(368az) Computer Vision for Real-Time Monitoring and Control of Chemical Processes

Research Interests: chemical engineering, high-throughput experimentation, computer vision, process chemistry, chemical automation, modeling and simulation.

Abstract:

Chemists heavily rely on visual cues for routine tasks. The Hein Lab aims to automate these tasks by introducing HeinSight, a computer vision system designed to automatically monitor and control diverse workup processes. Integrating flexible hardware, data-rich monitoring techniques, image analysis, and machine learning, HeinSight is a significant step towards achieving autonomous chemical operations. This presentation explores the evolutionary journey of HeinSight, detailing its applications in chemical operations, varying scales, and diverse contexts.

Initially focused on image analysis, HeinSight monitored and controlled liquid levels in the EasyMax ecosystem through edge detection. In advancement, HeinSight2.0 emerged in collaboration with Pfizer, incorporating image analysis and machine learning to simultaneously monitor multiple visual cues across the entire reactor viewing window. This advanced version found application in various workup processes, including solvent exchange distillation, antisolvent crystallization, evaporative crystallization, cooling crystallization, solid–liquid mixing, and liquid–liquid extraction.

Continuing innovation led to HeinSight3.0, featuring real-time monitoring of high-throughput vial analysis for liquid-liquid separation. The hardware system provides an affordable platform for camera-assisted vial analysis. The software combines turbidity analysis with machine learning, offering enriched data for optimizing conditions in liquid-liquid extraction through vision-guided feedback. Developed in partnership with the Aspuru-Guzik Lab at the University of Toronto, HeinSight3.0's versatility optimizes various unit operations (e.g., impurity, solvent, and catalyst removal).

HeinSight's ability to dynamically monitor and respond to experimental changes across scales and throughputs is crucial for future flexible automation workflows. This work underscores the significance of academic-industrial collaborations in shaping the labs of the future.