Parameter Identification of Multi-dimensional PBE Growth Model Based on PINN

This study focuses on the identification of multi-dimensional growth model parameters in continuous oscillatory baffle crystallizers using physics-informed neural networks (PINNs). Due to the fact that crystal growth parameters are varying with changing operational conditions, it is crucial to identify these parameters for better control results. The paper first employs a mechanistic model, the morphological and multi-dimensional Population Balance Model(PBM), to describe multi-dimensional crystallization processes. Subsequently, high-precision numerical solution algorithms are used to obtain crystal size distributions. A uniform design scheme is implemented to set crystal growth model parameters and generate training data for PINN. Additionally, the mechanistic model are incorporated into the loss function as a regularization term for network training. This approach guides the training process towards physically consistent solutions rather than purely data correlations. Finally, the PINN model is applied to the identification of growth model parameters in glutamic acid experiments, demonstrating its capability for parameter identification.