Inverse design of static mixers using generative diffusion models, computational fluid dynamics and 3D printing

The design of static mixers, critical components in continuous flow processes, has traditionally relied on iterative, expert-driven methods which have difficulty predicting complex fluidic processes and have limited opportunity for agile geometry and design optimisation. We move away from this conventional approach and present an inverse design framework that integrates generative diffusion models with computational fluid dynamics (CFD) and additive manufacturing to automate and accelerate static mixer development. By training a diffusion model on a dataset of CFD-simulated mixer geometries and their associated performance metrics (pressure drop, mixing efficiency, concentration, and porosity), we enable the generation of new mixer designs conditioned on target performance values. The model demonstrates strong predictive accuracy, with median relative errors below 25% across key metrics. Selected AI-generated designs were successfully fabricated via additive manufacturing. This approach enables rapid, performance-driven design of complex fluidic devices and offers a scalable pathway for custom mixer development and manufacture in chemical, pharmaceutical and process industries.