environment, and the safety of production is greatly
improved, which also increases the automation of the assembly line. In a complex
intelligent manufacturing environment, humanerobot interaction control technology will
have an important role in the coordinated operation of personnel and carrier platforms and
the operator control of the delivery platform. On the one hand, the automatic guided
carrier platform requires the operator to provide instructions to complete the loading,
transporting, and unloading operations of the carrier platform. On the other hand, the
automatic guided carrier platform needs to obtain operator instructions in a small space to
complete high-performance obstacle avoidance [110].
1.2.2.3.2 Intelligent operation and maintenance
To ensure the stability, safety, and reliability of the urban rail transit system, it is necessary
to achieve the intelligent manufacturing of rail vehicles, but also to ensure its intelligent
operation and maintenance and improve system efficiency. The intelligent operation and
maintenance framework includes parts such as the automatic identification of vehicles,
automatic equipment control, processing data collection and analysis, and vehicle status
detection, and is interconnected with the network.
Introduction 27](https://image.slidesharecdn.com/34088-250403165553-e555ad51/85/Robot-Systems-for-Rail-Transit-Applications-1st-Edition-Hui-Liu-46-320.jpg)
![In 2017, Tao et al. proposed a method for intelligent state-of-sales awareness and process
control through intelligent instrument and detection technology [111]. They discussed the
application mode of intelligent instrument and detection technology in intelligent
manufacturing systems, digitalized in combination with actual projects. The innovative
practice of production line transformation and construction puts forward the development
direction of intelligent instruments and detection technologies that meet the requirements
of intelligent manufacturing. Through the practical application of the project, this model
can effectively promote the realization of data acquisition and intelligent perception in
intelligent manufacturing systems and can provide effective data resources. In 2019, Xu
et al. proposed an AGV intelligent logistics scheduling model and response method based
on the request scheduling response model for the dynamic scheduling of intelligent
manufacturing workshop logistics [112].
1.2.3 Power management
Robots can do a lot of work instead of humans; they have an important role in the fields of
manufacturing, service, and rail transportation. Power management is mainly divided into
the robot’s power forecasting and autonomous recharging.
1.2.3.1 Power forecasting
The robot is powered by a lithium battery during the work process. How to ensure that the
robot does not fail owing to a low battery during the mission is a serious problem.
Traditional robots continuously detect their own power and perform recharging tasks when
they find that the battery is below the minimum threshold. This method is inapplicable to
the robot that is performing the task. For example, whether the robot can have enough
power to reach the next tower or the next-level recharging base station during the
inspection is an urgent problem to be studied. In addition, when the inspection personnel
uses the robot to inspect the transmission line, it is necessary to grasp the lifetime and
cruising range of the robot in real time, and then the next inspection plan can be
formulated according to the robot’s endurance capacity. Most research on the cruising
range of the inspection robot uses a rough estimation of the amount of power consumed
by the robot at different stages; thus, the accumulation method is not scientific. Many
scholars have begun to use the power time series data of the robot to establish a power
forecasting model, to estimate the power in the subsequent time period [113,114]. The
power forecasting model is built by data processing methods and machine learning.
1.2.3.2 Autonomous recharging
Most intelligent robots use battery packs to power them and their working hours are
generally shorter. The mobile robot that can self-charge to achieve long-term continuous
work when the battery power is reduced to a certain level has attracted the interest of
28 Chapter 1](https://image.slidesharecdn.com/34088-250403165553-e555ad51/85/Robot-Systems-for-Rail-Transit-Applications-1st-Edition-Hui-Liu-47-320.jpg)
![researchers. Research in the field of robotic autonomous recharging has begun. In the
mid-20th century, Walter developed the world’s first self-aware electric robot [115], which
was able to achieve autonomous recharging. Walter designed a recharging station with a
charger and illuminator installed. It not only avoids obstacles, it uses a light beam to reach
the recharging station for recharging. Autonomous recharging for mobile robots can be
mainly divided into contact and noncontact, which is major topic in the field of robotics
[116].
1.3 Scope of this book
Today, robotics and advanced rail transit are two major areas of development in the world.
This book starts with the field of rail transit manufacturing, operation, and maintenance. It
also introduces different types of robots and related technologies in detail. The book
includes eight chapters:
Chapter 1: Introduction.
This chapter first outlines rail transit robots. Then, it describes three basic issues of
robotics, including navigation, humanerobot interaction, and power control.
Chapter 2: Rail transit assembly robot systems.
This chapter first introduces progress in development and the key technologies of
assembly robots. Then, it introduces the main components of assembly robots: machinery
components, sensors, controllers, and actuators. Arm dynamics is a major research topic of
assembly robots. This chapter introduces mathematical models of forward dynamics and
inverse dynamics, discusses various trajectory planning algorithms in joint and Cartesian
space, and proposes an artificial neural network algorithm for the inverse dynamic
optimization calculation of the robot arm. Finally, the chapter presents the future direction
of intelligent assembly robots.
Chapter 3: Rail transit collaborative robot systems.
This chapter first gives a basic definition of collaborative robots. The chapter then
summarizes the development history, application field, and mode of collaborative robots.
The hardware composition and characteristics of the collaborative robot are summarized.
Finally, the basic concepts of feature extraction, object detection, and object tracking in
visual perception are introduced.
Chapter 4: Automatic guided vehicles in rail transit intelligent manufacturing
environment.
This chapter introduces progress in the development and types of AGVs in the rail transit
intelligent manufacturing environment. Then, the main components of AGVs are
Introduction 29](https://image.slidesharecdn.com/34088-250403165553-e555ad51/85/Robot-Systems-for-Rail-Transit-Applications-1st-Edition-Hui-Liu-48-320.jpg)
Robot Systems for Rail Transit Applications 1st Edition Hui Liu
- 1. Download the full version and explore a variety of ebooks or textbooks at https://ebookmass.com Robot Systems for Rail Transit Applications 1st Edition Hui Liu _____ Follow the link below to get your download now _____ https://ebookmass.com/product/robot-systems-for-rail- transit-applications-1st-edition-hui-liu/ Access ebookmass.com now to download high-quality ebooks or textbooks
- 2. We have selected some products that you may be interested in Click the link to download now or visit ebookmass.com for more options!. Unmanned Driving Systems for Smart Trains Hui Liu https://ebookmass.com/product/unmanned-driving-systems-for-smart- trains-hui-liu/ Wind Forecasting in Railway Engineering Hui Liu https://ebookmass.com/product/wind-forecasting-in-railway-engineering- hui-liu/ Optimization Models for Rail Car Fleet Management 1st Edition Milos Milenkovic https://ebookmass.com/product/optimization-models-for-rail-car-fleet- management-1st-edition-milos-milenkovic/ Tensors for Data Processing. Theory, Methods, and Applications Yipeng Liu https://ebookmass.com/product/tensors-for-data-processing-theory- methods-and-applications-yipeng-liu/
- 3. Emerging trends in applications and infrastructures for computational biology, bioinformatics, and systems biology : systems and applications 1st Edition Arabnia https://ebookmass.com/product/emerging-trends-in-applications-and- infrastructures-for-computational-biology-bioinformatics-and-systems- biology-systems-and-applications-1st-edition-arabnia/ Tensors for Data Processing 1st Edition Yipeng Liu https://ebookmass.com/product/tensors-for-data-processing-1st-edition- yipeng-liu/ Supply Chain Analytics: Concepts, Techniques and Applications 1st Edition Kurt Y. Liu https://ebookmass.com/product/supply-chain-analytics-concepts- techniques-and-applications-1st-edition-kurt-y-liu/ Good Robot, Bad Robot: Dark And Creepy Sides Of Robotics, Autonomous Vehicles, And AI 1st Edition Jo Ann Oravec https://ebookmass.com/product/good-robot-bad-robot-dark-and-creepy- sides-of-robotics-autonomous-vehicles-and-ai-1st-edition-jo-ann- oravec/ Philosophy, Travel, and Place: Being in Transit 1st ed. Edition Ron Scapp https://ebookmass.com/product/philosophy-travel-and-place-being-in- transit-1st-ed-edition-ron-scapp/
- 6. Robot Systems for Rail Transit Applications Hui Liu School of Traffic and Transportation Engineering, Central South University, Changsha, China
- 7. Elsevier Radarweg 29, PO Box 211, 1000 AE Amsterdam, Netherlands The Boulevard, Langford Lane, Kidlington, Oxford OX5 1GB, United Kingdom 50 Hampshire Street, 5th Floor, Cambridge, MA 02139, United States Copyright © 2020 Elsevier Inc. All rights reserved. No part of this publication may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopying, recording, or any information storage and retrieval system, without permission in writing from the publisher. Details on how to seek permission, further information about the Publisher’s permissions policies and our arrangements with organizations such as the Copyright Clearance Center and the Copyright Licensing Agency, can be found at our website: www.elsevier.com/permissions. This book and the individual contributions contained in it are protected under copyright by the Publisher (other than as may be noted herein). Notices Knowledge and best practice in this field are constantly changing. As new research and experience broaden our understanding, changes in research methods, professional practices, or medical treatment may become necessary. Practitioners and researchers must always rely on their own experience and knowledge in evaluating and using any information, methods, compounds, or experiments described herein. In using such information or methods they should be mindful of their own safety and the safety of others, including parties for whom they have a professional responsibility. To the fullest extent of the law, neither the Publisher nor the authors, contributors, or editors, assume any liability for any injury and/or damage to persons or property as a matter of products liability, negligence or otherwise, or from any use or operation of any methods, products, instructions, or ideas contained in the material herein. Library of Congress Cataloging-in-Publication Data A catalog record for this book is available from the Library of Congress British Library Cataloguing-in-Publication Data A catalogue record for this book is available from the British Library ISBN: 978-0-12-822968-2 For information on all Elsevier publications visit our website at https://www.elsevier.com/books-and-journals Publisher: Matthew Deans Acquisitions Editor: Glyn Jones Editorial Project Manager: Naomi Robertson Production Project Manager: Nirmala Arumugam Cover Designer: Matthew Limbert Typeset by TNQ Technologies
- 8. List of figures and tables Figure 1.1 Robots in the manufacturing, dispatch, and maintenance of rail transit. Figure 1.2 Role of robots in rail transit maintenance. Figure 1.3 Key problems of rail transit robot systems. Figure 2.1 Working steps of the assembly robot. Figure 2.2 Different types of assembly robots. Figure 2.3 Overall frame diagram of rail transit assembly robot system. Figure 2.4 Main components of assembly robot. Figure 2.5 Mechanical diagram of assembly robot: (A) base component, (B) rotating joint, (C) arm connecting component, (D) wrist joint, and (E) end effector. Figure 2.6 Trajectory planning algorithms. Figure 2.7 Process of using the artificial neural network (ANN) for inverse dynamics calculation. Figure 3.1 Multirobot collaboration. Figure 3.2 Humanerobot collaboration. Figure 3.3 The block diagram of collaborative robot system. Figure 3.4 Sensors for collaborative robots. Figure 3.5 Classification of end effectors. Figure 3.6 Feature extraction algorithms. Figure 3.7 The flow chart of the HOG algorithm. Figure 3.8 The flow chart of the SIFT algorithm. Figure 3.9 The flow chart of the LBP algorithm. Figure 3.10 Target detection algorithms. Figure 3.11 Target tracking algorithms. Figure 4.1 Main content diagram of automatic guided vehicles (AGVs). Figure 4.2 Navigation methods. AGV, automatic guided vehicle; LiDAR, light detec- tion and ranging; SLAM, simultaneous localization and mapping. Figure 4.3 Global path planning algorithms. Figure 4.4 Local path planning algorithms. Figure 4.5 Humanerobot interaction algorithms. Figure 4.6 Flowchart of the hybrid path planning model. KELM, kernel-based extreme learning machine; QPSO, quantum particle swarm optimization. Figure 5.1 The advantages of Autonomous rail Rapid Transit (ART). Figure 5.2 The overview diagram of Autonomous rail Rapid Transit (ART). ix
- 9. Figure 5.3 Schematic diagram of ART sensor fusion. Figure 5.4 The core of the pedestrian detection algorithm. Figure 5.5 The flow chart of Histogram of Oriented Gradient (HOG) feature þ Support Vector Machine (SVM) classification. Figure 5.6 The flow chart of the Support Vector Machine (SVM) classification. Figure 5.7 The flow chart of pedestrian contour extraction. Figure 5.8 Posture recognition process. Figure 6.1 Inspection robot structure. Figure 6.2 The railway applications of the inspection robots. Figure 6.3 Main components of the inspection robots. Figure 6.4 Rail transit inspection robot technologies. Figure 7.1 Advantages of dual-arm robots. Figure 7.2 Channel robots Trouble of Moving Electric Multiple Units Detection System diagram. Figure 7.3 Structure of ground track. Figure 7.4 Forming diagram of infrared image formation. Figure 7.5 Analysis process of visible image. HOG, histogram of orientation gradient. Figure 7.6 Intelligent analysis method of infrared thermal image. ANFIS, adaptive network-based fuzzy inference system; BP, backpropagation. Figure 7.7 Bogie fault diagnosis structure diagram. Figure 7.8 Flow of fault diagnosis. EMD, empirical mode decomposition; WPT, wavelet packet transform. Figure 7.9 The sigmoid function. Figure 7.10 The tanh function. Figure 7.11 The rectified linear unit function. Figure 8.1 Main components of rail transit inspection unmanned aerial vehicles (UAVs). GPS, global positioning system; IMU, inertial measurement unit. Figure 8.2 Scheduling system of an unmanned aerial vehicle (UAV). Figure 8.3 Flowchart of perception algorithm. ROI, region of interest. (A) The real-time image stabilization of the collected video is finalized. (B) The track region of interest is extracted from the perception image according to railway boundary regulations, to reduce unnecessary operations and speed the subsequent image processing rate. (C) The intrusion is identified based on a track map. Table 2.1 Advantages and disadvantages of three assembly methods. Table 2.2 Advantages and disadvantages of joint space trajectory planning and Carte- sian space trajectory planning. Table 3.1 Comparison of traditional industrial robots and collaborative robots. Table 3.2 Several mainstream collaborative robots and manufacturers. Table 5.1 Comparison of main technical solutions of ART and modern tram signal system. Table 5.2 Traditional pedestrian detection algorithms. List of figures and tables x
- 10. Table 5.3 Pedestrian detection algorithms based on deep learning. Table 5.4 The pedestrian posture recognition accuracy of BP neural network. Table 7.1 Composition of laser sensor. Table 7.2 Functions and characteristics of two-dimensional laser sensor. Table 7.3 Comparison of channel robot wireless recharging methods. List of figures and tables xi
- 11. Preface Rail transit is the lifeblood of many national economies and the backbone of transportation. Safe and efficient rail transit is based on highly reliable manufacturing and high-quality maintenance. Rail transit robots can replace manual repetitive tasks and improve the auto- mation level of the rail transit system. As a result, work efficiency can be improved, acci- dental failures due to human negligence can be avoided, and the safety of the rail transit system can be improved. Rail transit robots involve the intersection of automatic control, artificial intelligence, signal processing, pattern recognition, mechanical engineering, and transportation engineering. When applied to the rail transit system, the robot is faced with the key problems to be solved urgently. Therefore, rail transit robots are currently recognized as research hotspots among scientific problems. Based on research from the past 10 years, the author puts forward a framework of rail transit robot technology and completes the related work. This book covers seven mainstream rail transit robots, including assembly robots, collaborative robots, automated guided vehicles, autonomous rail rapid transit, inspection robots, channel robots, and inspection unmanned aerial vehicles. The key problems of robots are described in detail, including positioning navigation, path planning, humanerobot interaction, and power management, etc. For students and managers in related departments, this book can provide valuable information about rail transit robots. For researchers and doctoral students, this book can provide some ideas and encourage future research in rail transit robots. This book contains eight chapters: Chapter 1: Introduction This chapter first outlines the rail transit robot. Then the chapter describes three basic issues of robotics, including navigation, humanerobot interaction, and power control. Chapter 2: Rail transit assembly robot systems This chapter first introduces the development progress and key technologies of assembly robots, and then introduces the main components of assembly robots. After that, the dynamics models for the assembly robot systems are explained. Finally, the artificial neural network algorithm for the inverse dynamic optimization calculation of the robot arm is introduced. xiii
- 12. Chapter 3: Rail transit collaborative robot systems This chapter first gives the basic definition of a collaborative robot. Then the chapter summarizes the development history, application field and collaborative mode of collaborative robots. The components of the collaborative robot are summarized. Finally, the basic concepts of visual perception in human-robot collaboration are introduced. Chapter 4: Automatic Guided Vehicles (AGVs) in the rail transit intelligent manufacturing environment This chapter first introduces the development progress and types of the AGV in rail transit intelligent manufacturing environment. Then the main components of AGVs are introduced. After that, the key technologies and the applications of the AGV are introduced. Chapter 5: Autonomous rail Rapid Transit (ART) This chapter firstly introduces the hardware of ART. Then, the technologies of ART are introduced. Finally, the pedestrian detection algorithms of ART are introduced in detail. Chapter 6: Rail transit inspection robots This chapter first introduces the development history, function and main components of the rail transit inspection robot. Then, two key technologies to ensure that the inspection robots normally complete the inspection work, positioning methods, path planning methods, and handeeye vision system are introduced in detail. Chapter 7: Rail transit channel robot systems This chapter firstly introduces the development history and main components of the rail transit channel robot, including the ground rail, dual-arm robot, infrared thermometer, laser sensor, etc. Then the TEDS intelligent sensing system is described in detail. Finally, fault diagnosis algorithms based on deep learning models are introduced. Chapter 8: Rail transit inspection Unmanned Aerial Vehicle (UAV) This chapter first introduces the development history of the UAV and its applications in various fields. Secondly, it introduces the basic structure of fixed-wing UAVs, unmanned helicopters and rotary-wing UAVs. Various sensors are applied to rail transit inspection. Then the UAV technologies are described in detail. Finally, the applications of the UAV in the detection of rail transit intruding detection are introduced. Prof. Dr.-Ing. habil. Hui Liu Changsha, China November 2019 Preface xiv
