Research On Diagnosis Technology Of Diesel Engine Based On Support Vector Machine

With the development of marine diesel engine, the intelligent technology has been the main research direction, of which one of the three most core modules is engine condition monitoring and diagnosis. However, because of marine engine’s working condition, obviously we have to face a big problem that is terribly lack of typical fault samples. However, the intelligent diagnosis technology nowadays, such as neural network method, etc., are all based on a large amount of samples training, which badly prohibits the development of fault intelligent diagnosis obviously. As a new kind of effective classification method which is based on Statistical Learning Theory and developed in recent years, Support Vector Machine (SVM) has an excellent performace in solving fault classification problem in the situation of high cost with small samples. Therefore, it is necessary and significant to choose the application of SVM in fault diagnosis of marine diesel engine as a research direction.On the basis of analyzing SVM theory and fault test simulated, the paper carries out research on fault diagnosis technology for diesel engine based on SVM. The main research achievements are as follows:1. The basic theory of SVM, muti-classification model and model parameters optimization algorithm based on Particle Swarm Optimization (PSO) are discussed to lay the theoretical foundation for the application of SVM in fault diagnosis of diesel engine. Compared with traditional diagnostic method based on standard values and neural network method, the feasibility and superiority applying SVM in fault diagnosis of diesel engine with small samples is disscussed and the application step is pointed out.2. Diesel engine test bench is established and the test system including hardware and software is designed. Then, faults such as abnormal valve clearance, valve leakage and misfire, are simulated on test bench and instantaneous angular speed (IAS) signal and acoustic emission (AE) signal are measured. All these provide a reliable data sources and an analysis platform for the research on fault diagnosis technology for diesel engine.3. IAS fault diagnosis technology based on SVM is researched. Applying waveform analysis method, four IAS fault sensitive characteristic parameters, including the maximum change value of fluctuation ratio, the power fluctuation ratio, the working frequency and firing frequency ratio, the octave frequency and firing frequency ratio, are extracted as the feature vector. Then fault diagnosis model is established based on SVM. Test results show that the model can effectively diagnose misfire and underpower (valve leakage) with best classification performance under high load.4. AE fault diagnosis technology based on SVM is researched. Applying parameter analysis method, four time-domain fault sensitive characteristic parameters, including RMS of work cycle, RMS of intake valve closing, RMS of exhaust valve closing, RMS of combustion, are extracted. Applying wavelet packet analysis method, the frequency band energy ratio of S2, S3and S4after Daubechies8decomposition with level3to AE signal, are extracted. The seven parameters compose the feature vector. Then fault diagnosis model is established based on SVM. Test results show that the model can effectively diagnose abnormal valve clearance, valve leakage (even specific leakage level) and misfire with stable classification performance under different loads.5. Multi-method monitoring and diagnosis strategy of marine diesel engine is researched. Thermal parameter method, IAS method, indicator diagram of cylinder pressure, AE method, piston ring monitoring method, shaft vibration method, shaft power method and ship operation performance monitoring method are integrated. Application scope of each method is discussed, comprehensive monitoring and diagnosis model is established and multi-method strategy is researched. Engineering application in some ferries of line3and line4has proved its validity.