地動監測技術對於岩坡破壞案例與落石試驗之運用

本研究運用11個位於橫貫公路的岩坡破壞案例(北橫:1個; 中橫:4個; 南橫: 6個)，搭配13個臺灣寬頻地震網以及氣象局屬之測站，以及12個暫時性的地動監測站之資料來針對(1)案例定位、(2)訊號源分類、(3)量體推估，以及(4)時頻圖推估運動過程等議題進行探討。在定位方面運用相互相關函數法與振幅震源定位法，同時採用水平向與垂直向地動訊號，並依照5km定位標準差的門檻，可以將岩坡破壞案例的定位品質分為A、B、C三類，結果顯示，定位品質為A或B兩類的共有六個案例，其定位誤差最大為3.19km，因此，若有案例的定位品質為B以上，顯示該定位為可信賴之結果，另外，由於頻繁的地震事件也會有較佳的定位展現，所以需要進一步釐清訊號源的歸屬，本研究發現岩坡破壞訊號的延時受控於整體的運動過程，該現象會使得其地震持續時間規模(MD)的數值大於芮氏規模(ML)，藉由分析11個岩坡破壞案例與10個地震事件後，發現規模比(ML/MD)為0.85能有效的區分上述兩種不同的震動源。接著，本研究利用岩坡破壞目錄所提供之量體(V)與岩坡破壞之芮氏規模(ML)和震源位置處的振幅值(A0)來建立回歸關係式，得V=40,756A00.27和 Log(V)=0.67ML+3.49，上述兩迴歸式適用的量體推估範圍介於2,000m3至60,000m3之間。最後，結合案例的時頻圖與影片後發現，當大量的破壞材料(>5,000m3)順著邊坡向下運動時，時頻圖會出現低於2Hz的訊號;而當大塊體直接墜落於路面/坡面時，時頻圖會出現頻帶範圍固定，延時介於5至10秒的柱狀表徵，若塊體持續向下運動，則訊號頻帶的範圍會隨著塊體堆積而逐漸地縮小; 落石的彈跳以及墜落於路面，則會於時頻圖產生延時短暫的脈衝狀訊號，藉由頻帶固定且重複的脈衝狀特徵，或者是密集排列的脈衝狀訊號，可以知道落石是屬於單一岩塊於坡面彈跳，亦或是持續不斷的落石墜落案例。根據上述流程，針對未來發生的岩坡破壞案例，搭配即時訊號，便可以快速的進行定位分析、訊號源辨識、推估破壞的量體，以及從時頻圖判讀破壞材料的運動過程。然而，上述安裝於集水區流域內的地動監測網，僅適用於監測大於2,000m3量體的岩坡破壞案例，並無法針對小量體的落石案例進行監測，因此，本研究利用重複性的現地落石試驗，以及工務段所施作的攔石網試體試驗之資料來了解(1)量體推估、(2)攔石網試驗之地動訊號特性，以及(3)落石軌跡等三項議題。在量體推估方面，本研究發現試驗塊體重量與運動過程中訊號的頻率最低值具有關係性存在，但需要考慮使用測站周圍的地質材料是否一致，因為當撞擊物不一致時，會使得最低頻率產生改變，而降低迴歸式中的相關性。另外，從工務段攔石網試驗，得知試體於坡面彈跳產生的訊號頻率約介於18.94 Hz至25.78 Hz之間，會低於試體撞擊攔石網之頻率29.68 Hz至33.98 Hz。最後，本研究同樣將振幅震源定位法運用於落石軌跡的議題，結果發現該方法較不適用於落石於坡面彈跳點的定位，且產生誤差的原因主要受控於震波傳遞之路徑響應的影響，但對於落石墜落於地面的位置，僅運用架設於地面之地動測站，定位的結果和真實下落位置具有4.6m的誤差。為了進一步釐清落石於坡面上運動之軌跡，本研究利用位於落石運動區域與遠離試驗區的地動測站之高頻訊號(>70Hz)的比值(R)，該數值的突跳可反應出塊體的彈跳行為，因此，突跳的間隔即為兩彈跳間的時間差，而當一測站之高頻訊號的R值持續上升，代表落石的運動方向逐漸接近該測站，反之，則為遠離。低頻訊號(<40Hz)的R值對於地面測站而言，則可以清楚的反應最後掉落地面產生的訊號。從11個岩坡破壞案例與重複性實驗之結果，證實運用地動監測技術能有效且即時運用在監測道路岩坡破壞，同時也能針對高潛勢落石邊坡進行監測。

關鍵字

地動監測技術；岩坡破壞；落石；時頻圖表徵；落石試驗

並列摘要

Eleven rock slope failures(RSFs) along provincial highways (No.7:1, No.8:4, No.20:6) have been analyzed in this research. We used ground vibration signals produced by RSFs from 25 seismic stations to (i)locate RSFs, (ii)distinguish the signal source, (iii)construct regression for a volume estimate, and (iv)classify their physical process with their spectrogram features. First, the cross-correlation-based and amplitude-source-location methods with horizontal and vertical envelopes determine the location of RSFs. Then, according to location uncertainties of 5 km, we can categorize the location quality A, B, and C for each RSF. The result indicates that six events belong to location quality A or B. Their maximum location error is around 3.19km. It nominates that when an RFS presents location quality A or B, the result is reliable. Howerver, the frequent earthquake performs great location results as well; thus, distinguishing the signal sources from earthquake or RSF is necessary. Based on the analysis of 10 earthquakes and 11 RSFs, we found physical process of RSFs dominates the signal duration. The phenomenon inherits that their earthquake duration magnitude(MD) is higher than local magnitude(ML), and the magnitude ratio (ML/ MD) as 0.85 could be a threshold to distinguish these two different signal sources. Further, combine the failures volume(V) with ML and amplitude at source (A0), two regressions V=40,756A00.27 and Log(V)=0.67ML+3.49 have been built for volume estimation. Finally, while video records of RSFs correspond to seismic signals, the processes of giant boulders impacting the ground and massive material moving down on the slope present column-shaped and low bonds of frequency shift spectrograms, respectively. The impacts of rockfall directly link to the pulses of seismic signals. According to the process, we can offer warining within short time by real-time data transmission to show the location of RSF, recognize the signal source, estimate event volume and understand their physical process. However, the above station coverage can only monitor the event with a volume over 2,000m3. To classify the details of small rockfall linking to seismic signals, we present the repeat rockfall experiment to understand the issue about (i) volume estimation, (ii) seismic feature of rockfall in rockfall fences, and (iii) rockfall trajectory. First, we found that the lowest frequency during the process can be built regression with rock mass presenting negative relation. Nevertheless, the selection of the lowest frequency should be from the same impacted material. The contact between different materials contributes to different domain frequencies. Next, the signal frequency of rock bouncing on a slope between 18.94 Hz to 25.78 Hz is lower than rock contact with rockfall fences from 29.68 Hz to 33.98 Hz. Lastly, we utilize the amplitude-source-location method on the location of rock bouncing and falling points. The result indicates that the path effect controls the uncertainty of determining locations and implement poor constrain to the bouncing points. Nevertheless, the method performs well with only a 4.6m location error for the falling point on the ground. To classify rockfall trajectory on the slope, a ratio of the high-frequency signal (>70Hz) of stations on rockfall moving and non-moving areas reflects rock's dynamic. When the ratio rises substantially, the rock is closing to the station at the moving area. On the other hand, a low-frequency signals (<40Hz) ratio presents the final impact of rockfall. This research shows that the seismic technique can monitor the RSF along highways and can be applied to monitor the high sensitivity slope of rockfall.