| Project/Area Number |
22KF0087
|
|---|
| Project/Area Number (Other) |
22F21785 (2022)
|
|---|
| Research Category |
Grant-in-Aid for JSPS Fellows
|
| Allocation Type | Multi-year Fund (2023)
Single-year Grants (2022) |
|---|
| Section | 外国 |
|---|
| Review Section |
Basic Section 22020:Structure engineering and earthquake engineering-related
|
|---|
| Research Institution | The University of Tokyo |
|---|
Principal Investigator |
楠 浩一 東京大学, 地震研究所, 教授 (00292748)
|
|---|
| Co-Investigator(Kenkyū-buntansha) |
YEOW TREVOR 東京大学, 地震研究所, 外国人特別研究員
|
|---|
| Project Period (FY) |
2023-03-08 – 2024-03-31
|
| Project Status |
Completed (Fiscal Year 2023)
|
| Budget Amount *help |
¥2,300,000 (Direct Cost: ¥2,300,000)
Fiscal Year 2023: ¥1,100,000 (Direct Cost: ¥1,100,000)
Fiscal Year 2022: ¥1,200,000 (Direct Cost: ¥1,200,000)
|
|---|
| Keywords | Structural monitoring / Damage assessment / Collapse mode / Machine learning / Numerical modelling / Earthquake engineering / Residual displacement / Shake-table test / 構造ヘルスモニタリング / 振動台実験 / 構造応答 / 変形モード / 機械学習 / 構造解析 |
|---|
| Outline of Research at the Start |
Damage inspection is an intensive process that result in socio-economic losses. To address this, a structural health monitoring system was developed. Refinement is needed to identify building deformation mode. This research addresses this using machine-learning techniques.
|
| Outline of Annual Research Achievements |
(1) We performed shake-table tests of steel frame structures exhibiting various collapse modes (e.g., total yield, soft-story). This data was then used to verify the accuracy of numerical models to capture building response and the correct failure mode.
(2) We proposed building response features based on the cumulation of plastic displacement response for use in developing a machine-learning classification model.
(3) We developed a database containing acceleration and displacement response of RC frame buildings considering different number of floors, floor stiffness distribution and collapse modes using numerical simulations. Using this database and experimental results, we verified the importance of the feature proposed in (2). This work was awarded an “Excellent Presentation Award” at the 2022 JAEE Annual Meeting.
(4) Using the input features identified from (2), we trained a collapse mode machine-learning classification model and obtained 95% accuracy. We rained another model using response features which are easier to obtain (i.e., peak floor accelerations), but were only able to obtain 78% accuracy. This shows that using our proposed feature as an input into training the model results in greater accuracy.
(5) We performed analyses of single-degree-of-freedom systems exhibiting Takeda hysteretic behavior. This data was used to derive an expression for estimating residual displacements following seismic events. This expression was found to be more accurate compared to existing equations in the Specifications for Highway Bridges by the Japan Road Association.
|
