Intelligent Fault Diagnosis Of Diesel Engine Turbocharging System

Diesel Turbo System has been a failure-prone system thereby affecting the entireengine work. Only when a fault has occurred, the traditional fault diagnosis method can playa role. Obviously, this method can not satisfy people’s demands for fault diagnosis in today’shigh-tech development. With the development of artificial intelligence, people graduallytransfer the idea of solving problem on neural networks which has a good classificationperformance with learning ability. Neural network-based applications have achieved somesuccess to a certain extent.Firstly this paper attempts to use RBF neural networks, who has the best approximationfeature and no local minimum problem. But RBF neural networks can not afford to explaintheir reasoning basis. The networks can not ask the user to make the necessary inquiries, andwhen data is not sufficient, the neural network will not work. The fuzzy inference systembases on the "if-then", whose reasoning works easily be understood. Therefore, the T-Sfuzzy inference system and RBF neural network compose together fuzzy neural networks.However, the fuzzy system generates huge rules resulting in redundant network structure,thus affecting the training accuracy and time. In order to reduce the number of fuzzy rules,we consider clustering algorithm to cluster sample. In the proof of T-S fuzzy inferencesystem and RBF neural network based on the principle of equivalence, they form acluster-based T-SRBF fuzzy network. In the clustering algorithm selection, Firstly withsubtractive clustering we can determine the number of clusters, and then Fuzzy C-Meansclustering we can determine the center vector of network. The T-SRBF neural network withthe clustering algorithm method can automatically generate and adjust membershipfunctions and the number of fuzzy rules, effectively reducing the number of fuzzy rules.By analyzing the failure of turbo system to extract the sample of failure and then byusing matlab to compare the results of RBF neural network, T-SRBF neural network andbased clustering T-SRBF neural network, the last one has the best performance in terms ofaccuracy and training time. the network and expert system constitute the fault diagnosissystem of the turbo, and then we use java and matlab programming of the system in orderthat the theoretical results will be applied to practice.