Study on an improved intelligent visual monitoring method of water inrush in complex construction environment of tunnels and engineering application

doi:10.21203/rs.3.rs-3217472/v1

Download PDF

Article

Study on an improved intelligent visual monitoring method of water inrush in complex construction environment of tunnels and engineering application

https://doi.org/10.21203/rs.3.rs-3217472/v1

This work is licensed under a CC BY 4.0 License

Version 1

posted

You are reading this latest preprint version

As one of the typical geological hazards that is very difficult to prevent, the monitoring and early warning technologies and methods of water inrush are increasingly becoming a hot research problem during the deep-buried and long tunnels construction. Aiming at its unpredictability and strong suddenness, the intelligent analysis method based on machine vision was introduced to monitor the hazard intuitively. According to the characteristics of a strong interference of tunnel construction environment, a new frame difference and background subtraction visual analysis algorithm was proposed. Meanwhile, the effectiveness and practicability of the proposed algorithm were verified by the borehole water inrush video. Then, the proposed method was used to conduct video surveillance on PDK170 + 671 of the parallel heading. The occurrence location of water inrush was effectively identified, and the water inflow was predicted according to the pixel point area, which provides a solution for real-time and continuous monitoring of water inrush.

Physical sciences/Engineering/Civil engineering

Earth and environmental sciences/Solid earth sciences/Hydrogeology

Tunnel construction

water inrush

Inter-frame difference method

Intelligent visual monitoring

engineering application

With the rapid advancement of China's transportation power strategy, the construction focus of highways, railways, and water conservancy and hydropower projects is gradually shifting to the western karst mountainous area. As a result, high-risk karst tunnels continue to emerge. And more and more tunnels present the characteristics of long tunnel lines, largely buried depth, strong karst, high stress and water pressure, and complex geological conditions^1,2. The common geological hazards in the construction process of the tunnel and underground engineering include collapse, water and mud inrush, rock burst, and large deformation³. When unfavorable geological bodies such as the water-filled cave, water-rich faults, and underground rivers are encountered, the water and mud inrush hazard is very prone to occur^4–8. Because the water and mud inrush is a dynamic catastrophe system of multi-scale and multi-field coupling, its causes and catastrophe process are extremely complex, and it is the most frequent and harmful geological hazard^9,10. Once the disaster happens, it will not only lead to serious machine damage, schedule delay, and casualties but also cause the decline of groundwater level and the destruction of the ecological environment^11–13. Therefore, it is crucial to research the advance monitoring and early warning of water and mud inrush to achieve the effective active prevention of the hazard and avoid post-disaster treatment.

In recent decades, scholars at home and abroad have carried out a lot of research on the monitoring and early warning of water and mud inrush hazards^14–17. Bai et al.¹⁸ determined the temporal and spatial ranges of monitoring and early warning on delayed water inrush in deep rock faults based on the monitoring data of water pressure and water temperature and proposed the early-warning thresholds and judging criteria for the monitoring indicators of fault-type water inrush. And finally, the monitoring and early warning of water inrush at the working face of Wanglou coal mine 13301 was realized by using the optical fiber real-time monitoring system, which ensured the safe mining of the coal seam. Wang et al.¹⁹ established a wireless real-time monitoring system for the multiple precursor information of water inrush and proposed a time-series-based fusion method of multi-sensitive information, which was successfully applied to the underpass river section of the Yuelongmen tunnel on the Chengdu to Lanzhou railway. When it was found that microseismic signals such as rock failure and water flow would be generated during the water inrush, some scholars began to try to apply microseismic monitoring technology to water inrush²⁰. Jiang et al.¹¹ developed the microseismic equipment for water inrush monitoring in mines, and adopted an optimized monitoring network to monitor the dynamic activities of the water-conducting channel under the influencing of mining in real-time and continuously, which realized early warning of water inrush by accurately locating the water inrush channel. According to the energy accumulation-release phenomenon during the formation process of water inrush channel, Dou et al.²¹ put forward a time-divisional and zone-divisional microseismic monitoring method for water inrush in goaf based on a microseismic system. The risk of water inrush in different regions was analyzed in real-time. Zhang et al.²² and Sun et al.²³ studied the formation process of water inrush channels by microseismic monitoring and numerical simulation. Ma et al.²⁴ used a three-dimensional visualizable microseismic system to sense the emergence and expansion of micro-damage in real-time and calculated the floor disturbance depth under coal seam mining. A comprehensive analysis method of floor failure in coal seam was presented, which provides a reference for the evaluation and prevention of floor water inrush. In addition, geophysical methods are also widely used in water and mud inrush monitoring of underground engineering. Xu et al.²⁵ used the parallel network electrical CT technology to monitor the changes of apparent resistivity during floor water inrush in real-time and analyzed the whole change law of floor mining failure and water inrush channel formation, evolution, and water filling combined with the stress and water pressure obtained by numerical simulation. Li et al.²⁶ and Liu et al.²⁷ put forward a quantitative evaluation method of the time-lapse resistivity tomography to identify the evolution state of the water inrush channel based on the image grayscale correlation. Liu et al.²⁸ adopted the three-dimensional induced polarization method to monitor the changes in the polarization rate in real-time, which achieves timely warning of water inrush in hidden structures. The above research can be divided into two categories: (1) Lossy monitoring, mainly including the embedded sensor. This method can obtain the precursor information of surrounding rock in the process of water inrush, such as stress field, displacement field, and seepage field. However, the integrity of the rock mass structure is destroyed and the normal construction is disturbed. (2) Non-destructive monitoring, mainly including microseism. This method can realize the process monitoring of water inrush channel formation under the excavation, and the damage to rock mass is less. However, the monitoring information is single and the results are vulnerable to noise. Meanwhile, with the continuous tunnel excavation and working face formation, the spatial location of the water inrush is variable. The existing method cannot realize the dynamic monitoring of water inrush in the whole process of tunnel excavation. Therefore, it is urgent to study the dynamic monitoring method of water inrush in complex construction environments of tunnel engineering.

