Abstract: Granular column collapse, a representative process of geophysical flows such as landslides and debris flows, presents significant risks to infrastructure and ecological systems. The dynamics of these flows, including their initiation, propagation, and final deposition are governed by particle size distribution. However, a comprehensive understanding of the intrinsic mechanical controls, particularly for polydisperse systems, remains an open research challenge. This paper presents an Eulerian multiphase model to simulate the bidisperse granular collapse. The model incorporates an advanced anisotropic kinetic theory for granular flows, combined with a derived frictional rheology that explicitly accounts for dilatancy and contraction effects. Crucially, the constitutive model for the granular phase incorporates the influence of interstitial fluid on both collisional and frictional stress. The employed drag force encompasses mechanisms covering dilute to dense particle concentrations. Additionally, a turbulence model capable of resolving density stratification effect is introduced to accurately capture the interaction between the granular material and the fluid. The numerical model is validated against laboratory experiments of monodisperse and bidisperse granular collapses. The simulations show excellent agreement with experimental measurements across key variables: the final deposit morphology, the total runout distance, and particle segregation during the collapse of bidisperse granular column. This study highlights that the initial aspect ratio governs the granular segregation process: an increase in this results in a significantly longer total runout distance and notably intensifies particle segregation during flow. Furthermore, the analysis identifies that the interplay between dilatancy/contraction and particle collisions is a fundamental mechanism governing the flow dynamics and the resulting particle segregation patterns. The adopted frictional viscosity model in this study integrates shear effects with dilatancy/contraction behavior, enabling a more rational characterization of the frictional stresses between coarse and fine particles during collapse. Consequently, it significantly outperforms traditional frictional models that neglect these effects in predicting deposit morphology and the position of segregation interfaces. The numerical model developed in this study provides a reliable theoretical tool for revealing the mechanism of granular flow disasters and designing disaster mitigation projects.