Deep learning‐driven image analysis for online monitoring and optimization of oiling‐out spherical crystallization

Abstract Oiling‐out crystallization is a green, efficient route to spherical products with improved powder properties. However, key measurements and operational decisions often rely on offline tests and operator experience, causing low efficiency and high trial‐and‐error cost. Here, a deep learning workflow converts process images into quantitative metrics and decision triggers. Oiling‐out crystallization of ethyl vanillin was online monitored and images of droplets and spherical particles were analyzed by deep learning model. By extracting droplet size and number, quenching was triggered when oiling‐out equilibrium was detected, defining operation time. After spherical particles formed, particle size and 2‐dimensional sphericity were extracted online for quality evaluation. Results show that higher stirring speed accelerated equilibrium and reduced product size, while early partial sodium dodecyl sulfate addition at the same speed yielded smaller spheres and better powder properties. This workflow provides a scalable template for online optimization and intelligent control of oiling‐out crystallization.