Inverse optimization of spray drying by surrogate models trained on frozen-field CFD simulations

Spray dryers are highly dynamic systems in which complex airflow patterns, heat and mass transfer, and particle motion interact across multiple length and time scales. Accurately resolving these interactions with fully transient CFD-DPM simulations is computationally expensive, limiting their use for process optimization. This study addresses this challenge by developing and validating a frozen time-averaged airflow CFD-DPM approach for a co-current spray dryer and embedding it within a surrogate-based inverse optimization framework. Compared to fully transient simulations, the frozen airflow approach reproduces key product outputs with comparable accuracy while reducing computational time from approximately 72 hours to 15 minutes, corresponding to a speedup close to 300 for a pilot-scale co-current spray dryer. The workflow is used to train a polynomial surrogate model embedded in an inverse optimization framework. The differences between surrogate and CFD-DPM results remain small, with mean particle temperature deviations below 2 °C, particle diameter differences under 1.2 µm, moisture discrepancies below 0.024 kg kg⁻¹, and product yield differences below 0.01. The proposed framework can be used as a practical tool for examining the operating input space and for rapid spray dryer optimization without the need for expensive computational hardware. • Frozen airflow CFD reproduces transient spray dryer results with high accuracy. • Re-using frozen airflows reduces CFD runtime from days to minutes. • Particle properties are predicted accurately using time-averaged air fields. • Polynomial surrogate models capture CFD trends across operating conditions. • Surrogate-based inverse optimization enables rapid spray dryer design.