Predicting the fracture mechanism of cast AlSi 7 Mg wheel hub under radial impact with machine learning assistance

Abstract

The impact fracture performance of cast aluminum wheels is crucial for automobile safety, yet accurately simulating their fracture evolution remains challenging due to material inhomogeneity across components. This study investigates the impact fracture behavior and regional material heterogeneity of cast aluminum wheel hubs through an integrated experimental, theoretical, and computational framework. Axial impact tests using a drop-weight machine revealed a distinct dual-failure mode: significant plastic deformation in the rim and a symmetrical triple-crack distribution in the core, consisting of a superior vertical tensile crack and two inferior 45° crushing shear cracks. To characterize this behavior, six advanced fracture criteria were developed by incorporating the second principal stress (σ₂) and mean stress into Mohr–Coulomb and Hosford–Coulomb frameworks, effectively accounting for high-order stress effects and hydrostatic pressure sensitivity. Furthermore, a back-propagation neural network (BPNN) was employed to address casting-induced inhomogeneity by inversely identifying regional fracture energies for the rim, spoke, and core through the minimization of objective functions between experimental and finite element (FE) results. Comparative analysis underscores that mean–stress-based criteria, particularly those integrating σ₂ as a mechanical constraint, exhibit superior fidelity in predicting crack initiation and propagation. The proposed BPNN-FE framework establishes a robust methodology for the failure prediction of complex, heterogeneous cast components.