- 13. Acknowledgement The studies in the book are supported by the National Natural Science Foundation of China, the National Key R&D Program of China, and the Innovation Drive of Central South University, China. The publication of the book is funded by the High-level Postgraduate Text Book Project of the Hunan Province of China. In the process of writing the book, Mr. Zhu Duan, Mr. Jiahao Huang, Mr. Kairong Jin, Mr. Yu Xia, Mr. Rui Yang, Ms. Shi Yin, Mr. Ye Li, Mr. Guangji Zheng, Ms. Jing Tan, Mr. Huipeng Shi, Mr. Haiping Wu, Mr. Chao Chen, Mr. Zhihao Long, and other team members have done a lot of model verification and other work. These team members as mentioned have the same contribution to this book. xv
- 14. Nomenclature list # 2D Two-Dimensional 3C Computer, Communication, Consumer Electronic 3D Three-Dimensional A ABB Asea Brown Boveri AC Alternating Current ACMS Aircraft Condition Monitoring System ACO Ant Colony Optimization ADC Analog-to-Digital Converter ADU Automatic Drilling Unit AGV Automatic Guided Vehicle AGVS Automated Guided Vehicle System AMR Anisotropic Magnetoresistive ANN Artificial Neural Network ANIFS Adaptive Network-Based Fuzzy Inference System AR Augmented Reality ARM Advanced RISC Machine AP Access Point APF Artificial Potential Field API Application Programming Interface ARMA Autoregressive Moving Average ART Autonomous rail Rapid Transit ASK Amplitude Shift Keying ATC Automatic Train Control ATO Automatic Train Operation ATP Automatic Train Protection ATS Automatic Train Supervision AUC Area Under Curve xvii
- 15. B BFS Best-First Search BP Back Propagation BPNN Back Propagation Neural Network BRIEF Binary Robust Independent Elementary Features C CAD Computer-Aided Design CCD Charge Coupled Device CCOT Continuous Convolution Operators Tracker CIMS Computer Integrated Manufacturing System CNC Computerized Numerical Control CNN Convolutional Neural Network CNR China Northern Locomotive Rolling Stock Industry Group CPU Central Processing Unit CRF Conditional Random Fields CSM Correlation Scan Match CSR China Southern Locomotive Rolling Stock Industry Group CW Continuous Wave CWT Continue Wavelet Transform D DC Direct Current D-DCOP Dynamic Distributed Constraint Optimization Problem Dec-MDP Decentralized Markov Decision Process DEM Digital Elevation Model DFT Discrete Fourier Transform D-H Denavit-Hartenberg DLT Deep Learning Tracker DMPC Distributed Model Predictive Control DNFO Dynamic Network Flow Optimization DNN Deep Neural Network DOF Degree of Freedom DSP Digital Signal Processor DTW Dynamic Time Warping E ECO Efficient Convolution Operator EKF Extended Kalman Filter EKF-SLAM Extended Kalman Filter SLAM ELM Extreme Learning Machine EMD Empirical Mode Decomposition EMU Electric Multiple Units Nomenclature list xviii
- 16. F FAS Flexible Assembly System FAST Features from Accelerated Segment Test Faster RCNN Faster Regional Convolutional Neural Network FDD Frequency Division Duplexing FFT Fast Fourier Transform FMS Flexible Manufacturing System FSK Frequency Shift Keying FTP File Transfer Protocol G GA Genetic Algorithm GOA Grade of Automation GPRS General Packet Radio Service GPS Global Positioning System H HDT Hedged Deep Tracking HDFS Hadoop Distributed File System HMM Hidden Markov Model HOG Histogram of Oriented Gradient HRI Human-Robot Interaction HSB Hue-Saturation-Brightness I ID Identity Document IDIM-LS Inverse Dynamic Identification Model and Linear Least Squares Technique IFF Identification Friend or Foe IFR International Federation of Robots IGBT Insulated Gate Bipolar Translator IL Imitation Learning IMU Inertial Measurement Unit IMW Intelligent Manufacturing Workshop IMF Intrinsic Mode Function IoU Intersection over Union INS Inertial Navigation System ISM Industrial Scientific Medical J JFET Junction Field-Effect Transistor K KCF Kernelized Correlation Filters KF Kalman Filter KNN K-Nearest Neighbors Nomenclature list xix
- 17. L LAN Local Area Network LBP Local Binary Pattern LCD Liquid Crystal Display LiDAR Light Detection and Ranging LRR Long-Range Radar M MAE Mean Absolute Error MANET Mobile Ad Hoc Network mAP mean Average Precision MBTA Massachusetts Bay Transit Authority MDPs Markov Decision Processes MEEM Multiple Experts using Entropy Minimization MEMS Micro-Electro-Mechanical Systems MF Morphological Filter MIL Multiple Instance Learning MILP Mixed Integer Linear Programming MLP Multilayer Perceptron MSE Mean Square Error MTSP Multiple Traveling Salesman Problem MVB Multifunction Vehicle Bus N NFS Network File Systems NMS Nonmaximun Suppression NOMA Nonorthogonal Multiple Access NP Nondeterministic Polynomial O OFDM Orthogonal Frequency Division Multiplexing OGAePSO Optimum Genetic AlgorithmeParticle Swarm Optimization algorithm ORB Oriented FAST and Rotated BRIEF P PG Policy Gradient PLC Programmable Logic Controller POS Point of Sale PRM Probabilistic Road Map PSK Phase Shift Keying PSO Particle Swarm Optimization PTZ Pan-Tilt-Zoom PUMA Programmable Universal Machine for Assembly Nomenclature list xx
- 18. Q QPSO Quantum Particle Swarm Optimization QR Quick Response R RANSAC Random Sample Consensus RBF Radial Basis Function RBPF Rao-Blackwellized Particle Filter RCC Remote Center Compliance RCNN Regional Convolutional Neural Network ReCNN Regions with CNN features RDD Resilient Distributed Dataset ReFCN Region-based Fully Convolutional Networks RFID Radio Frequency Identification RGB Red-Green-Blue RGB-D Red-Green-Blue-Deep RL Reinforcement Learning RNN Recurrent Neural Network ROC Receiver Operating Characteristic curve ROI Region of Interest ROS Robot Operating System RPN Region Proposal Networks RRT Rapid-Exploration Random Tree RSSI Received Signal Strength Indication RTP Real-Time Protocols S SARSA State Action Reward State Action SCARA Selective Compliant Assembly Robot Arm SDA Stacked Denoising Autoencoder SEA Series Elastic Actuator SfM Structure from Motion SIA Swarm Intelligence Algorithm SIFT Scale Invariant Feature Transform SLAM Simultaneous Localization and Mapping SMT Surface Mount Technology SPP Spatial Pyramid Pooling SRDCF Spatially Regularized Discriminative Correlation Filters SRR Short-Range Radar SSD Single Shot multibox Detector SURF Speeded Up Robust Features SVM Support Vector Machine SMT Surface Mount Technology Nomenclature list xxi
- 19. T TCN Train Communication Network TCP Transmission Control Protocol TCSN Train Control and Service Network TDD Time Division Duplexing TEDS Trouble of moving EMU Detection System TOF Time of Flight U UAV Unmanned Aerial Vehicle UDP User Datagram Protocol UHV Ultrahigh Voltage UNECE United Nations Economic Commission for Europe USB Universal Serial Bus UWB Ultrawideband V VR Virtual Reality VRP Vehicle Routing Problem VSA Variable Stiffness Actuator W WiFi Wireless Fidelity WPT Wavelet Packet Transform WTB Wire Train Bus Y YOLO You Only Look Once Nomenclature list xxii
- 20. CHAPTER 1 Introduction 1.1 Overview of rail transit robots Railway transit is vital to the national economy. Research in Japan and China indicates that development of the railway can lead to economic growth in the railway field [1,2]. To stimulate economic development, governments are actively developing the railway transport industry. In such a large industrial system, the development of automation can improve efficiency and reduce costs, so the level of automation in railways should be increased. The increase in automation requirements in the rail transit system generates a pursuit for rail transit robot systems. In worldwide, many countries have proposed similar strategic plans to encourage the development of robots in the railway transit system. Taking China as an example, “A Country With a Strong Transportation Network” and “Smart Railway” are two important plans that encourage the development of automation of the rail transit system. Driven by these plans, many Chinese rail equipment manufacturers and operators have carried out extensive research in robotics. Zhuzhou CRRC Times Electric Co., Ltd. applied an automatic generating line to produce high-speed train converters [3]. CRRC Qishuyan Institute Co., Ltd. developed an intelligent manufacturing workshop for gear transmission systems for high-speed trains. CRRC Zhuzhou Institute Co., Ltd. combined automatic guided vehicle (AGV) technology with urban transportation equipment and designed autonomous rail rapid transit (ART) [4]. A variety of robot systems are employed. The use of robot systems in the rail transit system can be divided into three aspects: manufacturing, dispatch, and maintenance. In these three parts of rail transit, different kinds of robots have completely different roles, as shown in Fig. 1.1. 1.1.1 Rail transit robots in manufacturing “A Country With a Strong Transportation Network” points out that intelligent manufacturing is required for the rail transit system. In the modern production line, robots can greatly improve processing efficiency. Taking the production of a high-speed train gearbox as an Robot Systems for Rail Transit Applications. https://doi.org/10.1016/B978-0-12-822968-2.00001-2 Copyright © 2020 Elsevier Inc. All rights reserved. 1
- 21. example, in the process flow of a transmission gear, robot arms can result in the efficient transmission of gears between different machine tools; when welding the gearbox, the welding robots can improve machining efficiency; when assembling the gearbox, assembly robots can work with high-precision; when the train is assembled, AGV enables the gearbox to be transported quickly between workshops. Applying robots in manufacturing not only saves processing time and labor costs, it improves manufacturing quality. 1.1.1.1 Assembly robots As for assembly robots, the primary mission is to achieve high-precision positioning of the workpiece. According to previous research, assembly costs account for 50% of total manufacturing costs [5]. Assembly robot systems can also be divided into rigid assembly and flexible assembly robots. Rigid assembly robots are customized processing systems for specific workpieces in the traditional industrial environment. Rigid assembly robots have poor generalization. If the production line is replaced with processed parts, the equipment needs to be customized. Replacement of equipment will cause a great economic burden. Compared with rigid assembly robots, flexible assembly robots can design customized processing programs according to the workpiece. Flexible assembly robots are programmable, which can result in different assembly schemes for different workpieces. Flexible assembly robots are significant for a flexible assembly system. In current industrial development, flexible assembly robots are the focus of development [6]. In the following discussion, assembly robots refer to flexible assembly robots. Figure 1.1 Robots in the manufacturing, dispatch, and maintenance of rail transit. 2 Chapter 1
- 22. An assembly robot consists of four components: machinery components, sensors, controllers, and actuators. To bring about a complex workpiece track in the real assembly environment, assembly robots usually have more than four degrees of freedom (DOFs). Mainstream assembly robots can be divided into two types: selective compliant assembly robot arms (SCARAs) and six-DOF robots. SCARAs have four DOFs, which are commonly used in electronic assembly, screw assembly, and so on [7]. SCARAs are specially designed for assembly applications by Yamanashi University. SCARAs contain two parallel joints, which can assemble a workpiece in a specified plane. Compared with six-DOF robots, advantages of SCARAs are a higher assembly speed and precision; disadvantages are limited workspace. Commonly used control strategies for SCARAs contain adaptive control, force control, robust control, and so forth [8]. In state-of-the-art research on robot control, intelligent algorithms are employed to improve control performance [9]. Dulger et al. applied a neural network to control the SCARA [10]. The neural network was optimized by particle swarm optimization to improve performance. Son et al. adopted an optimized inverse neural network for feedback control [11]. To deal with disturbances in running, the parameters of the inverse neural network are updated by a back propagation algorithm. Luan et al. used the radial basis function (RBF) neural network to achieve dynamic control of the SCARA [12]. Six-DOF robots can locate the workpiece at almost any point. Thus, six-DOF robots can handle the assembly task of complex three-dimensional (3D) workpieces. The dynamics of six-DOF robots are basic for operating the robots. Zhang et al. considered the friction of the robots and used a hybrid optimization method to model the dynamics of the six-DOF robot [13]. After optimization, dynamic accuracy increased significantly. Yang et al. proposed a simulator for the dynamics of the six-DOF robot [14]. Robots with a large degree of freedom have large feasibility. However, too much freedom is uneconomical. To handle the trade-off between economy and feasibility, the DOF can be optimized for specific tasks. Yang et al. proposed an optimization method to minimize the DOF [15]. This optimization method can reduce the DOF and improve the use of the DOF. Assembly robots should cooperate with the ancillary equipment. The fixtures are vital equipment to ensure cooperation in performance. The fixtures can fix the relative position between the workpiece and the robot under load. If the precision of the location of the fixtures is low, no matter how accurate the positioning precision of the robot is, it cannot achieve high-precision assembly. Currently, the flexible fixture is a future development [16]. Lowth et al. proposed a unique fixture that can adjust the radial and angular adaptively [17]. Although auxiliary devices are applied for assembly robots, the results of assembly robots may still be unsuccessful. Avoiding unsuccessful assembly is particularly Introduction 3
- 23. important in electric connector assembly, because the electric connector is not a rigid component. To detect the unsuccessful assembly of the electric connector assembly, Di et al. proposed a hybrid detection system with a force sensor and camera [18]. The fault diagnosis and prognosis system of assembly robots guarantees assembly accuracy. There are many studies about fault diagnosis and prognosis systems. Huang et al. designed a classifier for the wiring harness robot [19]. That study modeled the manufacturing process was and calculated the fault with a fuzzy model. Baydar et al. introduced a diagnosis model with error prediction [20]. The proposed model integrated the Monte Carlo simulation, genetic algorithm, and so forth. The functions of the fault diagnosis and prognosis system for assembly robots should contain the main aspects as given in Choo et al. [21]: (a) The health states of the assembly robots are monitored in real time. The monitored data are logged into the dataset. The health features are extracted from the health states of the assembly robots. The faults and remaining useful life can be calculated according to the features. (b) According to the fault diagnosis and prognosis results, the assembly tasks are reas- signed to make sure the failed assembly robots are replaced by the fully functioning robots. The maintenance plans can be made to repair the failed robots. 1.1.1.2 Collaborative robots When applying the resulting classical industrial robot systems, interaction between robots and humans is limited. There are three reasons for this phenomenon: (a) Traditional industrial robots do not consider moving humans. If a collision occurs along a certain trajectory, it may cause great damage to humans. Therefore, the working area of the industrial robot is mostly separated from the working area of the human. (b) The weight and volume of traditional industrial robots are large, and it is difficult for humans to operate robots. (c) Reprogramming of robots is difficult and requires special programming tools for tuning. However, humanerobot collaboration can combine human creativity with the efficiency of robots and can amplify the flexibility of robots and further improve work efficiency. Under this demand, collaborative robots are born. Compared with traditional robots, collaborative robots have three main advantages: safety, ease of operation, and ease of teaching. Some robot manufacturers have launched collaborative robotics products. The robot company Universal Robot launched the UR3 collaborative robot [22]. This robot is the first truly collaborative robot. The UR3 collaborative robot is based on a six-DOF robotic arm and is flexible enough to achieve complex motion trajectories. In terms of safety, the UR3 4 Chapter 1
- 24. collaborative robot has a collision monitoring system that protects human safety by monitoring the joint position, speed, and power of the robot. KUKA Robotics has launched a collaborative robot, the LBR iiwa [22]. The robot’s seven- DOF design provides greater flexibility than traditional six-DOF robots, enabling more complex trajectories to cope with complex environments that work with humans. The shape of the robot is designed to be ergonomic and easy for humans to operate. The outer casing is made of aluminum alloy, which can reduce weight and improve operability. The robot is equipped with torque sensors at each joint to monitor collisions in real time. The teaching method of the robot is dragging, which reduces the technical threshold of the robot operator. ABB launched the robot YuMi [22]. The robot is highly safe and can achieve humanerobot interaction in a small space. To improve the performance of collaborative robots, AAB acquired Gomtec Robotics, which launched the collaborative robot Roberta [22]. The Roberta can handle higher load applications compared with YuMi. Franka Emika launched the Franka Collaboration [23]. Like the LBR iiwa, the robot has seven DOFs. It is also equipped with a torque sensor on each joint to enable collision monitoring. Rethink Robotics launched the two-arm collaborative robot Baxter and the one-arm collaborative robot Sawyer [24]. The two robots are exquisitely designed with high positioning accuracy and can be assembled with high precision. Safety in humanerobot interaction is essential for collaborative robots. Threats to the safety of collaborative robots can be divided into two aspects [25]: (a) The first kind of threat is from robots. During operation, the robot may collide with workers and cause injuries. To ensure the safety of employees, the robot needs to detect the location of workers in real time and determine whether the location of workers is in a safe position. If workers intrude on the safe area, the robot should immediately stop to avoid a collision. In case a collision between a person and a robot occurs, the robot needs to detect the collision in time and change torque to minimize damage to workers. Mohammed et al. proposed a collision avoidance system for collaborative robots [26]. This system used depth vision sensors to detect the position of the worker. Considering the virtual model of the collaborative robot, a collision could be detected. The collaborative robot could take measures to avoid collisions. In addition to collision detection, it is necessary to ensure the integrity of the collabora- tive robot control system during operation. Failure of any part of the sensors, control- lers, or actuators will lead to the failure of humanecomputer interaction, thus threatening the safety of workers. In addition, interaction with robots may cause mental stress to workers [27], which will increase the risk for a collision. Introduction 5
