automaticapproachforfastprocessinganddataanalysisof...

Research ArticleAutomatic Approach for Fast Processing and Data Analysis ofSeismic Ahead-Prospecting Method A Case Study inYunnan China

Wei Zhou 12 Lichao Nie 1 Fahe Sun 3 Xinji Xu 12 and Yi Zhang 12

1Geotechnical and Structural Engineering Research Center Shandong University Jinan 250061 China2School of Qilu Transportation Shandong University Jinan Shandong 250061 China3Shandong Hi-Speed Group Cp Ltd Jinan Shandong 250061 China

Correspondence should be addressed to Lichao Nie lichaonie163com

Received 17 May 2020 Accepted 25 August 2020 Published 14 October 2020

Academic Editor Carlo Rainieri

Copyright copy 2020 Wei Zhou et al )is is an open access article distributed under the Creative Commons Attribution Licensewhich permits unrestricted use distribution and reproduction in any medium provided the original work is properly cited

)e seismic ahead-prospecting method is useful to detect anomalous zones in front of the tunnel face However most existingseismic detection method is designed for drilling and blasting tunnel )e detection method should be improved to satisfy therapid tunneling of Tunnel BoringMachines (TBMs))is study focuses on reducing the time spent on seismic data processing andresult analysis )erefore to reduce the data processing time an automatic initial model establishment method based onsurrounding rock grade is proposed To reduce the time spent on result analysis and avoid subjective judgment a modified k-means++ method is adopted to interpret the detecting results and extracting anomalous zones )e efficacy of the developedmethod is demonstrated by field tests )e fractured zones such as cavity collapse and fissure are successfully predictedand identified

1 Introduction

More and more tunnels are constructed for efficienttransport in long-distance traffic [1] Tunnel Boring Ma-chines (TBMs) have been widely used in the tunnel industrydue to their advantages of higher efficiency and being safeand environmentally friendly compared to the traditionaldrilling and blasting methods Currently TBM excavationsare increasingly being employed in complex geologies suchas karst terrains and mountain areas However TBMs incurhigh risk and have poor adaptation when passing throughadverse geologies [2] If the distribution of adverse geologycannot be accurately predicted water and mud inrush mayoccur during excavation and eventually lead to delays andeven casualties Moreover water inrush may result in watertable dropping and lead to damage to the ecosystem [3] Toreduce the impact of water inrush on tunnel constructionand ecosystem adverse geology investigations should beconducted before grouting Geophysical exploration

methods provide a nondestructive means for detectinggeological structures in front of tunnel faces with advantagesof short detecting time and low cost and they are widelyused in engineering investigation [4 5] However thegeotechnical investigation performed on the ground surfacecan only provide rough profiles of the tunnel constructionarea rather than the exact geological conditions of the localarea in front of the tunnel face during the tunneling [4])erefore researchers try to perform geophysical explora-tion in tunnels for detecting anomalous zones in front of thetunnel face [2] Detecting in a tunnel not only improves thedetection by being closer to the anomalous zones but alsocan obtain surrounding rock lithology integrity and itsphysical properties as prior information for detection duringexcavation )ereby the safety risks during tunnel con-struction can be decreased by adjusting the tunnelingscheme according to the local geological conditions obtainedby the ahead-prospecting method )e seismic detectionmethod is one of the frequently used ahead-prospecting

HindawiMathematical Problems in EngineeringVolume 2020 Article ID 8947591 11 pageshttpsdoiorg10115520208947591

exploration methods which is relying on the detection ofelastic differences in the physical properties such as velocityof shear wave velocity of pressure wave and density )erehave been lots of developments in the seismic ahead-pro-specting method [6] In most cases data processing of theseismic ahead-prospecting method is based on the migrationmethod During the migration correction process the ve-locity needs to be calculated based on picking the first arrivalwave for the subsequent imaging procedure However firstarrival picking is time-consuming which can take up to20ndash30 of the total processing time [7]

To satisfy the requirement of TBM rapid constructionthe time spent on ahead prospecting should be reduced toless interference in the construction Fortunately seismicpropagation velocity is related to lithology and integrity ofrock and the distribution of lithology and integrity of rockalong the tunnel trunk can be roughly obtained by thesurface survey )erefore geological information obtainedfrom surface surveys can be used as prior information toestablish initial velocity models for migration correctionwhich can reduce time spent on data processing

Regarding the seismic imaging result analysis theimaging results are the distribution of geophysical prop-erties Usually geophysical experts should analyze theresults and interpret the results as a geological model(distribution of adverse geology such as faults and karstcaves) according to their experience )us the constructorswithout geophysical experience can easily understand thelocal geology conditions ahead of the tunnel face and thenadjust the construction strategy A common analysismethod is to involve geological experts to manually identifyareas with strong reflections However this method re-quires geophysical experts to analyze the imaging resultsaccording to their experience resulting in the accuracy ofmanual identification depending on the experience of theinterpreter )is makes manual analysis time-consumingand the division of anomalous zones boundaries in analysisresults rely on subjective judgment [8] Recently attemptshave been made to adopt automated data analysis methodsin the geological field to obtain objective data analysisresults within a relatively short time K-means algorithm isone of the effective unsupervised learning algorithms usedto solve the well-known clustering problem By adopting k-means++ method Li et al successfully interpreted theresistivity inversion results and obtained the boundaries ofhigh-risk areas [9] Di et al introduced k-means clusteranalysis into seismic data to accurately delineate theboundary of salt bodies [10] )erefore k-means has beenproved to be an effective tool to interpret geophysical re-sults Meanwhile when conducting manual data analysis itis necessary to comprehensively consider geological in-formation and geophysical detection results and analyzeand compare the detection results under similar engi-neering geological conditions to improve the accuracyHowever this process is time-consuming due to the dataanalysis of standards that rely on experience As opposed todetermining the anomalous zones by experience k-means++ algorithm is based on certain criteria which ismore efficient and objective than manual results analysis

