Explainable hybrid machine learning approach for mechanical performance of 3D-printed strain-hardening cementitious composites (3DP-SHCC)

Abstract

This paper presents an innovative and explainable AI framework for predicting the mechanical performance of three-dimensionally printed strain-hardening cementitious composites (3DP-SHCC), focusing on compressive strength (CS) and flexural strength (FS). A rigorously curated database of 202 mechanical performance records collected from state-of-the-art literature was used to develop AI models. To enhance robustness and overcome the limitations of conventional tuning methods, GBM was integrated with four metaheuristic optimization algorithms—GWO, WOA, HHO, and SSA—combined with five-fold cross-validation. The outcomes show that HHO-GBM achieved the highest accuracy for CS prediction (R2 = 0.951) during the test phase, while SSA-GBM performed best for FS prediction (R2 = 0.897). SHAP and ICE analyses identified binder composition, loading direction, and fiber parameters as key drivers. Additionally, a user-friendly, real-time decision-support interface was developed and validated using independent unseen mixtures. Overall, the proposed framework offers a reliable and engineering-ready tool for 3DP-SHCC design.