A combinatorial tool for monitoring flocculation processes: Using non-invasive measurements and hybrid deep learning assisted modelling

Abstract

Abstract Particulate processes have a wide range of applications in many different industries, from wastewater treatment to the pharmaceutical industry. Despite their extensive applications, control and monitoring of chemical and biochemical processes that contain solid particles are challenging due to the lack of fundamental understanding of the process mechanism and the limited availability of real-time process data. In this study, a hybrid multiscale framework is introduced for flocculation processes as a particulate process, and it is validated against experimental data resulting from the flocculation of silica particles. The variations of the particle size distribution are imposed by varying the pH in different experimental batches. In this study, an integrated hybrid deep learning approach combining deep learning with first principles is implemented to predict the future state of the process. The first-principles model combines a population balance model with surface properties of the particles calculated with computational chemistry, while the deep learning model is a deep neural network.