Gaoligongshan is a national nature reserve located inYunnan China To develop the economy a tunnel con-structed by TBM was built through Gaoligongshan In thispaper we focused on reducing the time-consumption ofseismic processing and result analysis to obtain the localgeological conditions in front of the tunnel face To reducethe time spent on data processing an initial velocity modelrelated to geological conditions is proposed)e concept wasto utilize the distribution of lithology and integrity of for-mations obtained from the surface survey to establish theinitial velocity model and then update it according to theexposed geology during the tunneling to ensure its accuracyIn addition an automated data analysis method based on k-means++ algorithm is proposed to cluster the seismic im-aging results for reducing time spent on result analysisFinally the proposed method was tested in Gaoligongshantunnel for preventing water inrush

2 Imaging and Results Analysis ofAnomalous Zone

)e original seismic data collected by the data acquisitionsystem are seismic traces but the distribution of theanomalous zones in front of the tunnel face can only beobtained after data processing In this section an automatedresult analysis method based on k-means++ method isadopted to the seismic imaging results for reducing the timespent on result analysis

21 Automatic Initial Model Establishment with GeologicalInformation )e initial velocity model is one of the im-portant factors affecting the migration correction quality[11] Krebs et al integrated various sources of subsurfacevelocity information such as surface seismic reflections [12]surface seismic direct arrivals well data and prior geologicinformation to estimate the velocity model Sala et alpresented a velocity building scheme using informationfrom borehole and laboratory data [13] In this study anautomated velocity model generation method based ondynamic geological information and historical wave velocitywas used to improve the imaging accuracy and reduce thedata processing time

During seismic detection the average value of the directwave velocity obtained by all source-detectors pairs isasserted as the average velocity of the elastic waveAccording to the lithology and classification of the rockmassbetween the geophones and shot points this value is thenrecorded as the direct wave velocity under this condition)e historical wave velocity is generated from the directwave velocity at each detection and the wave velocity datafrom the preliminary geotechnical investigation )e ve-locity model is generated from the historical wave velocitiesand then used as the initial model for the migrationcorrection

It has to be noted that a geologic longitudinal sectionmap can provide the lithology and classification distributionalong the tunnel trunk However tunnels can be locatedhundreds or thousands of meters below the ground surface

2 Mathematical Problems in Engineering

and the estimated geological conditions along the tunneltrunk may have deviations with the actual geological con-ditions To obtain a more accurate distribution of the for-mation the geologic longitudinal section map is revisedaccording to the excavation record If the actual revealedmileage of the geological structure or phenomenon revealedsuch as the lithological interface and weathering slot isdifferent from the mileage in the geologic longitudinalsection map the section will be corrected according to thedeviations between the revealed geological conditions andthe geologic longitudinal section map generated from theprevious survey on the surface ground )en the historicalwave velocities are automatically matched to generate theinitial model according to the lithology and rock mass typedistribution of the unexcavated longitudinal profile in frontof the tunnel Finally the velocity distribution in front of thetunnel face is obtained by using the geologic longitudinalsection map In this manner wave velocity models can beestablished automatically according to the distribution oflithology and rock mass types in the geologic longitudinalsection map rather than assigning the same velocity to allrock masses within the imaging region

A fixed finite-difference grid was established for mi-gration correction When processing the datasets acquiredby the acquisition system the velocity information is au-tomatically loaded from the records and assigned to the grid)en a 1D velocity model can be generated and used as theinitial velocity model for migration correction

22 Automated Result Analysis Based on k-Means++MethodIn this study a seismic ahead-prospectingmethod is adoptedto image the anomalous zones [14] Its main process includesband-pass filtering diffusion compensation wavefield sep-aration and migration correction )e workflow of theseismic data processing is presented in Figure 1 To obtainthe local geological conditions ahead of the tunnel face theresult analysis should be conducted to convert the imagingresults into geological units Anomalous zones such aslithological interface and fracture zone are usually of con-cern when interpreting and detecting the results of geo-logical conditions in front of the tunnel face )e commonresult analysis of seismic waves requires geophysical expertsto analyze the imaging results according to their experienceHowever the boundary of anomalous zones is difficult torecognize )erefore we adopted the k-means++ algorithm[15] to the reflection coefficient to automatically and quicklyclarify the area representative of adverse geological struc-tures based on the imaging results

K-means++ is an important technique for clustering Inthis study the k-means++ method is applied after obtainingthe migration correction results based on the automatedvelocity model For the result analysis problem the purposeis to minimize the function as follows [15]

empty 1113944minx minus c

2 (1)

Actually the imaging results consist of cells with a specificreflection coefficient We consider each cell as a sample and

the spatial coordinates and reflection coefficients of the cellswere used as parameters to cluster the cells Because themagnitude of the coordinate parameter is too large comparedto the reflection coefficient the coordinate parameters andreflection coefficient are normalized as follows first

Rnor Rori minus Rave

σ (2)

where Rnor is the normalized parameter of the samplesRoriis the original parameter Rave is the average value and σ isthe standard deviation

)en a scheme for calculating Euclidean distance isproposed )e distance calculated based on the coordinatesand reflection coefficient was used as the criteria for centroidselection and sample classification the shortest distance D(x) between a sample and the closest centroid is calculated as

D(x) s(x y z) minus c(x y z)2

+ λs(r) minus c(r)2 (3)

where s (x y z) is the spatial coordinates of the cell c (x y z)is the spatial coordinates of the centroid s (r) and c (r) are thereflection coefficient of the samples and centroid respec-tively and λ is a factor adjusting the weight of the reflectioncoefficient

)e probability that the sample is chosen as the nextcenter is as follows

p D

2(x)

1113936D

2(x)

(4)

1File formatting

Data inputIntegral transformation

2 Trace equalization

Read the observed data3

Raw data

Bandpass filtering4

Deconvolution

f minus k filteringτ minus p filtering

5

6Velocityanalysis

Initialvelocitymodel

Migration imaging7

Figure 1 Flowchart of the seismic processing method

Mathematical Problems in Engineering 3

Finally cells in the migration correction results areclassified according to the spatial position and the reflectioncoefficient )e workflow of the k-means++ is as follows

