Architecture for Artificial Intelligence in Operational Process ControlCopyright: © IAT
Digitalization is a driving force for the application of artificial intelligence in an industrial context. Especially in the last decade, new and optimized machine learning methods, corresponding tools and the associated powerful as well as more cost-effective hardware became available. But the utilization in industrial applications faces some challenges, especially due to constraints from the field of "Operational Technologies", such as availability, robustness, durability, maintainability, etc.
In order to close the gap between the new machine learning methods and their use in industrial production, models trained in this way must be integrated into current process control systems. Besides the pure technical integration, an adaptation to existing standards, their interfaces, protocols and behavior patterns is a core task in order to guarantee a uniform process control with and without AI support. At the Chair of Information and Automation Systems for Process and Material Technology we conduct research on the mapping of standardized state machines, such as ANSI/ISA-88 (PackML) and interfaces, which are developed especially in the context of a component-based control architecture. In doing so, the chair builds on its many years of experience with so-called process control components and continues to develop them in the context of Industry 4.0 in the BaSys4.0, BaSys 4.2 and BaSys4Brenner projects and demonstrators, among others. Research into the integration of results from machine learning, in particular reinforcement learning, by means of services (driving modes) in process control components will thus enable a step-by-step migration to a changeable but at the same time industry-compliant architecture for artificial intelligence in operational process control in the future.