Reinforcement effects of bonding Fe-SMA in steel bridge diaphragms based on machine learning
Year: 2025
Type: article
Abstract: Although bonded iron-based shape memory alloy (Fe-SMA) has been used for diaphragm reinforcements in practice, accurately evaluating the reinforcement effects of this technique remains challenging. This is due to the complex interaction of multiple factors, including material properties of Fe-SMA, component dimensions, crack-stop hole (CSH) diameters, crack lengths, and activation conditions of Fe-SMA. The study aims to develop an automatic and robust assessment framework for the reinforcement effects of cracked diaphragms repaired by the CSH combined with bonded Fe-SMA. Based on the experimental results, ABAQUS software is used to establish numerical models of the repaired diaphragms. These models are subsequently utilized to assess the impact of key parameters on reinforcement performance. Additionally, a dataset containing 160 samples is collected by experimental and numerical results. Five-fold cross-validation, combined with the African Vultures Optimization Algorithm, is employed to optimize the performance of the categorical boosting (CatBoost) and extreme gradient boosting (XGBoost) models for predicting the hole-edge stress. Furthermore, Shapley Additive Explanations (SHAP) analysis is applied to explore the interpretability of the optimized machine learning models. The results indicate that the observed strain during testing closely matches the numerical predictions , with a maximum average error of 12.39 %. The established numerical model provides a reasonable prediction regarding the mechanical behavior of repaired diaphragms. When the Fe-SMA thickness exceeds 2.0 mm or the activation-zone stress on the diaphragm surpasses 100 MPa, the hole-edge stress falls below the fatigue limit. The optimized CatBoost and XGBoost models demonstrate excellent predictive performance, achieving determination coefficients exceeding 0.968. The optimized XGBoost model exhibits a lower risk of overfitting, suggesting better generalization to unseen data. The findings of this study based on machine learning offer significant insights for optimizing repair strategies and advancing multidisciplinary integration, contributing to enhanced structural durability and safety in engineering applications .
Source: Structures
Authors: Yue Shu, Xuhong Qiang, Xu Jiang, Yapeng Wu
Institutions: Tongji University, Shijiazhuang Tiedao University
Language: English
Subfield: Building and Construction
Field: Engineering
Domain: Physical Sciences
Open Access status: hybrid
Estimated APC paid: $3,510
OpenAlex ID: W4409787952