AI‑guided design of 3D-printable gyroid spacers for scale‑resistant membrane distillation in sustainable water treatment.

Scaling of sparingly soluble salts remains a major barrier to deploying membrane distillation (MD) in sustainable water treatment. This work introduces an artificial intelligence (AI)-driven computational framework for designing scale-resistant feed spacers based on gyroid triply periodic minimal surfaces (TPMS) in direct-contact membrane distillation (DCMD). The framework bridges computational fluid dynamics (CFD), population-balance modeling of gypsum crystallization, a statistical design-of-experiments exploration of spacer geometry, and eight machine learning (ML) algorithms trained on three-dimensional (3D) transient CFD simulations that resolve coupled heat/mass transfer and gypsum scaling. Support vector regression (SVR) delivered the most accurate surrogates (R2 > 0.96 for all targets) and was embedded in a weighted multi-criteria optimization that prioritizes minimum wall shear stress (WSS) as a hydrodynamic proxy for scaling risk, together with flux, maximum WSS, and pressure-drop. Using the calibrated and validated CFD-population-balance model, the optimized gyroid spacer was predicted to achieve a 7.1-fold increase in minimum WSS, 68% lower scalant mass density, ∼200 min delay in gypsum nucleation, and 38.5% higher terminal flux than the reference gyroid, as well as a 99% reduction in scalant mass density and 167% higher terminal flux than a commercial ladder-type spacer. The trained SVR models are released as an open-source Python library and Excel calculator, enabling digital-twin style AI-assisted optimization of MD spacers for sustainable water management without requiring CFD or ML expertise.