A Condition Monitoring System for Electric Vehicle Batteries Based on a Convolutional Neural Network Using Thermal Image

Abstract

A new monitoring technique has been developed to evaluate the capacity and performance of Lithium-ion batteries batteries by utilizing two convolutional neural networks (CNNs) models, Deep convolutional neural network (DnCNN) and CNN with BFGS quasi-Newton optimization. The system utilizes thermal images of lithium-ion batteries as input for training and testing. DnCNN model is utilised to accurately calculate battery capacity and performance, and the performance is evaluated using mean squared error (MSE) and PSNR. The CNN-based training method employs the BFGS quasi-Newton algorithm to measure battery capacity accurately by evaluating the mean squared error (MSE) and regression results. The proposed condition monitoring system using thermal imaging and CNN models, specifically the CNN- BFGS quasi-Newton algorithm model, can accurately detect battery capacity with an accuracy rate of 98.5%, compared to the DnCNN model with an accuracy rate of 96.7%. The proposed system can address the critical issue of battery capacity and degradation in EVs, providing a more sustainable and efficient alternative for real-time applications.