Application of thermal error in machine tools based on Dynamic Bayesian Network

International Journal of Research in Engineering and Science (IJRES)
ISSN (Online): 2320-9364, ISSN (Print): 2320-9356
www.ijres.org Volume 3 Issue 6 ǁ June 2015 ǁ PP.22-27
www.ijres.org 22 | Page
Application of thermal error in machine tools based on Dynamic
Bayesian Network
Xu Yang，Mao Jian
College of Mechanical Engineering, Shanghai University of Engineering Science, 333 Long Teng Road,
Shanghai 201620, China
Correspondence should be addressed to Jian Mao; jmao@sues.edu.cn
Abstract: In recent years, the growing interest toward complex manufacturing on machine tools and the
machining accuracy have solicited new efforts in the area of modeling and analysis of machine tools machining
errors. Therefore, the mathematical model study on the relationship between temperature field and thermal error
is the core content, which can improve the precision of parts processing and the thermal stability, also predict
and compensate machining errors of CNC machine tools. It is critical to obtain the thermal errors of a precision
machine tools in real-time. In this paper, based on Dynamic Bayesian Network (DBN), a pioneering modeling
method applied in thermal error research is presented. The dependence of thermal error and temperature field is
clearly described by graph theory, and the fuzzy classification method is proposed to reduce the computational
complexity, then forming a new method for thermal error modeling of machine tools.
Keywords: machine tools; thermal error; DBN; fuzzy classification; modeling;
I. Introduction
Artificial intelligence techniques have been used for many years now in the field of thermal error modeling
in machine tools. The thermal model forms one of the most critical elements in machine tools. In the prediction
of thermal error, most research has focused on the principle of regression [1,2]
. Chen [3]
used a three-layer
Artificial Neural Network (ANN) with a supervised back-propagation training algorithm and the ‘sigmoid’
activation function to map the calibrated thermal errors to the temperature measurements. Zhejiang University in
2002 with the improved BP neural network on three-dimensional noncontact measurement system is analyzed in
thermal error modeling [4]
. Hong Yang and Jun Ni [5]
proposed a fuzzy model adaptation method is used to update
in the thermal error model under different processing conditions. The traditional thermal error model cannot
handle abnormal situations, and the structure and the choice of the type too dependent on experience. Research
on the application of the Bayesian Network in thermal error modeling of CNC machine tools on the thermal
error has opened up a new way, in order to better solve the complexity and defect of thermal error modeling [6]
.
R. Ramesh[7]
applied Hybrid Bayesian network to solve thermal error measurement and modeling problems in
machine tools. A brief explanation of the theory of Bayesian Network is presented as following:
The Bayesian Network (BN) is a graphical model consisting of nodes representing causes and effects in
real-world situations and a set of edges each of which connects two nodes, it has been used in medical
diagnostic system, software testing, web intelligent navigation, power system fault diagnosis, etc. A Bayesian
network for a set of variables  1 2
, , N
x x x x  , the joint probability distribution for x is given by
1 2
1
( , , ) ( ( ))i i
N
N
i
p x x x p x pa x

 
(1)
Where i
x denotes the variable and ( )i
pa x denotes the parents of node i
x . From the chain rule of
probability,
1 2 1 1
1
( , , ) ( , , )i
N
N i
i
p x x x p x x x 

  
(2)
BN is an effective tool for complex systems uncertainty reasoning and data analysis, but the thermal error
of machine tool research with differences between the working conditions are different, thus requiring thermal

