Abstract
ABSTRACT This study proposes a machine learning-based approach to predict the moment-rotation relationship of dovetail mortise-tenon joints in ancient wooden structures using Support Vector Regression (SVR). Finite element models were developed in ABAQUS and validated against experimental data to ensure accuracy. A dataset of 328 moment-rotation relationships was generated by varying material properties and joint dimensions. The SVR model was optimized through grid search and cross-validation, and its interpretability was analyzed using SHAP, revealing that geometric features such as tenon dimensions significantly impact bending resistance, while material properties play a lesser role. The SHAP analysis further demonstrated that the looseness gap negatively affects load-bearing capacity, whereas increasing parameters like tenon neck width or shoulder width enhances the bending resistance. The results show that the constructed SVR model can quickly and accurately predict the moment-rotation relationship of dovetail mortise-tenon joints, with determination coefficients exceeding 0.989 in the testing set under different deformation conditions, offering higher prediction accuracy, better generalization ability, and computational efficiency compared to traditional theoretical formulas.
Published Version
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call