Robust fault detection for gearbox failure

This contribution presents a methodology to detect material deterioration in periodic systems using model identification. To detect the failure, the relation between the model's parameters and failure's characteristics are analyzed. Thus, a robust and an efficient algorithm is proposed to detect failures. The algorithm leads to a residual parameter to identify anomaly, which allows an intuitive failure detection. The convexity of criterion, the variation of parameters within interval and the residual's expression allow to use optimisation algorithm or step evaluation of criterion, where the last procedure allows to simplify the implementation of the algorithm and aims to online utilization. The proposed approach is applied to gearbox system, where the classical analysis cannot bring clear information about deterioration. The efficiency of the proposed method is as well proved and the application is used to benchmark the algorithm's parameters.