Concept of a Machine Learning supported Cross-Machine Control Loop in the Ramp-Up of Large Series Manufacturing

Abstract

With the advancing digitalization of production plants, it becomes possible to use process data across machine boundaries. A machine can adapt its parameters to another machine-measured parameter to increase product quality. The present paper describes the design of an inter-machine control loop with machine learning techniques in order to improve the final quality output. The production ramp-up represents a special application case for this since at this point of time there is only limited knowledge about cause-effect relationships. For this purpose, the paper presents a method for analyzing these interrelations. On the one hand, simple linear regression is used to analyze the linear relationships; on the other hand, machine learning algorithms are used to analyze non-linear relationships. Two independent control loops form the overall control loop, which is capable of deriving holistic prognoses on upstream or downstream process effects.