台灣地震頻繁，若能在震後盡快得到各地區的災損情形，將會提升救災速度，目前震災緊急應變在第一階段僅能推估大致災損情況。本研究希望利用建物內智慧型手機結合眾包技術獲得個別建物受損程度，進而輔助第一階段警急應變之使用。
本研究主要利用手機配備之加速度計量測加速度資料，於觸發後數秒內判斷是否為地震，若為地震，記錄其加速度訊號，並利用Wi-Fi直連技術將手機進行點對點之時間同步，再上傳至伺服器，於伺服器端將訊號方向對齊後，利用系統識別之方法獲得結構自然基礎頻率。本研究假設可利用手機之位置資訊篩選出位於同一棟建築物之手機，且手機數量足夠且均勻分布於各樓層。
在系統識別方法中，選擇了唯輸出分析，包括：頻率域分解法(Frequency Domain Decomposition, FDD)、隨機子空間識別法(Stochastic Subspace Identification, SSI)，和輸入-輸出分析，包括：結合隨機與確定性子空間識別(Combined stochastic and deterministic Subspace Identification, CSI)、頻率響應函數(Frequency Response Function, FRF)等方法進行探討，並模擬智慧型手機感測器之品質及所獲得之訊號進行分析，探討在不同情況下適合之系統識別方法。
在無法得到地表訊號的情況下，僅能使用唯輸出的系統識別方法，但根據數值模擬結果，即使在接近完美的狀況下(時間同步、角度對齊、無雜訊)，FDD識別之頻率誤差接近30%、均方根誤差(Root Mean Square Error, RMSE)為8.2%，而SSI之頻率誤差高達55%、RMSE為11.8%，原因為頻率識別結果受到地震特性影響非常大，誤差過大而不可靠。因此假設在獲得地表訊號且各樓層皆有收到訊號之前提下，使用輸入-輸出分析識別結構自然頻率，其中是否同步對CSI影響較大，是否方向對齊對FRF影響較大。在時間同步且方向對齊使用CSI識別頻率之結果最佳，頻率誤差小於1%、RMSE為0.3%，FRF在時間同步且方向對齊之頻率誤差可小於6.5%、RMSE為2.78%。
本研究並利用鋼結構試體之振動台試驗，驗證所提之方法。試驗結果表明，在使用CSI在時間同步下方向對齊之頻率誤差為0.9%、RMSE為0.5%，而FRF在時間不同步下且方向對齊之頻率誤差為3.5%、RMSE為1.1%。其趨勢與數值模擬相近，因此未來運用上在時間同步下使用CSI，時間不同步下使用FRF應有其可行性。

In Taiwan, earthquakes happen frequently. If we can know the damage situation of each area right after an earthquake, emergency response for disaster relief will be faster . In the first stage of emergency response to the earthquake disaster , only a rough information of damage situation can be estimated. In this study, damage levels of individual buildings are anticipated to be estimated using smartphones based on the crowdsourcing technology , and hopefully this information is helpful for emergency response to the earthquake disaster .
The smartphones of steady state of crowds in the buildings are employed to measure the acceleration response during earthquakes. Once the smartphones were triggered by an excitation, an embedded artificial neural network (ANN) model would classify whether the event was earthquake or not within a few seconds. If the triggered motion was considered an earthquake event, then smartphones will record the whole vibration history of this event. After the earthquake event finished, the smartphones used Wi-Fi Direct connection technology to synchronize the time history if Wi-Fi Direct connection is established on two adjacent smartphones. After the vibration data are uploaded to the cloud server, orientation alignment of the vibration measured on different floors will be conducted. Finally, the fundamental natural frequency of the building will be estimated by system identification methods. This study assumes that it is possible to measure the acceleration response on all the floors in the same building.
The output-only system identification methods including Frequency Domain Decomposition (FDD) and Stochastic Subspace Identification (SSI), and the input-output ones including Combined stochastic and deterministic Subspace Identification (CSI) and Frequency Response Function (FRF) are employed in this study. The quality of acceleration response measured by the smartphones in practice is considered.
In case the ground motion is not available, only the output-only system identification methods can be used. However, according to the results of numerical simulations, even under ideal situation i.e. time is synchronized, orientation aligned, and without noise, the estimated fundamental natural frequency using the output-only system identification methods is not reliable because it will be mixed by the characteristic of ground motion.
Therefore, an algorithm to identify the ground motion signals from the measured time history of different smartphones is proposed. Based on the assumption that the ground motion signals are available, the input-output system identification methods can be used. The results of numerical simulation indicate that whether time synchronization has a great influence on the CSI approach and whether orientation alignment has a great influence on the FRF approach.
Shaking table tests of a four-story steel building were conducted to verify the proposed methods. The results show that the maximum and RMSE of frequency error using CSI with time synchronization and orientation alignment are 0.9% and 0.5% respectively, while the ones using FRF with time asynchronization and orientation alignment are 9.6% and 2.7% respectively. This trend is similar to the numerical simulation. Therefore, it is suggested to use CSI when time synchronization is valid, and use FRF instead when time synchronization is not available.

1. Ozer, E., M. Feng, and D. Feng, Citizen Sensors for SHM: Towards a Crowdsourcing Platform. Sensors, 2015. 15(6): p. 14591-14614.
2. Kong, Q., et al., Structural health monitoring of buildings using smartphone sensors. Seismological Research Letters, 2018. 89(2A): p. 594-602.
3. Yu, Y., et al., Initial validation of mobile-structural health monitoring method using smartphones. International Journal of Distributed Sensor Networks, 2015. 11(2): p. 274391.
4. Feng, M., et al., Citizen Sensors for SHM: Use of Accelerometer Data from Smartphones. Sensors, 2015. 15(2): p. 2980-2998.
5. Ozer, E. and M.Q. Feng, Synthesizing spatiotemporally sparse smartphone sensor data for bridge modal identification. Smart Materials and Structures, 2016. 25(8): p. 085007.
6. Han, R., et al., A cyber-physical system for girder hoisting monitoring based on smartphones. Sensors, 2016. 16(7): p. 1048.
7. Ozer, E., D. Feng, and M.Q. Feng, Hybrid motion sensing and experimental modal analysis using collocated smartphone camera and accelerometers. Measurement Science and Technology, 2017. 28(10): p. 105903.
8. Xie, B., J. Li, and X. Zhao, Research on damage detection of a 3D steel frame model using smartphones. Sensors, 2019. 19(3): p. 745.
9. Matarazzo, T.J., et al., Crowdsensing framework for monitoring bridge vibrations using moving smartphones. Proceedings of the IEEE, 2018. 106(4): p. 577-593.
10. Mei, Q. and M. Gül, A crowdsourcing-based methodology using smartphones for bridge health monitoring. Structural Health Monitoring, 2018. 18(5-6): p. 1602-1619.
11. Kong, Q., et al., Machine learning aspects of the MyShake global smartphone seismic network. Seismological Research Letters, 2019. 90(2A): p. 546-552.
12. Rochford, K., et al., MyShake: using human-centered design methods to promote engagement in a smartphone-based global seismic network. Frontiers in Earth Science, 2018. 6: p. 237.
13. Kong, Q., et al., Machine learning in seismology: Turning data into insights. Seismological Research Letters, 2019. 90(1): p. 3-14.
14. Hsu, T.-Y. and C. Nieh, On-site earthquake early warning using smartphones. Sensors, 2020. 20(10): p. 2928.
15. 劉政言, 基於智慧型手機量測層間位移之建築震後損傷評估研究, in 營建工程系. 2020, 國立臺灣科技大學: 台北市. p. 118.
16. 聶君平, 現地型強震預警APP: My Shake Alert, in 營建工程系. 2019, 國立臺灣科技大學: 台北市. p. 160.
17. Ammon, C.J., The isolation of receiver effects from teleseismic P waveforms. Bulletin-Seismological Society of America, 1991. 81(6): p. 2504-2510.