Autonomous characterisation of fouling-free feasible operating regions in continuous flow precipitation of ketoprofen nanoparticles

Continuous antisolvent precipitation enables efficient pharmaceutical manufacturing but is susceptible to fouling, which disrupts operation and compromises product quality. To address this challenge, we developed and validated a closed-loop, autonomous screening platform for identifying non-fouling conditions in continuous antisolvent precipitation, demonstrated for ketoprofen nanoparticle formation in a confined impinging jet reactor using ethanol as solvent, water as antisolvent, and Kollidon VA 64 (PVP/VA) as additive. The platform integrates real-time, pressure drop-based fouling detection with adaptive experimental design. Initially, it performs design of experiments to generate feasibility labels (fouling or non-fouling). A class balancing strategy then targets under-represented regions, and active learning with Gaussian Process (GP) classification selects subsequent experiments to reduce uncertainty in feasibility prediction. With an experimental budget of 50 precipitation experiments, the system autonomously mapped the boundary separating fouling and non-fouling regions across three key process variables: total volume flow rate (1–11 mL min −1 ), antisolvent-to-solvent volume flow rate ratio (4–10), and drug-to-additive mass ratio (0.2125–4.25). The resulting GP classifier achieved high prediction accuracy with an area under the receiver operating characteristic curve (AUC) greater than 0.8 in cross-validation experiments and an AUC of 1.0 in a set of eight follow-up validation experiments. The methodology revealed complex interactions between process conditions and fouling behaviour, identifying robust non-fouling operating windows. This autonomous framework provides a systematic and efficient approach for fouling detection and operating region characterisation in continuous pharmaceutical precipitation, reducing development effort and improving process reliability.