Evaluation of two Algorithms for Change Detection Based on Vibration Signals Processing
Change detection and diagnosis are important activities and research directions, in the field of system engineering and conditional maintenance of the equipment and industrial processes. The processed signals are coming from vibration generated by incipient faults in mechanical structures, e.g. bearings. Classical algorithms based on various version of CUSUM do not have enough performances to use intensively in real industrial application. The present work considers two new algorithms for change detection working on real industrial data of radial bearings. One is based on classical CUSUM criterion applied to the Renyi entropy. The second one is based on energy processing distributed over time-frequency region. The algorithms are tested on real recorded data. The results indicate good behavior and performance of the proposed algorithms, and define the rationale to implement them in commercial software product for change detection and diagnosis.