Spray drying is a critical unit operation in the food industry, directly determining product quality in terms of particle size distribution (PSD), moisture content, bulk density, and powder quality. Despite its industrial importance, the process remains difficult to control reliably: scorched and caked particles lead to product rejection and consumer complaints, while equipment fouling reduces plant run-time and increases downtime. Current approaches rely heavily on trial-and-error experimentation, which is costly and time-consuming, especially when adapting to new formulations or scales. Existing physics-based models face challenges in extrapolating beyond their calibration domain, and purely data-driven approaches lack physical interpretability and generalizability. There is therefore a pressing need for a robust, flexible, and predictive modeling framework that combines mechanistic understanding with data-driven adaptability.

This project aims to develop a hybrid modeling platform for spray drying that integrates a compartmentalized physics-based model incorporating Population Balance Modeling (PBM) and heat & mass transfer with machine learning (ML) algorithms. The resulting tool will predict product properties and process conditions inside the dryer, detect anomalies, and recommend optimal operating parameters with minimal experimental effort. The platform is designed to be transferable, scalable across different dryer geometries, and deployable as a Python-based software tool for routine use by Nestlé.

A 1.5-D compartmentalized model is developed, dividing the dryer into N vertical compartment pairs, each consisting of a core zone (downward hot jet) and an annulus zone (upward recirculation), as illustrated in Figure 1. Within each compartment, heat and mass balances are solved.

The model obtains a lot of unkonwn parameters including hydrodynamics, drying kinetics, and agglomeration. The challenge here to to determine the paramters in a way that makes the model scalable for different sizes and therefor we are trying to answer the flowing questions.