Research On Fault Diagnosis Of Piston Pump Based On VMD-CWT And CBAM-ResNet

The piston pump of switch machine is the core component of the transmission system of railway turnout electro-hydraulic switch machine.The reliability of its working state is the key to ensure the high speed and stable operation of the train.If the fault of the switch machine piston pump fails to be found in time during operation,it will affect the normal operation of the train,endanger the safety of the train,and cause a catastrophic accident.Therefore,it is very important to diagnose the fault of the switch machine piston pump.Taking the piston pump of switch machine as the research object,a fault diagnosis method of piston pump based on VMDCWT and CBAM-Res Net is studied by combining vibration signal processing technology with deep learning technology,which expands the fault information,realizes the adaptive extraction of fault features,and accurately identifies different fault types of piston pump.The primary topics of the study include the following three aspects :(1)A time-frequency feature extraction method of piston pump vibration signal based on VMD-CWT is proposed.Firstly,the original piston pump vibration signal is decomposed by variational mode decomposition(VMD)algorithm to obtain multiple intrinsic mode components.Each component contains different frequency information in the original signal,which greatly enriches the fault information.Secondly,each component is arranged from small to large according to the center frequency,and a new fault signal is formed after splicing and reconstruction.The coupling signal in the same time domain is decoupled into different time domains to enhance the fault information in the signal.Finally,the continuous wavelet transform(CWT)is used to transform the new fault signal into a two-dimensional timefrequency feature map to visualize the fault information in the signal,which is easier for subsequent fault diagnosis processing.(2)The residual neural network(ResNet)extracts the deep feature information in the CWT time-frequency feature map of the piston pump,and avoids the possible gradient disappearance phenomenon of the model during training through multiple residual connections.Then,the attention mechanism(CBAM)is introduced into the residual neural network to adaptively weight the multi-channel features and spatial features input by the network layer,and more accurately extract the fault features in the time-frequency feature map of the highdimensional piston pump to improve the accuracy of the fault diagnosis model and reduce the number of network training.(3)The CBAM-Res Net fault diagnosis model is used to identify the fault data set of the switch machine plunger pump.Firstly,the piston pump fault data set is divided into a training set and a test set.The VMD-CWT method is used to generate the corresponding twodimensional time-frequency feature images.The training set samples are input into the CBAMRes Net diagnostic model for training,and the test set samples are used for testing,achieving an accuracy of more than 99 %.Then,compared with Alex Net model,Res Net18 model and Alex Net(CBAM-Alex Net)model with attention mechanism,the comparison results verify the superiority and robustness of the model.At the same time,the performance of the CBAMRes Net fault diagnosis model is verified by using the centrifugal pump vibration data set of the Sant Longowal Institute of Engineering Technology.The experimental results fully prove that the model has strong generalization.