- 25. (b) The second kind of threat is from the industrial process. In the process of humane robot interaction, workers need close contact with the manufacturing process. The temperature of the manufactured workpieces can cause damage to workers. Fallen workpieces can also endanger workers. Therefore, it is necessary to fully consider the impact of machining parts on workers in the design process of collaborative robots. In addition, the unreasonable ergonomic design of the collaborative robot during mainte- nance will have an impact on safety. 1.1.1.3 Automatic guided vehicles According to the level of automation, manufacturing systems can be divided into three levels [28]. Manufacturing systems in the first level are manual. Those in the second level have small-scale automated manufacturing in which transportation is carried out manually. Manufacturing systems in the third level have large-scale automated manufacturing and use automated transportation. Manufacturing systems in the third level are also called flexible manufacturing systems (FMSs). According to the Material Handling Industry of America, only 20% of the time is spent on processing and manufacturing; the remaining 80% is used for storage, handling, waiting for processing, and transportation [29]. As factory automation increases, transportation efficiency between workstations needs to improve. In the FMS, the AGV can improve the use of space in the factory and the efficiency of transportation in the material handling system. Therefore, transportation costs can be reduced. The rail transit manufacturing system is a typical FMS. Transportation is an important part of the rail transit manufacturing system. In the rail transit processing environment, not only the transfer of workpieces between processes within the plant but also the free flow of workpieces between plants is required. In the traditional rail transit manufacturing environment, transportation inside the factory is carried out by gantry cranes and the transportation between factories is carried out by trucks. These modes of transportation have some disadvantages. There is a safety hazard when using gantry cranes to lift. If the lifting workpiece falls, it may cause serious safety accidents. The use of trucks carries a higher cost and is less efficient to transport. Improving the safety and economy of products in the transportation process is important for the development of the rail transit industry. In the current manufacturing environment, the AGV is an effective mode of transportation. The AGV is safer compared with gantry cranes and is more automated and more efficient for transportation than trucks. The typical structure of an AGV consists of sensors, chassis, a control unit, and so on [30,31]. During work, the sensor can determine the position of the AGV and transmit the current position to the control decision system, and the control decision system plans an 6 Chapter 1
- 26. optimal path. The chassis is driven by the controller’s control command to transport the workpiece to the designated position. To improve transportation performance, the model predictive control of the AGVs should be achieved. The remaining power and the working state of the AGV should be predicted to calculate the remaining life of the AGV, and the dispatch plan of the AGV can be optimized with consideration of these factors to improve the operational safety of the AGV. Popular data-driven forecasting methods contain statistical methods, intelligent methods, and hybrid methods. The statistical methods can discover the statistical rule of the data and generate an explicit equation for prediction. Commonly used statistical methods contain autoregressive moving average, Winner process, Gaussian process, and so on. Intelligent methods can generate better forecasting performance than statistical methods with the help of the strongly fitting capacity of the neural network. The Elman neural network, multilayer perceptron, and extreme learning machine (ELM) are the three most popular intelligent methods. However, the training process of these neural networks depends on the initial values in some way. If the initial value is unsuitable, training of the neural network may stop at the locally optimal solution. To improve the performance of intelligent prediction methods, the initial values can be optimized by optimization algorithms [32]. Hybrid prediction methods combine data processing algorithms with statistical or intelligent methods. The decomposition algorithms are proved to be effective [33]. The decomposition algorithms can divide the raw series into several more stationary subseries. Each group of subseries has a simpler fluctuation mode than the raw series, so it is more predictable. After optimization, the control command can be assigned to the AGVs in two different ways: static control and dynamic control [34]: (a) Static control. The control commands are assigned before the task. Once the AGVs receive the control command, the transportation path will not be changed until the AGVs receive another control command. This control scheme is simple and easy to operate. However, flexibility is weak. (b) Dynamic control. This control method can adjust the control commands according to the real-time state of the AGVs, so the task scheduling strategy is complex. The multi-AGV system is being studied worldwide. Compared with the single AGV, the advantages of the multi-AGV system are: (a) The multi-AGV system can cover a large area. The single AGV can achieve trans- portation only between points. In the modern manufacturing environment, the transportation task is far more complex than the point-to-point transportation. The multi-AGV system can build a transportation network and improve transportation efficiency. Introduction 7
- 27. (b) The multi-AGV system can execute the transportation task in parallel. A multi-AGV system can perform tasks simultaneously for a complex task. Thus, the multi-AGV system can greatly improve the efficiency of transportation. The dispatch and routing of the multi-AGV are important. The conflict-free function of the multiple AGVs is the bottleneck for the multi-AGV system. Draganjac et al. proposed a control algorithm for the multi-AGV system [35]. The proposed algorithm can detect the conflict between the AGVs and guarantee the safe operation of the multi-AGV by the priority mechanism. Miyamoto et al. proposed a conflict-free routing algorithm for the multi-AGV system [36]. Because of the limited memory space of each AGV, a heuristic algorithm was adopted for routing. Małopolski et al. considered the transportation system in the factory as a combination of squares [37]. Based on the square topology, a novel conflict-free routing algorithm for the multi-AGV system was proposed. The dispatch plan of the multi-AGV should be calculated to optimize the task waiting time, collision, load use, and so forth [38]. The optimization methods can be divided into single- and multiple-objective optimization. Single-objective optimization can optimize only one objective function. If the objective function consists of several objective subfunctions, these subfunctions should be combined as the weighted sum [39]. However, it is difficult to design these weights. Therefore, the generated optimization results might not be the global optimal solution. The multiple-objective optimization can balance the trade-off between different subfunctions and generate a Pareto front [40]. The Pareto front contains many solutions, each of which has both advantages and disadvantages. The final optimization results should be selected according to the expert. In this manner, the intelligent decision-making ability of humans can be used to improve the dispatch performance of the AGV. The environment of the factory is dynamic. The AGVs should be flexible enough to cope with the dynamic manufacturing system. There are many studies on the dynamic transportation system. Brito et al. proposed a dynamic obstacle avoidance algorithm for the dynamic unstructured environment [41]. In this algorithm, model predictive control is applied to improve control performance. This algorithm was verified in the environment with walking humans. Li et al. proposed an integrated algorithm for obstacle avoidance [42]. This algorithm can generate a path for the AGV by a model-predictive algorithm. The AGV can track the generated path reliably. 1.1.1.4 Manufacturing robots Manufacturing robots contain many types including welding robots, drilling robots, grinding robots, milling robots, and so on. Manufacturing robots can produce better machining performance than a classical computerized numerical control (CNC) machine. 8 Chapter 1
- 28. For example, it is proven that a workpiece polished by a manufacturing robot has a better surface quality than any CNC machine [43]. The better performance of the manufacturing robots is because the robots have more flexibility to make sure the tool is in the right position. Welding robots are one of the most widely used manufacturing robots. Difficulties of welding robots are [44]: (1) it is hard to observe the welding seam in a complex manufacturing task, (b) it is hard to obtain the absolute and relative locations of the workpieces, and (c) the trajectory is hard to track. Current commonly used location methods for the welding seam are based on optical sensors such as a depth camera and laser sensor [45,46]. Jia et al. proposed a welding seam location method and trajectory tracking algorithm [44]. The proposed location method was achieved using a laser scanner, which could obtain the location and direction of the pipe. The cubic spline was applied to fit the obtained welding seam. The velocity control was used to track the welding trajectory. Liu et al. proposed a trajectory planning algorithm for welding robots to cope with a single Y-groove welding task [47]. This study provided two different velocity planning algorithms. A key concern with drilling robots is positioning accuracy. An inaccurate hole can reduce the mechanical performance of the equipment. The positioning compensation method can be divided into two types: model-based and model-free [48]. Model-based methods can guide the robot to move according to the measured positioning error. The essentials of model-based methods are the dynamics and kinematics of the drilling robots. Model-based methods are time-consuming. Model-free methods can solve this drawback. Model-free methods build a model to describe the relationship between the positioning error and the robot’s joints’ parameters, which do not consider the dynamics and kinematics of the robots. Commonly used model-free methods contain the interpolation method [49], cokriging method [50], and so on. Neural networks are applied to compensate intelligently for positioning . With the help of the strongly fitting capacity of neural networks, these intelligent positioning compensation methods obtain good performance. Yuan et al. used ELM to predict positioning error and guide the robot to compensate for it [51]. Chen et al. used the RBF neural network to estimate positioning error. The bandwidth of the adopted RBF neural was fine-tuned. Positioning accuracy can be improved by more than 80% with these intelligent positioning compensation methods [52]. The grinding manufacturing is highly precise, so the positioning accuracy of the grinding robots requires attention. In real applications, the grinding robots may deviate from the preset track because of the disturbance and wear to the tools. Thus, grinding robots should have the ability to self-adjust, to ensure manufacturing quality. Control of grinding robots is more difficult than for CNCs because grinding robots have more DOFs. Huang et al. proposed an intelligent gear grinding system [53] in which the robot arm can detect the Introduction 9
- 29. actual trajectory by a vision sensor and adjust the trajectory adaptively. Two cameras are adopted for tool center point calibration. When coping with the complex grinding task, multiple grinding robots are necessary, because multiple grinding robots can cope with the complex manufacturing task more easily than a single grinding robot. Han et al. [54] proposed a multiple grinding robot system and introduced the trajectory planning method for the multiple-robot system. Experimental studies indicated that the multiple grinding robot system can generate a steadier manufacturing trajectory. The robot only needs five DOFs for the milling task; the reserved one is the DOF of the spindle of the milling cutter [55]. In the real application, six-DOF robots are applied for milling to improve flexibility. The stiffness of milling robots is a challenge in manufacturing. Milling robots usually have low stiffness. Robots may deform or vibrate when milling workpieces. There have been many studies on improving milling performance. Peng et al. optimized the stiffness of the robots in the feed direction with a seven-DOF robot [56]. Commonly used manufacturing robots are developed from six-DOF robots. Trajectory planning is a common technology for manufacturing robots. The trajectory planning algorithm can generate the optimal trajectory and the robot control algorithm can drive the robot to follow a predetermined trajectory [57]. The optimization targets of the trajectory are the position, velocity, accelerated velocity of each joint. The trajectory planning algorithm is the basis of the industrial robot control system. The optimization functions of the trajectory contain many aspects, including reduce execution time, consumed energy, and impact. 1.1.1.5 Loadingeunloading robots In the existing transportation manufacturing environment, processing each workpiece requires different procedures. The connection between processes requires the workpiece to be moved from one machine to another [58]. This repetitive work can be replaced by loadingeunloading robots. The application of loadingeunloading robots has many benefits. On the one hand, loading and unloading of robots is repetitive and consistent, which can avoid a decrease in work efficiency caused by worker fatigue. On the other hand, using the robots can parameterize the loading and unloading operation and the fault source can be quickly located when troubleshooting problems, which is convenient for improving the processing quality. Loadingeunloading robots have a wide range of applications. Liu et al. adopted a loadingeunloading robot for the CNC and designed a control system based on a programmable logic controller [59]. Zhang et al. developed a robot system for loadingeunloading [60]. The designed robot was able to separate good and bad workpieces. Fan et al. used the loadingeunloading robot for electric equipment detection 10 Chapter 1
- 30. [61]. With the help of the loadingeunloading robot, automatic and high-precision detection is achieved. The development of loadingeunloading robots shows a trend in intelligence. Computer vision and neural network technology have been widely applied [62]. The adoption of computer vision can increase the flexibility of loadingeunloading robots. Loadingeunloading robots can observe the posture of the workpiece and adjust the robots to grasp the workpiece from the appropriate position. The neural network is widely used in controller design with its strong generalization ability. Gu et al. proposed a visual servo loadingeunloading robot [63]. The camera on the manipulator can detect the location of the workpiece and help the robot capture the workpiece precisely. The design contains two steps: (1) extract the features of the observed image and target image and calculate the error between these features; and (2) input the error into the neural network and output the control command for the robot. In the proposed robot control structure, the fuzzy neural network was adopted as the controller. 1.1.2 Rail transit robots in dispatch The running stage of the rail transit system takes up the longest length and generates the highest cost in the whole life cycle of the rail transit system. With the help of robot technology, the rail transit system can achieve automatic operation, thus improving operational safety and reducing transportation costs. Autonomous driving is the most important form of automation in rail transit. Autonomous driving is the application of robot control technology to the railway train. The automatic train control of the railway train contains automatic train supervision (ATS), automatic train protection (ATP), and automatic train operation (ATO) [64]. The ATS can supervise the running states of the railway train. The ATP system can monitor the train’s running position and obtain its speed limit to ensure its running interval and safety. The ATO can accept output information of the ATS and ATP and generate control instructions [65]. The international standard International Electrotechnical Commission 62,290 defines four grades of automation (GOA) [66]: (a) GOA1. The train is able to monitor the train’s operating status continuously. Operation needs to be carried out by drivers. (b) GOA2. This level is self-driving with driver duty. The train is able to drive automati- cally through the signal system, but the driver is required to close the door and issue the command. (c) GOA3. There is no need to equip drivers on this level of the train; the train can auto- matically complete the whole process of operation, including outbound, pit stop, Introduction 11
- 31. switch door, and so on. However, there is still a need for onboard personnel to deal with emergencies on this level of train. (d) GOA4. No operator is required on this class of trains, and the train’s control system can automatically respond to unexpected situations. ART is an important example of the automation of rail transit operation; it is developed by CRRC Zhuzhou Institute Co., Ltd. The core technology of this train is virtual orbit tracking control technology. That is to say, the ART train does not need a true track, but rather, the marks on the road serve as tracks. Because the intelligent rail train does not have a physical track, existing positioning devices used in the subway train and high-speed train cannot be used. The ART adopts the global satellite positioning system as the main positioning method [4]. During the operation, inertial sensors and angle sensors installed in intelligent trains are used to monitor the movement posture and position of the train, thereby enabling a comprehensive perception of the state of the ART train. The monitored train status information is fed back to the controller of the intelligent rail train to control the intelligent rail train to run along the virtual track. Using this control technology, all wheels of the ART train can be driven along a virtual track [4]. This position accuracy can ensure the passing performance of the ART rail train in urban traffic. To improve operational efficiency, the ART train is also equipped with automatic driving technology [4]. The ART train can intelligently sense the environment and control the automatic operation of the train. 1.1.3 Rail transit robots in maintenance Robots have become the focus of modern rail transit equipment manufacturing and operation. With the increase in railway operating mileage and operational density, the workload of railway maintenance support is also increasing, and higher requirements are put forward for the maintenance and repair of railway infrastructure equipment. Currently, the maintenance of railway infrastructure equipment is mainly carried out by workers [67]. The main problems Figure 1.2 Role of robots in rail transit maintenance. 12 Chapter 1
