A Park's vector approach using process monitoring statistics of principal component analysis for machine fault detection

To ensure reliable operation of electric motors, various types of condition monitoring techniques are employed. Among them Electrical Signature Analysis (ESA) is a well-established technique for fault detection. ESA mainly consists of motor current signature analysis, instantaneous power signature analysis and extended Park's vector approach. While these techniques are very effective for fault detection but they require monitoring of frequency spectrum resulting in high number of computations. On the other hand, Park's vector approach (PVA) can detect faults without monitoring frequency spectrum providing high computational efficiency. For fault detection, principal component analysis (PCA) is often used in PVA to check deviation in circularity of Park's vector Lissajous curve by calculating principal values. While in this paper, PCA model based process monitoring statistics are utilized for detection of faults. In this way, PCA is only used for an initial model construction and then monitoring statistics are employed to check any deviation from the reference model. The validity of this technique is established by simulations carried out on three phase synchronous machine data. Our simulation results show that the inclusion of process monitoring statistics significantly improve the computational performance.