www.automation-mag.com
18
'26
Written on Modified on
Achieving Autonomous Operations Through Native Reinforcement Learning
Yokogawa’s latest distributed control system update utilizes advanced soft sensors to tackle hard-to-automate industrial processes.
www.yokogawa.com

CENTUM R7 integrated production control system conceptual image
Yokogawa Electric Corporation has developed CENTUM R7.02, a new version of its CENTUM integrated production control system with AI functions as standard. Part of the OpreX Control and Safety System lineup, the release supports AI Control and adds an AI Estimation function. With this version, the product name changes from CENTUM VP to CENTUM.
Autonomous Control AI for Hard-to-Automate Processes
Yokogawa’s autonomous control AI targets processes that are difficult to automate with conventional control methods and have so far required manual adjustment by operators, as well as control tasks that must balance several competing objectives. Even where disturbances cause large process fluctuations, the AI determines and implements the appropriate control actions directly. In 2022, the solution controlled a chemical plant autonomously for 35 consecutive days, and the plant has since operated stably for more than four years with energy savings and a lower environmental impact. The underlying reinforcement learning algorithm, Factorial Kernel Dynamic Policy Programming (FKDPP), was jointly developed by Yokogawa and the Nara Institute of Science and Technology (NAIST).
AI Control Models Executed on the DCS Platform
With AI Control, control models generated by the autonomous control AI run as a standard function within the CENTUM system environment. The models are downloaded and deployed through the CENTUM engineering environment, and operators monitor operating conditions in the familiar CENTUM operation and monitoring environment using graphics, trends, alarms and faceplates. Before deployment, the AI undergoes repeated learning cycles on a plant simulator that reproduces process behaviour in order to derive an optimal control model. After the customer has evaluated the model’s reliability and validity, it is applied to plant control together with functions that ensure safety.
AI Soft Sensors for Quality-Relevant Process Values
The AI Estimation function provides AI soft-sensor capabilities that estimate values such as component concentrations from process data including temperature, flow rate and pressure. Such values are important for assessing product quality and process conditions but are often difficult to measure directly and continuously in real time, and some require laboratory analysis. Machine-learning models estimate these values and indicate how much each process variable contributes to the result. Operators can view the estimated values alongside other operating information in the CENTUM operation and monitoring environment.
Target markets include refining and petrochemicals, chemicals, power, gas, pulp and paper, pharmaceuticals, food, iron and steel, metals, water and wastewater, and cement.
Additional Context
This section details technical specifications and competitive benchmarking not included in the original news release.
Neural-network-based soft sensors are also available on other distributed control systems. Emerson’s DeltaV Neural, for instance, creates virtual sensors using neural networks that execute directly in the DeltaV controller as a function block, with automated pre-processing, design, training and verification. CENTUM R7.02 offers comparable AI-based estimation, adding an indication of how much each process variable contributes to the estimated value, and combines it with the execution of reinforcement-learning control models generated by Yokogawa’s FKDPP-based autonomous control AI, which determines manipulated variables directly instead of only supplying estimated values to conventional control loops.
Edited by Aishwarya Mambet, Induportals Editor, with AI assistance.
www.yokogawa.com
Yokogawa Electric Corporation has developed CENTUM R7.02, a new version of its CENTUM integrated production control system with AI functions as standard. Part of the OpreX Control and Safety System lineup, the release supports AI Control and adds an AI Estimation function. With this version, the product name changes from CENTUM VP to CENTUM.
Autonomous Control AI for Hard-to-Automate Processes
Yokogawa’s autonomous control AI targets processes that are difficult to automate with conventional control methods and have so far required manual adjustment by operators, as well as control tasks that must balance several competing objectives. Even where disturbances cause large process fluctuations, the AI determines and implements the appropriate control actions directly. In 2022, the solution controlled a chemical plant autonomously for 35 consecutive days, and the plant has since operated stably for more than four years with energy savings and a lower environmental impact. The underlying reinforcement learning algorithm, Factorial Kernel Dynamic Policy Programming (FKDPP), was jointly developed by Yokogawa and the Nara Institute of Science and Technology (NAIST).
AI Control Models Executed on the DCS Platform
With AI Control, control models generated by the autonomous control AI run as a standard function within the CENTUM system environment. The models are downloaded and deployed through the CENTUM engineering environment, and operators monitor operating conditions in the familiar CENTUM operation and monitoring environment using graphics, trends, alarms and faceplates. Before deployment, the AI undergoes repeated learning cycles on a plant simulator that reproduces process behaviour in order to derive an optimal control model. After the customer has evaluated the model’s reliability and validity, it is applied to plant control together with functions that ensure safety.
AI Soft Sensors for Quality-Relevant Process Values
The AI Estimation function provides AI soft-sensor capabilities that estimate values such as component concentrations from process data including temperature, flow rate and pressure. Such values are important for assessing product quality and process conditions but are often difficult to measure directly and continuously in real time, and some require laboratory analysis. Machine-learning models estimate these values and indicate how much each process variable contributes to the result. Operators can view the estimated values alongside other operating information in the CENTUM operation and monitoring environment.
Target markets include refining and petrochemicals, chemicals, power, gas, pulp and paper, pharmaceuticals, food, iron and steel, metals, water and wastewater, and cement.
Additional Context
This section details technical specifications and competitive benchmarking not included in the original news release.
Neural-network-based soft sensors are also available on other distributed control systems. Emerson’s DeltaV Neural, for instance, creates virtual sensors using neural networks that execute directly in the DeltaV controller as a function block, with automated pre-processing, design, training and verification. CENTUM R7.02 offers comparable AI-based estimation, adding an indication of how much each process variable contributes to the estimated value, and combines it with the execution of reinforcement-learning control models generated by Yokogawa’s FKDPP-based autonomous control AI, which determines manipulated variables directly instead of only supplying estimated values to conventional control loops.
Edited by Aishwarya Mambet, Induportals Editor, with AI assistance.
www.yokogawa.com

