Fault Diagnosis And Trend Prediction Of Ship Slow Speed Diesel Engine Based On Data Driven Methods

The performance of slow speed diesel engine as the ship plant affect the whole shipping.The failure of the diesel engine will lead to economic losses caused by ship idles and seriously human casualties.Therefore,it is significative for marine diesel engine to make fault diagnosis and trend prediction.In this paper,marine slow speed diesel as the research object,the data is acquired by the sensors installed in the diesel engine based on a number of bulk cargo shipping.The fault diagnosis and trend prediction of the spare parts of the marine slow speed diesel engine,provide the crew with the running status and health of the diesel engine judging the safety of the ship.Considering the data acquired by various sensors and uploaded to the database,different force majeure factors and sensor noise may cause the loss of the data.Therefore,it is necessary to make data preprocess,which lays a good foundation for the next fault diagnosis and trend prediction.Secondly,for the marine diesel engine fault diagnosis.The data from the whole range of a bulk is selected to analyze and the speed interval is divided as the working condition.Then the host axial vibration,the host scavenging temperature and the host exhaust temperature are selected to build BP neural network model and ELMAN neural network model from the training sample,and in the measured sample the host exhaust temperature is predicted by the BP neural network model and ELMAN neural network model.The purposed of fault diagnosis is achieved by the comparison between predicted temperature and actual measured temperature.Further,multiple models are selected for trend prediction based on discrete wavelet transform for the single parameter from the marine diesel engine.These models are auto regressive,gray model,BP neural network and RBF neural network.In this chapter,the middle bearing temperature is as the example.The historical data of the each parameter in the same working condition fits the curve and is predicted by these multiple models.And these models are compared from the quickness and accuracy.Finally,the graphical user interface based on Matlab realizes the modularization of the content of the previous research and the corresponding requirements of the project cooperation,which includes the module of data dump and visualization,fault diagnosis module and data trend prediction module.The form of the modular integration algorithm and the show of platform simplify the data input and the call of the algorithm,so that these will facilitate the staff of ship based and shored based view and operation.