Font Size

HOME > No.28, Feb. 2022 > Using machine learning to measure building earthquake damage

Using machine learning to measure building earthquake damage

Facilitating evacuation immediately following an earthquake and the continued use of buildings Taiki Saito
Taiki Saito

Damage caused to municipal government office buildings by the 2016 Kumamoto Earthquakes significantly hindered the evacuation and reconstruction efforts which followed. We need to develop technology that enables us to inspect municipal government offices, fire departments and other hub buildings for disaster control activities immediately after an earthquake occurs. The Earthquake Disaster Engineering Research Laboratory in the Department of Architecture and Civil Engineering, Toyohashi University of Technology, has developed a method for instantaneously evaluating earthquake damage to a building from the readings of the building’s seismometer using machine learning technology. All city government offices in the Higashi-Mikawa area are already equipped with seismometers and a system for sharing the results of damage assessments by email immediately after an earthquake has been established. Applying the machine learning technology that has been developed will enable faster, more accurate damage assessment.

Municipal government offices, fire departments and other hubs responsible for implementing measures following an earthquake must be capable of assessing damage to their buildings immediately after an earthquake, to determine whether the building is still fit for purpose. To date, the methods for evaluating a building’s condition after an earthquake have basically been limited to visual inspections from outside of the building because of the potential of an aftershock to cause the building to collapse. For this reason, it has been difficult to assess internal building damage after a large earthquake. The research team has developed a method for remotely assessing the condition of a building during an earthquake based on the readings from the building's seismometer. This method uses observation records stored on the Internet-cloud to analyze the response of the structural model of the building and based on these results, assess damage. However, a highly accurate diagnosis required time-consuming analysis. So, the research team developed a method for immediately assessing building damage using machine learning technology without the need for a structural model of the building.

Damage Assessment Flow Using CNN Machine Learning
Damage Assessment Flow Using CNN Machine Learning

The method would remotely and immediately assess the level of earthquake damage (no damage, mild damage, moderate damage, severe damage, collapse) and whether the damaged building can continue to be used (safe, caution needed, dangerous) . It bases these assessents on images of the wavelet spectra of observed waveforms from seismometers installed in the building using the CNN (Convolutional Neural Network) machine learning method. Edisson Alberto Moscoso Alcantara, the lead author and doctoral student, explains damage assessments using the new method will be faster than the conventional method using a structural model of the building.

"Machine learning technology is rapidly spreading across the field of earthquake preparedness. Previously, assessing damage to a building was dependent on human experience. In the future, this will be automatically handled by AI. The goal of the research is to establish a method for remotely assessing the condition of a building right after an earthquake without having to send someone to the site. Initially, we were concerned about whether it would be possible to determine the extent of damage using only the waveform of the seismometer, but we found that we could determine the extent of damage with considerable accuracy by using wavelet spectra," says Professor Taiki Saito, the leader of the research team.

The method for assessing earthquake damage developed by the research team may be applicable irrespective of the differences in buildings, such as the number of floors or the structure of the building. A real-time seismic testing system developed by Toyohashi University of Technology is already being used in city government office buildings in the Higashi-Mikawa area. Hopefully, the new method will allow for a faster and more accurate seismic diagnosis, thereby contributing to the improvement of disaster preparedness in the region.


Edisson Alberto Moscoso Alcantara, Michelle Diana Bong and Taiki Saito (2021). Structural Response Prediction for Damage Identification using Wavelet Spectra in Convolutional Neural Network. Sensors 2021, 21(20), 6795;



齋藤 大樹




この方法では、CNN(Convolutional Neural Network)という機械学習の方法を用いて、建物に設置された地震計の観測波形のウェーブレットスペクトルの画像から、被害の程度(無被害、軽微な被害、中被害、大被害、倒壊)や継続使用の可能性(安全、注意、危険)を遠隔で直ちに診断するものです。これまでの構造モデルを用いた診断よりも、迅速な診断が可能になると、筆頭著者である博士後期課程のEdisson Alberto Moscoso Alcantaraは説明します。



Share this story

Researcher Profile

Taiki Saito
Name Taiki Saito
Affiliation Department of Architecture and Civil Engineering
Title Professor
Fields of Research Structural Engineering / Earthquake Engineering