From Eyes to Cameras: Computer Vision for High-Throughput Liquid-Liquid Separation

We present a modular, high-throughput (HT) automation platform for screening Liquid-Liquid Extraction (LLE) workup processes. Our automated hardware platform simultaneously screens up to 12 vials, and is coupled with a computer vision (CV) system for real-time monitoring of macroscopic visual cues. Our CV system, named HeinSight3.0, leverages machine learning and image analysis to classify and quantify multivariate visual cues such as liquid level, phase split clarity, turbidity, homogeneity, volume, and color. These cues, combined with process parameters like stir rate and temperature, enable real-time analysis of key workup processes (e.g., separation time, phase split quality, volume ratio of layers, color, and emulsion presence) to aid in the optimization of separation parameters. We demonstrate our system on three case-studies: impurity recovery, excess reagent removal, and Grignard workup. Our application of HeinSight3.0 on literature data also suggests high potential for generalizability and adaptability across different platforms and contexts. Overall, our work represents a significant step towards achieving end-to-end autonomous LLE screening guided by visual cues, contributing to the realization of a self-driving lab for workup processes.