AI‐Assisted Design of Custom Insoles for Plantar Pressure Redistribution via a Digital Twin Framework

Abstract

Foot health is essential for mobility and quality of life, with custom‐made insoles (CMIs) offering effective plantar stress redistribution. This study integrates finite element analysis (FEA) and machine learning (ML) to optimize CMI material and design. A total of 648 FEA simulations revealed that insole thickness, material stiffness, and soft tissue Young's modulus significantly influence plantar stress. Among the three tested thicknesses (5, 10, and 15 mm), the 10 mm insole showed the lowest average peak stress, suggesting a potential U‐shaped relationship that warrants further investigation. A two‐model ML framework is developed: Model 1 evaluated insole effectiveness with  test accuracy of 0.988 ± 0.009, and Model 2 predicted optimal insole parameters with high precision (Overall Root Mean Squared Error [RMSE] for Random Forest [RF] : 0.338 ± 0.061). SHAP (SHapley Additive exPlanations) analysis confirmed that model predictions aligned with FEA‐derived biomechanical patterns, enhancing interpretability and trustworthiness. This AI‐assisted digital twin framework enables data‐driven, personalized optimization of insoles while reducing trial‐and‐error costs. Future work will incorporate dynamic gait loading, viscoelastic material models, and in situ tissue measurements to further refine accuracy and clinical relevance. The framework provides a promising path toward scalable, precision‐designed insoles for both therapeutic and performance applications.