Casting CAE simulation has become an essential tool in modern manufacturing for optimizing mold designs and ensuring product quality. Among various numerical approaches, the Smoothed Particle Hydrodynamics (SPH) method with the oxide-film model for molten aluminum alloy[1] has gained significant attention. Unlike conventional grid-based methods, SPH naturally captures complex free-surface behaviors, such as splashing and merging of molten metal, which are critical for predicting casting defects in intricate geometries. The mold-filling analyses of casting processes using the particle-based SPH method have two problems: a huge number of particles and a very long calculation time. For example, in die casting processes[2], high-speed and high-pressure injection behavior is achieved by narrowing the ingate, and calculations must be performed with small particle size diameters corresponding to the narrowed gate thickness. Furthermore, in gravity castings, a cast filter is attached to the sprue[3]; numerical analyses must be performed with small particles corresponding to the filter opening diameter. The particle size is reported to be less than 1/16 of the flow path. In this study, we investigated a method for converting particle size during mold filling analysis. Our idea defines regions where particles of different resolutions overlap. Particles entering these regions undergo refinement or coarsening, enabling dynamic adjustment of resolution based on spatial requirements. This method was applied to simple shape processes of die casting and gravity casting with a filter. In gravity casting simulations, the flow behavior of the melt front shows good agreement with the single-resolution results. Also, in high-pressure die casting (HPDC) processes, a little discrepancy in flow behavior is observed, suggesting that further refinement of the conversion criteria is necessary for high-velocity flows. The numerical results show a significant reduction in memory usage compared to conventional single-resolution simulations.