Leveraging first-principles and empirical models for disturbance detection in continuous pharmaceutical syntheses

Abstract

A strategy for combining theoretical and empirical model predictions to enhance process monitoring and disturbance detection in continuous pharmaceutical manufacturing is investigated using the first two steps of ciprofloxacin. The first-principles component is a dynamic model that reads in process parameter data and returns a concentration prediction for each species in the system using well-established equations and numerical discretization. The input data for the dynamic model comes from low-cost and reliable sensors that are already commonly deployed in manufacturing scenarios, such as flowmeters and thermocouples, making the approach amenable to potential uniform deployment across numerous manufacturing sites. The empirical component is infrared spectra collected from an inline flow cell that feeds to a partial least squares regression model for product concentration. Process parameter disturbances were introduced while continuously collecting the outlet stream infrared spectra, reagent flowrates, reactor temperature, and running the theoretical and empirical prediction models. Post-processing included the application of changepoint analysis, which is a statistical method of determining changes in the mean of a given time-series dataset. Both types of disturbances were captured as changepoints in the theoretical and empirical model predictions and could be obtained more rapidly by analyzing the residuals between the two predictions. This indicates that the deployment of theoretical models along with empirical is a robust approach for rapidly detecting deviations in the process health, reducing the time that potentially out of specification material is sent downstream. Additionally, by comparing trends in the models with the process parameter data, root-cause analysis can be rapidly carried out for a given disturbance. This places emphasis on holistic process monitoring by incorporating characterization knowledge and understanding into the process along with applying all available data sources to ensure product quality.