Accelerated Parameter Optimization of Granular Models of Battery Manufacturing

Mesoscale particle simulations play a crucial role in investigating the manufacturing processes of electrochemical energy devices. They enable physical insights into the processes while building datasets to train AI surrogates for electrode microstructure and manufacturing optimization. Hence, the automation of parameter optimization for mesoscale simulations remains imperative. Optimization algorithms that do not require prior knowledge or experience simulating a specific chemistry/system are the ideal solution. One such optimizer is the Particle Swarm Optimizer (PSO). However, in the classical serial implementation, the evaluation time varies across different parameter sets. This means the optimizer has to wait for all simulations to finish at each iteration, which can lead to unnecessary use of computational resources. This study presents an optimization framework that implements Asynchronous PSO (APSO), enabling more efficient resource use to find the global optimum, i.e., parameter values that yield model outputs that agree with experiments. The effect of different APSO configurations on convergence was tested. The framework was then used to find the optimal parameters for two different mesoscale Discrete Element Method simulations. The data acquired from the automated search were then analyzed to study the models' behavior within the chosen parameter search space.