Fault Identification of Vehicle Automatic Transmission based on Sparse Autoencoder and Support Vector Machine

Abstract

Support vector machine(SVM) got a good classification ability, but the recognition accuracy was easily affected by the value of the kernel parameters. Aiming at this problem, sparse autoencoder(SAE) has its unique advantages in dealing with complex structured data, so the combination of sparse autoencoder and support vector machine(SAE+SVM) was proposed on the fault identification of vehical automatic transmission. Firstly, eight indicators such as engine speed, throttle opening, water temperature and so on are collected from acquisition automobile automatic transmission under 3 running conditions. The data was used as input dataset of the sparse autoencoding model to extract the features. Then the features was used for the fault classification and identification based on support vector machine. Compared with using support vector machine only, the experiment results showed that the recognition accuracy based on the combination of sparse autoencoder and support vector machine(SAE+SVM) was less affected by the value of the kernel parameters and got better recognition accuracy. So the combination of sparse autoencoder and support vector machine can be better used in the real-time fault identification and diagnosis of automatic transmission.