Remaining Useful Life Prediction for Reducer of Industrial Robots Based on MCSA

Abstract

Vibration signal-based analysis is widely used in fault diagnosis and reliability evaluation for electromechanical transmission system. Due to the structural design of system, the service environment, its accuracy requirements and other factors, it is difficult to collect vibration signals for condition monitoring in some cases. As a result, the Motor Current Signature Analysis (MCSA) now develops rapidly because it can minimize the damage to the mechanical system and save economic costs while maintaining the accuracy of condition monitoring. However, the fault information contained in the current signal is weak and easily omitted. It is particularly important to effectively reduce the noise of the original signal. In addition, most of the existing researches often used the current signal to analyse the fault of the reducer, the method for predicting the remaining useful life (RUL) of the reducer is limited. In this study, a life prediction framework is proposed based on MCSA for the harmonic reducer. Maximum Correlated Kurtosis Deconvolution (MCKD) and Completed Ensemble Empirical Mode Decomposition (CEEMD) are combined to de-noise and decompose the original current signal to obtain Intrinsic Mode Function (IMF). Then effective IMF components are extracted and dimensioned in multiple domains, the degradation index of the harmonic reducer is constructed, and the degradation stage of the entire life cycle is divided. BAS optimization algorithm is used to improve the accuracy and efficiency of BP neural network model so as to predict the RUL.