#BigIdeas2022 - Survive & Thrive HyperAutomation

Millions of manufacturing jobs will be transformed with HyperAutomation. Will you need to change jobs? Should you alter career plans? The 4th industrial revolution brings big opportunities for plants and people alike. Everything is changing but the driving force is not what you think. There is a new change agent. It has been going on for some time now. You’ve heard about it but maybe it didn’t quite register. It is everywhere under our noses but we don’t see it. Even following us around. I’m not thinking of computers, not data science, not the cloud, and not machine learning, There is something else much more pervasive that underpins a New Era of Automation and a new era of Artificial Intelligence (AI). I’m talking about sensors. I have lost track of how many sensors there are in a smartphone or a car. Moore’s law is about the number of transistors doubling, driving computational power. The new ‘law’ driving performance is about the number of sensors doubling. And how do we put HyperAutomation to good use for decarbonization and sustainability of plants, and how do you build a career around it in 2022? Here are my personal thoughts:

HyperAutomation

Manufacturing is not as automated as most people outside the manufacturing industries believe. There are a huge number of manual tasks involved in keeping a plant operational. Most of it related to maintenance. Many of these manual tasks are about gathering data and interpreting that data using it to prevent problems like equipment failure or dangerous incidents, and to run the plant more efficiently. Digital Transformation (DX), digitalization, Industrie 4.0, Industry 4.0, the 4th industrial revolution (4IR), or whatever you prefer to call it, including Industrial Internet of Things (IIoT) is all about automating these manual tasks in the plant. This HyperAutomation is enabled by sensors and it allows plants to run more reliably, more efficiently, safely, and produce at lower cost. One of the safety elements is improved social distancing in the plant along with ability to work from home enabled by sensors as much as by the Internet. Changing how work is done is the whole purpose of this additional automation. But since the duties of those working for the plant are changing, people must transform with new skills too.

With new powers comes new responsibilities

When most people hear the term "manufacturing" or “industrial revolution” they think of assembly lines such as for cars and toaster ovens because that’s what is shown on TV and in videos because fast moving items are visually impactful. Very few think of the process industries where the product is a fluid that flows hidden inside a pipe such as petrol & lube, chemicals, medication, clean water, power gas, beverages, but also steam for electric power, ore for steel and other metals, the pulp for paper, even foods. The largest group being chemicals including fertilizers, plastics, synthetic rubber, synthetic fiber, agrochemicals, paint, hygiene products, and other daily necessities. They are not shown much on TV or in videos because there is nothing visibly moving. These products are produced at a well site, on an offshore platform, in a refinery, power station, or plant. Most people would be surprised if they knew just how much manual work is going on in a process plant like an oil refinery. Sure, the production process itself is highly automated, but the supporting functions vital to keep the plant operational are very manual, for instance maintenance, reliability, integrity (corrosion and erosion), sustainability (energy efficiency and emissions), even occupational health and safety, but also some tasks in production and quality. The data collection and data interpretation are relatively easy to automate and this is a huge relief as it is a heavy burden. Automating the action to be taken to complete maintenance tasks will be very much harder so automating the actions will take very much longer, but guidance can be provided for the humans that do it manually.

That is, to stay competitive plants must deploy the automation for data collection and interpretation, as well as to guide human action. Personnel must obtain the skills to install, use, and support these automation tools.

Traditionally also associated with industry in the past was pollution. Therefore we also use automation for decarbonization of new transportation fuels like hydrogen and ammonia, power gas, and electric power. And we use automation for sustainability in production of chemicals, paper, steel, and other metals by improving efficiency and reducing losses. That is, automation including sensors is critical for the energy transition. The ambitious net zero carbon goals are made possible by automation of wind turbines, hydrogen production, ammonia synthesis, transportation, storage, and fuel cells. Coupled with every industrial revolution there has also been some form of energy transition; coal and steam, electricity, oil and gas, and now hydrogen.

