In a world's first, an AI technology has successfully controlled operations for an extended period at an industrial facility in Yokkaichi, Japan. This autonomous control AI solution from Yokogawa remains in use at the Yokkaichi Plant to the present day, controlling operations in a facility at this plant that previously had proven difficult to automate, achieving both energy savings and maintaining product quality.
In 2020, Yokogawa and the members of Production Team 3 at the Yokkaichi Plant commenced a joint demonstration project. The distillation column for the production of butadiene — a raw material used in synthetic rubber — required tightly controlled liquid levels and waste-heat/steam supply to efficiently separate high-purity butadiene from substances with close boiling points. Fluctuations in liquid level negatively impacted product quality, and large steam use wasted energy. Due to external factors such as weather-related ambient temperature changes, the plant's DCS and APC could not effectively optimize adjustments to two key valves, so plant operators controlled these valves manually.
The AI solution under consideration was Factorial Kernel Dynamic Policy Programming (FKDPP), a reinforcement learning-based AI algorithm jointly developed by Yokogawa and the Nara Institute of Science and Technology. The test progressed in three steps:
Step 1: Build a plant model of the actual plant on the plant simulator and let FKDPP self-learn on the simulator to generate an AI control model.
Step 2: Evaluate and refine the generated AI control model based on the customer's on-site knowledge and data. MVs presented by the AI were input by the operators into the DCS to control operations at the plant.
Step 3: Ensure safety, then control a real plant. The AI was constrained by rule-based limits, while existing DCS alarms, interlocks and other safety systems remained active. Operators could stop the AI and return to conventional DCS operation.
The AI control model realized AI-powered operations for 35 days (January 17 to February 21, 2022) at this plant — the first reported case of AI directly controlling outputs in a chemical plant. Over the following year, the system handled seasonal temperature changes of about 40°C and daily changes in feedstock composition while keeping the process stable and reducing steam use and related CO2 emissions by about 40% compared with manual control. The technology moved beyond demonstration into routine production, where it has now operated for more than four years through regular shutdowns and maintenance without model-related problems.
ENEOS Materials plans to apply the autonomous control AI solution to its downstream polymerization plant.
Source: Yokogawa Electric Corporation success story.
Details