Application of thermal error in machine tools based on Dynamic Bayesian Network
error model must have a strong ability to learn, and according to the current working update the status of the
model results. BN in dealing with real-time tracking data still has some limitations, not good with the
experimental data reflect updates to the model. This paper seeks to address the issue of developing a thermal
error model that account for a variety of operating conditions while yet generating accurate prediction. So it will
propose fuzzy classification and Dynamic Bayesian Network to build a thermal machine error model, a better
solution of dynamic thermal error modeling problems.
II. Fuzzy Dynamic Bayesian Network
Dynamic Bayesian Network (DBN) is a type of BN that can model time-series data to capture the fact that
time flows forward, which coupled with the time constraints on the properties of the original network structure.
It has the advantage both of model-based and data-based methods. And with the introduction of time factor, the
data on the state of the formation of different time, reflects the development change rule represented by the
variable, and its intuitive, high precision and adaptive for thermal error modeling of machine tools.
The actual processing for precision CNC machine tools, the thermal error generated by many factors and
there are many uncertain factors, also the relations among these factors are perplexing, so the requirements
of thermal error prediction model should not only be a strong inclusive of many complex factors, but also
the structure of flexible function. In order to account for the particularity of thermal error, we present the DBN
thermal error model herein is therefore very essential.
2.1 Dynamic Model
Although the pattern of heat transfer in machine tools is very complex, these are plausible cause and effect
relationships that might exist between the various elements of the machine when subjected to varying operating
conditions. Therefore, in order to carry out the research of complex system and model should be simplified [8]
:
1. The assumption of conditional probability in a finite time change process for all t is smooth. 2. To solve
the problems of thermal error in this study, assuming an environment with Markov nature, that every state heat
distortion temperature and sampling points are related to the time of sampling points on temperature and thermal
deformation states. Bayesian network structure inference process, making inferences and decisions based on
prior probabilities. 3. Assuming that the conditional probability of adjacent time process is steady, which
1( )t tp x x has nothing to do with the time t , you can easily get the transition probability 1( )t tp x x in
different time.
Based on the above assumptions, the establishment of DBN should be considered  1,B B : the priori net
1B ,the transfer net B ,the limited time in practical application stage 1,2，…T，then we can get:
     11 2 1 1
1
, ,
T
T B B tt
t
p x x x p x p x x 

 
(3)
If we use  1t tp x x  to represent the probability of the current state occurs when any previous time state is
variable. i
tx represents the value of the variable i in time t ，  i
tpa x denotes the parents of node i
tx . So,
joint probability distribution of DBN can be calculated for any node as the following shown:
       1
1:
1: 1 1
1 2 1
N T N
N i i i i
T B B t t
i t i
p x p x pa x p x pa x
  
  
(4)
The research of thermal error on precision machine tools showed us the parameters of sampling point
temperature and thermal deformation data had discrete features. In the experiment, the decision variable is
always discrete, and certain temperature observation is continuous, to solve such problems in pure discrete
Dynamic Bayesian Network have some difficulties. In this regard, this paper choose the calculation method of
discrete fuzzy DBN that fuzzy classification techniques combined with the discrete DBN [9]
to solve these
problems. It will not only make full use of the advantages of discrete DBN which has relatively faster inference
speed compared with continuous DBN, but also can achieve qualitative reasoning under continuous observation
without loss of information provided.
2.2 Fuzzy Classification
In the fuzzy theory, an element x is the degree of  A X belongs to the set A, this degree means

membership. The element x must also exit in certain membership with other sets, when 0 ( ) 1A A  . Fuzzy
set A is defined in x and its affiliated with A and  A X ,as the following:
 
 
( , )A i i
A
A i
A x x
N
x
N




(5)
First, the fuzzy classification process is divided the sample space into several subsets, which is a subset of
fuzzy sets. Fuzzy sets is the corresponding discrete DBN node status. Then defines membership function
according to actual condition, that membership of fuzzy classification is the characteristics of the response
variable, not a specific numerical size[10]
. Discrete BN with N hidden points and M observation points, along
with the time development can get the discrete DBN with T time slices, if the observed value is only one
combination of state, so this observation distribution of hidden variables are:
 
     
     
 
1 2 1 2
1 1 1
1 2 1 2 1 2 1 2
1 1 1 1 1 1
, ,
, ,, , , , , ,
, , , , , , , , , , , ,
1, , ; 1, , ; 1, ,
N N
T T T
N N M M
T T T T T T
j j k k
t t t t
t j t k
j j k k
t t t t
t j t kx x x x x x
p x x x x x x y y y y y y
p y pa y p x pa x
p y pa y p x pa x
t T j M k N