(1) )e purpose of adopting the k-means method is toclassify the migration correction into k geologicalunits with different rock mass integrity )ereforethe value of k is determined according to the geo-logical analysis which mainly depends on the pos-sible number of lithological types and the possiblenumber of rock quality classifications in front of thetunnel face

(2) A cell of migration correction is randomly selected asthe initial centroid )en the distance between thecentroid and all the remaining cells in the migrationcorrection results is calculated as in equation (3) andthe cell with the largest distance is selected as thesecond centroid Next the distance calculation andcentroid selected steps are repeated until k centroidsare determined

(3) )e distance between the centroids and the cellsexcept the k centroids is calculated separately andeach sample is assigned to the cluster containing thenearest centroid In this manner the cells of themigration correction results are assigned to k clus-ters However this classification result is not optimaland needs to be further optimized according to thedistance

(4) )erefore the location of the centroids of eachcluster is recalculated

(5) If the new centroids are the same as the previousones the k-means++ method is considered to beconvergent If the new centroids are different fromthe previous ones steps 3 to 5 are repeated until thek-means method becomes convergent

Finally the cluster representing the strong reflectionareas can be selected based on the statistical results of eachcluster)e spatial distribution of such cells belonging to thiscluster represents the location of anomalous zones such asfaults and karst caves

3 Field Test

31 Site Description )e proposed method was testedthrough application to an actual tunnel Gaoligongshantunnel is the longest railway tunnel in China with a length of34538 km 13 km of which was constructed using a TBM of9m diameter )e maximum depth of the tunnel floor fromthe ground surface is about 1155m)e tunnel site is locatedin the boundary of the Eurasian plate and Indian Oceanplate thus featuring complex geological conditions(Figure 1(a)) Meanwhile many valleys developed resultingin great topographic relief which makes it easy to collectgroundwater and then cause the rock below to be fractured)erefore obtaining the adverse geological conditions infront of the tunnel face through ahead-prospecting methodand then formulating a tunneling strategy are very impor-tant to ensure the safe and efficient construction of TBM

Actually the detection was performed practically alongthe whole line of the tunnel two tests are selected to verifythe proposed method A longitudinal section map illus-trating the geological conditions along the tunnel trunkaccording to the surface survey is described in Figure 2(c)Meanwhile the seismic observed configuration is as shownin Figure 3 To illustrate the applicability of the proposedmethod a test with a relative accurate surface survey in-vestigation and a test with biased survey investigation wereselected here

32 Field Test 1 at Mileage of D1k 225 + 313 )e first fieldgeological detection was conducted at the mileage of D1k225 + 313 According to the geotechnical investigationperformed on the ground surface none adverse geologycondition can be inferred at this mileage Howeveraccording to the actual excavation rock fissures cavitycollapse and alteration zones are developed as shown inFigure 4 )is indicates that the actual rock mass is con-sistently poor with the survey detection results

To determine geological conditions in front of the tunnelface and adjust the construction strategy for the sake ofconstruction efficiency and safety an experiment wasconducted at mileage of D1k 225 + 313 during the downtimeof tunnel excavation)e observed configuration is as shownin Figure 3 In both two tests the seismic wave is excited byan 8-pound hammer with a frequency of 700ndash1500Hzrecording sampling is 0125ms and the whole recordingtime is 250ms According to the corrected geologic longi-tudinal section map and rock exposed in the tunnel onlygranite is developed between mileages of D1k 225 minus 313 and225 + 213 and the parameters of the model are set toVp 3822ms )e migration correction results are shownin Figure 5 In order to recognize the location and shape ofthe anomalous zones more clearly the k-means++ methodwas adopted to further process the imaging results)e valueof k was set to 4 because there are 4 possible classifications ofrocks according to geological analysis )e descriptive sta-tistical parameters of each cluster are summarized in Table 1Cluster 3 contains both the higher positive reflection and thehigher negative reflection values of the entire imaging re-gion and the standard deviation of the cluster is also large)erefore cluster 3 represents the fragmental area becausethe statistical parameters of cluster 3 are consistent with thestrong positive and negative reflection characteristics of thefragmental area Because we focused only on anomalouszones the image of cluster 3 is as shown in Figure 6 )egeological sketch is drawn according to the excavatingrecords to verify the analysis results and is as shown inFigure 7

As shown in Figures 6 and 7 manual analysis results andautomated analysis results based on k-means++ algorithmare similar )e boundary of the anomalous zones is moredistinct in the k-means++ results Strong reflections fall intothree zones A1 A2 and A3 )ese three zones invade thetunnel trunk approximately at mileages of D1k 225 + 289 toD1k 225 + 277 D1k 225 + 265 to D1k 223 + 257 and D1k225 + 237 to D1k 225 + 213 Combined with the geological


D1k

223

+ 0

00

D1k

225

+ 5

00

D1k

225

+ 2

50

D1k

225

+ 0

00

D1k

224

+ 7

50

D1k

224

+ 5

00

D1k

224

+ 2

50

D1k

224

+ 0

00

D1k

223

+ 7

50

D1k

223

+ 5

00

D1k

223

+ 2

50

1250

1350

17501700165016001550150014501400

1350

Grade IV Grade V Grade II

Grade III

Granite+ +

Tunnel face duringexperiment 1

Tunnel face duringexperiment 2 Tunneling excavation direction

YunnanGaoligongshan tunnel

(a) (b)

(c)

Figure 2 Sketch of the cross-section illustrating (a) (b) the geographic location of the Gaoligongshan tunnel and (c) the distribution of therock lithology and integrity which is obtained according to both the exposed surrounding rock and survey investigation [16]

R1 ndash R4

R5 ndash R8

S1 ndash S3

S7 ndash S9

S4 ndash S6

S10 ndash S12

13m3m 1m

9m Tunnelface

Surrounding rock

SourcesReceivers

Figure 3 A lateral view of the observed configuration adopted to the seismic detection in Gaoligongshan tunnel

Alteration zone

(a)

Cavity collapse

(b)

