Stochastic modeling of particle structures in spray fluidized bed agglomeration using methods from machine learning

Agglomeration is an industrially relevant process for the production of bulk materials in which the product properties depend on the morphology of the agglomerates, e.g., on the distribution of size and shape descriptors. Thus, accurate characterization and control of agglomerate morphologies is essential to ensure high and consistent product quality. This paper presents a pipeline for image-based inline agglomerate characterization and prediction of their time-dependent multivariate morphology distributions within a spray fluidized bed process with transparent glass beads as primary particles. The framework classifies observed objects in image data into three distinct morphological classes--primary particles, chain-like agglomerates and raspberry-like agglomerates--using various size and shape descriptors. To this end, a fast and robust random forest classifier is trained. Additionally, the fraction of primary particles belonging to each of these classes, either as individual primary particles or as part of a larger structure in the form of chain-like or raspberry-like agglomerates, is described using parametric regression functions. Finally, the temporal evolution of bivariate size and shape descriptor distributions of these classes is modeled using low-parametric regression functions and Archimedean copulas. This approach improves the understanding of agglomerate formation and allows the prediction of process kinetics, facilitating precise control over class fractions and morphology distributions.