Automated prediction and design of PBL connectors using physics-integrated explainable machine learning and user-friendly web API

In this paper, an automated prediction and design framework of perfobond rib (PBL) connectors is developed using physics-integrated explainable machine learning. The key objective is to obtain a design and calculation formula with high precision based on the experimental datasets collected from the literature papers. To do this, as the first step, six machine learning (ML) models, namely Adaptive Boosting (AdaBoost), Decision Tree (DT), KNearest Neighbor (KNN), Random Forest (RF), Support Vector Machines (SVM), and Extreme Gradient Boosting (XGBoost), have been developed to predict the shear capacity of PBL connectors. Then, the SHapley Additive exPlanations (SHAP) algorithm is used to explain the input-output relationship of the best-performing model, XGBoost. Eventually, an effective formula for prediction of the shear capacity of PBL connectors was derived using physics-integrated explainable ML by combining the equation of surface fitting results and expanded lightweight Artificial Neural Network (ANN) model. Finally, a user-friendly Web API was developed to facilitate the prediction and design of PBL connectors for engineers.