Seismic risk assessment of human evacuations in buildings

Abstract

Major earthquakes may require people to evacuate immediately from buildings as recently observed in the 2015 Mw 8.3 Illapel earthquake in Chile. The building may suffer damage, thus affecting the evacuation process. Perhaps due to its apparent complexity, this interaction has not been taken into account when computing seismic risk variables that are intrinsically coupled, such as evacuation times and number of injured people. This limitation can be addressed by simulating the evacuation processes and the physical damage together using agent-based modelling. The evacuation of the building emerges from a set of rules that govern the interaction between agents and with their (damaged) physical surrounding. This research focuses first on modeling evacuations when no physical damage occurs, and uses real evacuation drills performed in a K-12 school and an office building as validation. The comparison was carried out under a low level of uncertainty in the initial conditions of the occupants, i.e., their initial positions and pre-evacuation times were relatively well known, resulting in prediction errors in total evacuation time of only 5.9% and 5.7% for the school and office building, respectively. The evacuation model is then extended to consider building damage and used in an integrated methodology to evaluate the seismic risk of building occupants. This assessment was divided into five steps: (i) seismic hazard, (ii) structural response, (iii) building damage, (iv) evacuation, and (v) risk assessment. First, probabilistic seismic hazard analysis was used to compute the frequency of different levels of local earthquake intensity, characterized herein by the spectral acceleration at the fundamental period of the structure. Ground motions accelerograms matching these intensities were then used in dynamic analyses of the inelastic structure to compute the building response. Story drifts and floor accelerations of the building were related to the damage of non-structural components (e.g., ceilings and partition walls) using appropriate fragility curves. The estimated damage state of the building was used to feed an agent-based evacuation model and assess the evacuation response of the building occupants in this new environment. The outputs of the model are probability distributions of different performance measures and losses, such as evacuation times and number of injured people. These results can better inform decision making processes to mitigate the consequences that future earthquakes will have on buildings and their inhabitants, as well as provide useful information in modeling other larger scale city evacuation scenarios.