Neural network based computational model for estimation of heat generation in LiFePO4 pouch cells of different nominal capacities

Significant variance exists in the nominal capacity of lithium ion (Li-ion) pouch cells used for commercial electric vehicle battery packs. Accurate estimation of heat generation in such cells is critical for designing battery thermal management system. However, multi-physics models describing thermal behaviour of these cells are too complex whereas other numerical models discount the effect of cell capacity on heat generation. This paper proposes a new computational model based on artificial neural network (ANN) for estimating battery heat generation rate with cell nominal capacity as one of its key inputs along with ambient temperature, discharge rate and depth of discharge. A custom-designed calorimeter is utilised for experimentally generating the training dataset for the ANN. Problem of data scarcity is addressed analytically and virtual samples are produced via enthalpy formulation for battery heat generation. Subsequently, the model is trained using Levenberg-Marquardt algorithm. Results disclose that a three-layered feedforward ANN with one hidden layer having six neurons is optimum for this application. The architecture of the trained ANN for accurately simulating thermal behaviour of LiFePO4 pouch cells of the nominal capacities from 8 to 20 Ah under varied conditions is exemplified.