At ARC Asia’s annual online industry forum titled Accelerating industrial digital transformation and sustainability from July 12 to 14, 2022, Yokogawa participated as a Gold Sponsor. This Forum saw the registration of over 1,600 delegates for both language streams – Japanese and English. During the session on AI and Machine Learning, Dr. Darius Ngo, Senior Vice President, Head of Digital Business Solution, Yokogawa Engineering Asia, shared an implementation of Artificial Intelligence (AI ) for real plan-based autonomous control using factorial reinforcement learning Kernel Dynamic Policy Programming (FKDPP). The FKDPP is a disruptive innovation that allows another dimension of control. This AI technology can be applied in energy, materials, pharmaceuticals and many other industries.
At the end of this session, Dr. Ngo joined the other speakers for the round table. This blog captures the key points of Dr Ngo’s presentation and his insights during the panel discussion. The entire session can be viewed on YouTube.
Process industry challenges
Process industries (oil refineries, petrochemicals, steel, water, etc.) require complex control of temperature, pressure and flow due to chemical reactions and other factors. Dr. Ngo explained this complex control scenario by giving the example of an oil refinery, from refining to processing and final assembly. The 4Ms that impact quality and production are:
Manufacturers are now turning to exploring advanced technologies, such as AI and ML, to empower operations. Since the launch of Industry 4.0, the scope of AI has expanded. Dr. Ngo gave a schematic representation of AI/ML in process control via a typically linear map over the control layers of the hierarchy. At level 1 (sensor level) itself, there is already an integrated AI; Level 2 deals more with control layers – IoT network, DCS, etc. At this point, the AI can be integrated into a reinforcement learning map (FKDPP) algorithm on the controllers. Levels 3 and above take advantage of applications such as visualization analytics – AI embedded in image analysis on field monitoring devices, robots, etc. In addition to this, there are a variety of applications and services for specific solutions; it is Yokogawa’s AI Platform Studio – Xperience and the AI platform responsible for creating AI algorithms for specific applications.
Kernel Factorial Dynamic Policy Programming (FKDPP)
The FKDPP algorithm was jointly developed by Yokogawa and the Nara Institute of Science and Technology (NAIST) in 2018. It was recognized at an IEEE International Conference on Automation Science and Engineering as the first Reinforcement learning based AI to the world that can be used in plant management. FKDPP was run on a simulator of a vinyl acetate manufacturing plant and operated valves to maximize product volume while ensuring quality and safety standards were met. Stable and optimized valve operation was obtained during 30 learning trials.
FKDPP uses a factorial policy model and a smooth policy update based on the factorial kernel by regularization with the Kullback-Leibler divergence between current and updated policies. Compared with the previous methods which cannot directly process a large number of actions, the method proposed by Yokogawa exploits the same number of training samples and achieves a better strategy for controlling the yield, quality and plant stability in vinyl acetate monomer (VAM).
Key features of FKDPP
- Can be applied to most control types
- Increases productivity
- Simple
- Explainable operation
- Security levels identical to those of conventional systems
Case study
In 2019, Yokogawa Engineering Asia successfully completed an experiment using a control training device. A three-tank level control system was set up via a laptop PC. Although the system can be controlled with conventional PID technology, FKDPP has been shown to reduce settling time by 50-70%, while preventing overshoot and maintaining reservoir water level. This was demonstrated by a video showing the differences between the 1st, 20th, 25th and 30th iterations of the reinforcement learning-based AI (FKDPP algorithm). The three basic steps from FKDPP model generation to actual control are: target definition, AI control model construction, and AI autonomous control.
Over the past three years, the effectiveness of the FKDPP algorithm has been tested and projects have been initiated with ENEOS Materials Corporation and NTT DOCOMO. Next, Dr. Ngo explained how FKDPP balances quality and energy savings. The media opined that FKDPP “can greatly contribute to empowering production, maximizing return on investment and environmental sustainability.”
Future prospects
In this context, Dr. Ngo spoke about Yokogawa’s vision from Industrial Automation to Industrial Autonomy (IA2IA). A survey of 534 decision makers in 390 manufacturing plants reveals that 42% of them believe that over the next three years, the application of AI to the optimization of factory processes will have a significant impact on industrial autonomy. The envisioned application of 5G, cloud and AI for industrial autonomy will enable optimal control anytime and from anywhere.
Perspectives
Dr. Ngo’s responses during the roundtable are summarized below.
Is the design suitable for the different interfaces?
Currently, the implementation is via the OPC interface; but in the long term, the company will integrate full visualization. Multi-vendor data will be pulled from a data lake and placed into the system.
Why was the pilot project on the chemical plant limited to 35 days?
For this chemical plant, there was routine maintenance on the 36th day, that’s why it was shut down on the 35th day. After that, when the factory was restarted, it was under AI control.
What is the time taken by FKDPP to learn the actions of the operator before putting in autonomous control?
Security is always the key linchpin of Yokogawa’s implementations. FKDPP learning came from factory history, including operator action simulation, to ensure safe autonomous control. The learning time depends on the complexity of the command. In this particular plant, the time required was short due to the adaptive processes implemented in FKDPP.
In the future, do you see AI replacing traditional PID?
We’re more interested in addressing what PID and APC can’t do and filling those gaps and improvising using AI on that. However, in the future it may happen. Even academia is trying to push the ideal of an all-encompassing AI, but intelligence should be based on the fundamentals of PID-like methodology. This is a time of transition – even academia will take time to adapt to AI as a control strategy rather than fundamental engineering.
#Artificial #intelligence #autonomous #control #explained #ARC #Industry #Forum #Asia