Fault detection and diagnosis in machining processes and rotating machinery using fuzzy approach and hidden Markov model

Abstract

Over the last three decades, the research for new fault detection and diagnosis techniques in machining processes and rotating machinery has attracted increasing interest worldwide. This development was mainly stimulated by the rapid growth of industrial technologies and the increase in complexity of machining and machinery systems. In this study, a two-stage monitoring system is proposed. The first stage consists of developing four generic condition indicators to detect transient and continuous anomalies. These indices are then fused to improve the fault detection using two fuzzy methods: a fuzzy-set based fusion and a fuzzy-rule based fusion. In the first approach a condition index is developed based on the min-max technique, whereas in the second method a fuzzy fused index (FFI) is defined using the Sugeno-style inference engine and a list of detection rules representing the knowledge gained through learning and experience of domain experts. Additionally, an index of deterioration is developed as part of the first fusion approach. The second stage forms the fault diagnosis part of the monitoring system. A discrete hidden Markov model is developed to identify the severity of the fault and to isolate its location. Methods are applied to two types of problems: tool wear/fracture and bearing faults. Results obtained from experimental evaluations show a minimum detection, classification, and isolation rate of 95%. Particularly, the FFI is found sensitive to fault severity and not susceptible to speed and load changes. On the diagnosis front, the training required to define the hidden Markov model does not exceed five seconds in both monitoring cases.