Machine learning algorithms for structural performance classifications and predictions: Application to reinforced masonry shear walls. (December 2019)

• Record Type: Journal Article
• Authors: Siam, Ahmad; Ezzeldin, Mohamed; El-Dakhakhni, Wael
• Is Part Of: Structures. Volume 22(2020)
• Page Start: 252
• Page End: 265
• Publication Date: 2019-12
• Subjects: Classification -- Clustering -- Exploratory data analysis -- Machine learning -- Structural performance -- Supervised learning -- Unsupervised learning -- Predictions; Structural engineering -- Periodicals; 624.1
• DOI: 10.1016/j.istruc.2019.06.017
• Languages: English
• ISSNs: 2352-0124

Abstract

Abstract: Current building codes and design standards classify different structural components according to their expected structural performance. Such classification is usually based on datasets of experimental results typically supplemented by analytical and/or numerical simulations. However, it is usually prohibitive to experimentally evaluate the influence of the typically large number of (and the wide numerical range of each of the) interacting design parameters, on the response of any one class of structural components. Subsequently, the current study builds on the recent advances in the area of machine learning (ML)—a class of artificial intelligence, to introduce a robust ML-based framework for performance prediction and classification of structural components. In order to demonstrate the use of the developed framework, a dataset of 97 reinforced masonry shear walls (RMSWs) is utilized. In this respect, the current study first conducts an exploratory data analysis to recognize the influence of the walls' geometrical and mechanical characteristics on the wall responses. Subsequently, an unsupervised learning algorithm is developed to cluster the walls based on their features. Finally, the training and validation datasets are used to further develop and validate a supervised learning algorithm to classify the walls and predict their lateral drifts according to their failure modes. The study is expected to introduce and demonstrate the capability of ML-based frameworks for future relevant studies within other structural engineering applications.