Figure 4 Surrounding rock exposed during excavation (a) Alteration zone with brown muddy at mileage of D1k 225 + 325 (b) cavitycollapse at the left arch shoulder at mileage of D1k 225 + 318


analysis the fractured zones at the different mileages areinterpreted as follows

(i) At mileages of D1k 225+ 289 to D1k 225+ 277 D1k225+ 265 toD1k 225+ 257 andD1k 225+ 237 to D1k225+ 213 three anomalous zones appear whichcontain strong reflectionsWe can thus expect that therock mass within this area is fractured as indicated bythe lower RMR value suggesting that the integrity ofthe rock mass is poor in the section Meanwhile the

anomalous zones are more concentrated at mileagesof D1k 225+ 237 to D1k 225+ 213 )us we canexpect that rock mass could be the worst within themileages of D1k 225+ 313 to D1k 225+ 213

(ii) At mileages of D1k 225 + 277 to D1k 225 + 265 andD1k 225 + 257 to D1k 225 + 237 the reflections arenot remarkable )us it is assumed that the rockmass of this zone is the same as that exposed in thetunnel face

20

A1

D1k 225 + 313 225 + 293 225 + 273 225 + 253 225 + 233 225 + 213

A2

A3

1

05

0

ndash05

ndash1

0

Z (m

)

Y (m)

ndash20

ndash200

20

X (m)

Figure 5 Seismic imaging results by manual result analysis at D1k 225 + 313)e detecting depth is 100m the x-axis is parallel to the tunnelaxis the y-axis is parallel to the tunnel face and the z-axis is perpendicular to the bottom of the tunnel

A1A2 A3

20

20

0

0

Z (m

)

ndash20

ndash20

Y (m)

D1k225 + 313 225 + 293 225 + 273 225 + 253

X (m)

225 + 233 225 + 213

Figure 6 K-means++ results based on the seismic detection results)e detecting depth is 100m the x-axis is parallel to the tunnel axis they-axis is parallel to the tunnel face and the z-axis is perpendicular to the bottom of the tunnel

Table 1 Descriptive statistical parameters of different clusters for experiment 1

Clusters Average Standard deviation Min Max MedianCluster 1 minus000002 00124 minus00352 00355 minus000003Cluster 2 minus000004 00725 minus01055 01055 minus00457Cluster 3 minus00023 01422 minus02313 02264 minus01060Cluster 4 minus00013 00332 minus00528 00525 minus00158


To verify the efficacy of the proposed method a geo-logical sketch map was drawn during the tunnel construc-tion)e automated analysis result and the geological sketchmap are compared in Figure 7 Some cavity collapse andfissure exposed during excavation between mileages of D1k225 + 313 and D1k 225 + 213 are as shown in Figure 8

33 Field Test 2 atMileage of D1k 223 + 563 )e second fieldgeological detection was conducted at the mileage of D1k223 + 563 According to the geotechnical investigationperformed on the ground surface the TBM tunnel passedthrough granite strata at this mileage When the tunnelreached the section between the mileages of D1k 223 + 600and D1k 223 + 580 the surrounding rock mass was relativelyintact coinciding well with the surrounding rock classifi-cation from geotechnical investigation However the rockmass became much more fractured in the section betweenmileages of D1k 223 + 580 and D1k 223 + 563 To determinegeological conditions in front of the tunnel face and adjustthe construction strategy for the sake of construction effi-ciency and safety an experiment was conducted at mileageof D1k 223 + 563 during the downtime of tunnel excavation)e observed configuration is as shown in Figure 3

According to the corrected geologic longitudinal sectionmap and rock exposed in the tunnel only granite is

developed between mileages of D1k 223 minus 563 and D1k223 + 463 and the parameters of the model are set toVp 3358ms )e migration correction results are shownin Figure 9 In order to recognize the location and shape ofthe anomalous zones more clearly the k-means++ methodwas adopted to further process the imaging results)e valueof k was set to 4 because there are 4 possible classifications ofrocks according to geological analysis )e descriptive sta-tistical parameters of each cluster are summarized in Table 2According to the method for judging the clusters corre-sponding to the broken area mentioned in experiment 1cluster 2 of experiment 2 corresponds to the broken areaBecause we focused only on anomalous zones the image ofcluster 2 is as shown in Figure 10 )e geologic sketch isdrawn according to the excavating records to verify theanalysis results and is as shown in Figure 11

As shown in Figures 9 and 10 manual analysis resultsand automated analysis results based on k-means++ algo-rithm are similar Moreover the anomalous zone boundaryof automated analysis results is more distinct Strong re-flections fall into three zones A1 A2 and A3 )ese threezones invade the tunnel trunk approximately at mileages ofD1k 223 + 563 to D1k 223 + 530 D1k 223 + 500 to D1k223 + 488 and D1k 223 + 482 to D1k 223 + 463 Combinedwith the geological analysis the fractured zones at thedifferent mileages are interpreted as follows

Anomalous zones(a)

(b)

Le foot oftunnel arch

Right foot oftunnel arch

D1k225 + 313

FissureCavity collapse

Underground water

225 + 293 225 + 273 225 + 253 225 + 233 225 + 213

20

0

ndash20

Figure 7 Comparison of the k-means++ result and the geological sketch map based on an actual excavation (a) k-means++ result atmileages of D1k 225 + 313 to D1k 225 + 213 and (b) sketch map of the actual excavation at mileages of D1k 225 + 313 to D1k 225 + 213


(i) At mileages of D1k 223 + 563 to D1k 223 + 530 D1k223 + 500 to D1k 223 + 488 and D1k 223 + 482 toD1k 223 + 463 three anomalous zones appear whichcontains strong reflections We can thus expect thatthe rock mass within this area is fractured as indi-cated by the lower RMR value suggesting that theintegrity of the rock mass is poor in the sectionMeanwhile the scale of the anomalous zone atmileages of D1k 223 + 563 to D1k 223 + 530 could bethe largest within these three anomalous zones

(ii) At mileages of D1k 223+530 to D1k 223+500 andD1k 223+ 488 to D1k 223+482 the reflections are notremarkable )us it is assumed that the rock mass ofthis zone is the same as that exposed in the tunnel face