In recent years, with the rapid development of artificial intelligence technology, machine vision as an important branch is gradually applied to target detection, unmanned driving, etc²⁹. Stent et al.³⁰ and Huang et al.³¹ applied the intelligent extraction technology of the characteristics based on image data set to the tunneling and underground engineering, such as lining diseases recognition³², surrounding rock classification³³, and water inflow prediction³⁴. Therefore, the machine version technology was introduced to establish a novel real-time dynamic monitoring method of water and mud inrush applicable to the complex construction environment of the tunnels. Then it was applied to the Chaoyang tunnel on the Guiyang to Nanning railway for the early warning of water and mud inrush. In the meantime, to verify the effectiveness and practicality of the proposed method, its obtained results were compared and analyzed with these of the traditional machine version algorithm.

The water and mud inrush is a geological hazard caused by encountering the water-bearing and water-conducting structures in the construction process of tunneling and underground engineering, such as fault fracture zone, water-filling karst caves, underground rivers, and karst pipelines. The research shows that one of the necessary conditions for water and mud inrush occurrence is the presence of water-bearing and water-conducting structures¹⁴. When the tunnel face is close to the disaster-causing structures of water and mud inrush, there will be some precursor phenomena different from the other geological hazards, which provides a theoretical basis for visual monitoring of water and mud inrush. The detailed visual monitoring information is shown in Table 1.

Table 1

Visual monitoring basis of tunnel water and mud inrush.
Classification	Description	Monitoring method
Lithology change	(1) Spotted iron-strain rust on surrounding rock exposed by free surface; (2) The wetting, softening and drenching of exposed lithology are serious; (3) Lithology mutation. For example, the dark black rock mass changes to white or variegated rock mass indicates that the weakly soluble rock mass into strongly soluble rock mass.	Machine version monitoring
Water inflow change	(1) The drill pipe has protrusion, water is ejected from the borehole, and the jet distance is large; (2) Water flows out along the rock fractures and the flow changes from small to large; (3) Water flow is turbid containing fine sand, silty fine sand and clay particles.	Machine version monitoring

The common machine version algorithms are the inter-frame difference method, background subtraction method, and optical flow method. All of the above methods identify the moving targets from the background of video sequence images, but their basic principles and advantages and disadvantages are different, as shown in Table 2.

As we all know, the environment inside the tunnel during construction is dark and the light changes easily due to dust, lighting, shelter, and other factors. Therefore, it is a great challenge to realize the effective visual monitoring of water and mud inrush. By comparing the above three methods, the inter-frame difference method was selected for the visual monitoring of water and mud inrush. Due to the defect that the extracted target contour is incomplete, the background subtraction method was introduced to improve the inter-frame difference method.

