Data-driven identification and modeling for cleaning of surfaces covered with film-like soils

In the food industry, processing plants undergo daily cleaning, using significant water and chemicals, often inefficiently due to unoptimized procedures. This thesis focuses on developing a classification tool and models to optimize cleaning processes through simulations. Soils are classified by their removal behavior, termed cleaning mechanisms: diffusive dissolution, cohesive separation, adhesive detachment, and viscous shifting. A model for diffusive dissolution already exists, and this research aims to create models for the other mechanisms.
Two classification strategies based on cleaning experiment videos are developed. The first analyzes small sub-regions with neural networks, while the second interprets videos as 3D images using advanced segmentation methods like U-Nets. The research also explores how various process parameters affect the cleaning mechanism, presented in a regime map. Models for cohesive separation, adhesive detachment, and viscous shifting are developed and validated using experimental data.
A combined cleaning model is created to simulate multi-step cleaning processes, considering temperature and sodium hydroxide concentration. Application of the model to a proteinaceous soil in a heat exchanger is demonstrated, with cleaning simulations completed in under ten minutes.