Research On Early Failure Prediction Method Of Aviation Hydraulic Pipeline Based On Deep Learning

Since the introduction of Prognostics and Health Management,various fields have begun to conduct "physical examination" on parts and various mechanical equipment.While ensuring the normal operation of the equipment,faults can be discovered and resolved in time.Especially in the key parts of the aviation field,the diagnosis and prediction of faults has become the focus of attention.Aviation hydraulic pipelines are one of the most important parts of aero-engine.Vibration data,combined with deep feature learning and complex nonlinear function approximation capabilities of deep learning,can accurately identify and predict early failures of aviation hydraulic pipelines.First,four methods of vibration signal data processing are introduced.Through the comparison and analysis of the four decomposition methods,the optimal adaptive white noise method is selected,and combine the signal obtained by decomposition with the deep residual network in deep learning.According to the characteristics of the vibration signal of the aviation hydraulic pipeline,the Res Net network structure is built,and the early fault diagnosis and prediction of the aviation hydraulic pipeline is designed Model,and select the corresponding activation function and loss function.Secondly,design an experimental plan for the early failure of aviation hydraulic pipelines,and conduct fluid-solid coupling experiments on a total of 14 groups of working conditions of hydraulic straight pipes and elbows that are normal and free of failures and two different types,different positions,and different faults.Collect the vibration signal,compare the vibration amplitude of different types,different faults and different locations where the fault occurs,and summarize the relationship between the amplitude of the hydraulic pipeline and whether the fault occurs and the location of the fault,through the time domain diagram and The frequency domain diagram found that the vibration amplitude of the early fault pipeline is slightly larger than that of the non-faulty pipeline,but it is not obvious,and the use of deep residual network for pipeline fault identification and prediction is better.Finally,using Matlab and Python as platforms,the collected vibration data was unified and standardized,and 14 groups of data were selected for CEEMDAN decomposition.Each group obtained the required 3 IMF components and imported them into the convolutional network and the residual network.In the model,four different network models are combined.By calculating the accuracy of model classification and recognition in the confusion matrix obtained from the network model training,the superiority of the CEEMDAN-Res Net model is verified,and the number of iterations and accuracy,The loss rate curve,when the CEEMDAN-Res Net model is used to perform fault prediction training iterations to 1200 times,the accuracy rate will reach 99.5% and continue to be stable,which verifies that the established CEEMDAN-Res Net model can identify and identify early faults in aviation hydraulic pipelines.The accuracy and feasibility of the forecast.