Sensors

Sensors pick up on physical properties like temperature and pressure but there are sensors for a wide variety of physical properties and fluid constituents. Sensors include microphones (sound sensors) and cameras (image sensors) as well as GPS (global position sensors), gyroscope (orientation), and magnetometer (compass). Modern sensors make data available in various digital formats converted into a standard format that is understood by software that does any analytics and presentation. The term Cyber-Physical Systems captures well how sensors are the interface between the real world of physical entities and events, and the virtual world of cyber software data processing and control. For instance, analytics software use measurements from vibration, acoustics, and ultrasound thickness sensors to predict equipment failure and corrosion so action can be taken in time to prevent production downtime or hazardous material spills.

Sensors is what has taken camera phones and palmtop computers to the level of smartphones we have today enabling location tracking and information based on location as well as orientation tracking and Augmented Reality (AR). Since we carry these digital devices wherever we go, sensors follow us everywhere. Appliances like cookers, refrigerators, washers, and dryers etc. have many sensors to run safely and automatically.

If you had the data of how many sensors there are in new models of cars, phones, or a refinery and plotted it over the years you would probably see an exponential trend doubling every few years, similar to Moore’s law for transistors.

Sensors in plants are very much more robust and reliable since they must operate at extremely high and low temperatures, high pressures and vacuum, with corrosive and abrasive fluids, and high vibration. Depending on the size, plants have tens of thousands of sensors, mostly for the process, as part of the process automation. As part of HyperAutomation for the 4th industrial revolution, plants are now adding hundreds or thousands more sensors on equipment and to monitor the plant environment. All operational disciplines use more data from more sensors. Corrosion engineering has its sensors and analytics. Process engineering has its sensors and analytics. Electrical engineering has their sensors and analytics. Reliability engineering has its sensors and analytics, and so on. Each discipline must be familiar with its sensors and analytics even though there is a plant instrumentation team to manage the shared sensor system and automation software framework.

Wireless sensors require no power cord and no data network wiring. Sensors do not need cloud storage so cost of operation is low. Electronic sensors have no moving parts so little or no maintenance costs. Many sensors are non-intrusive so can easily be moved to another location should requirements change in the future. Sensors can be installed high on pipe racks, below decking, or on pipeline trestle jetties where walking is not possible – but it is not a problem for permanent sensors. Permanent sensors provide data updates every hour, every minute, or every second as needed straight into analytics, storage, or other software. Updated more frequently and thus more predictive than humans or even robots can. Sensors are the foundation which automation including AI rests upon. Sensors free up personnel from rote, mundane, data collection to instead act on the insights from the data such as attending to equipment when failure is predicted.

New Era of Automation, not IT

The 4th industrial revolution is much more than computers and software. The most interesting fact is that coding (writing software) may not even be required because most of the software is already readymade available off-the-shelf because the automation is mature. But you need the skill to select the right ready-made software for each task. And you also need the skill to pick the right sensors, in some cases actuators, and in the future also robots. And you need the skill to maintain all this software and hardware once it has been deployed. So plant automation is a key job opportunity for the fourth industrial revolution. The approach to digitalization is not to form a new digital group or entity for data analytics because conflicts may arise when those that run the analytics tell the domain experts what to do. Instead, automation skills like sensors and software must be infused into all disciplines and operational departments that take care of their own analytics. This way reorganization is not required which is less disruptive for personnel.

Disrupt the competition, not your personnel

There is a lot of confusing information around digital transformation. You are led to believe you must have a digital twin, but you usually don’t. Or the software must run in the cloud, but it doesn’t have to. Or you must code APIs and apps, but you don’t. Or it has to be machine learning, but it doesn’t. This is why plants need more automation talent to not go down the wrong path with their digital transformation.

Automation software know-how

Just like you would not code your own word processor or spreadsheet software, you use readymade apps for this, there is also no need to code your own analytics and other automation software for your plant. Ready-made software for equipment condition and performance analytics, batch analytics, process analytics, corrosion and erosion analytics, vibration analytics, energy management, tank inventory management, and many more is available off-the-shelf.

