Multiple Target Machine Learning Prediction of Capacity Curves of Reinforced Concrete Shear Walls

Abstract

Reinforced concrete (RC) shear wall is one of the most widely adopted earthquake-resisting structural elements. Accurate prediction of capacity curves of RC shear walls has been of significant importance since it can convey important information about progressive damage states, the degree of energy absorption, and the maximum strength. Decades-long experimental efforts of the research community established a systematic database of capacity curves, but it is still in its infancy to productively utilize the accumulated data. In the hope of adding a new dimension to earthquake engineering, this study provides a machine learning (ML) approach to predict capacity curves of the RC shear wall based on a multi-target prediction model and fundamental statistics. This paper harnesses bootstrapping for uncertainty quantification and affirms the robustness of the proposed method against erroneous data. Results and validations using more than 200 rectangular RC shear walls show a promising performance and suggest future research directions toward data- and ML-driven earthquake engineering.

Keywords

• machine learning for capacity curve prediction
• multiple-target regression model
• clus
• shear wall database
• uncertainty quantification