Double-diffusive convection of NEPCM slurry in a stepped cavity with a slender inner Y-fin: ISPH modeling and physics-informed learning

Purpose The purpose of this study is to investigate unsteady double-diffusive natural convection and nano-encapsulated phase change material (NEPCM) melting in a porous stepped cavity containing a slender inner Y-shaped fin, with particular emphasis on coupled heat and mass transfer, phase change dynamics and thermodynamic irreversibility. Design/methodology/approach An incompressible smoothed particle hydrodynamics (ISPH) method is used to simulate the two-dimensional transient flow and transport phenomena. The formulation incorporates Darcy–Brinkman–Forchheimer porous resistance, Soret and Dufour cross-diffusion, viscous dissipation, inclined magnetohydrodynamic effects with Joule heating, nonlinear Rosseland thermal radiation and an Arrhenius-type chemical reaction. NEPCM melting is modeled using an effective heat-capacity ratio. System performance is assessed using first-law heat and mass transfer metrics alongside second-law measures based on entropy generation and Bejan number. A physics-informed stacked ensemble learning model is developed to predict the transient evolution of average Nusselt and Sherwood numbers. The novelty lies in the unified transient ISPH framework that simultaneously couples NEPCM phase change, porous double diffusion with cross-effects, inclined MHD with Joule heating, nonlinear radiation, chemical reaction and entropy generation, together with a physics-informed ensemble surrogate for fast prediction of Nu and Sh. Findings The results indicate that higher permeability intensifies circulation and accelerates NEPCM melting but increases non-thermal entropy generation due to enhanced fluid friction. Magnetic forcing suppresses convection, delays melting and shifts irreversibility toward thermal entropy production. Thermal radiation enhances heat penetration and enlarges the melted region, while a stronger chemical reaction weakens solutal buoyancy and localizes concentration gradients near the heated fin. The learning surrogate accurately reproduces transient heat and mass transfer trends with significantly reduced computational cost. Engineering correlations for Nu and Sh are provided for the quasi-steady regime, enabling rapid design-level estimates within the studied parameter ranges. Originality/value This work presents a unified ISPH-based and data-driven framework for analyzing transient double-diffusive convection with NEPCM phase change in complex porous cavities, providing combined first- and second-law insights supported by physics-informed learning. The integrated multiphysics model and the regression-ready outputs are positioned to support practical engineering optimization in porous thermal storage systems.