Numerical investigation of double diffusion of NEPCM around oscillating cylinders in a curved cavity using ISPH and machine learning

Purpose This paper aims to investigate transient double-diffusive convection and phase change of nano-encapsulated phase-change materials (NEPCM) in a porous curved cavity with two oppositely oscillating cylinders and to quantify how oscillatory actuation and boundary conditions govern heat and mass transfer. Design/methodology/approach The governing equations are solved using incompressible smoothed particle hydrodynamics (ISPH). An Extreme Gradient Boosting (XGBoost) surrogate is trained to predict the average Nusselt number (Nuavg) and Sherwood number (Shavg). Parametric sweeps span Darcy number (Da = 10−5–10−2), Soret number (Sr = 0.1–1.6), thermal radiation parameter (Rd = 0–5) and oscillation amplitude (A = 0.25–1.0), at fixed Ra = 104, Du = 0.05, Le = 5, N = 1, θf = 0.05 and ϕ = 0.05. Findings Boundary placement strongly regulates transport; the configuration with heated cylinders and adiabatic walls (Case C4) yields the largest Nuavg and Shavg. Increasing oscillation amplitude accelerates melting and enhances convection; larger Da improves mixing; radiation promotes thermal homogenization; and stronger Soret effects intensify solutal diffusion. The XGBoost surrogate reproduces ISPH outputs with near-unity (R2) and residuals below (2.0 × 10−4), achieving speed-ups greater than (103) compared with direct simulations. Originality/value Unlike earlier studies with stationary obstacles, this work shows how oscillatory forcing and boundary conditions jointly control latent-heat buffering and double-diffusive transport of NEPCM in porous curved cavities. The integrated ISPH–AI framework provides a compact and accurate route for real-time design and optimization of NEPCM-based thermal storage, cooling technologies and micro-scale heat exchangers.