Genetic algorithm based independent component analysis method in machine fault diagnosis

The performance of existing independent component analysis methods is highly affected by the non-linear contrast functions selected according to the distribution of original signals, and the separation results are unsatisfied. To solve this problem, an improved independent component analysis method based on genetic algorithm was proposed. Then simplified mutual entropy among signals was adopted as the optimization function. The probability of separated signals was estimated by histogram method, and the mutual entropy could be evaluated. The genetic algorithm was applied to find the optimum separation matrix to minimize the simplified mutual entropy. Simulation results show that the proposed independent component analysis method is more effective than FastICA. Finally, this method was applied to diagnose the fault of rolling bearing. The results show that modified independent component analysis method can separate the mixture of rolling bearing sound signal and electromotor signal well, and improve the diagnostic information quality.