Methodology for real-time prediction of structural seismic risk based on sensor measurements

Current earthquake early warning systems utilize p-wave data to predict the extent of an earthquake threat and issue warnings at a regional scale. In the assessment of seismic risk, we propose a methodology to go beyond ground motion prediction to consider the response of the structure itself. It is a localized real-time approach where we utilize the first 3 s of data from sensors mounted on a structure to infer the characteristics of the upcoming earthquake. These parameters are used to simulate ground acceleration histories and the structural response estimated under each input motion. A structure-specific warning can then be issued based on the predicted maximum structural response. The method enables probabilistic inference on the structural risk to the earthquake event. In this paper, we describe the proposed methodology and apply it to an example earthquake. We assess the accuracy of the method, compute its computational efficiency, and investigate its robustness to uncertainty in system parameters. Finally, we apply the method to several recorded earthquakes to demonstrate its generalizability. The approach does not require extensive knowledge of regional earthquakes or site characteristics. Such data, however, if available, can be easily incorporated to improve the efficiency and accuracy of the method.