A Machine-Learning-Based Seismic Vulnerability Assessment Approach for Low-Rise RC Buildings

Seismic vulnerability evaluation of existing buildings is essential to minimize the destructive impacts of earthquakes. Rapid visual screening (RVS) methods are simple and effective vulnerability assessment techniques to help quickly identify high-risk buildings for more detailed evaluations. Among various RVS methods, the Hassan–Sozen priority index (PI) is one of the simplest methods that can be used for low-rise reinforced concrete (RC) buildings. The PI relates simple, easily attainable geometric features of a building including number of stories, floor area, column area, and wall area to damageability. However, the relationship is overly simplified, and there is no absolute basis for defining damage classification boundaries that can be used to interpret the PI. Furthermore, given the lack of seismic parameters as inputs, the PI only allows for a relative evaluation of buildings in a specific region. To address these issues and develop a more broadly applicable RVS method, this study first proposes an improved PI evaluation method using machine learning techniques to define damage classification boundaries. Then, a new generalized RVS method is proposed that considers the PI input features and earthquake intensity measures to predict damage states. Data from six post-earthquake damage surveys (Duzce (1999), Bingol (2003), Nepal (2015), Taiwan (2016), Ecuador (2016), and Pohang (2017)) are used to train and evaluate the classification models. Two earthquake intensity features, modified Mercalli intensity and peak ground acceleration, are introduced to develop a new earthquake intensity aware RVS. The results of the proposed methodologies show a considerable improvement from the original PI with no judgment needed to define the damage classification boundaries.