Yokogawa Maps Five-Stage Path from Automation to Autonomy

Industrial AI is moving from concept to deployment in process plants, but not as a wholesale replacement for existing automation. Yokogawa Electric Corporation has outlined a maturity model that treats industrial autonomy as a disciplined evolution of measurement and control systems, introducing adaptive capabilities within bounded operating environments.

Dr. Rajeev Joshi, Deputy General Manager at Yokogawa, presented the company's IA2IA framework—Industrial Automation to Industrial Autonomy—at the ARC Advisory Group's 24th Annual Industry Forum in Bengaluru. The approach recognizes that industrial environments demand precision, reliability, and integration with existing safety systems in ways office AI applications do not.

Why it matters

Process plants cannot afford the trial-and-error approach common in consumer AI. Decisions affect production throughput, equipment integrity, product quality, and worker safety. Yokogawa's staged framework and deployment results at major facilities demonstrate that industrial autonomy can deliver measurable operational gains without bypassing human expertise or established control layers.

Five stages from semi-automated to autonomous

Yokogawa's maturity model progresses through five distinct stages: semi-automated, automated, semi-autonomous, autonomous orchestration, and autonomous operations. Each stage introduces additional adaptive capabilities and is designed to deliver measurable benefits before organizations advance further.

Traditional automation relies on predefined logic executed by PLCs, distributed control systems, PID controllers, and advanced process control. Operators intervene when conditions exceed what those systems were designed to handle. Industrial autonomy extends this foundation by adding learning and adaptive capabilities that respond to changing conditions while operating within secure boundaries.

Reinforcement learning for complex process control

Yokogawa's technical approach centers on Factorial Kernel Dynamic Policy Programming (FKDPP), a reinforcement-learning algorithm developed jointly with the Nara Institute of Science and Technology. The technology addresses multivariable and dynamic processes that remain difficult to control using PID or advanced process control alone.

FKDPP uses operational data to develop control policies that adapt as process conditions change. The technology is not intended to replace existing control systems but to introduce autonomous control in areas where conventional approaches still require frequent operator judgment.

Proven results at chemical and gas facilities

At an ENEOS Materials chemical plant in Japan, Yokogawa applied FKDPP to a distillation process that had previously required experienced operators to adjust a valve continuously. The AI-based controller manages competing objectives including product quality, liquid levels, yield, energy use, and temperature disturbances.

At Aramco's Fadhili Gas Plant, Yokogawa deployed autonomous control for the acid gas removal unit following a phased process: historical data collection and AI training, validation through simulation, introduction into non-critical process areas, and wider deployment after establishing operator confidence and safety integration. Initial results include 10–15 percent reduction in amine and steam usage, approximately 5 percent lower power consumption, improved process stability, and significantly reduced manual intervention.

Assess, pilot, scale

Joshi recommended starting with operational problems rather than asking where AI can be applied. Organizations should identify processes where better control could deliver meaningful value, assess whether AI is appropriate, validate through focused pilots, and scale only after establishing operational performance and operator confidence.

This approach acknowledges that industrial AI must coexist with established control systems, safety requirements, cybersecurity controls, and experienced personnel. Industrial autonomy shifts where human expertise is applied rather than removing people from operations. As repetitive interventions and complex monitoring become automated, operators can focus on optimization, exceptions, and higher-value decisions.

These details were first reported by Automation Watch, based on Joshi's presentation at the ARC Industry Forum.