Explainable machine learning aided optimization of masonry infilled reinforced concrete frames

Abstract

Machine learning has become a powerful tool in structural and earthquake engineering to accurately predict structural design parameters; however, these complex algorithms are highly inexplainable. In this work, two machine learning algorithms, i.e., Neural Network and eXtreme Gradient Boosting are used to predict the fundamental period of vibration of masonry infilled reinforced concrete frames. The input parameters considered for predicting the fundamental period of vibration of masonry infilled reinforced concrete frames are the number of storeys, opening ratio, span length, number of spans, and masonry wall stiffness. The model predictions are explained globally and locally, using the partial dependence plots, the accumulated local effects, and the game theory-based approach of Shapely values. The performance of the trained machine learning models is compared with expressions existing in the literature and building codes to predict the fundamental period of vibration of masonry infilled reinforced concrete frames. Finally, the trained machine learning models and the existing equation in the literature are used as surrogates to optimize the opening ratio and masonry wall stiffness of buildings present in the database using the genetic algorithm approach. Therefore, the adopted methodology results in a new database of buildings created together by machine learning and optimization algorithms, and the methodology can be used to optimize the opening ratio and masonry wall stiffness to achieve a targeted fundamental period. The dashboard for predicting and optimizing the fundamental period using the trained machine learning models are also developed as a part of this study.