Title:
Machine Learning Models for Predicting Bond Strength of Deformed Bars in Concrete
Author(s):
Vitaliy V. Degtyarev
Publication:
Structural Journal
Volume:
119
Issue:
5
Appears on pages(s):
46-56
Keywords:
DOI:
10.14359/51734833
Date:
9/1/2022
Abstract:
This paper proposes eight machine learning models for predicting the bond strength between straight deformed reinforcing bars and concrete under tensile load. The models included support vector regressor (SVR), k-nearest neighbors (KNN), random forest (RF), gradient boosting regressor (GBR), extreme gradient boosting (XGBoost), light gradient boosting machine (LightGBM), gradient boosting with categorical features support (CatBoost), and adaptive boosting (AdaBoost). The models were trained using a large data set of test results from the Joint ACI-ASCE Committee 408 database. The performance of the models was evaluated through a robust tenfold cross-validation method. Optimal parameters for each model were established through an extensive tuning process. The bond strengths predicted by the proposed models agree well with the experimental data. Their accuracy exceeds the accuracy of the available descriptive equations. The relative feature importance for predicting the bond strength was evaluated using the permutation and Shapley additive explanations (SHAP) methods for each model. Partial dependence of the bond force from the considered features was also presented and discussed.