Research On Fault Diagnosis Of Diesel Engine Using Pattern Recognition Method

Condition monitoring and fault diagnosis of diesel engine is crucial to improve the reliability of diesel engine power plant, and to meet the requirements of modern predictive maintenance and automatization. However, diesel engine is a typical reciprocating machine, its monitoring and diagnosis are very difficult because of its complex structure. Developing a advanced method of extracting the feature parameters of surface vibration signals and instantaneous speed of diesel engine, investigating the application of fuzzy clustering and neural network based pattern recognition, establishing an ANFIS system for fault diagnosis of diesel engine have been done in this dissertation.A lot of artificial diesel engine faults have been simulated and investigated. Surface vibration signals of diesel engine under 7 different conditions have been analyzed in time domain and frequency domain. Non-dimensional feature parameters have been extracted. Based on the study of the time-frequency distribution theory, PMH analysis of vibration signals has been carried out. Time-frequency features of the surface vibration signal of cylinder head and cylinder block have been got, corresponding to exhaust-valve leakage and piston-ring damage. Accordingly, feature parameters in time and frequency domain have been extracted, and combined with time-frequency feature for identifying the diesel engine condition.A simplified diesel instantaneous speed analyzing model has been developed based on the investigation of dynamic property of crankshaft flywheel system of diesel engine. Three feature parameters, which can be used to detect diesel engine's faults and locate fault cylinders, have been proposed, based on the simulation and signal analyses of the diesel instantaneous speed.Basic principles of fuzzy C mean value clustering has been studied, and its implementation algorithm has been developed. Effectiveness of the selected parameters on clustering has been investigated. The fuzzy clustering method is then applied to classify the diesel vibration feature vector, the practical application demonstrates that proper by selected feature vector can be used to identify the diesel engine conditions effectively. Moreover, the self-organization neural network with learning vector quantization (LVQ) algorithm has been used to perform pattern-recognition. This neural network, after training, can identify the different diesel fault conditions accurately.The fuzzy inference theory and algorithm of adaptive network-based fuzzy inference system have been studied, the ANFIS, based on which, has been developed for fault diagnosis of diesel engines. The ANFIS, after training, testing and examining the selected parameter vector samples, can accurately identify single fault and compound faults, such as high-pressure oil-pipe leakage, exhaust valve leakage, andpiston-ring damage. It is verified to be very practically useful.An effective strategy for fault diagnosis of diesel engine is also proposed in this dissertation. That is, the feature parameters of instantaneous speed are used to identify existence and location of fault, in the case that cylinder explosive pressures can't be measured directly. Then, the parameter vector extracted from the diesel surface vibration signals is used to find out the fault type or causes by means of the pattern-recognition method or ANFIS.