Mechanical fault diagnosis method based on Mahalanobis Distance and LMD

摘要

Against non-stationary characteristics of the mechanical fault vibration signal, this paper proposed a diagnosis based on LMD (Local Mean Decomposition, LMD) and Sensitive Threshold. This author adopted LMD to process the vibration signal and obtained a set of PF (Production Function, PF), adopted K-L (kullback-leibler) divergence to extract principal PF components, calculated their time-domain parameter indexes, combined them into a feature vector. Based on Mahalanobis Distance, this author took Mahalanobis Distance sensitive thresholds to represent different fault states, took the mean of multiple normal signal feature vectors as the standard feature vector, calculated the Mahalanobis Distance sensitive threshold of the unknown feature vector and the standard feature vector, and finally identified the fault states. The results showed that this method can effectively identify the mechanical fault, better than EMD (Empirical Mode Decomposition, EMD).