ENEOS Materials, a Japanese maker of synthetic rubber, thermoplastic elastomers and latex, uses an autonomous control AI from Yokogawa to run part of its Yokkaichi Plant, according to a Yokogawa customer success story. The system controls a butadiene distillation column, where liquid levels and the supply of waste heat and steam must be tightly controlled to separate high-purity butadiene from substances with close boiling points. Because of factors such as weather-related changes in ambient temperature, the plant's distributed control system and advanced process controls could not optimise two key valves, so operators had to adjust them manually.
The AI is Factorial Kernel Dynamic Policy Programming (FKDPP), a reinforcement-learning algorithm developed by Yokogawa and the Nara Institute of Science and Technology. The joint project, part of a METI-subsidised industrial safety programme, started in August 2020 and followed three steps: building a model of the plant on a simulator so FKDPP could learn and generate an AI control model; refining that model with on-site knowledge and data while operators entered the AI's proposed valve settings into the control system; and then, once safety was ensured, letting the AI control the real plant. Plant staff noted that safety instrumentation remained in place and that operators could stop the AI and return to normal control at any time.
The AI controlled operations for 35 days, from January 17 to February 21, 2022, which Yokogawa describes as a world first. After a scheduled maintenance shutdown, a more stable version of the model began operating in April 2022, and over the following year FKDPP ran the two valves efficiently, making ENEOS Materials the first company to achieve long-term autonomous control of a plant with this AI. The system has since become a routine part of operations, keeping the column level within plus or minus 5% of its target, stabilising production of high-quality butadiene, reducing operator workload and cutting steam usage by 40%. "A human cannot operate two valves at the same time, but an FKDPP can," a member of the production team said.
Details
- City
- Yokkaichi
- Organization
- ENEOS Materials
- Continent