This work aims to accurately map the internal temperature during cell operation using various implementation ideas. The investigations are carried out on modified cylindrical cells, in the centre cavity of which a temperature sensor has been integrated by opening the cell in a glove box and resealing it using a two-component adhesive, which serves as a reference point for the following investigations. The first step is to model the thermal parameters, specifically the heat capacity and thermal resistance, in order to understand the cell’s fundamental thermal behaviour and investigate the differences between internal and surface temperatures. Based on this, two approaches are pursued to determine or derive the internal temperature.
The first approach of this poster is based on a data-driven method in which a neural network is used to simultaneously estimate the internal temperature and the state of charge (SOC). The network is based on multi-task learning, which solves several similar tasks simultaneously and can thus enhance generalisation. The network is trained using a set of different driving cycles that represent various scenarios to track the temperature gradients in the cell during operation under different ambient temperature conditions. The second approach uses the correlation between impedance and temperature to determine the internal temperature of the cell. To do this, the impedance response to a current pulse is considered. GITT in charge and discharge direction is used to set up a reference lookup table for the required conditions. Finally, both methods are compared with each other, using the behaviour of the respective approaches during simulated driving cycles as a basis for evaluation.