Fault Detection and Diagnosis in Steel Industry: a One Class-Support Vector Machine Approach

Complexity of manufacturing systems and variability in anomalous operations make fault detection and diagnosis in industrial systems a challenging task. In steel industries characterized by high temperatures and pressures, elevated production speeds, and intense throughput, the early diagnosis of an incoming fault is highly relevant for both safety and economic reasons. However, expensive preventive maintenance and early substitution of equipment are largely adopted, hence strongly limiting the availability of data related to fault events and the applicability of standard machine learning methods. In this work, we present a one class-support vector machine (OC-SVM) approach to early detect anomalies in steel making plants; we validate our method using production data gathered from a steel making industry placed in the South of Italy and compare performance with a multivariate statistical method recently designed for the fault detection of the same plant. The study revealed that OC-SVM outperforms the statistical method, and also is able to predict breakdowns.