In-line Self-calibrated pH Monitoring System with Hyperspectral Imaging and Deep Learning

Performance Period: 7/15/2019 to 1/15/2021

This project aims at developing an automated system for in-line, non-invasive, self-calibrated pH and/or related DO and glucose measurement in pharmaceutical processing. It directly responds to the NIIMBL call topic 6.1 and the need from leading members of pharmaceutical industries. The system utilizes the hyperspectral imaging technique to capture spectrum information (900-2500nm) of samples and reference buffer media simultaneously. Through deep learning models, the system is expected to automatically extract spectrum information and make accurate in-line measurements. This hyperspectral imaging technique has advantages over conventional Fourier transform infrared spectroscopy in noise-resistance, low variation, high accuracy, and self-calibration functions.

The system at MRL 4 will go through a two-step development process with nine months each. First, the system including a universal adaptor (single-use or large batch) will be established. The deep learning model will be trained offline and pre-loaded to the system. Second, the prototype will be in-house tested in the UMD IBBR bioprocess facility. The potential impacts include labor savings, eliminating risk of microbial contamination, enhanced product quality and yield, in-line real-time pH monitoring and self-calibration.