Data-driven post-earthquake rapid structural safety assessment

Earthquake prone cities are exposed to important societal and financial losses. An important part of these losses stems from the inability to use structures as shelters or for generating economic activity after the event of an earthquake. The inability to use structures is not only due to collapse or damage; it is also due to the lack of knowledge about their safety state, which prohibits their normal use. Because a diagnosis is required for thousands of structures, city-scale safety assessment requires solutions that are economically sustainable and scalable. Data-driven algorithms supported by sensing technologies have the potential to solve this challenge. Several ambient vibration monitoring studies of buildings, before and after earthquakes, have shown that the extent of damage in a building is correlated with a decrease in the natural frequency. However, the observed worldwide data may not be representative of specific cities due to factors such as construction type, quality, material, age, etc. In this paper we propose a framework that is able to progressively learn the relationship between frequency shift and damage state as a small number of buildings in a city are inspected after an earthquake, and to use that information to predict the safety state of uninspected but monitored buildings. The capacity of the proposed framework to learn and perform prognosis is validated by applying the methodology to a city with 1000 buildings having simulated frequency shifts and damage states.

Mots clés

Resilience, Earthquake, Safety, Statistical Learning, Sensors, Condition assessment