Automating PAT Analysis and Building PAT Soft Sensors with Machine Learning

Описание: The use cases and adoption of PAT for pharmaceutical and bio-processing continues to increase. But the existing methods of extracting meaningful bio-informatics remain cumbersome and time consuming. Quartic.ai is using machine learning to accelerate this analysis.

The use of machine learning automates the process of analysis for spectral data, thereby bringing offline models to the production floor. By replacing currently used assays and tests with a solution like the PAT module, analysis times can be reduced from hours and days to minutes and seconds. Short term, consistency of both the calculated process characteristic and the overall process can be positively affected by the removal of manual sampling which is prone to errors and can introduce contaminations. Realtime soft sensors for key process attributes such as protein concentrations can be built with this module. Further, by combining other process variables with PAT data, predictive soft sensors can be built with machine learning models. Those ML models can be real-time estimates of unmeasurable attributes, final estimates of a CQA or KPI, or forecasts for the remaining batch trajectory.