Exploring the usefulness of available experimental data: a machine learning-focused approach to predicting seismic capacity

A fundamental understanding of the knowledge that can be gained from data-driven models within the realm of civil engineering will have a translational impact on the analysis, design, maintenance, and construction of civil infrastructure. With models firmly grounded in real-world data, material and structural properties and behavior can be precisely predicted from each other without computationally expensive analytical or empirical evaluations. Moreover, by integrating data-driven and physics-based models, transformative insights into the behavior of materials and structures their relation to one another can be discovered, actuating next-generation modeling approaches, experimental methods, empirical relations, designs, and construction methods. With the recent rapid development of machine learning (ML) techniques, various ML based approaches have been successfully applied in the general realm of structural engineering and have been validated as effective in reproducing accurate and robust results from experimental tests. In this presentation, a general framework for data-driven understanding of structural performance will be demonstrated and two novel machine learning models aimed at predicting the shear capacity and overall performance of reinforced concrete (RC) columns under reversed cyclic loading will be introduced. An exploration of various existing machine learning models and their application and suitability within the realm of civil engineering (CE) will be discussed, demonstrating the significant potential impact machine learning approaches could have on conventional CE methods and practices.  Furthermore, a comprehensive comparison of the novel machine learning approaches with traditional empirical and modeling approaches popularly used in the field.