(a) (b)

Figure 8 Surrounding rock exposed during excavation (a) Cavity collapse at mileage of D1k 225 + 265 (b) fissure exposed at mileage ofD1k 225 + 220

20

0

A1

A2A3

ndash20ndash20

20

0

D1k223 + 563 223 + 543 223 + 523 223 + 503 223 + 483 223 + 463

1

05

0

ndash05

ndash1



Clusters Average Standard deviation Min Max MedianCluster 1 00010 00446 minus00579 00579 00338Cluster 2 minus00017 00710 minus01183 00974 00570Cluster 3 minus00002 0009 minus00161 00161 minus00003Cluster 4 minus00021 00246 minus00351 00350 minus00171


20

A1

A2A3

0

Z (m

)

Y (m)

X (m)

ndash20ndash20

0

20D1k223 + 563 223 + 543 223 + 523 223 + 503 223 + 483 223 + 463

Figure 10 K-means++ results based on the seismic detection results conducted at D1k 223 + 563)e detecting depth is 100m the x-axis isparallel to the tunnel axis the y-axis is parallel to the tunnel face and the z-axis is perpendicular to the bottom of the tunnel

Anomalous zones(a)

(b)Le foot oftunnel arch


D1k223 + 563 223 + 543

FissureCavity collapseDripping water

223 + 523 223 + 503

240deg 45deg

223 + 483 223 + 463

20

0

ndash20

0 10 20 30 40 50 60 70 80 90



4 Conclusion

Existing seismic ahead-prospecting method for detectinganomalous zones ahead of the tunnel face may delay the TBMtunneling)erefore in this paper an automatic initial modelestablishment method with geological information is pro-posed to reduce time spent on data processing Moreover anautomated analysis method based on k-means++ method isproposed to facilitate and reduce time spent on data analysiswhich converts the seismic imaging results into severalgeological units By adopting the proposed methods theboundaries of the anomalous zones can be obtained whichcould help develop reinforcement strategies Moreover in-stead of analyzing and comparing the detection results undersimilar engineering geological conditions the anomalouszones can be identified according to the statistical parametersof different clusters in the k-means results In the field teststhe boundaries of the anomalous zones were obtained whichmatch well with the exposed rock )ough the detectingresults meet the engineering construction requirements inthe future methods such as full waveform inversion would bestudied to further improve our predicting results

Data Availability

)e data used to support the findings of this study areavailable from the corresponding author upon request

Conflicts of Interest

)e authors declare that there are no conflicts of interestregarding the publication of this paper

Acknowledgments

)e authors gratefully acknowledge the support of theScience amp Technology Program of Department of Transportof Shandong Province under Grant no 2019B47_2

References

[1] C Butscher P Huggenberger E Zechner et al ldquoImpact oftunneling on regional groundwater flow and implications forswelling of clay-sulfate rocksrdquo Engineering Geology vol 117no 3-4 pp 198ndash206 2011

[2] S Li L Nie B Liu et al ldquo)e practice of forward prospectingof adverse geology applied to hard rock TBM tunnel con-struction the case of the songhua river water conveyanceproject in the middle of jilin Provincerdquo Engineering vol 4no 1 pp 131ndash137 2018

[3] V Vincenzi A Gargini N Goldscheider et al ldquoUsing tracertests and hydrological observations to evaluate effects oftunnel drainage on groundwater and surface waters in theNorthern Apennines (Italy)rdquo Hydrogeology Journal vol 17no 1 pp 135ndash150 2009

[4] J Park K-H Lee J Park H Choi and I-M Lee ldquoPredictinganomalous zone ahead of tunnel face utilizing electrical re-sistivity I Algorithm and measuring system developmentrdquoTunnelling and Underground Space Technology vol 60pp 141ndash150 2016

[5] S H Hee S K Dae and J P Inn ldquoApplication of electricalresistivity techniques to detect weak and fracture zones duringunderground constructionrdquo Environ Earth Sci vol 60pp 723ndash731 2010

[6] Y Ashida ldquoSeismic imaging ahead of a tunnel face with three-component geophonesrdquo International Journal of Rock Me-chanics and Mining Sciences vol 38 no 6 pp 823ndash831 2001

[7] J I Sabbione and D R Velis ldquoAutomatic first-breaks pickingnew strategies and algorithms[J]rdquo Geophysics vol 75 no 42010

[8] A Asjad and D Mohamed ldquoA new approach for salt domedetection using a 3D multidirectional edge detectorrdquo AppliedGeophysics vol 12 no 3 pp 334ndash342 2015

[9] S Li S Xu L Nie et al ldquoAssessment of electrical resistivityimaging for pre-tunneling geological characterization - a casestudy of the Qingdao R3metro line tunnelrdquo Journal of AppliedGeophysics vol 153 pp 38ndash46 2018

[10] H Di M Shafiq G Alregib et al ldquoMulti-attribute k-meansclustering for salt-boundary delineation from three-dimen-sional seismic datardquo Geophysical Journal Internationalvol 215 no 3 pp 1999ndash2007 2018

[11] X Gong L Han J Niu X Zhang D Wang and L DuldquoCombined migration velocity model-building and its ap-plication in tunnel seismic predictionrdquo Applied Geophysicsvol 7 pp 265ndash271 2010

[12] J R Krebs L K Bear and J Liu ldquoIntegrated velocity modelestimation for accurate imagingrdquo SEG Technical ProgramExpanded Abstracts vol 22 2014

[13] P Sala N Tisato O A Pfiffner and M Frehner ldquoBuilding a3D geological near surface model from borehole and labo-ratory datardquo Geophysical Research Abstracts vol 14 2012

[14] B Liu L Chen S Li et al ldquo)ree-Dimensional seismic ahead-prospecting method and application in TBM tunnelingrdquo

Cavity collapse

(a)

Cavity collapse

(b)

Figure 12 (a) Cavity collapse at mileage of D1k 223 + 550 (b) Cavity collapse at mileage of D1k 223 + 495