- 32. are that (1) diversified test operations have higher requirements for operators, (2) it is inconvenient to have operators carry measurement equipment, and (3) manual inspections maintain labor intensity and quality is uncontrollable. As shown in Fig. 1.2, robots can have an important role in the maintenance stage of rail transit in this case. 1.1.3.1 Inspection robots The objectives of railway maintenance include tracks, instruments, and equipment next to railways, traction substations, pantographs, temperature of instruments, bogies, and foreign matter intrusion. According to the work scenario, the inspection area of rail transit inspection robots is mainly divided into traction substations and railway. 1.1.3.1.1 Inspection in traction substation The traction substation converts electric energy from the regional power system into electric energy suitable for electric traction according to the different requirements of electric traction on current and voltage. The resulting electricity is then sent to catenary lines set up along the railway line to power electric locomotives, or to the urban rail transit system to power electric subway vehicles. The inspection work is the top priority for the daily maintenance of the traction substation. However, because of insufficient staffing and other reasons, the inspection cannot meet the needs. In addition, with the increase in substations, operation, and maintenance, personnel are struggling to comply. It is also difficult to keep satisfactory working conditions and a strong sense of responsibility. Currently, the technology of intelligent robots is increasingly developing and the use of robots to replace manual inspections of equipment partially or completely has become the trend of future substations [68]. Substation intelligent inspection robots are equipped with advanced equipment such as infrared thermometers, high-definition cameras, and audio signal receivers. According to the preset inspection tasks, the entire station equipment and partitions are inspected in time. All task modules of meter reading, temperature measurement, sound collection, and functional inspection are processed together to identify and alert regarding discovered equipment abnormalities, defects, and hidden dangers in a timely manner. In addition, robots intuitively form charts and report weekly on operational data indicators of devices stored in the inspection so that maintenance personnel can perform equipment maintenance and fault analysis better. By using the inspection robots in the substation, the equipment in the substation can be inspected more reliably and the data analysis returned by the system can give the maintenance personnel a better understanding of the working state of the equipment. Recording and analyzing historical data can also significantly improve the predictability of equipment in the substation, providing more effective data support for the condition of maintenance and a status assessment of the equipment, which will directly reduce the possibility of equipment accidents [69]. Using inspection robots to inspect substation Introduction 13
- 33. equipment can not only keep field maintenance personnel away from the field equipment, which improves the safety and reliability of the inspection work, it provides a new method for spot inspection for the popularization of an unattended substation and truly improves the safety and reliability of substation equipment. 1.1.3.1.2 Inspection along the railway With the rapid development of rail transit and the continuous improvement in passenger high-speed processes, railway transport has been gradually heading toward function integration, information sharing, command concentration, and highly automated transition and transformation. The urgent demand for railway transportation safety poses new challenges to the railway traffic safety assurance system. To adapt to the new needs of railway development, it has become the focus of the construction of the railway traffic safety assurance system to build a fully covered and highly reliable railway intelligent monitoring system based on high information sharing [70]. The higher speed brings convenience to people, but it also increases the difficulty of preventing railway disasters. Although a high-speed train is equipped with a more advanced safety system, the large braking distance caused by high-speed driving reduces the flexibility of the train to respond to emergencies. In the event of an accident, a high-speed train often has more serious consequences than an ordinary train. Therefore, various factors such as the intrusion of foreign matter, natural disasters, and line failures may cause major railway accidents. Casualties and property loss caused by the intrusion are enormous and frequently occur. Therefore, research and development into railway foreign matter intrusion detection are of great significance to ensure the safe operation of trains [71]. The intrusion of foreign matter has the characteristics of irregularity, suddenness, and unpredictability. The fixed transportation mode of the operation route makes emergency measures that can be adopted limited when the train detects an intrusion on the road. If real-time detection of invasive foreign matter can be achieved and necessary measures are taken in time, property loss can be reduced as much as possible. The traditional foreign matter intrusion approach is to place sensors at key positions along the railway to monitor intrusion conditions in real-time. This method has many drawbacks, such as an unreasonable arrangement of the sensor and fixed monitoring position. With the rapid development of robotics, robots have entered the field of railway foreign matter invasion detection. The railway is inspected by an infrared sensor, a laser sensor, a smart camera, and an ultrasonic device. 1.1.3.2 Channel robots High-speed trains may experience poor performance or even failure of key components during long-term high-speed service. If a safety accident occurs, it will cause severe economic loss and even casualties. Therefore, how to monitor and diagnose its operating 14 Chapter 1
- 34. status has become an important research topic [72]. Among them, the bogie is a main components of the train. The basic structure of the bogie mainly includes the wheel set, suspension system, frame, auxiliary suspension system, and so on. Therefore, it is very important to research fault diagnosis for the high-speed train, especially the diagnosis of bogie faults [73]. With the complex diversification of high-speed rail operating conditions and the increase in operating mileage, mechanical wear and aging of vehicles are gradually accelerating, which causes safety hazards for train operation. Among them, the train axle is an important component to support the running of the train. During the driving process, almost all of the weight of the train and the impact caused by vibration are assumed. Therefore, the axle is one of the most vulnerable parts of the train. As a result of improper assembly of the axles and excessive operation of the axles during maintenance, the weight of axles will be aggravated. When the axle temperature rises abnormally, movement between the axle and bearings will be worse, friction and wear will be aggravated, and smoothness will be reduced. In this case, if the vehicle continues to run without emergency treatment, hidden safety hazards will occur and the train will be delayed or even derailed. Therefore, the health status of the bogie is directly related to the safety of the vehicle operation. The fault warning and diagnosis of the axle of the high-speed train are an urgent problem to be solved. With the rapid development of sensor technology, a large amount of real-time monitoring data generated during the operation of high-speed trains has been collected. Through the analysis and mining of historical monitoring data, the potential correlation and periodicity can be explored to provide a basis for train fault diagnosis. By analyzing the monitoring data collected during the train’s operation, it is important to identify the health status and the potential safety hazards that may exist at the time of the early warning, which is greatly significant for ensuring the safe operation of the train and improving the reliability of the train’s operation. Bogie fault diagnosis includes fault vibration signal processing and fault identification. Among them, fault vibration signal processing includes time-domain analysis, frequency- domain analysis, and time-frequency analysis. Fault identification uses the expert system, pattern recognition, neural networks, deep learning, and so forth [74,75]. Because the high-speed train bogie has nonlinear characteristics, its vibration signal is a nonlinear, nonstationary signal. Therefore, applying effective fault vibration signal processing methods to extract state information and constructing a fast and effective bogie fault diagnosis method is critical for accurately evaluating the safety state of the train. With the rapid development of robotics, more and more trains are on the line and rail transit channel robots have emerged. Compared with industrial robots and inspection robots, rail transit channel robots have been in development for a short time, but they form Introduction 15
- 35. a certain scale. Channel robots are mainly based on intelligent technology and integrate various sensing equipment. It is dedicated to the collection and diagnosis of fault information. The biggest contribution of rail transit channel robots is to replace on-site workers to check the train’s status. The staff of the inspection station can carry out remote control operations through network technology and can reliably monitor the site in real time, which can objectively monitor the equipment and meet the requirements of flexibility, high precision, and antiinterference. In general, rail transit channel robots need to meet the following requirements: (a) The channel robot belongs to the mobile robot. Unlike the fixed industrial robot, it needs to move according to the actual task. Therefore, the navigation control system needs to have a control module to perform robot motion control. Of course, the move- ment of the channel robot can be along the guide rail or autonomous navigation. (b) The robot needs to carry a variety of sensors to obtain information about the detected target, laying the foundation for intelligent fault diagnosis. (c) The navigation system needs to be modularized to make it easy to replace modules and easy to maintain. Communication between modules must be completed. (d) The channel robot needs to upload collected information and the corresponding fault type, fault location, and so on to the prebuilt historical fault database. With the contin- uous increase in fault information, the database has gradually grown. Data mining can be used to predict the fault and life of train equipment and to identify potential dangers in advance. Rail transit channel robots mainly perform a fault diagnosis of trains. Generally speaking, the workflow is as follows: (1) the maintenance task is carried out along the ground track to the checkpoint; (2) the vehicle maintenance information is collected by the onboard sensor including the visible light camera, the laser sensor, and the infrared camera, and some sensors are installed on the robot arm; (3) the collected image for fault diagnosis analysis is uploaded; and (4) the fault diagnosis results are reported and stored, including the fault types, fault problems, fault degrees, and so on. In addition to the growing database, the big data analysis module will perform data mining on train bogie equipment failure information to predict potential equipment failures. In addition, the robot continuously detects its own power and performs a recharging task once the minimum battery threshold is reached. Channel robots with a trouble of moving electric multiple units (EMU) detection system (TEDS) intelligent sensing system have been used for EMU fault diagnoses, aiming to use the rail transit channel robots to complete train status detection and fault diagnosis [76]. The hardware structure of a channel robot mainly includes a ground rail, dual-arm robot, laser sensor, visible light camera, infrared camera, server, and recharging device. Among them, the function of the robot guide is to drive the robot and make it move along the 16 Chapter 1
- 36. preplanned route to expand the working radius of the industrial robot. The arm of a two-armed robot is similar to that of an industrial robot. The difference is that a two- armed robot can perform more complex tasks and control is more difficult. Laser sensors, visible light cameras, infrared cameras, and other in-vehicle sensors belonging to a channel robot are used to collect different types of data. The background server is mainly used for the channel robot control, data storage, fault diagnosis, and big data processing. The self-recharging technology of the recharging device is an important means to ensure the normal operation of the channel robot. Currently, the more advanced one is noncontact recharging. Compared with contact recharging, noncontact recharging has the advantages of a small footprint and low cost. The software of a rail transit channel robot mainly includes intelligent visible light image analysis, infrared thermal imaging temperature analysis, location detection, and big data analysis. Among them, visible image intelligent analysis and infrared thermal image analysis are used for fault diagnosis of the train. Visible light image analysis mainly includes image registration and change detection algorithms, and the train fault is found by comparing images. Infrared thermal imaging temperature analysis mainly includes fault location and image segmentation, fault location image recognition, and classification. The former uses image processing technology and the latter relies on machine learning methods. Location detection mainly uses laser ranging technology to determine the location of train faults. Big data analysis is mainly based on the train bogie equipment fault information database; it uses data mining and machine learning and predicts the potential fault of the equipment to avoid accidents. 1.1.3.3 Unmanned aerial vehicles for inspection Railway accidents are sudden and random, so regular inspections of railways must be rigorous and critical. With the continuous development of unmanned aerial vehicle (UAV) technology, the demand for civilian UAVs in many countries is growing and UAVs have an increasingly important role in many fields. Research scholars have a major interest in developing new types of UAVs that can fly autonomously in different environments and locations and perform a variety of tasks [77]. Therefore, various types of the UAVs with different sizes and weights have been invented and designed for different situations and scenes. For example, UAVs slowly began to replace some manual work of railway operations and maintenance and have been successfully applied in the field of railway safety. Combining existing technical skills of UAVs with the daily safety requirements of railway tracks, UAVs can inspect for daily monitoring of railway tracks. Rail transit inspection UAV inspection refers to the use of the UAVs equipped with visible light cameras and infrared sensors to detect the operation status of railway equipment and real- time weather and discover the hidden dangers of railway train operation. Equipment along the railway includes pantographs, power lines, and so on. Introduction 17
- 37. 1.1.3.3.1 Pantograph The pantograph and catenary technology can connect electrical energy from a contact network to an electric locomotive [78]. As the speed of the train increases, the vibration characteristics of the catenary and the pantograph are particularly important for the stability and safety of the train. To obtain a stable current, the contact pressure should be maintained between the contact slide of the pantograph and the contact line. Receiving current from the traction net by the pantograph is a process in which several mechanical movements occur simultaneously during the high-speed running of a high-speed train. The pantograph and the locomotive are closely connected. When the locomotive is driving fast, the pantograph is also rubbing quickly and moving forward relative to the contact line. Because of the structural characteristics of the pantograph, it maintains tension, so it generates vibration in the vertical direction. To ensure the pressure value, the pantograph is also in contact with the catenary, and high-speed lateral vibration occurs. The catenary has an inherent vibration frequency that creates a traveling wave propagating along the catenary. These kinds of movements occur at the same time and are likely to cause separation between the nets [78]. In the case of high-speed driving, these types of motion are superimposed, and the accuracy of construction of each line is different. This makes it difficult to maintain stable contact pressure between the pantograph and the contact line. When the contact pressure is zero, it is likely that mechanical disengagement will occur, which is the pantograph-catenary offline. When the electric locomotive runs at high speed, contact between the pantograph and the high-voltage catenary is good, which is equivalent to a closed circuit with an inductive load, and the load operates normally. When the net is separated, it is equivalent to disconnecting the inductive load circuit. Because of its circuit characteristics or overvoltage, overvoltage between networks is high enough to rupture the air around the gap, which will cause a spark discharge and then an arc discharge. Whether it is spark discharge or arc discharge, the locomotive current at this stage will be greatly affected. At this stage, not only will overvoltage be generated, electromagnetic noise will be disturbed. In addition, high heat released by the arc will melt the metal and cause wear on the pantograph and catenary. When there is an offline situation between the pantograph and catenary, safe and stable operation of the high-speed railway will face great hazards: unstable operation of the electric locomotive, severe wear of the pantograph slide, deterioration of the traction motor rectification conditions, and generated overvoltage in the main circuit. According to statistics, railway accidents caused by failure of the traction power supply system accounted for 40% and 30% of accidents in 2009 and 2010, respectively [79]. The quality of the pantograph-catenary system has become an important prerequisite for speeding and improving operational reliability [79]. Currently, the real-time dynamic monitoring system of the pantograph-catenary system has been installed on newly built trains, and the test system has been installed in the operation EMUs. The system is 18 Chapter 1
- 38. installed on top of the train and can detect the working state of the pantograph in real time. The main method is to collect image information through the camera for analysis and processing, and the pan/tilt can adjust the shooting angle of the camera. Nevertheless, the monitoring effect of the bow monitoring system will be limited by the fixed installation position of the camera and the range of camera angles, so that the pantograph can be monitored from only one direction and the pantograph cannot be fully fault- monitored. With the gradual rise of UAV technology, a train pantograph monitoring system based on the UAV inspection has emerged. Depending on the flexibility of the UAVs, it is theoretically possible to achieve all-around detection of the state of the pantograph and avoid blind spots in the pantograph failure. Moreover, considering simultaneous work in different areas, multiple sets of parallel UAVs can be used for detection to improve efficiency. 1.1.3.3.2 Power lines The overhead transmission line is a kind of steel frame structure that erects the wire and keeps it away from the ground. The safety of overhead transmission lines affects people’s normal life, is related to economic production in the country, and even poses a threat to national security. Long-distance transmission lines are characterized by long line distances and high voltage levels. Transmission lines are also called overhead transmission lines. The transmission capacity of the lines is large, the transmission distance is long, and they are in the open air for a long time. They are often affected by the surrounding environment and natural changes such as snow, lightning, ice, external forces, birds, and so forth. An ultrahigh voltage (UHV) transmission line has high voltage, a long line, a high pole tower, a large conductor, and a complex geographical environment. It puts high requirements on operation and maintenance [80]. Traditional line maintenance and overhaul operations have complicated processes, high risks, a poor monitoring work environment, and low labor efficiency. If there is an emergency trip, power failure, and nature force majeure, only low-end inspection equipment such as a telescope and night vision device can be used. Therefore, the manual operation and maintenance mode are unable to cope with the safe operation and maintenance of the UHV over a long distance. 1.1.3.3.3 Foreign matter invasion detection Compared with the land inspection robots introduced earlier, the UAVs can detect foreign matter intrusion with the obvious advantages of wide detection range and emergency response [81]. When the railway is hit by a major natural disaster, the advantages of inspection UAVs are highlighted. For example, serious mudslide disasters may occur along the railway line, posing a serious hazard to the railway subgrade. If a mudslide is not found and taken emergency measures in time, it will lead to serious traffic accidents. In general, the occurrence of mudslide disasters is accompanied by strong rainfall. In this case, the train driver’s line of sight is seriously degraded and it is difficult to detect the danger in time during heavy rain. Historically, there have been train crashes in which the Introduction 19
