The Earthquake Disaster Engineering Laboratory of Toyohashi University of Technology has developed a method to determine the damage status of a building immediately and with high accuracy using machine learning technology from the records of seismographs installed in the building.It is expected to be useful for evacuation immediately after an earthquake and for determining the continued use of buildings.
In the 2016 Kumamoto earthquake, several city halls were damaged and became a major obstacle to evacuation and restoration.At city halls and fire departments, which are the bases for earthquake disaster prevention, it is necessary to analyze the damage situation of buildings immediately after the earthquake and quickly determine whether they can be used continuously.Until now, because of the danger of building collapse due to aftershocks, diagnosis was limited to visual inspection in principle, and the damage situation inside the building was unknown.
Therefore, the research team has developed a technology to install a seismograph in a building and remotely evaluate the soundness of the building from observation records at the time of an earthquake.In this method, the degree of damage is diagnosed by analyzing the seismic response of the structural model of the building using the observation records stored in the Internet cloud, but the analysis took time.
Therefore, we have developed a method to immediately determine the damage status of a building by using a machine learning method called CNN (convolutional neural network) without using a structural model of the building.From the image of the wavelet spectrum of the observation waveform of the seismograph installed in the building, the degree of damage (no damage, minor damage, medium damage, major damage, collapse) and the possibility of continuous use (safety, caution, danger) can be determined. Immediate diagnosis remotely.It is possible to make a quicker diagnosis than before, and it can be applied even if the number of floors and structure of the building are different.
The developed real-time seismic diagnosis system is already in operation at the city hall in the Higashi Mikawa area of Aichi prefecture.This will enable quick and highly accurate diagnosis, and is expected to improve the disaster prevention capabilities of the region.
Paper information:[Sensors] Structural Response Prediction for Damage Identification Using Wavelet Spectra in Convolutional Neural Network
