All News & Insights

ML Journal

Dialogue: Yokogawa’s Autonomous Ambition

Last year’s 35-day field test of autonomous operations marked a new phase in the development of industrial AI control, believes Yokogawa Digital CEO, Hiroaki Kanokogi.  

“I believe, as human beings, we can solve major problems with new digital technologies like AI. And that’s a great opportunity for everyone on the planet.” 

Dr. Hiroaki Kanokogi, President and CEO, Yokogawa Digital Corporation


In March last year, Tokyo-based Yokogawa Digital announced the successful completion of a 35-day landmark trial using a new AI learning system to autonomously control a complex chemical plant process in collaboration with Japanese chemicals and materials company, ENEOS Materials Corporation (formerly JSR Corporation).

In our latest interview with an industry thought-leader, Yokogawa Digital’s President and CEO, Dr. Hiroaki Kanokogi, talks to the MLC’s Executive Editor Paul Tate about what the 35-day autonomy trial achieved, the future opportunities and challenges of autonomous adoption across manufacturing, the impact on human roles in future AI-empowered environments, and how new AI technologies may help to solve some of the world’s greatest problems.

Q: What excites you most about your role at Yokogawa Digital?
Digital transformation excites me, and that covers many things. To begin that process requires the collection of data from multiple fields. My biggest interest is how to then utilize that data, not only passively, but also actively. I see passive AI usage to be things like the simple analysis of data. But I would like to do more, to actively use that data to control and manage plants, or supply chains, or anything. That’s the kind of activity that excites me very much.

Q: What was the autonomous factory test project with ENEOS designed to achieve?
A: The ENEOS autonomous factory field test aimed to address two key things. The first was to save waste energy and reduce the total amount of energy used in a complex chemical manufacturing process while ensuring product quality and providing both economic and environmental benefits for the company. The second was to free up human expertise for higher value tasks.

The specific focus of the project was the autonomous control of values and temperature controls in a distillation column in the ENEOS plant. Typically, these are driven by PID (Proportional Integral Derivative) controls to help manage continuous variations. That’s OK, but controlling and reusing the waste heat is the difficult part. It’s a very interesting and good idea to do so, but waste heat can be very unstable and not appropriate for some specific manufacturing processes as it can lead to quality problems, like being out of range of key production specifications. ENEOS had tried various control management solutions but not had good results. So, an operator needed to be assigned to the task of constantly monitoring and manually adjusting the distillation column.

In the ENEOS test, we were able to use our AI reinforcement learning system to quickly learn the optimum conditions in the distillation columns and solve the problem, saving energy using waste heat and producing a good quality product. The test ran continuously for 35 days with the AI system successfully controlling the valves and temperature controls instead of the human expert and without any human intervention. Experts like that are a precious and expensive resource today and being assigned to a manual operation is a waste of their potential skill, especially in times of a labor shortage. But by using the autonomous AI approach, without any human intervention, it freed the expert to focus on higher value tasks, so it helped to solve two of the company’s problems.

“When you apply AI in an oil plant or a chemical plant, safety must be the primary concern. So, in this test we established a specific process to assure safe AI control in a real-world situation where there are some potentially dangerous hazards.”


Q: What key technologies did you use in the test project?
A: One of the key technologies was our AI reinforcement learning system, called FKDPP (Factorial Kernel Dynamic Policy Programming), which was specifically designed for plant control. This was supported by multiple sensors which provided the key process data to drive the AI system.

The second aspect of the trial, which is just as important, was how to ensure the safety of the autonomous system. When you apply AI in an oil plant or a chemical plant, safety must be the primary concern. So in this test, we established a specific process to assure safe AI control in a real-world situation where there are some potentially dangerous hazards. Establishing that safety process, and how to confirm the safe behavior of the AI, was also a very important deliverable from the ENEOS field test.

