Research On Vibration Fault Detection Of Water Pump Bearing Based On CNN-LSTM

As the core equipment of water conservancy hubs,water pump units play an indispensable role in water regulation,irrigation,and drainage.Bearings are crucial components for the operation of water pumps and are also one of the components that are prone to malfunctions.Therefore,timely and effective diagnosis of the working status of water pump bearings has important engineering significance.At present,most of the models for vibration fault detection of water pump bearings are based on traditional neural network models,and this traditional method shows great limitations in the face of increasingly complex and diverse pump fault signals.Therefore,aiming at the problem of vibration fault detection of water pump bearings,this paper introduces a deep learning network,combines the convolutional neural network(CNN)and the Long Short-Term Memory(LSTM),and carries out parameter adjustment and network improvement,and proposes a 2D-CNN-LSTM vibration fault detection model of water pump bearings based on deep learning.For the vibration fault detection of water pump bearings,the following research has been carried out:(1)This article constructs a dataset,and conducts preprocessing work on bearing signals.Four time-frequency conversion methods were use(?)convert time-domain signals into timefrequency images,and the results were compared.Finally,the time-frequency image of wavelet transform was used as the preprocessing method for time-frequency input.(2)Introducing a deep learning network,a CNN-LSTM water pump bearing fault detection model was proposed,and the pre processed vibration signals were input.The experimental results show that the 2D-CNN-LSTM model based on time-frequency maps has the highest fault diagnosis accuracy.(3)This article has carried out parameter optimization and network improvement work.Compared with other network models,the 2D-CNN-LSTM model proposed in this paper has the highest accuracy and fastest convergence speed,providing an effective method for fault detection of water pump bearings.(4)In order to test the generalization of the network proposed in this paper,transfer learning was carried out.Under the premise of constant parameters,the data set in the target domain is used to directly input the network training,and the results show that the network proposed in this paper can also obtain good recognition accuracy in the variable condition transfer learning.(5)A system platform for analyzing and diagnosing the fault data of rolling bearings in water pump units has been designed.This platform can complete vibration signal reading,data signal processing time-frequency visualization,and intelligent fault diagnosis functions,and has good practical value.