Design and prediction of soft-to-hard transitions using bioinspired hierarchical-gradient designs and hybrid stacking machine learning

Abstract

The fabrication of soft-to-hard transition phases poses significant challenges due to the disparity in mechanical properties across the interface. Among all soft-to-hard natural tissues, the tendon-to-bone interface is particularly complex, exhibiting both hierarchical and gradient structural characteristics. This study aims to design, fabricate, and optimize bioinspired structures that replicate tendon-to-bone interfaces by investigating their fundamental relationships with their natural counterparts. To achieve this, various designs featuring simple and hierarchical architectures with negative Poisson's ratio (NPR) were integrated with simple and gradient positive Poisson's ratio (PPR) structures to mimic the physical properties of enthesis. The results demonstrated that the novel hierarchical-gradient designs enhanced Young's modulus and failure force by up to 58.1 %. The finite element method (FEM) was employed to accelerate the prediction of mechanical properties, and a hybrid stacking machine learning (HSML) model trained on FEM results further improved the prediction accuracy, achieving an error of 2 %. The HSML method outperformed traditional approaches like decision trees and convolutional neural networks (CNNs) on small datasets, highlighting its suitability for such applications. Additionally, this study demonstrates that mimicking the energy-absorbing interface between natural soft and hard tissues significantly improves both Young's modulus and failure force in these complex structures.