Research On The Method Of Fault Diagnosis Of Marine Power Equipment Based On Deep Convolutional Neural Networks

Health assessment and fault diagnosis are part of fault prediction and health management technology.By analyzing the status information of the equipment,the most appropriate evaluation method is selected to evaluate the health status of the equipment,then the fault is diagnosed and predicted,and scientific maintenance suggestions are given according to the assessment and prediction results.As the power equipment is an important equipment of the ship,it is very important to evaluate its health status and predict its faults,find out the hidden danger of the equipment in time and maintain it,so as to ensure the safe navigation of the ship.In this paper,the ship's power equipment was taken as the research object.Model of health evaluation was established,and the fault diagnosis model based on one-dimensional CNN and deep convolution neural network was established by using the impact vibration signal of the equipment.The adaptability of the model under the condition of noise interference and load mutation was analyzed and optimized.The main work is as follows:(1)This paper analyzed the main structure and common faults of the marine power plant system.The health evaluation index system of marine diesel engine,which is the main equipment of power system,was established.The optimization method of index system based on subjective and objective weighting method was proposed.The least square support vector machine(LSSVM)algorithm based on the improved gray wolf algorithm optimization(IGWO)was proposed to evaluate the health status of equipment.The case results showed that the average accuracy of the algorithm was 99.4%,which verified the effectiveness of the above index system and evaluation algorithm.(2)Due to reasons such as high experience for experts and low universality of traditional fault diagnosis method of feature extraction,dimensionality reduction and feature classification,an end-to-end fault diagnosis method was studied.Taking the rolling bearing fault of the main shaft of marine diesel engine as the research object,an intelligent diagnosis method based on the deep convolutional neural network was proposed.First,a one-dimensional CNN fault diagnosis model with only one convolution layer was proposed.Combined with data enhancement,the case results showed that the diagnosis accuracy of the model was more than 97%.(3)In view of the problem that noise interference and load change will reduce the performance of fault diagnosis model.We improved the network depth and convolution kernel size of the first layer,and proposed WB-DCNN model with large convolution kernel in the first layer and small convolution kernel in other layers.The Batch Normalization(BN)was introduced to improve the training speed of WB-DCNN.In order to further improve the domain adaptive ability of convolutional neural network,the residual module was introduced and the Res14-DCNN diagnosis model based on the residual module was proposed.In order to solve the problem that neural network model is difficult to analyze,visualization of neuron visualization and classification process were carried out.The example analysis showed that Res14-DCNN model has high anti noise performance and adaptive ability of variable load.(4)Aiming at the problem that the noise interference will not disappear,but the variable load problem often occurs,the self encoder was studied,and the De-noising Auto-Encoder based on the Deep Convolution neural network(DCDAE)was proposed to de-noising the data.The enhanced Res14-DCNN fault diagnosis method based on DCDAE was proposed.The example analysis showed that the method has a high recognition rate under the noise interference,and the fault diagnosis ability was better than other models in the variable load problem under the noise interference.