Q: What makes the FKDPP AI algorithm different from other approaches to autonomous operations?
A: There are many existing AI control technologies, but they are predominantly based on PID theory, traditional control theory. That means the AI’s capabilities are limited within the PID theory. On the other hand, FKDPP was developed to control valves or other equipment directly without PID or traditional control theories. So FKDPP has the potential capability to achieve things that go beyond the traditional limits. Such a dream is very attractive to me.

Q: How would you measure the trial’s success?
A: In the early discussion phase of the experiment, the quality of the final product was the most important metric, plus the energy savings that could be achieved. At first, we expected some human intervention might be necessary, but as it turned out, no direct human action was required at all. The only human involvement was in helping to design the safety layers before the AI implementation, and generally overseeing the progress of the experiment. So, that was an additional successful result. Overall, we achieved the required quality control, as many of the energy savings as possible, and without any human intervention. And the trial ran continually for 35 days, which is a really good duration – not several hours, not three days, but 35 days. It was a great success, and we believe this is a world first for autonomous AI control.

Q: What lessons did you learn about implementing autonomous operations from the trial?
A: It was the first time we had attempted such a large-scale commercial application. We’d previously experimented with AI autonomy on smaller systems, like controlling the air conditioning system in a Yokogawa factory, but we needed to show that it could be used at a larger scale, like in the ENEOS chemical plant. The test proved that AI control could be used successfully for real commercial use at scale. That was a big lesson for us. The other key lesson was how to assure safety with autonomous systems. We worked closely with experts at ENEOS to develop the safety process and that was a great experience and another great lesson for us. It means that our AI control approach has moved to a phase of safe, practical use, and is no longer just in the test phase.

“The skillset required for humans in plants or factories will change after autonomous systems become more widespread, so they can then focus on higher value tasks where the AI helps empower them to be more creative and innovative.”


Q: Where do you go from here? What are your plans for applying this AI control technique elsewhere?
A: The ENEOS test was just the first step. The AI reinforcement learning approach has a very general capability that can be applied to many different kinds of tasks, so we are now discussing further areas where we can apply the same technology.  There are many areas where there could be positive trade-offs in real plants, so we are now looking to apply the FKDPP system in a variety of other fields. We want to prioritize the big issues first, and to apply our AI to the most important and most essential parts of plant where some trade off exists and there is room to improve. That could be a in a plant, or in supply chain management, or in energy usage optimization, or many other areas. We’re talking to a number of companies, but these are still confidential conversations right now.

Q: How do you see the adoption of industrial autonomous operations developing over the next few years to 2030?
A: Yokogawa conducted two surveys about autonomous operations over the past couple of years involving more than 500 C-level people. We found that many companies are very interested in the potential of autonomous approaches, and some are already exploring autonomous aspects in their industry sectors. Their motivation is often around traditional drivers like cost reduction or efficiency improvement, but they are also focusing on new issues like meeting sustainability goals, or coping with the labor shortage, or creating more resilient supply chains, and they are looking for new ways to solve these problems.

What may hinder that autonomous development, in my opinion, are perhaps some concerns and fears about adopting such a new and mostly unknown technology. To mitigate those situations, I think we need a strong vision or strategy for the future of industrial AI autonomy, and maybe a trustworthy industrial standard or regulatory environment that will help those companies move forward. And of course, technology to solve cybersecurity safety concerns too.

Q: Where do you expect the adoption of autonomous approaches to be most prevalent as we approach 2030?
A: Our research indicates that all industries and sectors have some degree of interest in autonomous operations. But there are several areas where traditional control approaches have some difficulties. For example, oil refining and other major process industries require high levels safety and that can be a big challenge when adopting advanced technologies. But we have now proved the AI reinforcement learning can be applied safely to such process tasks, so we expect other companies may begin adoption very quickly. Or take the bio industry, for example. The bio industry has some critical challenges in controlling bioreactors and other key processes where AI control, like reinforcement learning, could be very promising. And there are also areas of discrete manufacturing too, where AI control can help support traditional factory automation systems in areas where very precise controls are needed.

