A Modular Digital Twin for Autonomous Crystallization Experiments: AI-Based Image Analysis, Process Control, and Automation

We present a modular Digital Twin framework for crystallization experiments that integrates real-time control, data acquisition, and AI-based image analysis. The system is built around a custom graphical user interface (GUI) that allows researchers to control and monitor hardware components—including pumps, stirrers, temperature modules, and valves—enabling fully synchronized and programmable experiments.

In parallel, a dedicated image analysis GUI automates crystal detection and characterization using a YOLO-based object detection model trained on a synthetic dataset generated through procedural rendering of crystal-like geometries under varied lighting and contrast conditions. The model identifies crystals in real time and quantifies features such as aspect ratio, orientation, and count evolution throughout the process.

Together, these interfaces form a flexible and extensible digital twin environment, linking hardware control with live visual feedback and quantitative analysis. Its modular design enables seamless integration of future components such as Bayesian optimization and closed-loop feedback control for fully autonomous crystallization.