Robust Statistics-based Anomaly Detection in a Steel Industry

Abstract

This paper reports the outcome of an industrial research project on data-based anomaly detection in a steel making production process. Namely, the study aims to assess a fault detection strategy for rotating machines in the hot rolling mill line. Due to the adopted intense and expensive preventive maintenance program, available data enclose only few samples of fault events, avoiding efficient application of classical data driven anomaly detection models. We developed an automatic two-step strategy which combines two statistical methods. Namely, the combination of Reweighted Minimum Covariance Determinant estimator and Hidden Markov Models helped to identify actual conditions in a drive reducer of a hot steel rolling mill and automatically isolate signs of decreasing performance or upcoming failures.