Continuum Damage Modeling of Biaxial Fatigue Failure in Whole Bone: A Hybrid Approach with Machine Learning Integration

Abstract

Repetitive loading of bone is associated with microdamage accumulation and material property degradation that may ultimately result in fatigue fracture. Our previous work used continuum damage mechanics (CDM)-based finite element (FE) modeling to predict stiffness loss and fatigue failure in whole bone; however, this model did not account for inter-specimen variability in fatigue behaviour and multiaxial loading effects, which limited its applied efficacy. In this study, we refined the CDM-based FE model to predict experimental fatigue-life measurements from 21 whole rabbit tibiae subjected to cyclic axial compression with different magnitudes of superposed torsion. Machine learning (ML) methods were used to predict damage parameters from initial, undamaged, conditions, which were then used to predicted stiffness degradation and fatigue failure with the CDM-based FE modeling pipeline. Regression analysis was used to compare CDM-based FE fatigue-life predictions with experimental measurements. A random forest ML model predicted specimen-specific damage parameters with high accuracy (R2 = 0.85) and the CDM-based FE models demonstrated remarkable predictive capability, explaining up to 91% of the variance in fatigue-life measurements. Stiffness degradation profiles also followed a similar trend to experimental measurements; however, this agreement worsened with the level of superposed torsion suggesting additional refinements to the model may be necessary. These findings demonstrate the efficacy of integrating ML with CDM-based FE modeling to predict fatigue life and stiffness degradation. The observed agreement with experimental measurements suggests the modeling framework may provide valuable information regarding the mechanisms of fatigue fracture in whole bone