But you have to be able to pick the right software. For software, technologies like XML and JSON file formats come to mind, but because they give vendors almost infinite flexibility, they are used differently by each programmer, so further costly custom programing specifically for your plant is required to integrate such software which is what we must avoid. You must know how to pick software with interfaces based on more complete open technology standards like OPC-UA, HART-IP, and Modbus/TCP. This way you can easily bring in data from all kinds of plant systems without the need for proprietary connector software or custom coding to APIs or web services or scripting to messaging protocols. Software based on industrial standards is faster and lower cost to deploy and also to migrate in the future if need be.

Artificial Intelligence (AI) analytics is key to interpreting the data from sensors such as predicting failure or pinpointing overconsumption. There are two main branches of AI: data science and engineered analytics. Data science such as Machine Learning (ML) and Deep Learning (DL) including Artificial Neural Networks (ANN) are the form of AI we hear most about. We hear about DL for face, voice, and image recognition etc. ML is statistics-based analytics we hear about for assessing the probability of credit card fraud and predicting loan default. On the other hand, engineered analytics includes rule-based AI which is based on well-known cause & effect (C&E) and first principles (1P). This includes most use-cases in a plant such as predicting equipment failure, fouling, and product processing upsets. Engineers excel at this sort of analytics problems. ML is statistics-based applied to big data sets from a large population over many years used where C&E and 1P are not established such as for complex human behavior including predicting utilities consumption like power, gas, and water. That is, it is when problems are not tied directly to physics, chemistry, and mechanics – such as in the case of human preferences and other behavior which is when you need the expertise of your data scientists. You must know how to pick the right type of analytics for the use-case that you need to solve. In a plant this is mostly rule-based analytics. Readymade apps with built in domain expertise. In a plant there is little or no need for coding in statistical programming languages like ‘R’. That is, plants want deterministic analytics not probabilistic. Also, you must know how to formulate your analytics problems well. It is easy to get swept away by the romantic notion such as “gathering data in a single place to uncover correlations and new insights” or “you already have all the data, you just need to do something with it” but that is technology for technology’s sake, not solving real plant problems. Start with a workshop to uncover real plant problems, and then solve those with the right analytics and sensors. That is, the rigor of engineering applied to analytics.

The term “digital twin” is so cool eager vendors apply it to all kinds of software to the point where the term has become broad and almost meaningless so now when somebody says “digital twin” you must always ask what they mean. It is typically accompanied by an illustration with Tron lines. I too was guilty of ‘twin washing’ at some point. Strictly speaking a digital twin is software that looks and behaves like its physical counterpart, like twins. That is, a digital twin is a software model used for simulation of the 3D appearance and dynamic behavior of a machine or a whole plant. In the case of a plant, one part of the digital twin model simulates the physical aspects of the plant for structural and piping engineering etc. the other part simulates the process chemistry and thermodynamics used for training and process engineering. The whole idea of a digital twin is to simulate a plant or a machine so you can trial and error on this virtual model in ways that you cannot risk on its physical original. A digital twin model is very costly to build and a big commitment to maintain up to date as the physical plant changes, so you only want to commit to a digital twin if you really need it such as for Operator Training Simulator (OTS) or Virtual Reality (VR) training. Analytics if done right does not require a digital twin because it does not require simulation because analytics non-invasively monitors the plant. You must know if a digital twin is needed and when it is not.

Cloud is another good example of technology which is great for some use-cases but not all. There are many use cases in a plant where cloud computing or storage is not the way to go. This was discovered soon after cloud was introduced into plants. This is how “edge” or on-premises was created to tackle the use-cases in plants where cloud is not a good solution. Therefore do not force fit every application into the cloud. Non-real-time transactional office applications where folks can accept a lag works fine for cloud. We have all experienced how opening a presentation file from the cloud takes longer than from the local hard disk. And if the Internet connection is down for some a while it is not the end of the world for office work. Real-time use-cases such as condition monitoring of various types of rotating machinery require edge computing on servers that are on-premises. You must know which software can be run in the cloud and when to run it on-premises/edge

When you do use cloud there are a three ‘as a service’ models to choose from: Infrastructure as a Service (IaaS) basically meaning the computer hardware where you install software, Platform as a Service (PaaS) meaning a proprietary toolkit for system integrators to code your own apps for you, and Software as a Service (SaaS) meaning readymade software. IaaS and SaaS are pretty low risk as you can switch vendors without much difficulty. PaaS is higher risk because apps built in one PaaS platform do not work in another. You must know how to pick the right cloud subscription model.