Q: Will human roles change over the next few years as a result of increasing autonomous adoption?
A: It’s a very good question. I think initially, people will work closely together with AI and autonomous systems to monitor and supervise their behavior, so if something goes wrong, there is always someone there who can intervene. In that first phase, it will be important that workers have the right skills to be able to work collaboratively with such autonomous systems, and they will need to learn how to do that. But after that, as autonomous adoption becomes more pervasive, I think human workers will be able to focus more on entirely different high value optimization tasks which AI control can’t do, like inventing some new control method or a new chemical formula. So, maybe the skillset required for humans in plants or factories will change after autonomous systems become more widespread, so they can then focus on higher value tasks where the AI helps empower them to be more creative and innovative.

“I think we need a strong vision or strategy for the future of industrial AI autonomy, and maybe a trustworthy industrial standard or regulatory environment that will help those companies move forward.”


Q: Looking ahead, what would you highlight as the greatest challenges and opportunities for manufacturing industry for the rest of the decade?
A: Digitally empowered AI systems will be one of the most promising technologies to move us forward to the next generation of manufacturing. There are lots of opportunities here. It’s almost impossible to imagine how many major opportunities there will be in the future based on these new digital approaches. But there are also challenges. The adoption of new technologies is not always easy, so a strong leadership vision and clear strategy will be important, supported by evolving industry standards. Cybersecurity and safety concerns will also be a challenge in a more autonomous world. We already have a number of problems today, from achieving sustainable development goals to the labor shortage. But I believe, as human beings, we can solve these problems with new digital technologies like AI. And that’s a great opportunity for everyone on the planet.

Q: What key leadership skills, attributes, and roles do you feel that senior industry executives now need to lead successfully in an increasingly autonomous world?
A: In a digital world, strong leaders will need to understand not only the business objectives and challenges but the potential of the technology too. Technological advances are happening very fast today, so leaders will need to know what technologies are available and what they can do. More than that. Digital transformation (DX) will connect people in many different divisions of a manufacturing organization. That means leaders also need strong people skills. I believe human empathy will become more and more important because leaders with technology skills will need to convince and engage many people in these different divisions. In other words, I think the leaders of the future will need to become tech savvy humanists to succeed in the digital world.

Q: Finally, if you had to focus on one thing as a watchword or catchphrase for the future of manufacturing, what would that be?
A: It’s about moving from industrial automation to industrial autonomy. We think this vision can take manufacturing much farther in the future.  M

Fact File: Yokogawa Electric Corporation
HQ: Musashino, Tokyo, Japan
Industry Sector: Electrical equipment, industrial automation, software
Sales: $2.94 billion (¥389.9 billion – FY 2021)
Net Income: $163 million (¥21.3 billion 2021)
Employees: 17,258 Employees
Presence: 61 Countries
Production Sites: 18 manufacturing sites worldwide

EXECUTIVE PROFILE: Dr. Hiroaki Kanokogi
Title: President and Chief Executive Officer, Yokogawa Digital Corporation
Nationality: Japanese
Education: BSc degree in quantum electronics / physics, University of Tokyo; Masters’ degree in quantum electronics, University of Tokyo; PhD in quantum electronics, University of Tokyo
Languages: Japanese, English
Previous Roles Include:
– General Manager, Yokogawa Product Headquarters, Yokogawa Electric Corporation
– Head of Information Technology Center, Yokogawa Product Headquarters, Yokogawa Electric Corporation
– Development Manager, Microsoft Product Development Limited
Other Industry Roles/Awards/Board Memberships:
– Co-inventor, FKDPP Algorithm, Plant Control AI


About the author:
Paul Tate
is Co-founding Executive Editor and Senior Content Director of the NAM’s Manufacturing Leadership Council.

View More