The Study On Diesel Engine Fault Diagnosis Based On Information Fusion

The diesel engine fault diagnosis has become one research focus at home and abroad. There is much available information in fault diagnosis. Only to take full advantage of the useful information to diagnose the fault of the equipment can we improve the accuracy and reliability of fault diagnosis.In this paper the research status of the diesel engine fault diagnosis, and the necessity and feasibility applying information fusion to the diesel engine fault diagnosis are discussed. In addition, the vibration signals of diesel engine are studied to lay the foundation for the experiment.In order to reflect the operation state of the diesel engine in many ways, first carry out the characteristic class information fusion to diesel engine vibration signals with neural network method. The principle and the algorithm of BP neural network and RBF neural network are studied. Hybrid hierarchy genetic algorithm with hierarchical genetic coding scheme is introduced, by which the hidden structure, hidden center, the output value of linear weights width of RBF neural network are optimized. Using the three methods for fault diagnosis and a simulation in Matlab is carried out to compare the results of the diagnosis, which verifies the high speed of training and high accuracy superiority of the RBF neural network optimized by genetic algorithm.In addition, in order to avoid the problem that a single sensor information is not accurate, genetic RBF neural network and D-S evidence theory are combined, and the results of the characteristic class fusion are constructed to mass function, to diagnose the fault of the diesel engine in the decision level, for the vibration signals and pressure signals, will greatly improve the diagnostic reliability of the results.Finally, as the evidences are simply combined when the information is fused, the weighted evidence theory is presented. If many pieces of evidence are combined, the amount of conflict between evidences is at first evaluated by both evidence distance and conflicting belief, and every piece of evidence is given a weight coefficient according to its amount of conflict with the others. In addition, the belief function of each piece of evidence based on its weight coefficient is modified, the result of the experiment shows the effectiveness to the conflict evidences and the improvement of the diagnostic accuracy of the results.