Custom coding apps for analytics and visualization etc. is very time consuming and requires programming skills which makes it very expensive to have your own specialized software made as multiple iterations of development and testing is required. The apps may never mature as project runs out of time and budget. This is a familiar scenario for ERP system apps. Companies spend millions deploying and maintaining their ERP systems. Custom coding also creates a costly vendor “lock-in” dependency with the system integrator. It is easy to get swayed by the idea that software coded “just for you” is a better fit than readymade off-the-shelf software, but it is not, and the “ERP way” was many drawbacks. You must know how to spot when apps aren’t readymade but untested. Readymade is key to low-cost software.

Ready-to-work (prêt à travailler) software

That is, in a plant you need a lot of software know-how, but it is plant automation software, very different from the IT software know-how they use in the office, even in the office at the same site as the plant.

Sensor know-how

Data input is the fundamental limitation of computers. And without proper data there is no proper analysis. No algorithm is going to uncover information which is not there in the data, this must be a first principle of data science. You must sense the early symptoms in order to predict a problem. Data comes from sensors or manual observation. Without sensors there is no continuous flow of data updates. Without data there is no analytics and no information. So if you want to make better decisions based on new information you need to install more sensors. Sensors are also used for automatic control loops introduced in the 3rd industrial revolution.

Everybody wants the information but nobody wants to enter the data

You must know how to pick the right sensors for the problems you want to solve because there are many types of flow meters, corrosion sensors, level sensors, vibration, constituent detectors or analyzers, and many more, based on different principles of operation, each ideal for certain use-cases but not others. These are core automation skills but there is also an element of specialized domain knowledge around vibration, corrosion, and analyzers etc.

Actuator know-how

Automatic valves and actuators are key parts of an automatic control loop for fluids. Think of valves as a tap and an actuator as the hand that turns it. Automatic dampers also have actuators. This too was introduced in the previous industrial revolution. Actuators are the brawn of automation. There are many types including pneumatic, electric, and hydraulic as classified by power source. Each one with its unique advantages. You must know how to pick the right type of actuator, valve, and damper etc. for each control loop.

Robot know-how

Once the data has been gathered by sensors, interpreted by analytics, and a decision has been made, ultimately an action must be carried out by a human, or in the future by a robot. When we think about getting things done we think of using our hands. In most process plants the production is highly automated so manual labor is mostly around plant maintenance such as lubrication, filter changes, adjustments, cleaning, and worn parts replacement for equipment around huge plants. Robots are not yet capable of doing this. So here I’m not talking about the fixed location robot arm with grip found along factory assembly lines. Fixed robot arms are not useful in process plants. I’m referring to robots which can move around. But I am also not talking about robots that move around only carrying cameras and other sensors. Taking photos of mechanical gauges is of limited use since each gauge can only be visited so often. Also most inspection is not visible to the naked eye, or camera, but require sensors for vibration or ultrasound etc. Going inside tanks where people shouldn’t go and permanent sensors cannot be installed is useful. But the breakthrough for robots will be when robots can both move around and grip to perform tasks like lubrication, adjustments, cleaning, and replacement. This is when robots will become useful in the plants but we are not there yet.

AR know-how

For now, it takes a human to lubricate, adjust, clean, and replace something. However, automation such as Augmented Reality (AR) help personnel in these tasks. First, AR helps personnel locate the specific piece of equipment to overhaul among thousands of others in the plant. Some of these are small and hard to find, so AR guidance speeds up this process. Second, AR provides step-by-step guidance how to performance task which is helpful since there are hundreds of types of equipment, each with their own quirks.

