Research On Fault Diagnosis Method Of Aviation Hydraulic Pipeline Based On BN-1DCNN

Aviation hydraulic pipeline is used as a device to transport hydraulic oil,fuel oil,lubricating oil and other oil bodies between aircraft engine and various components,so that it will be in complex vibration for a long time,in the safety and reliability of aircraft and other important aspects of serious risks.In this thesis,the aviation hydraulic pipeline is taken as the research object.According to the working environment and vibration characteristics of the aviation hydraulic pipeline,the nonlinear and lossless Kalman filter is selected to filter the collected vibration signals.Through the design and construction of the comprehensive test platform,the vibration experiments of 15 different working conditions are carried out,and the vibration signals are analyzed in the time-frequency domain.The vibration signal is denoised by the lossless Kalman filter,and the BN-1DCNN model is designed and optimized to realize the accurate identification and diagnosis of aviation hydraulic pipeline mixing faults.Firstly,considering the influence factors of aviation hydraulic pipeline working environment and external environment interference and vibration,by comparing and analyzing the working principle and noise reduction effect of various traditional filters and nonlinear lossless Kalman filter,the lossless Kalman filter is selected to filter the acquired vibration signal,and UT transform and Kalman filter are integrated.It can achieve noise reduction and interference elimination in the vibration signal of aviation hydraulic pipeline,and retain the characteristic mode of the original signal,and obtain effective and reliable and real vibration data.Secondly,by analyzing the structural features of the original CNN network model,random forest model,XGBoost model,network operation mode and the function of each network layer,the BN-1DCNN network model is designed and optimized,and the original convolution kernel is changed into a one-dimensional convolution kernel to realize the direct convolution of data.Moreover,the BN layer and Dropout layer are added to the network model to prevent the overfitting problem of the model and greatly improve the model’s ability of computing and processing one-dimensional big data.A one-dimensional convolutional neural network(BN-1DCNN)model based on batch normalization is proposed to lay a theoretical foundation for improving the accuracy of aviation hydraulic pipeline mixed fault diagnosis and recognition.Then,through the design and construction of aviation hydraulic pipeline comprehensive test platform,vibration experiments of 15 different conditions were carried out.The time-frequency domain analysis was carried out on the collected vibration signals.The results showed that the vibration amplitude of both straight pipe fault and bent pipe fault would increase,especially for single pipe single fault,the vibration amplitude increased obviously,and the amplitude of straight pipe fault was slightly larger than that of bent pipe fault.The time-frequency domain analysis method can basically identify the single fault of straight pipe and bend pipe,but it is difficult to effectively and accurately identify and diagnose the multiple faults of single pipe(mixed fault).At the same time,the lossless Kalman filter is used to reduce the noise of the collected experimental vibration data,and remarkable results are obtained.Finally,by inputting the vibration data after noise reduction of lossless Kalman filter into a variety of excellent learning models,the accuracy of each model for hybrid fault diagnosis of aviation hydraulic pipelines is calculated through analysis.By comparing BN-1DCNN,original CNN model,BP neural network,SVM model,XGBoost model and random forest model in terms of model stability,running time and accuracy,it can be seen that the accuracy of diagnosis and recognition of mixed faults of aviation hydraulic pipelines by BN-1DCNN model reaches the highest99.9%,and the running time of the model is the shortest 5 minutes.It is enough to prove that compared with other models,the designed BN-1DCNN model has a great breakthrough in stability,running time,accuracy and other aspects.It can accurately identify and diagnose the mixed faults of aviation hydraulic pipelines,and has a certain reliability.It provides scientific data reference for intelligent diagnosis of aviation hydraulic pipeline fault.