Physics-Informed Deep Learning for Modelling Particle Aggregation and Breakage Processes

Particle aggregation and breakage phenomena are widely found in various industries such as chemical, agricultural and pharmaceutical processes. In this study, a physics-informed neural network is developed for solving both the forward and inverse problems of particle aggregation and breakage processes. In this method, the population balance equation is directly embedded in the loss function of a neural network so that the network can be trained efficiently and fulfil physical constraints. For the forward problems, solutions of population balance equations are obtained through the optimization of the neural network where the predictions well match the analytical solutions. In the inverse modelling, the data-driven discovery of model parameters of population balance equations is investigated. The sensitivity regarding the selection of different neural network structures is also investigated. The developed population balance equations embedded with neural network approach is promising for solving inverse problems of particle aggregation and breakage processes with noisy observation data.