Research On EHA Devices Fault Diagnosis And Prognostics Based On Deep Learning

Electro-Hydrostatic Actuator(EHA)has been widely used due to its strong anti-interference capability and high control efficiency.Affected by working conditions,EHA key devices inevitably fail,but the current fault diagnosis and maintenance measures are relatively lagging behind.Therefore,based on the Prognostics and Health Management(PHM)technology,the key to carry out the EHA intelligent fault diagnosis and prediction method of study,in order to further implement maintenance support capability of EHA to lay a certain theoretical basis.The main research contents are as follows:Firstly,determine the failure mode to study the EHA.From mechanical structure and working principle of EHA,through the analysis of the failure mode of EHA key components,and from the EHA is easy to implement fault simulation and the perspective of probability level,filter blockage,gas leakage,IGBT short circuit as a follow-up study of three failure modes.Secondly,the deep learning method for fault diagnosis and prediction is determined.According to the characteristics of the deep neural network architecture,combining with the nature of the neural network can approximate any continuous function,the analysis of four major in depth study of network model,the comparison advantage,defect and application of the model instance,Stacked Denoising AutoEncoder(SDAE)and Bi-directional Long Short-Term Memory(BiLSTM)two methods of deep learning as a follow-up research method in this paper.Thirdly,the fault diagnosis methods based on SDAE for different degrees of oil filter blockage and gas leakage in pressurized oil tank are studied.Taking EHA filter blockage,gas leakage faults of different degrees as the object,this paper analyzes the modeling process of filter blockage,gas leakage based on SDAE.By setting up an EHA experimental platform,simulation experiments were carried out on filter blockage and gas leakage of different degrees.According to the data set collected in the experiment,the influence of hyper-parameter of SDAE including the number of hidden layers,the optimizer and the number of neuron nodes in each hidden layer on the diagnostic performance was analyzed,and the optimal hyper-parameter of the model were determined.Finally,the SDAE model is compared with the traditional neural network method to verify that the SDAE method can be more accurately applied to the fault diagnosis of EHA with different degrees of filter blockage and gas leakage.Finally,a fault prediction method based on BiLSTM is proposed for EHA filter blockage and IGBT short circuit fault prediction.Based on BiLSTM model structure,the flow of filter blockage and IGBT short circuit fault prediction modeling was established.According to the IGBT accelerated aging experimental data set of NASA and the filter plugging simulation data set,the optimal prediction parameters of the model,such as time step and state vector,were determined.The results are compared with the typical time series prediction method to verify its applicability in the field of EHA key device prediction.