- 39. accident occurred because heavy rains caused mudslides to ruin railway bridges [82]. Railway inspection UAVs can take pictures of the railway along with the high altitude and determine whether the railway is normal along the way through wireless transmission technology and image processing technology. 1.2 Fundamental key problems of rail transit robot systems Key problems of rail transit robots are positioning navigation and path planning, humanerobot interaction, and power management, (Fig. 1.3). 1.2.1 Navigation 1.2.1.1 Development progress The development of robotics has gradually expanded from the industrial field to rail transit, services, medical, entertainment, fire protection, and other fields. The application of robot technology has brought tremendous productivity and economic benefits to society. How to make robots more efficient and serve humans more intelligently is the focus of scientific research. The positioning and navigation of robots have been the core emphases of content since the development of robotics. The positioning of mobile robots is a prerequisite for all of their actions. The accuracy of positioning determines the accuracy of the robots’ feasible operation. It is the basis for ensuring the robot navigation system. The positioning of rail transit robots is to use sensors to sense the surrounding environment and then determine its position and posture in the workspace. In the case of positioning, robots start the navigation operation and move to the target autonomously through a certain guiding method. Path planning means that the autonomous mobile robots can intelligently plan the relatively shortest and least Figure 1.3 Key problems of rail transit robot systems. 20 Chapter 1
- 40. time-consuming path for the target task. Path planning can independently interact with the intelligent environment through onboard equipment such as an automatic elevator ride, intelligent access control, and automatic obstacle avoidance function. Under the premise of ensuring the safety of robots, people, and the work environment, the target task is completed. 1.2.1.2 Methodologies 1.2.1.2.1 Positioning The key to the positioning and path planning of mobile robots lies in two points: “Where am I?” and “Where am I going?‘”. The former is positioning and the latter is path planning. Mobile robot positioning technology solves the first problem: that is, it achieves the precise positioning of robots. If a robot can self-navigate, the first condition is that the robot knows where it is. There are several basic tasks for navigation: (1) global positioning, sense the position, posture, and environment information of the robot; (2) path planning, plan an accessible path to the target point from the environmental model; and (3) motion control, select the motion control method, and move to the destination along the planned safe path. Many scholars have made great progress in the field of positioning [83]. The main research directions of mobile robot positioning technology include simultaneous positioning and map construction, path planning, and motion control. In the process of robot navigation, positioning is the first problem to be solved; the posture information and current state of robots need to be known. Existing positioning methods of intelligent mobile robots mainly include relative and absolute positioning [84]. Absolute positioning can obtain the posture and position information of a robot directly through the sensor carried by the mobile robot itself. Relative positioning requires the determined original position. Sensors for a robot can obtain the concrete position of the distance and direction to the target point. According to the current point moving distance and the direction of rotation angle, the robot can be aware of its posture and direction. Absolute positioning methods include a magnetic compass and global positioning system (GPS). Relative positioning can be divided into the dead reckoning method based on the inertial sensor and the dead reckoning method based on the odometer. According to the different types of sensors used to perceive the environment, navigation of mobile robots mainly includes GPS, gyroscope, inertial, photoelectric coding, magnetic compass, laser, and ultrasonic. 1.2.1.2.2 Path planning The path planning problem of mobile robots is a hot spot in the field of mobile robot navigation research [85]: mobile robots can find an optimal or near-optimal path from the starting state to the target state that avoids obstacles based on one or some performance indicators (such as the lowest working cost, the shortest walking route, the shortest Introduction 21
- 41. walking time, etc.) in the motion space. Generally speaking, the path planning selects the shortest distance of the path: that is, the shortest length of the path from the starting point to the target point as a performance indicator. The mobile robot path planning method can be divided into two types according to the known degree of environmental information: path planning based on global map information or local map information [86]. These two path planning methods are referred to as global path planning and local path planning. The global path planning method can generate the path under the completely known environment (the position and shape of the obstacle are predetermined). With the global map model of the environment where the mobile robots are located, the search is performed on the established global map model. The optimal algorithm can obtain the optimal path. Therefore, global path planning involves two parts: establishment of the environmental model and the path planning strategy. Path planning requires building an environmental map. Environmental map construction refers to the establishment of an accurate spatial location description of various objects in the environment in which the robot is located, including obstacles, road signs, and so on: that is, the establishment of a spatial model or map. The purpose of constructing the environmental map is to help the mobile robot plan an optimal path from the starting point to the target point in the established environment model with obstacles. There are many mature methods for establishing an environment model for mobile robot path planning. Existing basic environment models mainly include a grid decomposition map, quad split graph, visibility graph, and Voronoi diagram. After the environmental map is built, global path planning is carried out. Algorithms of global path planning are mainly divided into two types: heuristic search methods and intelligent algorithms. The initial representation of the heuristic search is the A* algorithm developed by the Dijkstra algorithm. The A* algorithm is the most commonly used heuristic graph search algorithm for state space. In addition to solving problems based on state space, it is often used for the path planning of robots. Many scholars have improved the A* algorithm and obtained other heuristic search methods [87,88]. Intelligent algorithms have lots of studies, including ant colony [89], particle swarm [90], genetic [91], bat [92], simulated annealing [93], and so forth. Different from the global path planning method, the local path planning method assumes that the position of the obstacle in the environment is unknown, and the mobile robot perceives its surrounding environment and its state only through the sensor. Because the global information of the environment cannot be obtained, the local path planning focuses on the current local environment information of the mobile robot and uses the local environment information obtained by the sensor to find an optimal path from the starting point to the target point that does not touch the obstacle in the environment. The path planning strategy needs to be adjusted in real time. Commonly used methods for local 22 Chapter 1
- 42. path planning include the rolling window [94], artificial potential field [95], and various intelligent algorithms [96]. 1.2.1.3 Applications Positioning and path planning are applied to almost all mobile robots, the most widely used of which are the AGV robots and rail transit inspection robots. 1.2.1.3.1 Carrying automatic guided vehicle robots To reduce labor costs and storage costs, logistics sorting has gradually shifted from manual to automated. Intelligent unmanned storage has gradually become a hot spot for many scholars [97]. Unmanned factories and unmanned warehousing have become a trend in development. The intelligent logistics system began to appear, which aims to reduce handling, management, communication, and labor costs. All links in picking and unloading are carried out in an orderly manner without manual intervention. AGV robots have become the main method of automated warehousing. Performing collision-free path planning for multiple AGVs and performing tasks is a difficult problem for the AGV group. 1.2.1.3.2 Rail transit inspection robots The main method of inspection in rail transit is to use human power to conduct inspections. However, the work environment of rail transit equipment is complex and conditions are harsh. In case of bad weather, there is a personal safety risk for personnel conducting a manual inspection. At the same time, the efficiency of manual inspections is low, depending on business proficiency and the subjective initiative of the inspection personnel. The long-term labor of mechanical repetition easily causes negligence, so rail transit inspection robots have emerged as the times require, and will be used in the rail transit line and station equipment. New methods of inspection are presented and manual inspections will be phased out. Inspection robots can use various sensors equipped to detect the rail transit and rail transit vehicle equipment through various advanced technologies autonomously or with assisting inspectors. The use of inspection robots can detect the work conditions of rail transit equipment under severe conditions and ensure the normal and orderly operation of rail transit. This is important for the future realization of the full automation of rail transit. [98]. 1.2.2 Humanerobot interaction 1.2.2.1 Development progress Humanerobot interaction is a technology that studies people, computers, and their interactions [99]. It is a focus of competition in the information industry. Its extensive application will profoundly change the manufacturing, construction, and transportation industries. So far, humanerobot interaction technology has evolved to the interface of Introduction 23
- 43. gestures and language, combining the powerful processing power of computers. The design of the current humanerobot interaction system has achieved a lot of results, such as the handwritten Chinese character recognition system, Chinese speech recognition system, and gesture recognition. In terms of handwritten Chinese character recognition, the most advanced system has been able to identify 28,000 Chinese characters after years of research and development [100]. If written at a speed of 12 Chinese characters per minute, the recognition rate of the system is nearly 100% [100]. In the aspect of sign language recognition and synthesis, the Institute of Computing Technology, Chinese Academy of Sciences has successfully developed a Chinese sign language recognition and synthesis system based on multifunctional perception [101]. 1.2.2.2 Methodologies 1.2.2.2.1 Humanerobot interaction based on gestures Gestures are a natural and intuitive mode of communication. Vision-based gesture recognition is a key technology that is indispensable for the realization of a new generation of humanerobot interaction. Gestures are a variety of actions produced by a human hand or arm. They include static and dynamic gestures. Because of the diversity and ambiguity of gestures and the complex deformation of human hands, vision-based gesture recognition is a multidisciplinary and challenging research topic. To find a breakthrough, the use of gestures in interpersonal communication is required. In the vision-based gesture recognition system: (1) the video data stream is acquired by one or more cameras; (2) the system detects whether there is a gesture in the data stream according to the interaction model; and (3) if so, the gesture is segmented from the video signal, and the gesture model is selected for gesture analysis. The analysis process includes feature detection and model parameter estimation. In the recognition process, gestures are classified according to model parameters and gesture descriptions are generated as needed. Finally, the system drives specific applications according to the generated descriptions. The gesture model is important for the gesture recognition system, especially for determining the recognition range. The model selection depends on the specific application. A simple and rough model may be sufficient for a given application. However, to achieve natural humanecomputer interaction, a complex gesture model must be established. In this way, the recognition system can respond correctly to most gestures made by users. From the current literature, most gesture modeling methods can be summarized as gesture modeling based on the apparent model and gesture modeling based on the 3D model. The task of the gesture analysis is to estimate the parameters of the selected gesture model. The analysis generally consists of feature detection and parameter estimation. In the feature detection process, the subject that makes the gesture must be located. 24 Chapter 1
- 44. Depending on the conditions used, positioning techniques can be classified as color-based, motion-based, and multimode. Most color-based positioning techniques rely on histogram matching or the use of skin training data to create a lookup table. Color-based positioning techniques have significant drawbacks (i.e., skin color changes under different lighting conditions). Gesture recognition is the process of classifying a trajectory (or point) in a model parameter space into a subset of that space. The static gesture corresponds to a point in the model parameter space, and the dynamic gesture corresponds to a trajectory in the model parameter space. Therefore, their identification methods are different. Static gesture recognition algorithms include recognition based on classical parameter clustering techniques and recognition based on nonlinear clustering techniques. Unlike static gestures, dynamic gestures involve time and space. Most dynamic gestures are modeled as a single trajectory in the parameter space. Because of the different habits of different users, the difference in speed and proficiency in gestures will cause nonlinear fluctuations on the time axis of the trajectory. Eliminating these nonlinear fluctuations is an important problem that must be overcome by dynamic gesture recognition technology. Considering the different processing of the time axis, existing dynamic gesture recognition technology can be divided recognition based on hidden Markov model [102], recognition based on dynamic time warping [103], and so forth. 1.2.2.2.2 Humanerobot interaction based on human skeleton The development of human pose estimation has become increasingly practical. In the fields of gait analysis, humanerobot interaction, and video surveillance, human pose estimation have broad application prospects. The mainstream human body pose estimation algorithm can be divided into traditional and deep learning-based methods. Traditional methods are generally based on the graph structure; they design a 2D body part detector, use the graph model to establish the connectivity of each component, and continuously optimize the graph structure model to estimate the human body posture combined with the relevant constraints of human kinematics. Although traditional methods have high time efficiency, the extracted features are mainly histogram of oriented gradient [104] and scale-invariant feature transform features [105]. These features cannot fully use image information; thus, the algorithm is subject to different viewing angles, occlusions, and inherent geometric ambiguities of the image. At the same time, because of the single structures of the traditional models, when the posture of the human body changes greatly, the traditional models cannot be accurately characterized and expressed. There are multiple feasible solutions to the same data; that is, the results of attitude estimation are not unique, which limits the scope of application of traditional methods. On the other hand, most traditional methods are based on digital Introduction 25
- 45. images such as depth images for feature extraction. However, because the acquisition of depth images requires professional acquisition equipment, the cost is high, so it is difficult to apply to all application scenarios. In addition, the acquisition process needs to synchronize the depth cameras of multiple perspectives, which reduces the impact of the occlusion problem. These factors will make the acquisition process of human posture data complicated and difficult. In contrast, monocular cameras are more common, although the color images they collect are susceptible to environmental factors such as illumination. Neural networks can be used to extract more accurate convolution features than artificial features and improve the performance of the forecast. Because of this, human body pose estimation methods based on deep learning have been deeply studied [106e108]. The human body pose estimation method based on deep learning mainly uses a convolutional neural network (CNN) to extract the human body pose features from the image. Compared with artificial design features of the traditional method, the CNN can obtain more rich features of semantic information as well as obtain different types of multiscale and multiple-type human joint point feature vectors. 1.2.2.2.3 Humanerobot interaction based on speech recognition Speech recognition is the process of converting human sound signals into words or instructions. Speech recognition is based on speech. It is an important research direction of speech signal processing and a branch of pattern recognition. The research of speech recognition involves many subject areas such as computer technology, artificial intelligence, digital signal processing, pattern recognition, acoustics, linguistics, and cognitive science. It is a multidisciplinary comprehensive research field. Different research areas have emerged based on research tasks under different constraints. According to the requirements of the speaker’s way of speaking, these areas can be divided into isolated words, connected words, and continuous speech recognition systems. According to the degree of dependence on the speaker, these areas can be divided into speech recognition systems for the specific person and nonspecific person. According to the size of vocabulary, they can be divided into small vocabulary, medium vocabulary, large vocabulary, and infinite vocabulary speech recognition systems. From the perspective of the speech recognition model, the theory of the speech recognition system is based on pattern recognition. The goal of speech recognition is to transform the input speech feature vector sequence into a sequence of words using phonetic and linguistic information. According to the structure of the speech recognition system, a complete speech recognition system includes a feature extraction algorithm, acoustic model, and language model and search algorithm. The speech recognition system is essentially a multidimensional pattern recognition system. For different speech recognition systems, the specific recognition methods and techniques used by people are different, but the basic principles are the same. The collected speech signals are sent to feature 26 Chapter 1
