Description of flow regime transitions with deep‐learning‐based polydisperse image method in a gas–liquid stirred tank

Abstract The characterization of flow regime transitions in gas–liquid stirred tanks is important for process optimization. Conventional studies, however, have primarily characterized steady‐state flow regimes, overlooking the evolution of gas‐phase polydispersity that significantly influences the transition dynamics. This work addresses this gap by introducing a deep‐learning‐based polydisperse particle detection system for efficient size‐resolved bubble characterization. The results reveal that regime transitions are mechanistically controlled by a hydrodynamic decoupling between bubble classes, with large‐bubble dynamics playing a key role. The flooding‐to‐loading transition is characterized by the radial dispersion of large bubbles, marked by a sharp surge in their local occurrence frequency, with its growth rate increasing more than fivefold. The transition to full recirculation coincides with axial entrainment of large bubbles below the impeller, causing a more than threefold increase in local gas holdup. This study establishes a mechanistically grounded framework linking microscopic bubble phenomena to macroscopic flow pattern transitions.