Research Of Electronic Controlled Engine Fault Diagnosis Based On Artificial Neural Networks

Automobile exhaust emissions not only pollute the atmosphere, but also affect human health,with the improvement of human living standards, the use of automobiles is increasing year by year,which makes the vehicle emissions increase, and increasing air pollution, therefore, it is necessary to do the vehicle validity fault diagnosis.From the traditional mechanical to mechatronics development, the proportion of electric technology in the automobile is higher and higher, making the electronic controlled engine structure more complex by the improvement of engine electronic controlled system technology, no doubt making the fault diagnosis more difficult. In modern society, although the technical level of the vehicle maintenance personnel is generally improved,they often can not refer to the fault location when they find there are something wrong with the exhaust data,which prolonging the time of vehicle maintenance,and indirectly making automobile emissions increase. In order to reduce the air pollution of cars, proposing a fault diagnosis method of the electronic controlled engine based on neural network.Neural network can imitate the function of human brain to deal with the collected data,and hasing many kinds of diagnosis model,this paper uses its BP neural network 、 Elman neural network and SOM neural network.Collecting the training data before network diagnosis, this paper takes Beijing Hyundai Elantra car as the experimental object, keeping the engine at idle speed condition, setting up some fault assumption for the electronic controlled system of engine,collecting engine fault data by automobile exhaust gas analyzer and fault diagnosis instrument and other tools,and making the collected data normalized treatment,then doing fault diagnosis study.(1) In order to ensure the reliability of the fault diagnosis,the test data and the validation data of all training methods are the same.Using the fastest descent learning method and LM method to do fault diagnosis, the test accuracy of the fastest descent learning method is 38%, the test accuracy of the LM method is 99%,so the learning and training effect of LM method is good,verifying the LM method, and its verification accuracy is 92%.Using Elman neural network to do fault diagnosis, the test accuracy of Elman neural network is 97%,and its verification accuracy is83%,although the result of the Elman network is better than the tradition BP method,the LM method of BP network is better than Elman neural network.(2) Using SOM network and SPSS to do fault diagnosis of electronic controlled engine, the test accuracy of SOM neural network is 83%,and its verification accuracy is 81%. The test accuracy of SPSS is 68%,and its verification accuracy is 69%,so SOM network is superior to spss.(3) Finally,the comparison of BP network、Elman network、SOM network and SPSS showes that the LM method of BP network is the best method in the diagnosis model of three kinds of neural network, at the same time,the diagnosis effect of neural network is much higher than spss.In a word, the research of electronic controlled engine fault diagnosis based on artificial neural network has far-reaching significance.