Abstract
This study presents a wireless edge-intelligent control framework for crystallization processes that integrates ultrasound tomography (UST) with deep reinforcement learning (DRL). Ultrasound tomography delivers non-invasive in-situ images of opaque media which makes it possible to observe precipitation fronts and spatial heterogeneity directly. A dedicated measurement platform acquires raw signals reconstructs images compresses salient tomographic features and forwards them to an edge controller for real-time inference. The reconstructed images serve as the state for a Proximal Policy Optimization agent that autonomously modulates cooling dosing and stirring to shorten cycle time while safeguarding product morphology. Simulation results indicate fast policy convergence and more than a twenty percent reduction in average crystallization time relative to baseline controllers. The principal contribution is the use of tomographic images as the decision substrate which elevates tomography from purely diagnostic use to a core component of closed-loop control. A wireless and modular architecture underscores suitability for Industry 4.0 settings and points toward adaptive image-drive

This publication has 8 references indexed in Scilit: