Real-time hysteresis identification in controlled structures based on restoring force reconstruction and Kalman filter
| Project/Area Number |
21K14284
|
|---|
| Research Category |
Grant-in-Aid for Early-Career Scientists
|
| Allocation Type | Multi-year Fund |
|---|
| Review Section |
Basic Section 23010:Building structures and materials-related
|
|---|
| Research Institution | Kyoto University (2022-2023)
Tohoku University (2021) |
|---|
Principal Investigator |
郭 佳 京都大学, 農学研究科, 准教授 (50868081)
|
|---|
| Project Period (FY) |
2021-04-01 – 2024-03-31
|
| Project Status |
Completed (Fiscal Year 2023)
|
| Budget Amount *help |
¥4,550,000 (Direct Cost: ¥3,500,000、Indirect Cost: ¥1,050,000)
Fiscal Year 2022: ¥1,300,000 (Direct Cost: ¥1,000,000、Indirect Cost: ¥300,000)
Fiscal Year 2021: ¥3,250,000 (Direct Cost: ¥2,500,000、Indirect Cost: ¥750,000)
|
|---|
| Keywords | Kalman filter / Autoencoder / Force identification / Unsupervised learning / Deep learning / Deep neural network / Force reconstruction / Linear multistep method / Data-driven approach / Identify hysteresis / Bayesian estimation / 構造ヘルスモニタリング / カルマンフィルタ / 復元力時刻歴推定 |
|---|
| Outline of Research at the Start |
計算機処理能力の向上と地震観測網の拡充に伴い、観測記録を活用した新たな技術の開発が目ざましい．本研究では、計測箇所の限られた地震観測データのみで免震・制振装置の性能変化や損傷をリアルタイムに推定・検知する新たな方法を開発する．この手法は、観測点数が少なくても高い推定精度が得られることと、復元力モデルを予め用意する必要がないことが特徴である．この手法の有効性と推定精度の検証は、計測装置が豊富に設置された既存の建物の観測データと、振動台実験から得られる実験データを用いて行う．
|
| Outline of Annual Research Achievements |
This year, we have enhanced the algorithms developed in previous years and demonstrate the effectiveness of the proposed method in estimating the local hysteric behaviors of dynamic systems with limited measurement data, as well as its ability to successfully overcome numerical instabilities. Specifically, we proposed a computationally practical approach for recovering the local hysteric behavior based on the combination of Kalman filter and unsupervised autoencoders. In the autoencoder, the measurement data is encoded and the feature of its noise level is first learned in the space of the latent variable. Then another neural network is used to predict system full states with ground motion and latent variable as input. These predictions are used for measurement data reconstruction by a physical guided decoder with the linear state equation from the Kalman filter. The improved algorithm is validated by a numerical example of a multi-story building with dampers on each floor and an experimental test conducted on a two-story base-isolated shear structure. In conclusion, the above-mentioned approach demonstrates computational stability with less measured data, making it more promising for real-world structural health monitoring systems.
|
Report
(3 results)
Research Products
(6 results)
