Crystallization Process Optimization Using Hybrid Tomographic Imaging and Deep Reinforcement Learning for Sustainable Energy Systems

Crystallization is a fundamental unit operation in chemical, pharmaceutical, and energy industries, where strict control of crystal size distribution (CSD) is essential for ensuring product quality and process efficiency. However, the nonlinear dynamics of crystallization and the absence of explicit functional relationships between process variables make effective control a significant challenge. This study proposes a hybrid approach that integrates process tomography with deep reinforcement learning (RL) for adaptive crystallization control. A dedicated hybrid tomographic system, combining Electrical Impedance Tomography (EIT) and Ultrasound Tomography (UST), was developed to provide complementary real-time spatial information, while a ResNet neural network enabled accurate image reconstruction. These data were used as input to a reinforcement learning agent operating in a Simulink-based simulation environment, where temperature was selected as the primary controlled variable. To evaluate the applicability of RL in this context, four representative algorithms: Actor–Critic, Asynchronous Advantage Actor–Critic, Proximal Policy Optimization (PPO), and Trust Region Policy Optimization, were implemented and compared. The results demonstrate that PPO achieved the most stable and effective performance, ensuring improved control of CSD and improved control proxies consistent with potential energy savings. The findings confirm that hybrid tomographic imaging combined with RL-based control provides a promising pathway toward sustainable, intelligent crystallization processes with enhanced product quality and energy efficiency.