THERMAL IMAGE-BASED BATTERY CELLS FAULT DETECTION IN ELECTRIC VEHICLES USING CNN MODEL

Abstract

This study compares and evaluates the performance of four Convolutional Neural Network (CNN) models, namely ResNet152, VGG19, DenseNet201, and Inceptionv3, for thermal images-based detection of faults in electric vehicle (EV) battery cells utilizing temperature. A dataset comprising thermal images of battery cells with various fault types and severities is collected and preprocessed for model training. Transfer learning is applied to train the CNN models using pre-trained weights on large-scale image datasets. The trained models are assessed using evaluation metrics that include precision, recall, F1-score, and accuracy, while their computational efficiency is evaluated in terms of inference time and memory usage. Results show promising performance for all four CNN models in detecting faults in battery cells. DenseNet201 achieves the highest accuracy, followed by ResNet152, VGG19, and Inceptionv3. Inceptionv3 demonstrates superior computational efficiency. These findings aid researchers and practitioners in selecting an appropriate CNN model for thermal image-based fault detection in EV battery cells, considering the balance between accuracy and computational efficiency.