A study on input ground motion processing platform for evaluating seismic fragilities using Deep Learning Phase Determination Model

The 9.12 Gyeongju earthquake(Sep. 12, 2016, Ml 5.8) and the Pohang earthquake(Nov. 15, 2017, Ml 5.4) have occurred in the Korean Penisula, resulting in emphasizing the stability of nuclear power plants. For safety evaluation, it is necessary to study the earthquake vulnerability caused by input ground motion. The input ground motion can be obtained from the earthquakes, and it is essential to acquire good quality and many samples input ground motion database for accurate evaluation. In this study, we tried to develop a platform that can automatically generate a ground motion database from past or real-time waveforms. To determine the detailed time window for data processing, deep learning-based earthquake detection, and phase-picking models were used. A voting method was conducted on these models to increase reliability in various environments. The platform produces a RotD50 5% damped pseudo-spectral acceleration, peak ground acceleration, and meta information related to site, hypocenter, and path. It also provides a web service to confirm generated data and meta information. The database generated by the platform could be used as input ground motion data to evaluate the safety of operating power plants and could be applied as fundamental data for the seismic design of planned nuclear power plants.