Research On Fault Diagnosis Of Fan Based On Wavelet Analysis And PSO-BP Neural Network

Fan usually works continuously in industrial production,and it's in unattended condition.once the fan failure,will not only cause the equipment damaged,causing huge economic losses,but also may lead to serious safety accidents.Therefore,it is of great significance for the condition monitoring and fault diagnosis of fan.Wavelet analysis is an effective method of signal analysis,it has the characteristics of adaptive time-frequency.In this paper,the centrifugal fanl is taken as the research object,the wavelet analysis and PSO-BP neural network are used to analyze the audio signal,the fault diagnosis of fan is realized,a fan fault diagnosis system is developed.The main research results and conclusions of this paper are as follows:(1)Based on the study of wavelet analysis theory,an improved threshold function wavelet denoising method is proposed.By conducting MATLAB simulation experiments,the results show that compared with the traditional hard threshold and the soft threshold function denoising method,the method obtains a higher SNR,eliminates most of the noise in the sound acquisition process,and it is helpful to extract the characteristic parameters of the signal.(2)By studying the mapping relationship between fan fault type and characteristic parameters,it is found that there is a serious aliasing phenomenon in the single-band reconstruction algorithm,and it can not extract the characteristics of the signal accurately.Based on this,this paper proposes an anti-aliasing single-band reconstruction algorithm to extract the energy distance of each frequency band,takes it as a fan fault diagnosis characteristics,and verify the effectiveness of anti-aliasing energy distance parameter through simulation experiments.(3)A rotor test rig has been designed and built.The simulation experiments and analysis have been carried out on normal,unbalanced rotor,rotor misalignment,dynamic and static rub impact signals of the centrifugal fan.According to the BP neural network convergence is not stable,easy to fall into local extremum shortcomings,this paper proposes the use of particle swarm algorithm to optimize BP neural network initial weights and thresholds,establishes a PSO-BP neural network model.The parameters of the model are designed,the model is applied to the fan fault diagnosis,the results show that: In the training phase,the PSO-BP neural network can converge quickly,and it is superior to the BP neural network in terms of training duration,training steps and convergence error;When the test data is diagnosed,the PSO-BP neural network can accurately identify the above four conditions,and realize the nonlinear mapping of the eigenvector and the fault type.(4)A set of fan fault diagnosis system which integrates sound collection,wavelet denoising,time domain analysis,feature extraction,fault diagnosis is designed.And the improved threshold function denoising algorithm,anti-aliasing single-band reconstruction algorithm,PSO-BP neural network are applied to the system.The reliability of the system is verified by multiple sets of experimental data.