Machine learning assisted multiscale modeling of composite phase change materials for Li-ion batteries’ thermal management

• Convolutional neural networks (CNNs) are trained to predict effective thermal properties of composite phase change materials. Finite element simulations used prepare the dataset for CNN training. Heat generations during a battery operation simulated. can accurately evaluate pack's management system. In this work, we develop combined convolutional and finite method (FEM) examine materials (CPCMs) consisting paraffin copper foam. approach, first CPCM microstructures modeled using FEM next image with corresponding is created. The subsequently train test CNN's performance, which then compared performance popular network architecture classification tasks. predicted employed define material pack. heat generation electrochemical response Li-ion cell charging/discharging simulated by developing Newman's model. Thermal achieved latent paraffin, foam enhancing conductivity. multiscale model finally developed investigate effectiveness these models estimated Our results confirm that on basis system an excellent accuracy in comparison original models.