Detecting abrupt changes based on dynamic analysis of similarity for rotating machinery fault prognosis

Detecting abrupt changes of dynamic structure of mechanical systems by its condition-based time series data is an important basis for fault prognosis. Segmenting time series at abrupt change points can classify the different dynamic structures and determine when the underlying model has changed. A novel method based on the exponent dynamical cross-correlation factor is presented to detect abrupt change points. Ideal time series is used to evaluate the performance of the proposed method. CWRU vibration signal data analysis of bearings using the presented method show that the load changes have no significant effect on the dynamic characteristics and fault defects have strongly influence on dynamic characteristics of rotating machinery.