Detection and evaluation of load oscillations in induction motors based on MCSA

Abstract

Load oscillations of induction motors can be the consequence of electrical or mechanical faults. One of the most important techniques to diagnose load oscillations is the so-called motor current signature analysis (MCSA). It has won widespread popularity owing to its non-intrusive and cost-effective performance. The common practice of MCSA is to detect the characteristic harmonic components of stator current caused by load fluctuations. However, in certain scenarios, the characteristic components can be hardly tracked from the current spectrum. Power frequency interference, nonstationary randomness of the sampled signals, and other factors could all limit the use of classical current spectrum analysis. The presented solution is the joint application of variational mode decomposition (VMD), delay addition method, multiple signal classification (MUSIC), and neural network. Specifically, VMD, delay addition and MUSIC are combined to decompose the raw signals, eliminate the power frequency leakage interference and extract the weak oscillation features with enhanced visualization. Meanwhile, a neural network is generated to describe the statistical features of the processed signals, and further realize the automatic evaluation of the load oscillation severity level. Theoretical analysis and experimental verification have been conducted to support this article. The results suggest that the proposed algorithm contributes to a more sensitive detection of the presence of load oscillations, and a more reliable evaluation of their severity with respect to traditional MCSA techniques.