Optimization of HMM Based on Adaptive GAPSO and Its Application in Fault Diagnosis of Rolling Bearing

Aiming at the problem that the parameter learning algorithm of hidden Markov model tends to converge to local optimal solutions, the genetic particle swarm optimization algorithm is proposed to optimize the initial parameters of hidden Markov model, with an adaptive parameter adjustment method is adopted to improve its optimization performance. The fault diagnosis experiments rolling bearings in different conditions are carried out. The constructed mixed domain feature set is input HMM after dimension reduction. The proposed method is analyzed from aspects of search capability and convergence speed. The experimental results show that the diagnosis accuracy of the proposed method for normal, inner ring failure, outer ring failure and roller failure can all reach 100%. Compared with PSO optimized HMM, the classification accuracy is improved by 28.57% and the classification dispersion is improved by 268.58% at the highest, which proves the effectiveness of the method.