- 46. extraction. The module processes and the obtained speech features are sent to the model library module. The speech pattern matching module identifies the segment speech according to the model library, and finally obtains the recognition result. 1.2.2.3 Applications In the field of rail transit, rail transit equipment manufacturing and rail transit operation and maintenance are the main components that determine whether the rail transit system can operate safely, efficiently, and economically for a long time. Introducing humanerobot interaction technology into rail transit equipment manufacturing will change the traditional equipment manufacturing production mode and transform the original staff to work together with production tools, which will greatly improve the quality and efficiency of equipment manufacturing. In the field of operation and maintenance, the friendly interface achieved by high-level humanerobot interaction will enhance the passenger’s ride experience from the inbound to the outbound service cycle and will greatly improve the management level of the rail transit system [109]. 1.2.2.3.1 Intelligent rail transit equipment manufacturing Intelligent industrial equipment has become a trend in the development of the global manufacturing industry. The robotic group used for welding replaces a separate welder to perform collaborative and high-precision welding and assembly operations. The automated guided carrier platform connects manufacturing modules in an intelligent manufacturing environment, based on unmanned and intelligent obstacle-avoidance delivery platforms. It can be shuttled within a complex environment, and the safety of production is greatly improved, which also increases the automation of the assembly line. In a complex intelligent manufacturing environment, humanerobot interaction control technology will have an important role in the coordinated operation of personnel and carrier platforms and the operator control of the delivery platform. On the one hand, the automatic guided carrier platform requires the operator to provide instructions to complete the loading, transporting, and unloading operations of the carrier platform. On the other hand, the automatic guided carrier platform needs to obtain operator instructions in a small space to complete high-performance obstacle avoidance [110]. 1.2.2.3.2 Intelligent operation and maintenance To ensure the stability, safety, and reliability of the urban rail transit system, it is necessary to achieve the intelligent manufacturing of rail vehicles, but also to ensure its intelligent operation and maintenance and improve system efficiency. The intelligent operation and maintenance framework includes parts such as the automatic identification of vehicles, automatic equipment control, processing data collection and analysis, and vehicle status detection, and is interconnected with the network. Introduction 27
- 47. In 2017, Tao et al. proposed a method for intelligent state-of-sales awareness and process control through intelligent instrument and detection technology [111]. They discussed the application mode of intelligent instrument and detection technology in intelligent manufacturing systems, digitalized in combination with actual projects. The innovative practice of production line transformation and construction puts forward the development direction of intelligent instruments and detection technologies that meet the requirements of intelligent manufacturing. Through the practical application of the project, this model can effectively promote the realization of data acquisition and intelligent perception in intelligent manufacturing systems and can provide effective data resources. In 2019, Xu et al. proposed an AGV intelligent logistics scheduling model and response method based on the request scheduling response model for the dynamic scheduling of intelligent manufacturing workshop logistics [112]. 1.2.3 Power management Robots can do a lot of work instead of humans; they have an important role in the fields of manufacturing, service, and rail transportation. Power management is mainly divided into the robot’s power forecasting and autonomous recharging. 1.2.3.1 Power forecasting The robot is powered by a lithium battery during the work process. How to ensure that the robot does not fail owing to a low battery during the mission is a serious problem. Traditional robots continuously detect their own power and perform recharging tasks when they find that the battery is below the minimum threshold. This method is inapplicable to the robot that is performing the task. For example, whether the robot can have enough power to reach the next tower or the next-level recharging base station during the inspection is an urgent problem to be studied. In addition, when the inspection personnel uses the robot to inspect the transmission line, it is necessary to grasp the lifetime and cruising range of the robot in real time, and then the next inspection plan can be formulated according to the robot’s endurance capacity. Most research on the cruising range of the inspection robot uses a rough estimation of the amount of power consumed by the robot at different stages; thus, the accumulation method is not scientific. Many scholars have begun to use the power time series data of the robot to establish a power forecasting model, to estimate the power in the subsequent time period [113,114]. The power forecasting model is built by data processing methods and machine learning. 1.2.3.2 Autonomous recharging Most intelligent robots use battery packs to power them and their working hours are generally shorter. The mobile robot that can self-charge to achieve long-term continuous work when the battery power is reduced to a certain level has attracted the interest of 28 Chapter 1
- 48. researchers. Research in the field of robotic autonomous recharging has begun. In the mid-20th century, Walter developed the world’s first self-aware electric robot [115], which was able to achieve autonomous recharging. Walter designed a recharging station with a charger and illuminator installed. It not only avoids obstacles, it uses a light beam to reach the recharging station for recharging. Autonomous recharging for mobile robots can be mainly divided into contact and noncontact, which is major topic in the field of robotics [116]. 1.3 Scope of this book Today, robotics and advanced rail transit are two major areas of development in the world. This book starts with the field of rail transit manufacturing, operation, and maintenance. It also introduces different types of robots and related technologies in detail. The book includes eight chapters: Chapter 1: Introduction. This chapter first outlines rail transit robots. Then, it describes three basic issues of robotics, including navigation, humanerobot interaction, and power control. Chapter 2: Rail transit assembly robot systems. This chapter first introduces progress in development and the key technologies of assembly robots. Then, it introduces the main components of assembly robots: machinery components, sensors, controllers, and actuators. Arm dynamics is a major research topic of assembly robots. This chapter introduces mathematical models of forward dynamics and inverse dynamics, discusses various trajectory planning algorithms in joint and Cartesian space, and proposes an artificial neural network algorithm for the inverse dynamic optimization calculation of the robot arm. Finally, the chapter presents the future direction of intelligent assembly robots. Chapter 3: Rail transit collaborative robot systems. This chapter first gives a basic definition of collaborative robots. The chapter then summarizes the development history, application field, and mode of collaborative robots. The hardware composition and characteristics of the collaborative robot are summarized. Finally, the basic concepts of feature extraction, object detection, and object tracking in visual perception are introduced. Chapter 4: Automatic guided vehicles in rail transit intelligent manufacturing environment. This chapter introduces progress in the development and types of AGVs in the rail transit intelligent manufacturing environment. Then, the main components of AGVs are Introduction 29
- 49. introduced. Next, the key technologies in AGVs are introduced. This part includes the navigation methods of AGVs, the path planning of AGVs, humanerobot interaction methods, and the AGV path planning application in rail transit intelligent manufacturing environment. Chapter 5: Autonomous rail rapid transit systems. This chapter introduces the hardware of the ART, which can be divided into railway-based vehicle basic components and various sensors based on the ART trains. Then, the technologies of the ART are introduced, including road traffic interaction, navigation, communication, and scheduling management. Finally, pedestrian detection algorithms for the ART are introduced in detail, including traditional pedestrian detection algorithms and those based on deep learning. Chapter 6: Rail transit inspection robots systems. This chapter introduces the development history, function, and hardware structure of rail transit inspection robots. The hardware structure is described in detail and explained in principle, which mainly includes the drive wheel, laser sensor, pan/tilt, infrared thermal imager, binocular vision camera, ultrasonic sensor, radio-frequency identification sensor, sound collection device, and wireless recharging device. Then, two key technologies to ensure inspection robots normally complete the inspection work, navigation methods, and the handeeye system are introduced in detail. Chapter 7: Rail transit channel robot systems. This chapter introduces the development history and hardware composition of rail transit channel robots, including the ground rail, dual-arm robot, infrared thermometer, and laser sensor. Then, the chapter describes the TEDS intelligent sensing system in detail, including visible light image intelligent analysis, infrared thermal image temperature warning, location detection, big data analysis, and autonomous recharging. Finally, fault diagnosis algorithms based on deep learning model and related principles are introduced. Chapter 8: Rail transit inspection unmanned aerial vehicle systems. This chapter introduces the development history of UAVs and their applications in various fields. Next, it introduces the basic structure of fixed-wing UAVs, unmanned helicopters, and rotary-wing UAVs. Various sensors are applied to rail transit inspection. Then, UAV technologies are described in detail, including communication methods, data collection methods, and scheduling methods. The communication mode, UAV condition monitoring, positioning and navigation algorithm, path planning algorithm, and mission assignment method are introduced. Finally, applications of UAVs in detecting of rail transit intrusion are introduced. 30 Chapter 1
- 50. Other documents randomly have different content
- 51. At the same moment the raven sailed past the door over the abyss, and uttered its hoarse cry; they heard the frozen heather bend beneath steps, and the fool appeared on the narrow terrace; he was wan and haggard, and cried, looking toward the fire: "Marc-Dives, try to leave this soon; I warn you. The fortifications of my domains must be free from such vermin. Take your measures accordingly." Then perceiving Jean-Claude, he knit his brows. "Thou here, Hullin?" said he. "Art thou yet far-sighted enough to accept the proposals I deigned to make thee? Knowest thou that the alliance I offered is the only means of saving thyself from the destruction that broods upon thy race?" Hullin could not avoid laughing. "No, Yegof," he replied; "my sight is not yet clear enough; it is dazed by the honor you offer me. But Louise is not yet old enough to marry." The fool seemed at once to grow more gloomy and thoughtful. He stood at the edge of the terrace, his back to the abyss, as if in his own hall, and the whirling of the raven around his head disturbed him not in the least. At length he raised his sceptre, and said, frowning: "I have twice demanded her, Hullin; twice thou hast dared to refuse me. Once more shall the demand be renewed—but once—dost hear?—and then the decrees of fate shall be accomplished." And turning upon his heel, with a firm step and haughty carriage, notwithstanding the steepness of the descent, he passed down the rocky path.
- 52. Hullin, Marc-Dives, and even the acrid Hexe-Baizel, burst into peals of laughter. "He is a fool!" said Hexe-Baizel. "I think you are not altogether wrong," sneered the smuggler. "Poor Yegof is losing his head entirely. But listen, Baizel; you will begin at once to cast bullets of all calibres; I am off for Switzerland. In a week, at latest, the remainder of our munitions will be here. Give me my boots." Drawing on the last, and wrapping a thick red woollen scarf about his neck, the smuggler took from a hook on the wall a herdsman's dark-green coat which he threw over his shoulders; then, covering his head with a broad felt hat and seizing a cudgel, he cried: "Do not forget what I say, old woman, or if you do, beware! Forward, Jean-Claude!" Hullin followed his host without even bidding Hexe-Baizel farewell, and she, for her part, deigned not to see her departing guest to the door. When they had reached the foot of the cliff, Dives stopped, saying: "You are going to the mountain villages, are you not, Hullin?" "Yes; I must give the alarm." "Do not forget Materne of Hengst and his two sons, and Labarbe of Dagsbourg, and Jerome of Saint-Quirin. Tell them there will be powder and ball in plenty; that Catherine Lefevre and I, Marc- Dives, will see to it." "Fear not, Marc; I know my men." They shook hands warmly and parted, the smuggler wending his way to the right toward Donon, Hullin taking the path to the left
- 53. toward the Sarre. The distance was rapidly widening between them, when Hullin called out: "Halloo, Marc! Tell Catherine, as you pass, that all goes well, and that I have gone among the villages." The other replied by a nod, and the two pursued their different ways. Chapter VI. An unusual agitation reigned along the entire line of the Vosges; rumors of the coming invasion spread from village to village. Pedlars, wagoners, tinkers, all that wandering population which is constantly floating from mountain to plain, from plain to mountain, brought each day budgets of strange news from Alsace and the banks of the Rhine. They said that every town was being put in a state of defence; that the roads to Metz, to Nancy, Huningue, and Strasbourg, were black with army and provision wagons. On every side were to be seen caissons of powder, shells, and shot, and cavalry, infantry, and artillery hurrying to their posts. Marshal Victor, with twelve thousand men, yet held the Saverne road, but the draw-bridges of all the fortified towns were raised from seven in the evening until eight in the morning. Things looked gloomy enough, but the greater number thought only of defending their homes, and Jean-Claude was everywhere well received. The same day, at about five in the evening, he reached the top of Hengst, and stopped at the dwelling of the hunter-patriarch, old Materne. There he passed the night; for in winter the days are short and the roads difficult. Materne promised to keep watch over
- 54. the defile of Zorn, with his two sons, Kasper and Frantz, and to respond to the first signal that should be made from Falkenstein. Early the next day, Jean-Claude arrived at Dagsbourg to see his friend Labarbe the wood-cutter. They went together to the hamlets around, to light in all hearts the love of country. Labarbe accompanied Hullin to the cottage of the Anabaptist Nickel, a grave and respectable man, but they could not draw him into their glorious enterprise. He had but one reply to all their arguments. "It is well," he said; "it is doubtless right; but the Scriptures say that he who takes up the sword shall perish by the sword." He promised, however, to pray for the good cause, and that was all they could obtain of him. They went thence to Walsch, where they found Daniel Hirsch, an ancient gunner in the navy, who promised to bring with him all the men of his commune. Here Labarbe left Jean-Claude to pursue his route alone. For a week more our brave friend wandered over the mountains, from Soldatenthal to Leonsberg, from Meienthal to Voyer, Cirey, Petit-Mont, Saint-Sauveur, and the ninth day he found himself at the shoemaker Jerome's, at Saint-Quirin. They visited together the defile of Blanru, after which Hullin, entirely satisfied with the results of his journey, turned once more toward his village. Since two o'clock in the afternoon he had been pressing on at a brisk pace, thinking of the life of the camp, the bivouac, the crash of battle, marches and countermarches—all those details of a soldier's life which he regretted so often and which he now looked forward to with ardor. The twilight shadows had begun to fall when he discovered the village of Charmes, afar off, with its little cottages, from which curled wreaths of light-blue smoke, scarcely perceptible against the snow-covered mountain-side, its little gardens with their fences, its slate-covered roofs, and to the left
- 55. the great farm-house of Bois-de-Chênes, and below, in the already dark ravine, the saw-mill of Valtin. And then, without his knowing why, a sadness filled his heart. He slackened his steps; thoughts of the calm, peaceful life he was losing, perhaps for ever, floated through his mind; he saw his little room, so warm in winter and so gay in spring, when he opened his window to the breezes from the woods; he heard the never- changing tick of the village clock; and he thought of Louise—his good little Louise—spinning in silence, her eyes cast down, or, mayhap, singing in her pure, clear voice at evening. Everything in his home arose before his eyes: the tools of his trade, his long, glittering chisels, the hatchet with the crooked handle, the porringers of glazed earthenware, the antique figure of Saint Michael nailed to the wall, the old curtained bed in the alcove, the lamp with the copper beak—all were before him, and the tears forced their way to his eyes. But it was to Louise—his dear child, his Louise—that his thoughts turned oftenest. How she would weep and implore him not to expose himself to the dangers of war! How she would hang upon his neck and beg him not to leave her! He saw her large, affrighted eyes; he felt her arms around him. He would fain deceive her, but deceit was no part of Jean-Claude's character; his words only deepened her grief. He tried to shake off his gloom, and, passing by the farm of Bois- de-Chênes, he entered to tell Catherine that all went well, and that the mountaineers only awaited the signal. Fifteen minutes later, Master Jean-Claude stood before his own door. Before opening it, he glanced through the window to see what Louise might be doing. She was standing in the alcove, and seemed busily arranging and rearranging some garments that lay
- 56. upon the bed. Her face beamed with happiness, her eyes sparkled, and she was talking to herself aloud. Hullin listened, but the rattling of a passing wagon prevented his hearing her words. He pushed open the door and entered, saying: "Louise, here I am back!" She bounded like a fawn to him and threw her arms around his neck, exclaiming: "It is you, Father Jean-Claude! How I have been waiting for you! How long you were gone! But you are home again, at last." "My child—many things"—said the good man, putting away his staff behind the door—"many things kept—" But his heart was too full; he could say no more. "Yes, yes, I know," cried Louise, laughing. "Mother Lefevre told me all." "How is that! You know all and only laugh? Well, it proves your good sense. I expected to see you weep." "Weep? And why, Father Jean-Claude? Oh! never fear for me; I am brave. You do not know me." Her air was so prettily resolute that Hullin could not help smiling; but his smile quickly disappeared when she added: "We are going to have war; we are going to fight, to defend the mountains!" "We are going! We are going!" exclaimed the good man, astounded. "Certainly. Are we not?" she asked, her smile disappearing at once.