 
 
 
  
 
  
  
     
  
(6)
In this formula,
k
tx represents a value of
k
tX ，value of
j
ty for observation variable
j
tY , ( )j
tpa y denotes
the parents of note
j
ty . As according to get fuzzy DBN inference formula is:
 
     
     
1 2
1 1
1 2 1 2
1 1 1
1 2 1 2 1 2 1 2
1 1 1 1 1 1
, ,
, ,
, ,, , , , , ,
, , , , , , , , , , , ,
( )
1, , ; 1, , ;
M
T
N N
T T T
N N M M
T T T T T T
j j k k
t t t t
t j t k j j
t t
tjy y y j j k k
t t t t
t j t kx x x x x x
p x x x x x x Y Y Y Y Y Y
p y pa y p x pa x
p Y y
p y pa y p x pa x
t T j M k
 
 
     
   
  
 
 
 
  

  
     
  1, , N 
(7)
Where ( )j j
t tp Y y is the value of
j
tY membership in continuous observation that belongs to each
states.
III. Development of DBN Model on Thermal Error
Aiming to improve the effectiveness and accuracy of abnormal situation management in complex process
system, thermal error should be studied and modeled in a scientific and systematic way. This paper proposed a
DBN framework for this purpose. The overall workflow is shown in Figure 1.

Optimization of temperature sensor location
Stage Ⅰ:Temperature points
development
Analysis of thermal dynamic process
Determine the location of temperature sensor
Develop temperature points
Stage Ⅱ:DBN development
Node determination
Develop structure
Determinate parameters
Parameters learning with condition monitoring data
Stage Ⅲ:DBN applied in thermal error
Figure 1.
DBN framework of thermal error research
3.1 Temperature Points Development
A number of studies on domestic and international machine tools show that: one of the primary factors that
influence machine tool accuracy is the rise in temperature of the critical elements of the machine. The data
available from the machine on a real-time basis is the temperature values as measured by the thermal sensors
mounted at critical locations on the machine. Comprehensive expertise this paper establishes a model of 14
nodes [11]
, respectively: X,Y and Z axle nuts and rail temperatures, motor temperature, main shaft bearing
temperature, after the bearing temperature, the temperature of the machine bed, spindle thermal error, then
obtained Dynamic Bayesian Network model for thermal error of precision CNC shown in Figure 2.

Thermal
error
d(L)
The X axis nut
T1
The Y axis nut
T2
The Z axis nut
T4
Main front
axle bearing 3
T12
Main front
axle bearing 2
T10
Main front
axle bearing 1
T8
The motor
T3
Main front
axle back
bearing 1
T7
Main front
axle back
bearing 2
T13
The X axis
guide
T5
The Z axis
guide T9
The Y axis
guide T6
environmental
temperature
T11
Figure 2. DBN model structure for thermal error of precision CNC
3.2 DBN Applied in Thermal Error
The above model and the discrete DBN inference algorithms, constitutes the thermal error of DBN. DBN in
addition to the network structure, also need to define the parameters. For discrete DBN, the conditional
probability is an expert knowledge which reacts experts views of the causal relationship among
the connected nodes in the network. State transition probability between two time slices is random probability
that changes over time[12]
.
The specific calculation model can be expressed according to the conditional probability between expertise
and nodes. The transition probabilities of DBN can be acquired by the priori knowledge of experts. Using fuzzy
classification to divide the experimental measurements of state parameter domain, in this paper respectively
divided into five states {1, 2, 3, 4, 5}.
According to the formula (7), to calculate probabilities:   131 2
, ,p d L T T T which means we can
obtain conditional probability at the corresponding ( )d L of different states when ( 1,2, ,13)i
T i   take
any state. Through the judgment of values 1 2 3 4 5, , , ,p p p p p to decide the state of thermal error area, then
finally predict thermal error value.
IV. Conclusions
(1) DBN structure is different from the other fitting modeling theory, it comes from the probability of data,
and combined with the specific language of graph theory clearly express the dependent relationships between
the various factors affecting the thermal error. And applied the fuzzy classification and membership the concept
of the measurement data and other scientific probability distribution, reducing the computational complexity,
concluding the probability distributions , making the numerical prediction and modeling has the high-precision
thermal deformation characteristics;
(2) DBN model is widely used in the present study in the field of threat estimation and target assignment,
etc. Applied it to the machine tools thermal error modeling is the current research in the field of thermal error for

