(441e) Process Development and Control Strategies with PAT and Soft-Sensors: A Solvent-Swap Case-Study

In this study, a streamlined methodology was developed using API-agnostic NIR chemometric models calibrated with carefully prepared solvent mixtures. This approach enables the availability of NIR model for endpoint determination before laboratory and manufacturing campaigns. The calibrated models were evaluated by comparing NIR-based water content predictions with KF measurements in laboratory and manufacturing samples containing API. Results demonstrate a good agreement between NIR and KF predictions, validating the effectiveness of this fast-track NIR model development workflow and its potential application in distillation IPCs. This approach notably reduces analytical workloads and accelerates process cycle times, enhancing operational efficiency. Furthermore, integrating NIR data with a soft sensor based on first-principles modeling enables tighter, real-time composition control. Overall, the proposed workflow offers a predictive, efficient, and scalable framework for solvent-swap distillation, advancing both process robustness and sustainability in pharmaceutical manufacturing.

When designing a solvent swap operation, several variables need to be taken into consideration simultaneously: a) the solvent chosen for the swap which impacts the performance of downstream operation, vapor-liquid equilibrium (VLE) including possible existence of azeotropes, and sustainability; b) operating pressure and reactor volume which need to be defined considering VLE, equipment constraints, and product solubility and stability; c) temperature gradient between heating jacket and reactor temperatures during distillation that impacts operation time and d) solvent feeding modes – put & take and continuous feed that will define fresh solvent consumption and distillation path. During a batch campaign, composition monitoring and process control are essential for meeting defined purity criteria and guiding the process within the desired operating window, thereby ensuring operational efficiency and reliability.

In this work, a systematic and model-driven workflow was developed to streamline the process development and a control strategy implemented based on process analytic technology (PAT). Determination of the best solvent to perform the swap distillation is based on the relative volatility, API solubility and existence or not of azeotropes [1]. The top promising solvents are then considered for the in-silico DoE to determine the design spaces with best compromise between fresh solvent consumption and distillation times are achieved. Thus, this strategy predicts solvent volumes and distillation times based on operating pressure, jacket temperature, and solvent feeding mode to be applied. Afterwards, the most promising design space predicted by the model is validated through laboratory experiments.

For in-process control (IPC) during distillation, standard solvent quantification techniques are commonly employed to confirm that specified purity criteria are met before progressing to subsequent steps. However, these analytical techniques have limitations; they are time-consuming, require chemical reagents (GC), and are destructive. Near-infrared (NIR) spectroscopy offers rapid, non-destructive analysis of liquid samples with minimal or no sample preparation.

References:

[1] Gentilcore, M. J. Reduce Solvent Usage in Batch Distillation. Chemical Engineering Process 2002, January, 56–59.