Artificial intelligence for self-control

In the annual online ARC Asia Industry Forum titled Accelerating industrial digital transformation and sustainability On July 12-14, 2022, Yokogawa participated as a gold sponsor. This forum saw more than 1,600 delegates register for the two-language track – Japanese and English. In the session on Artificial Intelligence and Machine Learning, Dr. Darius Ngo, Senior Vice President, Head of Digital Enterprise Solutions, Yokogawa Engineering Asia, Inc., Implementation of Artificial Intelligence (AI) for autonomous control based on an actual plan using reinforcement-factor learning, shared dynamic policy programming for the kernel ( FKDPP). FKDPP is a disruptive innovation that allows for different dimensions 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 panelists. This blog covers the main points of Dr. Ngo’s presentation and his views in the panel discussion. The entire session can be watched at Youtube.

Manufacturing Challenges

Process industries (oil refineries, petrochemicals, steel, water, etc.) require complex control of temperature, pressure and flow rate due to chemical reactions and other factors. Dr. Ngo explained this complex control scenario by giving an example of an oil refinery from refining to processing and final assembly. The 4Ms that affect quality and production are:

Artificial intelligence

Manufacturers are now turning to exploring cutting-edge technologies, such as artificial intelligence and machine learning, for process autonomy. Since the launch of Industry 4.0, the focus of AI has expanded. Dr. Ngo provided a schematic representation of AI/ML in process control via a typical linear application to hierarchical control layers. At level 1 (the sensor level) there is already artificial intelligence built in; Level 2 deals more with the control layers – IoT network, DCS etc. At this point, the AI ​​can be integrated into the Reinforcement Learning Card (FKDPP) algorithm on the consoles. Level 3 and above take advantage of applications such as visualization analytics – AI embedded in image analytics on field monitors, robots, etc. Furthermore, there are a variety of applications and services for specific solutions; This is Yokogawa’s AI platform studio – Xperience and the AI ​​platform responsible for creating AI algorithms for specific applications.

Working Kernel 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 the IEEE International Conference on Automation Science and Engineering as the world’s first reinforcement learning-based artificial intelligence that can be used in plant management. FKDPP ran on a simulator of a vinyl acetate manufacturing plant and operated valves to increase product volume while ensuring quality and safety standards were adhered to. Stable and improved valve operation was achieved in 30 learning experiments.

The FKDPP uses a factorial policy model and a seamless update based on kernel-wise factors through regulation with the Kullback-Leibler differing between existing and updated policies. Compared with previous methods that cannot directly process a large number of procedures, the proposed method from Yokogawa makes use of the same number of training samples and achieves a better control strategy for vinyl acetate monomer (VAM) monomer production, quality and stability.

Salient Features of FKDPP

  • It can be applied to most types of control
  • Increases productivity
  • basic
  • Explainable process
  • Security levels are similar to conventional systems

Case Study

In 2019, Yokogawa Engineering Asia successfully piloted a control training device. Three-cabin level control system was set up via a laptop computer. Although the system can be controlled using traditional PID technology, it has been shown that FKDPP can reduce settling time by 50-70 percent, while at the same time preventing overflow and maintaining the water level in the tank. This was demonstrated by a video showing the differences between the first, twenty, twenty-fifth and thirty iterations of AI-based reinforcement learning (FKDPP algorithm). The three basic steps from creating the FKDPP model to the actual control are: goal setting, AI control model building, and AI independent 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 spoke about how FKDPP balances quality and energy savings. The media opined that FKDPP “can contribute significantly to production autonomy, maximizing return on investment, and environmental sustainability.”

a future vision

In this context, Dr. Ngo spoke about Yokogawa’s vision of Industrial Automation to Industrial Independence (IA2IA). A survey of 534 decision makers in 390 factories revealed that 42 percent believe that in the next three years the application of AI to improving operations in the factory will have a significant impact on industrial autonomy. The envisioned implementation of 5G networks, cloud, and artificial intelligence for industrial autonomy will allow optimal control anytime and anywhere.

Point of views

Below we summarize Dr. Ngo’s responses during the panel discussion.

Is the design suitable for different interfaces?

Nowadays the implementation is done through the OPC interface; But in the long run, the company will integrate the full visualization. Multi-vendor data will be pulled from the data lake and put into the system.

Why was the pilot project in the chemical plant limited to 35 days?

Routine maintenance of this chemical plant was carried out on the 36th day and therefore it was discontinued on the 35th day. Then when the factory restarted, it was in control of the AI.

How long does it take for the FKDPP to learn the operator’s actions before setting independent control?

Safety is always the main focus of Yokogawa’s applications. FKDPP was learned from the plant historian, including operator action simulation, to ensure safe autonomous control. Learning duration depends on the complexity of the control. In this particular plant, the lead time was short due to the adaptive processes implemented in FKDPP.

Going forward, do you see AI replacing traditional PID?

We are more interested in tackling what PID and APC cannot do, filling in these gaps, and improvising with AI at that. However, this may happen in the future. Even academia is trying to push the ideal of all-inclusive artificial intelligence, but the intelligence must be based on the fundamentals of the PID type methodology. This is a transitional period – even academia will take time to adapt AI as a control strategy rather than core engineering.

Leave a Reply

Your email address will not be published. Required fields are marked *