- 57. "I must leave you for some time, my child." "Leave me? Oh! no, no. I will go with you; it is agreed. See, my little bundle is all ready, and I am making up yours. Do not be uneasy; let me fix everything, and you will be satisfied." Hullin stood stupefied. "But, Louise," he cried, "you are dreaming. Think, my child! We must pass long winter nights without a roof to cover us; we must bear hardship, fatigue, cold, snow, hunger, and countless dangers! A musket-ball would mar my pretty bird's beauty." "You are only trying your little Louise," cried she, now in tears, and flinging herself upon his neck. "You will not leave me here alone." "But you will be better here; you will have a good fire and food. Besides, you will receive news of us every day." "No, no! I will go with you; I care not for cold. And I have been shut up here too long; I want the fresh air. The birds are out; the redbreasts are out all winter; and did I not know what hunger was when a child? Mother Lefevre says I may go; and will you whom I love so much be more cruel than she?" Brave Jean-Claude sat down, his heart full of bitter sorrow. He turned away his head that she might not see the struggle going on within, while Louise eagerly continued: "I will be safe; I will follow you. The cold! What is the cold to me? And if you should be wounded—if you should wish to see your little Louise for the last time, and she not be near to take care of you— to love you to the last! Oh! you must think me hard-hearted!" She sobbed; Hullin could hold out no longer. "Is it indeed true that Mother Lefevre consents?" he asked.
- 58. "Yes, yes, oh! yes, she told me so; she said, 'Try to get Father Jean-Claude to let you; I am satisfied.'" "Well," said the sabot-maker, smiling sadly, "I can do little against two. You shall come! It is agreed." The cottage echoed with her cry of joy, and with one sweep of her hand her tears were dried, and her face, like an April sky, beamed in smiles. "You are a little gypsy still," cried Hullin, shaking his head. "Go trap a swallow." Then, drawing her to him, he continued: "Look you, Louise: it is now twelve years since I found you in the snow. You were blue with the cold, poor child; and when I brought you to the fire and warmed you, the first thing you did was to smile at me, and since that day your smile has ruled old Jean- Claude. But let us look at our bundles," said the good man with a sigh. "Are they well fastened?" He approached the bed, and saw in wonder his warmest coats, his flannel jackets, all well brushed, well folded, and well packed. Then in Louise's bundle were her best dresses and her thick shoes. He could not restrain a laugh, as he cried: "O gypsy, gypsy! It takes you to pack up." Louise smiled. "Then you are satisfied with them?" she asked. "I must be; but in the midst of all this fine work, you did not think, I'll wager, of getting ready my supper."
- 59. "That is soon done," said she, "although I did not know you would return to-night, Papa Jean-Claude." "That is true; but get something ready quickly; no matter what, for my appetite is sharp. In the meantime I will smoke a pipe." "Yes, smoke a pipe." He sat at the corner of his work-bench and drummed dreamily upon it. Louise flew to right and left like a veritable fairy, kindling up the fire, breaking eggs, and in the twinkle of an eye she had an omelette ready. Never had she looked so graceful, so joyous, so pretty. Hullin leaned his cheek upon his hand and gazed at her gravely, thinking how much firmness, will, resolution, there was in that little form, light as an antelope, but decided as a cuirassier. In a moment she had laid the omelette before him on a large plate, ornamented with blue flowers, a loaf of bread, his glass, and his bottle of wine. "There, Father Jean-Claude, eat your supper." The fire leaped and crackled in the stove, throwing ruddy stains on the low rafters, the stairs half in shadow and the large bed in the alcove, and lighting up the poor dwelling so often made joyous by the merry humor of the sabot-maker and the songs of his daughter. And Louise would leave all this without regret to brave the wintry woods, the snow-covered paths, and the steep mountain-side, and all for love of him. Neither storm, nor biting wind, nor torrents staid her. She had but one thought, and that was to be near him. The repast ended, Hullin arose, saying: "I am weary, my child; kiss me for good-night." "But do not forget to awake me, Father Jean-Claude, if you start before daybreak."
- 60. "Rest easy; you will come with us," he answered, as he climbed the narrow stairs. All was silence without, save that the deep tones of the village clock told the hour of eleven. Jean-Claude sat down and unfastened his shoes. Just then his eyes fell upon his musket hung over the door. He took it down, slowly wiped it, and tried the lock. His whole soul was in the work in which he was engaged. "It is strange—strange! The last time I fired it was at Marengo— fourteen years ago, and it seems but yesterday." Suddenly the frozen snow crunched beneath a foot-fall. He listened. Two taps sounded upon the window-panes. He ran and opened the door, and the form of Marc-Dives, his broad hat stiff with ice, emerged from the darkness. "Marc! What news?" "Have you warned Materne, Jerome, Labarbe?" "Yes, all." "It is none too soon; the enemy are advancing." "Advancing?" "Yes; along their whole line. I have come fifteen leagues since morning to give you warning." "Good. We must make the signal: a fire upon Falkenstein." Hullin's face was pale, but his eyes flashed. He again put on his shoes, and two minutes after, with his cloak upon his shoulders and his staff firmly clinched in his hand, he opened the door softly, and with long steps followed Marc-Dives along the path to Falkenstein.
- 61. Chapter VII. From midnight until six o'clock in the morning a flame shone through the darkness from the summit of Falkenstein. All Hullin's friends, and those of Marc-Dives and Mother Lefevre, with high gaiters bound around their legs, and old muskets upon their shoulders, trooped in the silence of the woods to the gorges of the Valtin. The thought of the enemy pouring over the plains of Alsace to surprise their glens and defiles nerved every heart and arm. The tocsin at Dagsbourg, at Walsch, and at Saint-Quirin ceased not to call the country's defenders to arms. Imagine the Jaegerthal, at the foot of the old burg, in the early morning hour, when the giant arms of the trees begin to break through the shadows, and when the approach of day softens somewhat the intense cold of the night. The snow lies deep upon the ground. Imagine the old saw-pit with its flat roof, its heavy wheel glittering with icicles; a fire of sawdust shining from within, but paling before the morning twilight, and around the fire fur caps and slouched hats and dark faces crowded together; further on, in the woods, and along the winding valley, were other fires lighting up groups of men and women seated on the snow. As the sky grew brighter friends began to recognize each other. "Hold! There is Cousin Daniel of Soldatenthal. You here too?" "As you see, Heinrich, and my wife too." "What! Cousin Nanette! But where is she?" "Yonder, by the large oak, at Uncle Hans's fire." They clasp each other's hand. Some slept, some piled branches and broken planks upon the fires. Flasks passed around, and those who
- 62. had warmed themselves made way for their shivering neighbors. But impatience was gaining upon the crowd. "Ah!" cried one, "we have not come here only to stretch our legs. It is time to look around, to agree upon our movements." "Yes! yes! let us organize and elect our leaders!" cried many. "No; all are not yet here. They are yet coming from Dagsbourg and Saint-Quirin," replied others. Indeed, as day advanced, the pathways of the mountain seemed full of people. There were already some hundreds in the valley— wood-cutters, charcoal-burners, and others—without counting the women and children. Nothing could be more picturesque than that halt in the snow, at the bottom of a defile covered to the clouds with high firs; to the right, valley following valley as far as the eye could reach; to the left, the ruins of Falkenstein, reaching, as it seemed, to the sky; and before you groups of thickly bearded men with gloomy brows, broad square shoulders, and hands callous from labor. Some of them, taller than their fellows, were red-haired and white-skinned, and seemed strong as the oaks of the forest. Of this number were old Materne of Hengst and his two sons, Frantz and Kasper. These three, armed with short Innspruck rifles, their high gaiters of blue canvas with leather buttons reaching above the knee, their bodies covered with hare-skin jackets, and their slouched hats pushed far back upon their heads, did not deign to approach the fire. Since one o'clock they had sat upon the felled trunk of a fir by the border of the brook, their eyes constantly on the watch, and their feet buried in the snow. From time to time the old man would say to his sons: "What are they shivering for yonder? I never saw a milder night at this season; it is a fine hunting night; the brooks are not yet frozen."
- 63. Every hunter as he passed pressed their hands, and then joined his fellows, who formed a separate band, among whom but few words passed, for silence is one of the great virtues of the chase. Marc-Dives, standing in the middle of another group, over whom he towered by a head, talked and gesticulated, now pointing to one part of the mountain, now to another. Opposite him was the old herdsman Lagarmitte, in his gray smock-frock, his dog at his side. He was listening open-mouthed to the smuggler, and from time to time gravely nodded his head. The remainder of the group was composed of wood-cutters and workmen with whom Marc had daily dealings. Between the saw-pit and the first fire sat the shoemaker Jerome of Saint-Quirin, a man between fifty and sixty years of age; his eyes were sunken, his face long and brown, and his yellow beard descended to his waist; his head was covered with an otter-skin cap; and as he leaned forward upon a heavy knotted staff, in his long woollen great-coat, he might easily have been mistaken for some hermit of the wilds. Whenever any one approached with news, Father Jerome slowly turned his head and listened with bent brows. Jean Labarbe sat motionless, his elbow resting upon his axe-helve. He was a pale man, with an aquiline nose and thin lips, and exercised a great influence over the men of Dagsbourg by the resolution and force of his character. When those around him cried out for action, he simply said, "Wait; Hullin has not arrived yet, nor Catherine Lefevre. There is no hurry," and all around became quiet. Piorette, a little, dry, thin, energetic man, with eyebrows meeting over his nose, and a short pipe between his teeth, sat at the threshold of the saw-mill, and gazed with a quick but thoughtful eye at the scene.
- 64. Nevertheless, the impatience increased every minute. A few village mayors in cocked hats called upon their people to deliberate. Happily the wagon of Catherine Lefevre at last appeared, and a thousand enthusiastic shouts arose on all sides. "Here they are! They have come!" Old Materne stood up upon the trunk of a tree and then descended, gravely saying: "It is they." Much excitement now prevailed; the scattered groups collected. Scarcely could the old woman be seen distinctly, seated upon a truss of straw with Louise by her side, when the echoes rang with the cry: "Long live France! Long live Mother Catherine!" Hullin, behind, his musket strapped upon his back, was crossing the field of Eichmath, grasping hands and saluting his friends: "Is it you, Daniel? Good-morning, Colon!" "Ha! Things look stormy, Hullin." "Yes, yes; we shall soon have lively times. You here, old Jerome! What think you of the state of affairs?" "All will yet go well, Jean-Claude, with God's help." Catherine, when she arrived in front of the saw-mill, ordered Labarbe to open the little cask of brandy she had brought from the farm-house. Hullin, approaching the fire, met Materne and his two sons. "You come late," said the old hunter.
- 65. "True, but there was much to be done, and too much yet remains to be done to lose more time. Lagarmitte, wind your horn." Lagarmitte blew until his cheeks seemed bursting, and the groups scattered along the path, and at the skirts of the wood hastened to assemble, and soon all were collected before the saw-mill. Hullin mounted a pile of logs, and spoke amid the deepest silence: "The enemy," said he, "crossed the Rhine the night before last. He is pressing on to our mountains to enter Lorraine. Strasbourg and Huningue are blockaded. In three or four days at most the Germans and the Russians will be upon us." A shout of "Long live France!" arose. "Ay, long live France!" cried Jean-Claude; "for, if the allies reach Paris, all our liberties are gone! Forced labor, tithes, privileges, and gibbets will flourish once more. If you wish that they should, let the allies pass." A dark scowl seemed to settle on every man's face. "I have said what I have to say!" cried Hullin, pale with emotion. "As you are here, you are here to fight!" "Ay, to fight!" "It is well; but one word more. I would not deceive you; I see among you fathers of families. We will be one against ten—against fifty. We must expect to perish! Therefore, let those whose hearts may grow faint ere the end comes, go. All are free!" Each in the crowd looked round to see his neighbors' faces, but no one left his place. Jean-Claude spoke in a firmer tone: "No one moves! All are ready for battle! A chief—a leader—must be named, for in times of danger everything depends on order and
- 66. discipline. He whom you shall appoint must be obeyed in all things. Reflect well, for on him depends the fate of every one of us." So saying, Jean-Claude descended from his tribune, and earnest voices began at once to whisper in the crowd. Every village deliberated separately; each mayor proposed his man; time passed; Catherine Lefevre burned with anxiety and impatience. At length she could contain herself no longer, and rising upon her seat she made a sign that she wished to speak. "My friends," said she, "time flies; the enemy is advancing. What do we need? A man whom we can trust; a soldier acquainted with war, and knowing how to profit by the strength of mere positions. Well, why not choose Hullin? Can any among you name a better? I propose Hullin!" "Hullin! Hullin!" cried Labarbe, Dives, Jerome, and many others. "Let us have a vote!" Marc-Dives, climbing the pile of logs, shouted in a voice of thunder: "Let those who are opposed to having Jean-Claude Hullin for our leader, raise their hands!" Not a hand rose. "Let those who wish Jean-Claude Hullin to be our chief, raise their hands!" Every hand rose. "Jean-Claude," said the smuggler, "you are the man. Come hither. Look!" Jean-Claude mounted the logs, and seeing that he was elected, said calmly:
- 67. "You name me your chief. I accept. Let old Materne, Labarbe of Dagsbourg, Jerome of Saint-Quirin, Marc-Dives, Piorette the sawyer, and Catherine Lefevre enter the saw-mill. We will hold a council, and in twenty minutes I will give my orders. In the meantime let every village detail two men to go to Falkenstein with Marc-Dives for powder and ball." Chapter VIII. Those whom Hullin named met in the hut attached to the saw-mill around the immense chimney. A sober sort of merriment seemed to play about the face of more than one. "For twenty years I have heard people talking of these Russians and Austrians and Cossacks," said old Materne, smiling, "and I shall not be sorry to see one at the muzzle of my rifle." "Yes," answered Labarbe; "we shall see enough of them at last, and the little children of to-day will have many a tale to tell of their fathers and their grandsires. And how the old women of fifty years hence will chatter of it at evening around the winter fire!" "Comrades," cried Hullin, "you know the country—you know our mountains from Thann to Wissembourg. You know that two grand roads—the imperial roads—traverse Alsace and the Vosges. Both starting from Bâle, one runs along the Rhine to Strasbourg, and enters Lorraine by Saverne. Huningue, Neuf-Brisach, Strasbourg, and Phalsbourg defend it. The other turns to the left to Schlestadt. Leaving Schlestadt, it enters the mountains, and passes on to Saint-Dié, Raon-l'Etape, Baccarat, and Lunéville. The enemy would like to force the passage of these two roads, as they are the best for cavalry, artillery, and wagons; but, as they are well defended, we need not trouble our about them. If the allies lay siege to the cities upon them, the campaign will be dragged out to a great length, and we shall have nothing to fear; but this is not probable.