Journal of Geotechnical and Geoenvironmental Engineeringvol 143 no 12 2017

[15] D Arthur and S Vassilvitskii ldquok-means++ the advantages ofcareful seedingrdquo in Proceedings of the Eighteenth AnnualACM-SIAM Symposium on Discrete Algorithms pp 1027ndash1035 Philadelphia PA USA January 2007

[16] C Chen and G Wang ldquoDiscussion on the interrelation ofvarious rock mass quality classification systems at home andabroadrdquo Journal of Rock Mechanics and Geotechnical Engi-neering vol 21 no 12 pp 894ndash900 2002


exploration methods which is relying on the detection ofelastic differences in the physical properties such as velocityof shear wave velocity of pressure wave and density )erehave been lots of developments in the seismic ahead-pro-specting method [6] In most cases data processing of theseismic ahead-prospecting method is based on the migrationmethod During the migration correction process the ve-locity needs to be calculated based on picking the first arrivalwave for the subsequent imaging procedure However firstarrival picking is time-consuming which can take up to20ndash30 of the total processing time [7]

To satisfy the requirement of TBM rapid constructionthe time spent on ahead prospecting should be reduced toless interference in the construction Fortunately seismicpropagation velocity is related to lithology and integrity ofrock and the distribution of lithology and integrity of rockalong the tunnel trunk can be roughly obtained by thesurface survey )erefore geological information obtainedfrom surface surveys can be used as prior information toestablish initial velocity models for migration correctionwhich can reduce time spent on data processing

Regarding the seismic imaging result analysis theimaging results are the distribution of geophysical prop-erties Usually geophysical experts should analyze theresults and interpret the results as a geological model(distribution of adverse geology such as faults and karstcaves) according to their experience )us the constructorswithout geophysical experience can easily understand thelocal geology conditions ahead of the tunnel face and thenadjust the construction strategy A common analysismethod is to involve geological experts to manually identifyareas with strong reflections However this method re-quires geophysical experts to analyze the imaging resultsaccording to their experience resulting in the accuracy ofmanual identification depending on the experience of theinterpreter )is makes manual analysis time-consumingand the division of anomalous zones boundaries in analysisresults rely on subjective judgment [8] Recently attemptshave been made to adopt automated data analysis methodsin the geological field to obtain objective data analysisresults within a relatively short time K-means algorithm isone of the effective unsupervised learning algorithms usedto solve the well-known clustering problem By adopting k-means++ method Li et al successfully interpreted theresistivity inversion results and obtained the boundaries ofhigh-risk areas [9] Di et al introduced k-means clusteranalysis into seismic data to accurately delineate theboundary of salt bodies [10] )erefore k-means has beenproved to be an effective tool to interpret geophysical re-sults Meanwhile when conducting manual data analysis itis necessary to comprehensively consider geological in-formation and geophysical detection results and analyzeand compare the detection results under similar engi-neering geological conditions to improve the accuracyHowever this process is time-consuming due to the dataanalysis of standards that rely on experience As opposed todetermining the anomalous zones by experience k-means++ algorithm is based on certain criteria which ismore efficient and objective than manual results analysis

Gaoligongshan is a national nature reserve located inYunnan China To develop the economy a tunnel con-structed by TBM was built through Gaoligongshan In thispaper we focused on reducing the time-consumption ofseismic processing and result analysis to obtain the localgeological conditions in front of the tunnel face To reducethe time spent on data processing an initial velocity modelrelated to geological conditions is proposed)e concept wasto utilize the distribution of lithology and integrity of for-mations obtained from the surface survey to establish theinitial velocity model and then update it according to theexposed geology during the tunneling to ensure its accuracyIn addition an automated data analysis method based on k-means++ algorithm is proposed to cluster the seismic im-aging results for reducing time spent on result analysisFinally the proposed method was tested in Gaoligongshantunnel for preventing water inrush

2 Imaging and Results Analysis ofAnomalous Zone

)e original seismic data collected by the data acquisitionsystem are seismic traces but the distribution of theanomalous zones in front of the tunnel face can only beobtained after data processing In this section an automatedresult analysis method based on k-means++ method isadopted to the seismic imaging results for reducing the timespent on result analysis

21 Automatic Initial Model Establishment with GeologicalInformation )e initial velocity model is one of the im-portant factors affecting the migration correction quality[11] Krebs et al integrated various sources of subsurfacevelocity information such as surface seismic reflections [12]surface seismic direct arrivals well data and prior geologicinformation to estimate the velocity model Sala et alpresented a velocity building scheme using informationfrom borehole and laboratory data [13] In this study anautomated velocity model generation method based ondynamic geological information and historical wave velocitywas used to improve the imaging accuracy and reduce thedata processing time

During seismic detection the average value of the directwave velocity obtained by all source-detectors pairs isasserted as the average velocity of the elastic waveAccording to the lithology and classification of the rockmassbetween the geophones and shot points this value is thenrecorded as the direct wave velocity under this condition)e historical wave velocity is generated from the directwave velocity at each detection and the wave velocity datafrom the preliminary geotechnical investigation )e ve-locity model is generated from the historical wave velocitiesand then used as the initial model for the migrationcorrection

It has to be noted that a geologic longitudinal sectionmap can provide the lithology and classification distributionalong the tunnel trunk However tunnels can be locatedhundreds or thousands of meters below the ground surface







2 (1)




σ (2)







p D

2(x)

1113936D

2(x)

(4)

1File formatting




Raw data

Bandpass filtering4

Deconvolution


5

6Velocityanalysis


Migration imaging7










3 Field Test







D1k

223

+ 0

00

D1k

225

+ 5

00

D1k

225

+ 2

50

D1k

225

+ 0

00

D1k

224

+ 7

50

D1k

224

+ 5

00

D1k

224

+ 2

50

D1k

224

+ 0

00

D1k

223

+ 7

50

D1k

223

+ 5

00

D1k

223

+ 2

50

1250

1350

17501700165016001550150014501400

1350


Grade III

Granite+ +




(a) (b)

