Development of an Artificial Neural Network Based Fault Diagnostic System of an Electric Car

Abstract

The objective of this work is to develop a fault diagnostic system of an electric car based on artificial neural networks (ANN). Data from an on-board data acquisition system capable of measuring a number of parameters during the electric car operation are used to train an artificial neural network. The car’s monitoring system using the computational power of modern portable personal computers, user-friendly data input and output, and fullscreen editor capabilities is used for fault diagnosis. The ANN was trained to predict the temperature of the two motors of the electric car in order to detect any fault. The training data were learned by the ANN with an excellent accuracy. The results obtained for the training set are such that they yield coefficients of multiple determination (R2-values) equal to 0.9912 and 0.9939 corresponding to the values of the temperatures of the two motors respectively. Completely unknown data were then used for validation of the network. The correlation coefficients obtained in this case were equal to 0.954 and 0.987 for the temperature of the two motors respectively, which are very satisfactory. The fault diagnostic system developed compares the measured and predicted temperatures from the two motors and gives an “error” when a difference greater than a user defined tolerance is obtained. A polling routine was developed which sums-up the error signals and only gives a “fault” when 10 consecutive error reading are recorded. In this way false error conditions, which might arise from erroneous data recorded from the thermocouples and/or from wrong predictions of the network, are avoided