Reinforced Concrete (RC) shear wall is one of the most important earthquake-resisting structures that is able to bear a horizontal shear force. The capacity curve and global stiffness reduction of reinforced shear walls are vital for understanding the properties and behaviors of the RC shear walls. Traditional approaches to obtain capacity curve are conducting experiments on shear walls or building finite element models to analyze them. However, these approaches are costly and time-consuming, especially conducting experiments. Meantime, degradation of core shear wall's flexural stiffness is vital to understand the natural frequency shift of the damaged shear walls. But it is hard to capture, often necessitating complex finite element analyses (FEAs). Therefore, this study seeks to provide efficient approaches to quickly obtain capacity curve using multi-target machine learning, and global stiffness reduction of U-shaped RC shear wall using cell network-based formulas. Importantly, both developed approaches are investigated to require only the easy-to-collect property information of shear walls. The acquirement of capacity curve and the remaining flexural stiffness of shear walls will help improve the quality of structural design.

The thesis is structured as follows. CHAPTER 1 introduces previous applications of machine learning in civil engineering domain and background of multi-target prediction model. CHAPTER 2 to CHAPTER 3 introduce the approach to predict capacity curve of shear wall using multi-target regression model. And CHAPTER 4 present computational implementation of cell network to predict remaining stiffness of shear wall. CHAPTER 3 and CHAPTER 5 separately illustrate limitation and future work of applications of multi-target prediction and cell network-based formulas.