Machine Learning‐Enabled Field Prediction and Inverse Design of Functionally Graded Materials Inspired by Tendon–Bone Enthesis

Functionally graded materials (FGMs) provide an effective strategy for mitigating stress concentrations at mechanically mismatched interfaces, but inverse design of optimal spatial gradients remains challenging. Here, a mechanics‐informed machine learning framework is developed for field prediction and inverse design of FGMs inspired by the tendon–bone enthesis. A 3D finite element model (FEM) is established for a representative enthesis region, in which local material properties are governed by the mineralization scale, mean fibril orientation, and angular dispersion via a multiscale continuum formulation. A composition‐ and microstructure‐informed risk factor is introduced to evaluate local failure susceptibility. FEM results demonstrate that graded mineralization and fibril dispersion substantially reduce stress concentrations and the risk of failure compared with discontinuous interfaces. To accelerate design exploration, a convolutional neural network‐based field predictor (CNNFP) is trained as an efficient FEM surrogate to map spatial material descriptor fields to stress and risk factor fields. Validation against FEM ground truth demonstrates that CNNFP captures full‐field distributions, peak values, and peak location characteristics with high fidelity. By integrating CNNFP with a kernel‐based gradient optimization framework, smooth and physically admissible gradient architectures are inversely designed and FEM‐validated, thereby establishing an AI‐assisted pathway to translate biological gradient architectures into mechanically resilient FGMs.