Research Of Automobile Engine Fault Diagnosis Based On GA-BP Neural Network

The formation of engine fault is directly related to the mechanical condition and the combustion cylinder as well as the load situation, and it's difficult to find these fault by the regular service method for its'unpredictability what easily creates terrible accidents.For the great ability about identifying and associative memory, neural network can rapidly study and train multiple-input-output model what has complex as well as non-linear relation. It is the inevitable trend for developing of intelligent fault diagnosis to applied neural network combined with other intelligent algorithms to fault diagnosis equipment.Firstly, this paper made an in-depth study about the mechanism of excessive exhaust gas. With analysis of exhaust gas components'relative content under six different fault models, this paper established a non-linear relationship between fault reasons with the exhaust gas composition. Based on the theory that neural network which has three layers can simulate all the non-linear relationship with a high degree of precision, according to the perspective of pattern recognition, the question of fault diagnosis is translated into the question how to use neural network classifier to classify the six fault models.Comparing the BP neural network classifier with the RBF neural Network classifier for classification of multi-mode effects, found that BP neural network classifier can achieve higher precision and need less neurons.Against to the shortcoming that BP neural network easily falls into the local minima, raised an algorithms combination composed of genetic algorithms and BP algorithms to train the neural network.Used genetic algorithm which has great global searching capability to globally optimize the structure and the weights as well as the right threshold value of BP neural network, then used BP algorithm which is the error back-propagation algorithms to optimize the right threshold value.Finally, made the final optimal weights value as the threshold value of the ultimate fault diagnosis neural network, and built fault diagnosis optimization model based on GA-BP neural network. The results showed that, fault diagnosis optimization model had better effect, such as precision and adaptability, comparing to the BP neural network.This paper used MATLAB software to simulate the fault diagnosis model and analyzed the simulation result. With the genetic algorithm optimization result this paper built BP neural network with 4-9-3 structure, and then compared the simulation results of fault diagnosis model respectively based on BP neural network and GA-BP neural network. The results showed that, GA-BP neural network is superior to the BP neural network on the effect of fault diagnosis.BP neural network reach convergence in about 4,233 steps,and the average error for the testing sample is 0.2096 as well as the correctly diagnostic rates was 83.33%,while the GA-BP neural network reach convergence in about 458 steps, and the average error for the testing sample is 0.1123 as well as correctly diagnosis rates was 91.67%.