(c)


R1 ndash R4

R5 ndash R8

S1 ndash S3

S7 ndash S9

S4 ndash S6

S10 ndash S12

13m3m 1m

9m Tunnelface

Surrounding rock

SourcesReceivers


Alteration zone

(a)

Cavity collapse

(b)







20

A1

D1k 225 + 313 225 + 293 225 + 273 225 + 253 225 + 233 225 + 213

A2

A3

1

05

0

ndash05

ndash1

0

Z (m

)

Y (m)

ndash20

ndash200

20

X (m)


A1A2 A3

20

20

0

0

Z (m

)

ndash20

ndash20

Y (m)

D1k225 + 313 225 + 293 225 + 273 225 + 253

X (m)

225 + 233 225 + 213










Anomalous zones(a)

(b)



D1k225 + 313


Underground water

225 + 293 225 + 273 225 + 253 225 + 233 225 + 213

20

0

ndash20






(a) (b)


20

0

A1

A2A3

ndash20ndash20

20

0

D1k223 + 563 223 + 543 223 + 523 223 + 503 223 + 483 223 + 463

1

05

0

ndash05

ndash1





20

A1

A2A3

0

Z (m

)

Y (m)

X (m)

ndash20ndash20

0

20D1k223 + 563 223 + 543 223 + 523 223 + 503 223 + 483 223 + 463


Anomalous zones(a)



D1k223 + 563 223 + 543


223 + 523 223 + 503

240deg 45deg

223 + 483 223 + 463

20

0

ndash20

0 10 20 30 40 50 60 70 80 90



4 Conclusion


Data Availability




Acknowledgments


References















Cavity collapse

(a)

Cavity collapse

(b)












2 (1)




σ (2)







p D

2(x)

1113936D

2(x)

(4)

1File formatting




Raw data

Bandpass filtering4

Deconvolution


5

6Velocityanalysis


Migration imaging7










3 Field Test







D1k

223

+ 0

00

D1k

225

+ 5

00

D1k

225

+ 2

50

D1k

225

+ 0

00

D1k

224

+ 7

50

D1k

224

+ 5

00

D1k

224

+ 2

50

D1k

224

+ 0

00

D1k

223

+ 7

50

D1k

223

+ 5

00

D1k

223

+ 2

50

1250

1350

17501700165016001550150014501400

1350


Grade III

Granite+ +




(a) (b)

(c)


R1 ndash R4

R5 ndash R8

S1 ndash S3

S7 ndash S9

S4 ndash S6

S10 ndash S12

13m3m 1m

9m Tunnelface

Surrounding rock

SourcesReceivers


Alteration zone

(a)

Cavity collapse

(b)







20

A1

D1k 225 + 313 225 + 293 225 + 273 225 + 253 225 + 233 225 + 213

A2

A3

1

05

0

ndash05

ndash1

0

Z (m

)

Y (m)

ndash20

ndash200

20

X (m)


A1A2 A3

20

20

0

0

Z (m

)

ndash20

ndash20

Y (m)

D1k225 + 313 225 + 293 225 + 273 225 + 253

X (m)

225 + 233 225 + 213










Anomalous zones(a)

(b)



D1k225 + 313


Underground water

225 + 293 225 + 273 225 + 253 225 + 233 225 + 213

20

0

ndash20






(a) (b)


20

0

A1

A2A3

ndash20ndash20

20

0

D1k223 + 563 223 + 543 223 + 523 223 + 503 223 + 483 223 + 463

1

05

0

ndash05

ndash1





20

A1

A2A3

0

Z (m

)

Y (m)

X (m)

ndash20ndash20

0

20D1k223 + 563 223 + 543 223 + 523 223 + 503 223 + 483 223 + 463


Anomalous zones(a)



D1k223 + 563 223 + 543


223 + 523 223 + 503

240deg 45deg

223 + 483 223 + 463

20

0

ndash20

0 10 20 30 40 50 60 70 80 90



4 Conclusion


Data Availability




Acknowledgments


References















Cavity collapse

(a)

Cavity collapse

(b)














3 Field Test







D1k

223

+ 0

00

D1k

225

+ 5

00

D1k

225

+ 2

50

D1k

225

+ 0

00

D1k

224

+ 7

50

D1k

224

+ 5

00

D1k

224

+ 2

50

D1k

224

+ 0

00

D1k

223

+ 7

50

D1k

223

+ 5

00

D1k

223

+ 2

50

1250

1350

17501700165016001550150014501400

1350


Grade III

Granite+ +




(a) (b)

(c)


R1 ndash R4

R5 ndash R8

S1 ndash S3

S7 ndash S9

S4 ndash S6

S10 ndash S12

13m3m 1m

9m Tunnelface

Surrounding rock

SourcesReceivers


Alteration zone

(a)

Cavity collapse

(b)







20

A1

D1k 225 + 313 225 + 293 225 + 273 225 + 253 225 + 233 225 + 213

A2

A3

1

05

0

ndash05

ndash1

0

Z (m

)

Y (m)

ndash20

ndash200

20

X (m)


A1A2 A3

20

20

0

0

Z (m

)

ndash20

ndash20

Y (m)

D1k225 + 313 225 + 293 225 + 273 225 + 253

X (m)

225 + 233 225 + 213










Anomalous zones(a)

(b)



D1k225 + 313


Underground water

225 + 293 225 + 273 225 + 253 225 + 233 225 + 213

20

0

ndash20






(a) (b)


20

0

A1

A2A3

ndash20ndash20

20

0

D1k223 + 563 223 + 543 223 + 523 223 + 503 223 + 483 223 + 463

1

05

0

ndash05

ndash1





20

A1

A2A3

0

Z (m

)

Y (m)

X (m)

ndash20ndash20

0

20D1k223 + 563 223 + 543 223 + 523 223 + 503 223 + 483 223 + 463


Anomalous zones(a)



D1k223 + 563 223 + 543


223 + 523 223 + 503

240deg 45deg

223 + 483 223 + 463

20

0

ndash20

0 10 20 30 40 50 60 70 80 90



4 Conclusion


Data Availability




Acknowledgments


References















Cavity collapse

(a)

