Demonstrating Containerization of Model Predictive Control for Modular Plants

Model predictive control has the potential to increase yields in the process industry, but its deployment is limited by high computational cost. In multi-purpose, modular plant concepts, it is particularly difficult to predict the whole variety of products for which a specific module will be used throughout its lifespan. During module manufacturing, this results in the challenge of allocating the correct computing, memory, and communication capacities to a module so that a model-predictive control application with currently unknown hardware requirements can be used later. The difficulty of this task often prevents the desirable deployment of model predictive control in production environments. The aim of this publication is therefore to demonstrate a simplified deployment and reconfiguration of model predictive control applications for modular plants through containerization. After an overview of the current state of the art, we present the underlying automation architecture based on modular plant concepts combined with a cloud-like automation approach enabled by containerization of the control applications. The feasibility of the chosen approach is demonstrated by implementing a proof-of-concept demonstrator. Subsequently, lessons learned from the demonstrator are discussed and future improvement potentials are identified, which will be the subject of future research.