Research On Multi-Source Intelligent Fault Diagnosis Method Of Sea Water Pump Under The Influence Of Multiple Factors

Intelligent fault diagnosis is the prototype of the future ship’s intelligent operation and maintenance system and "worry-free management".Fault diagnosis and conditionbased maintenance are performed by monitoring the data of equipment sensors.The sea water pump is an important equipment in the ship’s auxiliary machinery,which needs to be monitored and diagnosed in real time to ensure the safe navigation of the ship.The marine sea water pump equipment was taken as the research object.The fault characteristic mechanism of sea water pump was studied,and the fault diagnosis model based on "end-to-end" deep neural network was established by using the multi-source of the equipment.The anti-noise and generalization of the model under multi-factors such as limited data,noise interference and complex conditions were analyzed.The main work is as follows:(1)Fault modeling and mechanism research about sea water pump system were carried out.For several common faults,the simulation results are compared with theoretical results to provide a reliable theoretical basis for the fault detection of the sea water pump system.Aiming at the low efficiency of mechanical weak failure detection under the background of ship noise,a method for detecting weak characteristics of stochastic resonance of shapeable monostable structure was proposed.Taking the depth of the well,the radius of the well and the steepness of the well wall as parameters.The refined model was constructed to ensure the matching of the potential structure and the fault characteristics.The weighted kurtosis index was used as a measure,the effective detection and qualitative analysis of the fault could be realized.(2)Take the vibration and shock signal as the starting point,the convolutional neural network(WDCNN)fault diagnosis model with a wide convolution kernel in the first layer was proposed.Experiments show that the recognition rate can reach more than 97%.The Siamese neural network model based on WDCNN was proposed to solve the problem of limited training data.By using sample pairs and shared weights to learn to distinguish the similarity of fault classes.The impact of noise interference and variable operating conditions on model performance were studied.Case studies show that the Siamese neural network method was more effective for fault diagnosis when the amount of data was limited.(3)Aiming at the problem of low diagnostic accuracy caused by variable conditions in the previous chapter,the multi-scale convolution residual network(MRL-CNN)fault diagnosis model was established.Take the traction motor of the sea water pump system as the object,and extract the fault features in multiple scale spaces.And the residual network was combined to improve the stability and expressive ability of the network.The results showed that the classification performance of this model was the best.The average recognition rate reached more than 95% under variable working conditions.(4)In order to further improve the diagnostic performance of the sea water pump,the anti-noise performance and generalization performance of sea water pump fault diagnosis under the influence of multiple factors such as noise,load,speed and so on were discussed.A Gated Dilated Capsule Neural Network(GDCNN)model was proposed.This method expanded the receptive field by dilated convolution,and combined the LSTM gated structure and the capsule network to better retain the spatial position information of the signal sequence.The results proved that the GDCNN model has a high recognition rate under noise interference.The results of Example II showed that the generalization performance of the M-RL-CNN model is the best.