- 68. After having summoned Huningue to surrender, and Belfort, Schlestadt, Strasbourg, and Phalsbourg, on this side of the Vosges, and Bitche, Lutzelstein, and Sarrebrück, on the other, they will fall upon us. Now, listen. Between Phalsbourg and Saint-Dié there are several defiles practicable for infantry, but only one for cannon, that is, the road from Strasbourg to Raon-les-Leaux, by Urmatt, Mutzig, Lutzelhouse, Phramond, and Grandfontaine. Once masters of this road, the allies can debouch in Lorraine. This road passes us at Donon, two leagues hence, to our right. The first thing to be done is, to establish ourselves upon it at the place most favorable for defence—that is, upon the plateau on the mountain; to break down the bridges, and throw heavy abatis across it. A few hundred large trees, with their branches, will do the work, and under their cover we can watch the approach of the foe. All this, comrades, must be done by to-morrow night, or by the day after, at the latest. But it is not enough to occupy a position and put it in a good state of defence. We must see that the enemy cannot turn it." "That is just what I was thinking," said old Materne. "Once in the valley of the Bruche, and the Germans can bring their infantry to the hills of Haslach, and turn our left; and there is nothing to hinder their trying the same movement upon our right, if they gain Raon-l'Etape." "Yes; but to prevent their doing either, we have only to occupy the defiles of the Zorn and of the Sarre on our left, and that of Blanru on our right. We must defend a defile by holding the heights, and, for that purpose, Piorette will place himself, with a hundred men, on the side of Raon-les-Leaux; Jerome, on Grosmann, with the same number, to close the valley of the Sarre; and Labarbe, at the head of the remain on the mountain, will overlook the hills of Haslach. You will choose your men from those belonging to the villages nearest your stations. The women must not have far to come to bring provisions, and the wounded will be nearer home. The chiefs of each position will send me a report each day, by a messenger, on foot, to Donon, where will be our headquarters. We
- 69. will organize a reserve also; but it will be time enough to see to that when our positions are taken, and no surprise from the enemy is to be feared." "And I," cried Marc-Dives, "am I to have nothing to do? Am I to sit with folded arms while all the rest are fighting?" "You will superintend the transporting of our munitions. No one among us understands keeping powder as you do—preserving it from fire and damp—or casting bullets and making cartridges." "That is a woman's work," cried the smuggler. "Hexe-Baizel can do it as well as I. Am I not to fire a shot?" "Rest easy, Marc," replied Hullin, laughing; "you will find plenty of chances. In the first place, Falkenstein is the centre of our line—our arsenal and point of retreat, in case of misfortune. The enemy will know by his scouts that our wagons start from there, and will probably try to intercept them. Shots and bayonet-thrusts will not be wanting. Besides, we cannot confide the secret of your cave to the first comer. However, if you insist—" "No," said the smuggler, whom Hullin's' reflections upon the cave touched at once. "No; all things well considered, I believe you are right, Jean-Claude. I will defend Falkenstein." "Well, then, comrades," cried brave Jean-Claude, "we will warm our hearts with a few glasses of wine. It is now ten o'clock. Let each one return to his village, and see to the provisions. To-morrow morning, at the latest, the defiles must be occupied." They left the hut together, and Hullin, in the presence of all assembled, named Labarbe, Jerome, and Piorette chiefs of the defiles; then he ordered those who came from the Sarre to meet, as soon as possible, near the farm of Bois-de-Chênes, with axes, picks, and muskets.
- 70. "We will start at two," said he, "and encamp on Donon, across the road. To-morrow, at daybreak, we will begin our abatis." He kept old Materne, and his two sons, Frantz and Kasper, by him, telling them that the battle would surely begin on Donon, and sharp-shooters would be needed there. Mother Lefevre never seemed so happy. She mounted her wagon, and whispered, as she embraced Louise: "All goes well. Jean-Claude is a man. He astonishes me, who have known him forty years. Jean-Claude," she cried, "breakfast is waiting, and a few old bottles which the Austrians will not drink." "Good Catherine, I am coming." But as he struck the horses with the whip, and as the mountaineers had just begun to scatter on their way to their villages, they saw, on the road to Trois-Fontaines, a tall, thin man, mounted upon a red mare; his hare-skin cap, with a wide peak, pulled well down upon his head. A great shepherd-dog, with long black hair, bounded beside him; and the skirts of his huge overcoat floated like wings behind him. "It is Dr. Lorquin, from the plain," exclaimed Catherine; "he who at the poor for nothing; and that is his dog Pluto with him." It was indeed he, who rushed among the crowd, shouting: "Halt! stop! Halt, I say!" His ruddy face, large, quick eyes, beard of a reddish-brown, broad, square shoulders, tall horse, and dog, in a moment appeared at the foot of the mountain. Gasping for breath, he shouted, in his excitement: "Ah the villains! They wanted to begin the campaign without me! They shall pay for it!"
- 71. And, striking a little box he carried at his crupper, he continued: "Wait awhile, my fine fellows, wait awhile! I have some things here you'll want by and by; little knives and great ones—round and pointed ones—to cut out the bullets and canister your friends yonder will treat you to." So saying, he burst into a gruff peal of laughter, while the flesh of his hearers crept. After this agreeable pleasantry, Dr. Lorquin said gravely: "Hullin, your ears should be cut off! When the country was to be defended, was I to be forgotten? It seems to me that a surgeon might be useful here, although may God send you no need of one!" "Pardon me, doctor; it was my fault," replied Hullin, pressing his hand. "For the last week I have had so many things to think of that some escaped me, in spite of myself. But a man like you need not be called upon by me to do his duty." The doctor softened. "It is all well and good," he cried; "but by your fault I am here late. But where is your general? I will complain to him." "I am general." "Indeed!" "And I appoint you surgeon-in-chief." "Surgeon-in-chief of the partisans of the Vosges. Very good, Jean- Claude." And, approaching the wagon in which Catherine was seated, the doctor told her that he relied upon her to organize the hospital department. "Very well," she answered; "forward. You dine with us, doctor."
- 72. The wagon started, and all the way the brave doctor laughingly told Catherine how the news of the rising reached him; how his old house-keeper Marie was wild with grief, and tried to keep him from going to be massacred by the Kaiserliks; the different episodes of his journey from Quibolo to the village of Charmes. Hullin and Materne and his sons marched a few paces in the rear, their rifles on their shoulders; and thus they reached the farm of Bois-de- Chênes.
- 73. Catholicity And Pantheism. Number One. Introductory. Man is made for truth. The ray of intelligence beaming from his countenance and kindling his looks with life marks his superiority over all inferior creation, and loudly proclaims this fact. Intelligence must have an object; and what can this object be but truth? As a necessary consequence from this fact, it follows that error can be nothing else than fragments of truth; ill-assorted, improperly joined together. Error does not consist in what logicians call simple ideas, or self-evident propositions; but in complex ideas, the result of a long chain of syllogisms. Another consequence, closely allied to the first, is, that the greater the error, the more universal and more widely spread, the more particular truths it must contain. Or, if it does not contain a greater number of partial truths, it must have the power of apparently satisfying a real and prevalent tendency of our mind, otherwise it would never exert dominion over the intelligence; or else it must possess the secret of awakening and alluring a true and imperative aspiration of our nature. It is through these views that we have been enabled to explain to ourselves the prevalence of Pantheism. The simple utterance of the word Pantheism, the Deity of everything, would seem to carry its refutation with it, so plain and evident is its falsehood, so glaring its absurdity. Pantheism, however, has been the universal error in time and space. In India, Persia, China, Greece, Rome, Pantheism flourished; now under a religious, and then under a philosophical form. After the Christian era it was the religion or system of those who did not understand the Christian dogmas as taught by the church; and the
- 74. fathers of the first centuries, in battling against Gnosticism, Eclecticism, and Neoplatonism, were struggling with this old error of the world—Pantheism. Depressed for awhile by the efforts of the doctors of the church, it arose with fiercer energy under the forms of all those heresies which attacked the dogma of the Incarnation of the Word. In the middle ages there were many philosophers who held Pantheism; and in modern times, since the dawn of the Reformation, it has become the prevalent, the absorbing error of the world. Always the same as to substance, it assumes every variety of form: now you see it in a logical dress, as in the doctrine of the German school; again it takes a psychological garb, as in that of the French school with Cousin at its head; or it assumes a social and political form, as in the Pantheism of Fourier, Leroux, Saint Simon, and all the progressists of every color or shade; and finally, it puts on a ghostly shroud, as taught by the American spiritualists. Under whatever garb it may appear, it penetrates and fills all, and pretends to explain all. It penetrates philosophy, natural science, history, literature, the fine arts, the family, society and the body politic, and religion. It holds its sway over all, and exhibits itself as having the secret of good and evil. How is this to be explained? If the falsehood of Pantheism be so evident, whence is it that it is the universal error in time and space, and has made such ravages in man's intelligence? The greater its falsehood, the more inexplicable becomes its prevalence. Has the nature of man changed? Has his intelligence lost its object? It is true, man's intelligence is not perfect. Since the fall it is weakened and obscured, but doubtless it has not ceased and could not cease to be intelligence; truth has not ceased to be its natural essential object. How, then, are we to explain the prevalence of so mighty an error? By the fact that it is a system which by its generality seems to satisfy a supreme tendency of our mind, and to appease one of the most imperative cravings of our souls. Man's intelligence has a
- 75. natural tendency to synthesize, that is, to bring everything into unity. This tendency arises both from the essential oneness of the mind and from the nature of its object. The object of the mind is being or reality in some form or other. That which does exist cannot even be apprehended, and hence cannot be the object of the mind. To understand and to understand nothing is, at the same time, the affirmation and the negation of the understanding. Now, if the object of the intelligence, in order to be known and understood by the said faculty, must represent itself under the form of being or reality, it is under this respect necessarily one. Under whatever form it may exhibit itself, under whatever quality it may be concealed, it must always be reality or being, and, as such, one. But if being, reality, or unity, taken in the abstract, was the sole object of the intelligence, there would be an end to all its movement or life. All science would be at an end, because science is a process, a movement; and movement is not possible where an abstraction is the sole object of the mind. Being and unity, then, abstractly considered, would be the eternal stupor of the mind. This cannot be so, however. Intelligence is action, life, movement. Now, all this implies multiplicity; hence the object of the intelligence must also be multiple. But does not this second condition also destroy the former, which requires that the object of the intelligence should be one? Here reason finds a necessary, though, as we shall see, only an apparent contradiction, both in the logical as well as ontological order. In the logical order, because the intelligence seems to require unity and multiplicity as the conditions without which its action becomes impossible. In the ontological order, or the order of reality, because if the object is not at the same time one and multiple, how can those conditions of the mind be satisfied? The intelligence, then, in order to live, must be able to travel from unity to multiplicity in an ascending or descending process, and to do so, not arbitrarily, but for reasons resting on reality.
- 76. In this lies the life of the intelligence; science is nothing but this synthetical and analytical movement. Let the mind stop at analysis or multiplicity, and you will give it an agglomeration of facts of which it can neither see the reason nor the link which connects them: and hence you place it in unnatural bonds, which, sooner or later, it will break, it matters not whether by a sophistical or a dialectic process. On the other hand, let it stop at unity, and you condemn it to stupor and death. The foregoing ideas will explain the fact how a particular error will either have a very short existence or fall into the universal error of Pantheism. For in this, so far as we can see, lies the reason of the universal dominion of Pantheism. Because it proposes to explain the whole question of human knowledge, it takes it up in all its universality, and the solution which it sets forth has all the appearance of satisfying the most imperative tendency of our mind. To be enabled to explain the numberless multiplicity of realities, no matter how, and, at the same time, to bring them into a compact and perfect whole, strikes to the quick the very essence of man's intelligence and allures it with its charms. If this be not the main reason of the prevalence of Pantheism, we acknowledge we do not understand how such a mighty error could ever take possession of man's mind; we are tempted to say that human understanding was made for falsehood, which is to deny the very notion of intelligence. What Pantheism proposes to do for the mind it also promises to accomplish for the soul. There is, in man's heart or soul, impressed in indelible characters, a tendency after the infinite, a craving almost infinite in its energy, such is the violence with which it impels the soul to seek and yearn after its object. To prove such a tendency were useless. That void, that feeling of satiety and sadness, which overwhelms the soul, even after the enjoyment of the most exquisite pleasure, either sensible or sentimental; the phenomenon of solitaries in all times
- 77. and countries; the very fact of the existence of religion in all ages and among all peoples; the enthusiasm, the recklessness and barbarity which characterize the wars undertaken for religion's sake; the love of the marvellous and the mysterious exhibited by the multitude; that sense of terror and reverence, that feeling of our own nothingness, which steals into our souls in contemplating the wide ocean in a still or stormy night, or in contemplating a wilderness, a mountain, or a mighty chasm, all are evident proofs of that imperious, delicious, violent craving of our souls after the infinite. How otherwise explain all this? Why do we feel a void, a sadness, a kind of pain, after having enjoyed the most stirring delights? Because the infinite is the weight of the soul—the centre of gravity of the heart—because created pleasures, however delightful or exquisite, being finite, can never quiet that craving, can never fill up that chasm placed between us and God. The pretended sages of mankind have never been able to exterminate religion, because they could never root out of the soul of man that tendency. I say pretended sages, because all real geniuses have, with very few exceptions, been religious; for in them that tendency is more keenly and more imperiously felt. This is the second reason of the prevalence of Pantheism. To promise the actual and immediate possession of the infinite, nay, the transformation into the infinite, is to entice the very best of human aspirations, is to touch the deepest and most sensitive chord of the human heart. Both these reasons we have drawn a priori; we might now prove, a posteriori, from history, how every particular error has either fallen into Pantheism or disappeared altogether. But since this would carry us too far, we will exemplify it by one error— Protestantism. The essence of Protestantism lies in emancipating human reason from dependence on the reason of God. It is true that at its dawn
- 78. it was not proclaimed in this naked form, nor is it thus announced at the present time; but its very essence lies in that. For if human reason be made to judge objects which God's reason alone can comprehend, man is literally emancipated from the reason of God. What does this supreme principle of Protestantism mean, that every individual must, by reading the Bible, find for himself what he has to believe? Are the truths written in the Bible intelligible or superintelligible; that is, endowed with evidence immediate or mediate, or are they mysteries? If they be purely intelligible, endowed with evidence mediate or immediate, there is no possible need of the Bible, for, in that case, reason could find them by itself. If they be mysteries, how can reason, unaided by any higher power, find them out? It will not do to say, They are written in the Bible, and reason has merely to apprehend them. Suppose a dispute should arise as to the right meaning of the Bible; who is to decide the dispute? Reason? Then reason must grasp and comprehend mysteries in order to decide the dispute. For none can be judge unless he is qualified thoroughly to understand the matter of the dispute. From this it is evident that to make reason judge of the faith is to make it judge of the mysteries of the infinite, and, therefore, is to emancipate the reason of man from subjection to the reason of God. Hence, Protestantism was rightly called a masked rationalism. It soon threw off the mask. The human mind saw that it can never be emancipated from the reason of God unless it is supposed to be independent, and it could never be supposed independent unless it was supposed equal to the reason of the infinite. The result of all this is necessarily Pantheism. And into Pantheism Protestants soon fell, especially the Germans, who never shrink from any consequence if logically deduced from their premises.
- 79. Welcome to our website – the perfect destination for book lovers and knowledge seekers. We believe that every book holds a new world, offering opportunities for learning, discovery, and personal growth. That’s why we are dedicated to bringing you a diverse collection of books, ranging from classic literature and specialized publications to self-development guides and children's books. More than just a book-buying platform, we strive to be a bridge connecting you with timeless cultural and intellectual values. With an elegant, user-friendly interface and a smart search system, you can quickly find the books that best suit your interests. Additionally, our special promotions and home delivery services help you save time and fully enjoy the joy of reading. Join us on a journey of knowledge exploration, passion nurturing, and personal growth every day! ebookmasss.com