همایش بین المللی هوش مصنوعی

صفحه اصلی / دومین همایش بین المللی هوش مصنوعی

Interpretable Ensemble Learning for Predicting the Non-linear Moment Capacity of Bolted Extended Endplate Moment Connections

نویسندگان :

Matin Alizadeh¹ ُS.Mohammad Hosseini² .Mahmoud. R Shiravand³

1- دانشگاه شهید بهشتی 2- دانشگاه شهید بهشتی 3- دانشگاه شهید بهشتی

کلمات کلیدی :

Bolted Extended Endplate Connection،Ensemble Learning،Explainable AI،Seismic Performance Prediction

چکیده :

The Bolted Extended Endplate Moment resisting connection (BEEP) is one of the prequalified moment-resisting steel connections specified in AISC 358 and European codes. Its reliable seismic performance is supported by extensive experimental and analytical studies. However, conventional analytical models often fail to capture the complex interplay of component yielding, bolt fracture, and stiffness degradation. In this paper, an interpretable machine learning (ML) framework is proposed to predict the nonlinear behavior of BEEP connections. A wide range of experimental studies was reviewed, analyzed, and systematically assembled into a comprehensive database. An ensemble learning framework, integrating Random Forest, XGBoost, HISTGM, KNN and LightGBM models, was trained to predict the connection's peak and ultimate strength. High predictive accuracy was achieved on unseen test data. Crucially, the interpretability of these models is addressed using SHapley Additive exPlanations (SHAP), providing clear, physically-consistent interpretations that link geometric and material parameters to the connection performance. The result is a transparent, accurate, and rapid tool for preliminary assessment and design check of this connection.