A Fast Seismic Assessment Technique for Reinforced Concrete Buildings: Machine Learning-based Hassan Index

Abstract

Assessing large inventories of reinforced concrete structures in urban areas with high seismicity is a daunting task that requires tools that can be applied quickly to produce reliable results. The first goal should be to identify the most vulnerable structures requiring fast intervention. Existing assessment standards are often too complex for this purpose. Shiga et al. [1] and Hassan and Sozen [2] have proposed more efficient assessment options based on simple geometric parameters. The question addressed here is whether machine learning (ML) algorithms trained to use the same parameters can match field observations better. Survey data from 1,320 low- and mid-rise RC buildings are used to train and test an algorithm for seismic vulnerability classification. The algorithm was able to produce modest improvements in vulnerability classification, relative to the Hassan Index and for the dataset used in its training and testing. Yet, for a separate and independent dataset, the algorithm produced nearly the same quality of results as its empirical counterpart, which can be used using simple arithmetic. Nevertheless, the exercise presented suggests that it is worth compiling more data to keep exploring the possibilities ML offers.