基于特征重构深度置信网络的车用电机复合故障定位研究

The faulting mechanism of permanent magnet synchronous motor (PMSM) is complicated. Meanwhile, the occurrence of each kind of motor fault type is usually related to other faults. Therefore, the compound fault location of PMSM is difficult to be researched. Besides that, it’s common for PMSM to operate under variable speed conditions in automotive applications. Thus, realizing the fault location under variable operating conditions would be of great value in engineering applications. In order to improve the compound fault location accuracy of PMSM under variable speeds, a novel PMSM compound fault location method based on feature reconstruction and the deep belief network (DBN) is studied. Firstly, based on the vibration characteristic analysis of PMSM, the compound fault characteristic components of PMSM under rotor eccentricity, bearing inner ring fault, and the faulting mechanism of the compound fault are clarified. Next, the Vold-Kalman filtering order tracking (VKF-OT) and the angle-domain reconstruction algorithms are employed to extract the compound fault feature component and reconstruct the signal, thus the influence of variable speeds on the compound fault feature could be weakened or eliminated. Finally, the reconstructed signal is used to construct the fault location model based on DBN, and the compound fault of PMSM under variable speeds can be diagnosed. The experimental results show that the compound-fault-location accuracy based on the feature reconstruction and DBN can reach 99.5%.