Research On Fault Diagnosis Of Marine Diesel Engine Based On Kernel-based Learning Theory

Marine diesel engine is the key equipment of ship power plant. If it does not work well,the operation of marine will be affected. And also, the economic loss and damage of otherkey equipments, even the personal safety will be affected. The condition monitoring and faultdiagnosis for marine diesel engine are helpful to find and eliminate the faults of marine dieselengine promptly and effectively. It has important significance towards improving the securityand reliability of marine diesel engine, lowing equipment maintenance expenses, reducingeconomic loss and avoiding major accidents.Marine diesel engine is a typical complex system. Its structure and operational principleresult the complexity of fault symptom. The relationship between fault reasons of marinediesel engine and characteristic parameters is extremely complicated and nonlinear. Therelationship among characteristic parameters is also strong coupling and nonlinear. Therefore,a nonlinear method should be adopted for the condition monitoring and fault diagnosis ofmarine diesel engine.On the basis of summing up and drawing the previous research results, combined withthe unique advantages in solving nonlinear problems of kernel-based learning theory, thisdissertation researches the condition monitoring and fault diagnosis technology for the marinediesel engine in detail. The research contents and contributions of this dissertation are statedas follows:1. Using the advantage of kernel component analysis for nonlinear monitoring, a newcondition monitoring for the fuel system in marine diesel engine is proposed. Firstly, by thekernel principal component analysis of the normal sampling data set of fuel system andcalculating the monitoring statistics and their control limits, a condition monitoring model isbuilt. Secondly, the condition monitoring model is used to detect the fault of fuel system inmarine diesel engine. The results of condition monitoring in the fuel system of certain typemarine diesel engine verify the effectiveness of this method.2. Combining with the advantages of kernel principal component analysis for extractingnonlinear feature and the higher recognition rate of support vector machine, a new fault diagnosis method for the fuel injection system in marine diesel engine is proposed. Firstly, bythe kernel principal component analysis of training sample data set, the nonlinear principalcomponents are extracted. The nonlinear principal components can reflect the fault state offuel injection system in marine diesel engine. Secondly, the support vector machines aretrained by using the extracted nonlinear principal components, and a fault diagnosis model isbuilt. Finally, the unknown fault samples of fuel injection system in marine diesel engine arediagnosed by the fault diagnosis model. The results of fault diagnosis for fuel injectionsystem in a certain type marine diesel engine show that the several common faults in fuelinjection system can be identified accurately by using the method.3. According to the fuzzy and nonlinear features of turbocharged system in marine dieselengine, a fault diagnosis method based on kernel-based fuzzy clustering is proposed. Firstly,kernel-based fuzzy clustering is applied to classify the faults of historical data set in order toget the clustering centers. And a fault diagnosis model for turbocharged system in marinediesel engine is built. Secondly, the unknown fault samples of turbocharged system in marinediesel engine are diagnosed by the fault diagnosis model. The results of fault diagnosis forturbocharged system in a certain type marine diesel engine show that the several commonfaults in turbocharged system can be identified accurately. By introducing the concept offuzzy logic, the diagnosis results of this method are more close to practice and objective.4. According to the differences between intelligent diesel engine and traditional dieselengine, a fault diagnosis method for intelligent diesel engine based on multiclass kernel fisherdiscriminant analysis is proposed. Kernel fisher discriminant analysis has many advantagessuch as high precision of discriminant, short computing time, etc. The parameters aredetermined by using leave-one-out cross-validation method. The results of fault diagnosis fora certain type marine diesel engine show that the method has the advantages such as lowcomputational complexity, short time consuming, high diagnosis accuracy, etc. Therefore,this method is very suitable for the real-time fault diagnosis of intelligent marine dieselengine.This dissertation mainly researches the kernel-based learning theory, proposes andimproves the kernel-based classification method, and builds a series of condition monitoringand fault diagnosis models for the subsystems in marine diesel engine. These methods allhave their own advantages and can satisfy the requirements of fault diagnosis for the different subsystems.