Cavity collapse

(b)







D1k

223

+ 0

00

D1k

225

+ 5

00

D1k

225

+ 2

50

D1k

225

+ 0

00

D1k

224

+ 7

50

D1k

224

+ 5

00

D1k

224

+ 2

50

D1k

224

+ 0

00

D1k

223

+ 7

50

D1k

223

+ 5

00

D1k

223

+ 2

50

1250

1350

17501700165016001550150014501400

1350


Grade III

Granite+ +




(a) (b)

(c)


R1 ndash R4

R5 ndash R8

S1 ndash S3

S7 ndash S9

S4 ndash S6

S10 ndash S12

13m3m 1m

9m Tunnelface

Surrounding rock

SourcesReceivers


Alteration zone

(a)

Cavity collapse

(b)







20

A1

D1k 225 + 313 225 + 293 225 + 273 225 + 253 225 + 233 225 + 213

A2

A3

1

05

0

ndash05

ndash1

0

Z (m

)

Y (m)

ndash20

ndash200

20

X (m)


A1A2 A3

20

20

0

0

Z (m

)

ndash20

ndash20

Y (m)

D1k225 + 313 225 + 293 225 + 273 225 + 253

X (m)

225 + 233 225 + 213










Anomalous zones(a)

(b)



D1k225 + 313


Underground water

225 + 293 225 + 273 225 + 253 225 + 233 225 + 213

20

0

ndash20






(a) (b)


20

0

A1

A2A3

ndash20ndash20

20

0

D1k223 + 563 223 + 543 223 + 523 223 + 503 223 + 483 223 + 463

1

05

0

ndash05

ndash1





20

A1

A2A3

0

Z (m

)

Y (m)

X (m)

ndash20ndash20

0

20D1k223 + 563 223 + 543 223 + 523 223 + 503 223 + 483 223 + 463


Anomalous zones(a)



D1k223 + 563 223 + 543


223 + 523 223 + 503

240deg 45deg

223 + 483 223 + 463

20

0

ndash20

0 10 20 30 40 50 60 70 80 90



4 Conclusion


Data Availability




Acknowledgments


References















Cavity collapse

(a)

Cavity collapse

(b)











20

A1

D1k 225 + 313 225 + 293 225 + 273 225 + 253 225 + 233 225 + 213

A2

A3

1

05

0

ndash05

ndash1

0

Z (m

)

Y (m)

ndash20

ndash200

20

X (m)


A1A2 A3

20

20

0

0

Z (m

)

ndash20

ndash20

Y (m)

D1k225 + 313 225 + 293 225 + 273 225 + 253

X (m)

225 + 233 225 + 213










Anomalous zones(a)

(b)



D1k225 + 313


Underground water

225 + 293 225 + 273 225 + 253 225 + 233 225 + 213

20

0

ndash20






(a) (b)


20

0

A1

A2A3

ndash20ndash20

20

0

D1k223 + 563 223 + 543 223 + 523 223 + 503 223 + 483 223 + 463

1

05

0

ndash05

ndash1





20

A1

A2A3

0

Z (m

)

Y (m)

X (m)

ndash20ndash20

0

20D1k223 + 563 223 + 543 223 + 523 223 + 503 223 + 483 223 + 463


Anomalous zones(a)



D1k223 + 563 223 + 543


223 + 523 223 + 503

240deg 45deg

223 + 483 223 + 463

20

0

ndash20

0 10 20 30 40 50 60 70 80 90



4 Conclusion


Data Availability




Acknowledgments


References















Cavity collapse

(a)

Cavity collapse

(b)












Anomalous zones(a)

(b)



D1k225 + 313


Underground water

225 + 293 225 + 273 225 + 253 225 + 233 225 + 213

20

0

ndash20






(a) (b)


20

0

A1

A2A3

ndash20ndash20

20

0

D1k223 + 563 223 + 543 223 + 523 223 + 503 223 + 483 223 + 463

1

05

0

ndash05

ndash1





20

A1

A2A3

0

Z (m

)

Y (m)

X (m)

ndash20ndash20

0

20D1k223 + 563 223 + 543 223 + 523 223 + 503 223 + 483 223 + 463


Anomalous zones(a)



D1k223 + 563 223 + 543


223 + 523 223 + 503

240deg 45deg

223 + 483 223 + 463

20

0

ndash20

0 10 20 30 40 50 60 70 80 90



4 Conclusion


Data Availability




Acknowledgments


References















Cavity collapse

(a)

Cavity collapse

(b)










(a) (b)


20

0

A1

A2A3

ndash20ndash20

20

0

D1k223 + 563 223 + 543 223 + 523 223 + 503 223 + 483 223 + 463

1

05

0

ndash05

ndash1





20

A1

A2A3

0

Z (m

)

Y (m)

X (m)

ndash20ndash20

0

20D1k223 + 563 223 + 543 223 + 523 223 + 503 223 + 483 223 + 463


Anomalous zones(a)



D1k223 + 563 223 + 543


223 + 523 223 + 503

240deg 45deg

223 + 483 223 + 463

20

0

ndash20

0 10 20 30 40 50 60 70 80 90



4 Conclusion


Data Availability




Acknowledgments


References















Cavity collapse

(a)

Cavity collapse

(b)







20

A1

A2A3

0

Z (m

)

Y (m)

X (m)

ndash20ndash20

0

20D1k223 + 563 223 + 543 223 + 523 223 + 503 223 + 483 223 + 463


Anomalous zones(a)



D1k223 + 563 223 + 543


223 + 523 223 + 503

240deg 45deg

223 + 483 223 + 463

20

0

ndash20

0 10 20 30 40 50 60 70 80 90



4 Conclusion


Data Availability




Acknowledgments


References















Cavity collapse

(a)

Cavity collapse

(b)







4 Conclusion


Data Availability




Acknowledgments


References















Cavity collapse

(a)

Cavity collapse

(b)







automaticapproachforfastprocessinganddataanalysisof...

Documents