New Era of Artificial Intelligence (AI)

Sensors also herald a new era in Artificial Intelligence (AI). Normally data analysts spent weeks and months cleaning GB and TB of historical data and looking for statistical correlation in data points by trying out different machine learning algorithms. But strong correlation is not found in this algorithm training, and not with the advance notice needed to become predictive because the right data points are not there in the historical data in the first place since the early warning symptoms were not being measured. Sensors dramatically change the premise of analytics. Sensors make AI so much easier, more predictive, and more robust.

Thanks to sensors picking up on symptoms, AI can move from statistical probability to simple cause & effect relationships which can be pre-programmed as rules in readymade software. Rule-based AI is so much simpler to deploy than algorithms that require historical data, cleansing, and algorithm training.

Thanks to sensors measuring very frequently, they pick up on symptoms very early. This makes analytics using these direct measurements very predictive.

Thanks to sensors picking up on symptoms, AI can move from statistical probability to deterministic cause & effect relationships thereby also reducing false positives and missed events.

The Double-Life of Sensors

As Moore’s law slows down we need to find new avenues to improve product performance. Part of the solution is combating software bloat caused by successive layers of abstraction by spending more time producing efficient code requiring less CPU power, memory, and other resources. The other big part of it is to use sensors to provide new analytics and other capabilities not possible before. A third key enabler is to move software into specialized hardware much like what has been done for graphics processors and floating-point arithmetic before that.

The world of instrumentation with sensors is changing the world of computing, not only the other way around. And this trend will continue. Self-driving vehicles need even more sensors to be safer for those around us and to drive us all the way. But cyber-physical systems are physical so therefore cost doesn’t follow Moore’s law. Don’t expect industrial sensors to become dirt cheap. The cost of miniature sensing elements and silicon chips is reducing but the complete product sensor assembly to handle the high pressures and other extremes does have a significant cost. Key is to buy reliable sensors that can operate many years without any attention. Reliable sensors help a good ROI long term.

The number of sensors double every few years. A modern plant has many more sensors than a plant built ten years ago. So how long does it take to double the number of sensors in a given product or plant? I don’t know, but it is more than 2 years.

Moving Robots

A robot moving around with a camera and other sensors is a simple robot application but doesn’t live up to the expectation of a robot doing manual work with their arm grips and cannot compare to the update frequency and repeatability of permanently installed sensors. If you want frequent data updates it is far better to install permanent sensors and cameras because robots are not in the same place all the time so intermittent and sudden events are missed. The exception is where you can’t install permanent sensors, such as cameras and sensors for inspection on the inside of a tank and other vessels. This is a good application for moving robots. So robots to replace human eyes, ears, and nose is not the way.

What we really need robots for is to take the place of human hands. When robots can perform simple maintenance tasks like lubrication, change oil filters, replenish fluids, and cleaning etc. using their arms and grips is when they become really useful. This will enable more autonomous plants ideal for offshore platforms, mines, oil & gas fields, and other remote sites. The more dangerous or monotonous the job, the greater the need for robots.

Machines like pumps will also have to be designed in such a way that they become robot friendly. These robots will probably not be using tools designed for human hands. New robot friendly tools will be developed. Robots may change their grip to a grease gun, oil filter wrench, or some other tool snapped directly onto its arm. New standards will have to be established for robots to be able to use tools on machines. Tasks may not be carried out the way humans do it,

Career in Automation

As part of the 4th industrial revolution plants need to deploy more automation to remain competitive. This is particularly important for existing plants to compete with newer plants being planned right now. A Digital Operational Infrastructure (DOI) for HyperAutomation to automate many of the numerous manual tasks in the plant. Therefore plants need to hire more people skilled in automation. Not just software, but also sensors, actuators, and industrial networking. The industry needs automation engineers and technicians to design, install, and maintain ever larger automation systems. As more and more tasks get automated, the line of automation itself looks like a good career.

When many jobs are transformed by automation, automation is the best job

The 4th industrial revolution means a new era in automation. A 4th era of automation. Well, that’s my personal opinion. If you are interested in the detail implementation of digital transformation in the process industries click “Follow” by my photo to not miss future updates. Click “Like” if you found this useful to you and to make sure you keep receiving updates in your feed and “Share” it with others if you think it would be useful to them. Save the link in case you need to refer in the future.