The Research Of Gearbox Fault Diagnosis Based On RBF-PF And Wavelet Neural Network Of Particle Swarm Optimization

Gear transmission is the main form of mechanical transmission and the basis of the machine parts.Its quality,performance,service life directly affect the technical and economic indexes of the whole machinery and equipment.Operation condition of gear box in general is relatively complex and bad so that gear box in the mechanical equipment is prone to mechanical failure.Most commonly occurs of the fault types in the gear transmission system is gear crack.Gear crack is imperceptible and harmfull.Fault diagnosis is generally divided into two parts: the extraction of fault characteristic signal classification and diagnosis.In the actual fault diagnosis of gearbox fault signal is often mixed with other noise signal so that are we cannot directly obtain the gear box fault vibration signal in the process and need to the corresponding signal de-noising pretreatment.Particle filter algorithm is based on monte carlo simulation to achieve the optimal bayesian estimation filter method.As gaining the estimated posterior probability function,its basic principle is simulating system model for getting each moment of the corresponding system's amples.Based on the Radial Basis Particle Filter(Particle Filter of Radial Basis Function)filters the signal noising as pretreatment.Particle Filter of Radial Basis Function Radial eliminates estimation error caused by the process noise,improving the prior probable precision of density distribution,avoiding the lack of particle.The particle filter performance has improved.Particle Swarm Optimization algorithm is based on Swarm intelligence Optimization method and come from the study of artificial life and birds feed on behavior.It is a kind of optimization algorithm interaction among individuals through complex behaviors such as cooperation and competition to search the global optimal solution.Its main characteristic have the simple concept,fast convergence rate,better global and local convergence ability.In the process of optimization,the practical engineering problem should be converted to the corresponding mathematical model.We understand the meaning of the mathematical model,the significance and the value range of each parameter.Then initializing the particle swarm optimization algorithm parameters,the engineering problems of adaptive value function is put forward.Finally,the accuracy of fitness function evaluates optimize parameters till satisfying the precision.As traditional gradient descending method of wavelet neural network(BP algorithm)easy to fall into local minimum,slow convergence speed,i puts forward the particle swarm optimization wavelet neural network learning algorithm.It optimizes the various parameters of wavelet neural network in the gearbox fault diagnosis experiment.This optimization algorithm reduces the iteration times and improves the convergence precision,so it is an effective training algorithm.This paper combines radial basis particle filter with particle swarm optimization wavelet neural network in the study of gearbox fault detection.It not only makes full use of the advantages that radial basis particle filter is good for noise reduction while processing nonlinear non-gaussian system,but also makes full use of particle swarm optimization to reduce the number of iterations and improve the precision.