Decision Fusion for Fault Classification in Industrial Processes

This paper proposes an ensemble-based method that combines decisions produced by diverse machine learning classifiers, each with distinct pattern representations. The proposed method exploits the advantages of each classifier as a complementary information source to efficiently classify faults in industrial processes and provide plant operators with unified decisions. It uses the behavior knowledge space (BKS) method and updates its corresponding lookup table based on the F1-scores calculated for every single classifier. In this paper, the method is validated on a pulp mill process benchmark using simulated data. The obtained results demonstrate that the accuracy of the proposed ensemble is higher than the accuracy of every single classifier on this challenging application.