Study On The Noise Reduction Method Of Artillery Chamber Pressure Signal Based On Convolutional Neural Network

Chamber pressure as an important index of gun development,chamber pressure change directly affects the accuracy of the gun,if the chamber pressure is abnormal,can also cause the chamber accident,therefore high quality chamber pressure signal is particularly important,but the chamber pressure signal accompanied by various noise,such as testing equipment generated system noise and some external environment such as random noise caused by high impact,the existence of the noise,the structure of the strength of the shell component design and gun rack strength and stiffness design influence.Therefore,the denoising of the bore pressure signal is an important part of the data processing.To this end,this paper studies the denvoloising method of chamber pressure signal by convolutional neural network in traditional wavelet transform,Internet big data and hotspot deep learning of artificial intelligence.The specific work is as follows:(1)To address the problem that the different designs of wavelet base function,threshold function,threshold calculation rules and decomposition layers in traditional wavelet threshold denoising will affect the denoising performance,this paper simulated and analyzed these parameters affecting wavelet threshold denoising one by one,and used SNR and RMSD to determine the most appropriate denoising parameters for each influencing factor.The simulation results show that the de-noising parameters are chosen to keep the associated characteristics of the primitive signal and to remove the noise well.The result verifies the effectiveness of the appropriate de-noising parameters of the algorithm on the de-noising of the bore pressure signal.(2)For the traditional wavelet threshold denoising algorithm requires manual extraction and workload,this thesis ueses the hollow convolutional neural network,and improve on the basis of Dn CNN,using residual learning,BN and Adam to the network depth,training speed and network parameters optimization,realize a chamber pressure signal denoising algorithm based on empty convolutional neural network.By synthetic data and different degree of Gaussian white noise to test,and the denoising results with WT,EMD and Dn CNN comparison,adopting the SNR and mutual relationship number to quantify the denoising effect,the results show that the proposed algorithm denoising capability outperforms several others,it can remove the noise and retain the original effective signal to a large extent,which verifies the high efficiency of the algorithm to denoising the chamber pressure signal.(3)For the hole convolution neural network needs to be extracted from the large amount of data,facing the problem of large parameters,for wavelet transform threshold denoising cause signal distortion,false Gibbs phenomenon,this paper combines good time-frequency localization property of the WT and hole convolutional neural network from learning,self-adaptability,strong robustness advantage,on the basis of empty convolution,puts forward a kind of algorithm.To verify the superiority of this algorithm,with synthetic data and different amounts of random noise test,and the results with the SNR and the correlation and other methods of empty convolution,the results indicate the proposed algorithm suppresses the noise more thoroughly,retains the highest fidelity,verify the superiority of denoising chamber pressure signal.(4)To verify the applicability of the proposed algorithm for chamber pressure signal noise reduction,using the real measured in the range chamber pressure signal noise reduction,and noise reduction results and WT,EMD,Dn CNN and hollow convolution network,comparison,the experimental results show that the wavelet transform + cavity convolution noise effect of the best,verify the algorithm of the noise reduction of chamber pressure signal.