Design of a fault diagnosis system based on a bank of filter—observers with application to a hot rolling mill

Abstract

This paper proposes a new method for actuator/sensor faults detection, isolation and magnitude estimation. By establishing new theoretical results relating to fault diagnosis and by defining a specific application domain, the research work described in this paper aims at developing a model-based fault detection filter combined with a conventional Luenberger observer. Based on an appropriate gain, the developed technique enables one to supervise a system through an accurate bank of residuals within a generalized observer scheme. The approach is evaluated by diagnosing a series of actuator/sensor faults in a speed control loop of a hot rolling mill used to carry out multi-pass forward—forward or forward—reverse rolling. The performance of the technique is illustrated via a wide range of simulations in the ‘faulty’ case and a practical experiment in the ‘fault-free’ case using a real rolling mill.