A framework for smart control using machine-learning modeling for processes with closed-loop control in Industry 4.0

Anomaly detection for processes with closed-loop control has become a widespread need in Industry 4.0 shop floors. A major challenge in monitoring such processes arises from the unknown dependencies among monitored observations, as these dependencies may change dynamically and with high frequency. Motivated by these considerations, a novel framework is proposed for self-adaptive smart control using adaptive machine-learning models. On the one hand, data driven machine-learning algorithms can deal with patterns and dependencies within the data that were not necessarily known in advance. On the other hand, the recurrent self-adaptive mechanism triggers the need to switch to a new type of machine-learning model to capture and reflect new dependencies among monitored observations resulting from changes in the process. The proposed framework and the associated case study described in this paper could serve as a firm basis for implementing self-adaptive smart process control in Industry 4.0 shop-floor processes with closed-loop control.