Resonance-Based Sparse Signal Decomposition Based on Genetic Optimization and its Application to Composite Fault Diagnosis of Rolling Bearings

Abstract

The selection of approximate values for the weight coefficients of the objective function in the existing RSSD with large subjective randomness reduces the advantages of this method for mechanical fault diagnosis. To solve this deficiency, a new method, objective function optimization based on a genetic algorithm and the split augmented Lagrangian shrinkage algorithm applied to RSSD, is proposed. This method utilizes the global optimization ability of genetic algorithms to adaptively optimize each element value of the weight coefficient matrices of the objective function of RSSD and achieve the optimal value of the objective function in the range of the desirable weight coefficients. Thus, this method adaptively realizes a sparse decomposition of the high- and low-resonance components according to the input signal and minimizes the information leakage in the process of signal decomposition. Finally, the proposed method was applied to diagnose a rolling bearing with composite faults of the inner and outer races and used to effectively extract the composite fault characteristics of the rolling bearing vibration signal. Accurate diagnosis of the early composite fault validated the practicability of the proposed method.