another attempt and application, which also provides a new effective way to study CNC machine tool thermal
error.
(3) Thermal error model based on DBN takes full use of the prior knowledge and sample data, and can
dynamically predict thermal error according to the time of machine tools, and with the increase of sample to
obtain data updated. It can reflect the change of working condition of machine tools in the process of reasoning,
to make it better meet the needs of real-time compensation.
In summary, this paper use DBN structure for thermal error of machine tools to make a series of useful
analysis and exploration, get the corresponding conclusions. Based on DBN structure in the study of thermal
error is novelty, this paper is for further study on the experimental data and model combination.
Acknowledgements
We would like to acknowledge the support of the National Natural Science Foundation of the People’s
Republic of China (No. 51175322).
References
[1] R. Ramesh, M.A. Mannan, A.N. Poo. Error compensation in machine tools—a review. Part I: Geometric, cutting-force induced
and fixture dependent errors. International Journal of Machine Tools and Manufacture 40 (2000) 1235–1256.
[2] R. Ramesh, M.A. Mannan, A.N. Poo. Error compensation in machine tools—a review: Part II: Thermal errors. International
Journal of Machine Tools and Manufacture 40 (2000) 1257–1284.
[3] J.S. Chen. Fast calibration and modeling of thermally induced machine tool errors in real machining. International Journal of
Machine Tools and Manufacture 37 (2) (1997) 159–169.
[4] Fu Longzhu, Chen Yaliang, Di Ruikun .BP neural networks compensation dimensional surface non-contact measurement system
thermal deformation error research [J]. Electrical and Mechanical Engineering, 2002,19 (4): 55-58.
[5] Hong Yang, Jun Ni. A daptive model estimation of machine tool thermal errors based on recursive dynamic modeling strategy[J].
International Journal of Machine Tools&Manufacture,45(2005):1-11.
[6] Bai Fuyou. Thermal error of CNC machine tools based on Bayesian Network modeling research [D]. Master degree thesis of
Zhejiang University, 2008.
[7] R. Ramesh, M.A. Mannan, A.N. Poo, S.S. Keerthi. Thermal error measurement and modeling in machine tools. Part II. Hybrid
Bayesian Network—support vector machine model. International Journal of Machine Tools & Manufacture 43 (2003) 405–419.
[8] Xiao Qinku, Gao Song. Bayesian Network in the application of intelligent information processing [M]. Beijing: National Defense
industry press, 2012.
[9] Shi Jianguo, Gao Xiaoguang. Dynamic Bayesian Network and its application in autonomous intelligence operations [M]. Beijing:
The Publishing House of Ordnance Industry, 2008.12.
[10] Chen Haiyang, Nie Hongying, Pan Jinbo. Traffic light autonomous intelligence decision based on Dynamic Bayesian Network
[J]. Journal of Xi'an Polytechnic University, 2014,28 (4): 474-479.
[11] Yang Jianguo. Integrated CNC machine tools error compensation technology and its application [D]. Shanghai Jiao Tong
University doctoral thesis, 1998.
[12] Wu Tianyu, Zhang An, Li Liang. Estimation of the Air Combat Threat based on Fuzzy Dynamic Bayesian Network [J]. Fire
Control and Command Control, 2009,34(10):56-59.

Application of thermal error in machine tools based on Dynamic Bayesian Network

More Related Content

What's hot (19)

Viewers also liked (18)

Similar to Application of thermal error in machine tools based on Dynamic Bayesian Network (20)

More from IJRES Journal (20)

Recently uploaded (20)

Application of thermal error in machine tools based on Dynamic Bayesian Network