University of British Columbia
Vision-guided experimental control in self-driving labs
Abstract
dc:descriptionPharmaceutical purification processes rely on visual cues for decision-making, including color changes during titration, liquid levels during distillation, and solid formation during crystallization. Despite their importance, these visual signals are rarely incorporated into automated systems. As a result, chemists must manually perform time-consuming tasks based on visual observation which slows down purification across the pharmaceutical pipeline. This thesis addresses this challenge through the development of HeinSight, a computer vision system designed for real-time monitoring and control of chemical processes. By integrating classical image analysis with neural network models, HeinSight extracts multiple visual cues, including phase identity (solid, liquid, air), color, turbidity, and volume, to enable automated monitoring across diverse purification workflows. Additionally, HeinSight enables vision-guided feedback control across diverse hardware platforms, supporting purification processes that span early-stage screening, iterative optimization, and reactor-scale scale-up. This thesis presents three platform developments. First, HeinSight integrates image analysis and machine learning for multi-visual cue detection and is applied to automated reactor-scale systems to enable vision-based control of widely used purification processes: distillation, crystallization, liquid–liquid separation, and mixing. Second, a hierarchical detection framework expands monitoring to multiple vessels simultaneously, enabling high-throughput experimentation for accelerating liquid–liquid separation screening. Third, computer vision is incorporated into a closed-loop robotic system to enable serial, iterative refinement, where experiments are prepared, executed, analyzed, and autonomously refined based on visual feedback. This platform is applied towards refinement of solubility and drug formulation. Collectively, this thesis establishes a scalable framework for vision-guided experimental control and represents a significant step toward autonomous purification across the pharmaceutical pipeline.
Degree
thesis:*- Name thesis:degree_name
- Doctor of Philosophy - PhD
- Level thesis:degree_level
- doctoral
- Discipline thesis:degree_discipline
- Chemistry
- Grantor dc:publisher
- University of British Columbia
- Year dc:date
- 2026
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- El-khawaldeh, Rama
- Contributors dc:contributor
-
- Hein, Jason
Rights
dc:rights- Statement dc:rights
-
- Attribution-NonCommercial-NoDerivatives 4.0 International
- Language dc:language
- eng
Identifiers
dc:identifier.*- Handle dc:identifier
- http://hdl.handle.net/2429/95556