Fault Diagnosis Of Gearbox Based On Particle Filter

In the non-gaussian and nonlinear system, Particle filter has showed great superiority. According to the characteristics of PF, a new method was put forward in which PF was applied in gear box vibration signals for noise reduction: two measurements are made on the same fault vibration signal, and the two results were respectively as estimates of the state equation and measurement equation of observation.According with the actual situation of gear box, the noise of the gearbox vibration signal was defined for a range of random noise. And the noise signal was extracted from fault vibration signal by using the theory of singular value noise reduction.the state transition equation and measurement equation of the signal was built by using Noise signal range and two measuring results.The distribution characteristics of the signal could be respectively reflected wth Singular spectrum entropy, power spectrum entropy and wavelet energy spectrum entropy in time domain and frequency domain.Therefore, a new way was put forward in which PF was applied in gear box vibration signals for noise reduction: three kinds of entropy was as a vibration signal feature vector for pattern recognition.The search speed of Particle swarm algorithm(PSO) is faster, related parameters is less, but it is easy to fall into local optimum and premature convergence, and it may not come to the global optimal. Aim at the defects of PSO, an improved particle swarm optimization algorithm is put forward:particles were Initialized if their states were over the range;the optimal solutions of population was Changed based on the optimal solutions of individual.And the new way was applied to the parameters optimization of Support vector machine(SVM),and the optimized SVM was applied to fault diagnosis of gearbox.