The Research Of Marine Diesel Engine Fault Diagnosis Based On SVM And Oil Monitoring

Marine diesel engines are the most important for the safety and reliability of ships. Due to the severe working condition and the complexity of their structures, the failure rate of engines is very high. Among the cause of failure, the percentage of the wear fault is about37.5%and it is the major cause to fail an engine. Therefore, the information getting from the oil monitoring based on the theory of tribology can directly reflect the system operating condition. Only by recognizing and analyzing the information of oil, it can be used to diagnose the fault of machines effectively. In this paper, Support Vector Machine is used to diagnose the wear fault of marine diesel engines, which is based on the traditional ferrography technology and spectral technology.Firstly, the weights of each index of spectral technology and ferrography technology are ascertained by AHP, and the weighted characteristic indexes of oil monitoring technology for the wear fault diagnosis are screened out based on the calculation of AHP, then the characteristic indexes set for wear fault diagnosis based on oil monitoring technology is established by combined the result of weighted characteristic indexes. The wear fault diagnosis of marine diesel engine can be done more precisely by the use of the characteristic indexes set.Secondly, multi-classification SVM model of spectral analysis and direct reading ferrography of marine diesel engine is established based on the characteristic indexes set of fault diagnosis. And Grid-search method and GA are used to compare the accuracy rate of different kernel functions and to optimize the parameters of SVM. Eventually, the wear pattern of marine diesel engine is recognized. Beside, SVM regression model, which is combined with time series analysis, is established. And Grid-search method and GA are used to optimize its parameters and dimension of time series. Eventually, the wear trend of marine diesel engine is predicted based on oil monitoring technology.Finally, characteristic indexes set of analytical ferrography is expanded by using quantitative characters instead of qualitative indicators in diagnosis table of ferrography. And on the basis of the set, wear pattern of marine diesel engine is recognized by the multi-classification SVM model of wear particle analysis which is optimized. Besides, some problems concerned with the recognition process are discussed in this paper.