Particle Filter Signal Processing Based Diesel Engine Fault Diagnosis

Diesel engine is a common power equipment, it is normal work directly related to thewhole equipment operation state. Condition monitoring and fault diagnosis of diesel engine iscrucial to improve the reliability of diesel engine power plant,it can avoid unnecessary loss.The working conditions of the diesel engine is very bad, the background noise of theacquisition signal is big, and part of the useful signal is covered with strong noise, so noisereduction processing is required before the signal analysis.The particle filtering technique is a newly developed model-based state estimationtechnique, based on the in-depth study of the principle of particle filter,this paper apply it tothe diesel engine vibration acceleration signal noise reduction processing. The use of particlefiltering techniques for noise reduction need to know the model of the signal and noisestatistics, the approach of this article is: first,using the EMD smoothing processing on thevibration acceleration signal, taken decomposition IMF component for effective signal, andthen establish the AR model, use the coefficients of this model as the coefficients of theparticle filter’s state equation ; in the understanding of the wavelet transform noise reductionprinciple, the wavelet transform threshold noise reduction ideology is used to extract the noisesignal, the extracted noise signal is used as the particle filter observation equation.On the basis of the above theoretical analysis, do the vibration signal pre-processing,including detrended, removing the singular point, signal smoothing, AR modeling. Using theparticle filter algorithm based on the importance of the neural network weights adjustmentNNWA-PF to do the vibration Signal de-noising;Then, extract the wavelet packet energyspectrum;Last,use the energy spectrum as the BP neural network feature vectors classificationand identification of failure modes. In this paper, a particle filter noise reduction data andprocessed data were used to BP neural network to diagnose. After the training, testing anddiagnosis.,the results suggest that, after the diagnosis of NNWA-PF denoised data isbetter,and also proved that a neural network-based particle filter in noise reduction is better.