Utilizing multilayer perceptron for machine learning diagnosis in phase change material-based thermal management systems

Electric vehicles encounter significant challenges in colder climates due to reduced battery efficiency at low temperatures and increased electricity demand for cabin heating, which impacts vehicle propulsion. This study aims to address these challenges by implementing a thermal management system utilizing Phase Change Materials (PCMs) and validating the performance of a Multilayer Perceptron (MLP) model in predicting PCMs behavior and battery temperature distributions. The study employs an MLP model trained with 160 samples of diverse heat inputs, including pulsating, constant, wiener, discharging, and random temperatures. The model uses these temperatures as inputs and liquid fractions as target values. Performance evaluation is conducted using the MATLAB platform and is benchmarked against existing approaches, such as Long Short-term Memory (LSTM), spatiotemporal convolutional neural network (CNN), and pooled CNN-LSTM. The MLP model's accuracy in predicting PCMs phase transitions is validated by comparing predicted liquid fractions with numerically obtained values. Additionally, this study forecasts temperature distributions within a standard battery pack under various discharge scenarios, considering the performance of commercial lithium-ion batteries. The proposed MLP model demonstrates high efficacy, achieving a correlation of up to 0.999 and root mean squared error below 0.013 compared with numerical results.