Machine learning based design of reinforced concrete shear walls subjected to earthquakes

Abstract

Civil engineering structural components are classified according to their projected structural performance in the present building code regulations and design standards. These building design codes are largely based upon previous experimental results of thousands of samples tested to failure and validated with analytical solutions. Machine Learning techniques (ML) is a subset of Artificial Intelligence (AI) that facilitates classification and prediction of structural performances for a broad spectrum of complex structures with greater accuracy. Machine learning models have the potential to make reliable predictions with the help of algorithms. Thereby, saving a tremendous amount of time and resources invested in experimental investigations of large structural components such as shear walls and columns. The ML algorithms can learn from the available data, deduce underlying inter-relationships, make inferences and detect patterns based on previous experience. In the present work, various ML algorithms were implemented to identify the influence of geometrical as well as mechanical characteristics. Database of 393 specimens of reinforced concrete shear walls with rectangular (R), flanged (F) and barbell (B) cross-sections are adopted for the analysis. Shear walls are fundamentally classified into four failure categories which include flexure or due to bending, shear, intermediate flexure-shear and sliding due to shear. The objective of this paper is to classify and predict the shear strength, flexural strength as per the Indian standard code provisions and failure modes of shear walls with the help of ML techniques. Algorithms such as KNearest Neighbors, Naive Bayes, Decision Tree, Random Forest, AdaBoost, LightGBM, XGBoost and Cat-Boost is implemented using Python. Highest accuracy of 85% is achieved on the test set by Random Forest, 83% by CatBoost and 81% by LightGBM boosting algorithms. It is observed that input variables such as aspect ratio (lw/tw), characteristic strength of concrete in compression (f ck ), characteristic yield strength of steel (f y ), percentage of steel (ρ), web vertical reinforcement, horizontal reinforcement, boundary element reinforcements play a vital role in governing the shear strength (V u ) and flexural strength (M u ) of shear walls.