3.1 Improved inter-frame difference method

To overcome the defect that the traditional inter-frame difference method under the influence of noise is prone to “hole” or “ghost”, a novel visual monitoring method based on the inter-frame difference method and background subtraction method was proposed to improve the accuracy of water and mud inrush monitoring. That is, the current continuous frame f (x, y, n) is differentiated from the previous continuous frame f (x, y, n-1) and the established background model. The improved formulas are as follows:

$$R\left( {x,y,n} \right)=\left\{ {\begin{array}{*{20}{c}} 1&{if\left| {f\left( {x,y,n} \right) - f\left( {x,y,n - 1} \right)} \right|>T} \\ 0&{else} \end{array}} \right.$$

$$R^{\prime}\left( {x,y,n} \right)=\left\{ {\begin{array}{*{20}{c}} 1&{if\left| {f\left( {x,y,n} \right) - B\left( {x,y} \right)} \right|>T} \\ 0&{else} \end{array}} \right.$$

$$D\left( {x,y,n} \right)=R\left( {x,y,n} \right) \cap R^{\prime}\left( {x,y,n} \right)$$

Where B (x, y) is the established background model, D(x, y, n) is the intersection of R(x, y, n) and R’(x, y, n-1).

Table 2

Comparative analysis of three common machine vision algorithms.
Algorithm	Definition	Formula	Advantage	Disadvantage
Background subtraction method	The differential image is obtained by subtract the input current frame image from the previously established continuously updated background model to extract the foreground moving target.	$D\left( {x,y,t} \right)=\left\{ {\begin{array}{*{20}{c}} 1&{if\left\| {F\left( {x,y,t} \right) - F\left( {x,y,{b_t}} \right)} \right\|>T} \\ 0&{else} \end{array}} \right.$	Its calculation is simple, target contour is complete, moving information is accurate and rich, and it can meet the real-time video detection.	The background model need to be constantly updated and the monitoring results are vulnerable to the interference from background, lighting and other factors
Inter-frame difference method	The differential image is obtained by subtracting the pixels of the consecutive two-frame or three-frame images in time, and then the moving target is extracted by comparing the absolute value of gray difference with the set threshold.	$D\left( {x,y,t} \right)=\left\{ {\begin{array}{*{20}{c}} 1&{if\left\| {F\left( {x,y,t} \right) - F\left( {x,y,t+1} \right)} \right\|>T} \\ 0&{else} \end{array}} \right.$	It is easy to implementation and sensitive to moving targets, and it has fast operation speed and strong ability to resist the interference from light, shadow and other environmental factors	The too fast moving speed or the large moving displacement of the target between adjacent frames easily generates the “hole” or “ghost”, so that the complete target image cannot be extracted
Optical flow method	The spatial moving object is projected onto a two-dimensional plane image, and then the displacement vector of each pixel is calculated by using the temporal variation and correlation of optical flow of consecutive frame sequences in the video image	$\frac{{\partial F\left( {x,y,t} \right)}}{{\partial x}}\frac{{dx}}{{dt}}+\frac{{\partial F\left( {x,y,t} \right)}}{{\partial y}}\frac{{dy}}{{dt}}+\frac{{\partial F\left( {x,y,t} \right)}}{{\partial t}}=0$	It has high detection accuracy and is applicable to camera fixed and moving scenes	The calculation of optical flow field is complicated, it is vulnerable to light and other factors

Where (x, y) is the coordinate of any pixel point of the image, F(x, y, t) is the gray value of the current frame image at time t in the video sequence, F(x, y, b_t) is the gray value of the background image at time t, F(x, y, t + 1) is the gray value of the frame image at time t + 1, and T is the binarization threshold of the image.

3.2 Verification

To test the visual monitoring effect of the proposed method, the monitoring video of water inflow from the advanced geological borehole in a tunnel construction was selected, and the proposed method was used to recognize the image characteristic of water and mud inrush. The recognition results were as shown in Fig. 1, where the yellow represents the recognized water flow. It can be seen that the contour of the water flow is accurately recognized. When the water flow changes, the yellow marking area will also change. Therefore, using the water flow marking area to measure the change in water flow is feasible during the water inrush. In addition, the traditional inter-frame difference method was also used to process and analyze the above monitoring video, as shown in Fig. 2. The marked contour of water flow was incomplete. By comparing Fig. 1 and Fig. 2, it is found that the contour of water flow and marking area recognized by the proposed method are significantly better than these of the traditional method. Therefore, it is proved that the proposed method has good practicability and the visual monitoring effect is improved.

4.1 Engineering background of Chaoyang tunnel

The Chaoyang tunnel is an important control project of the new Guiyang to Nanning Railway, which is located in Chaoyang town, Libo County, Guizhou Province. The starting and ending mileage of the tunnel is DK159 + 802 ~ DK172 + 536 with a total length of 12734m and a maximum buried depth of 432m. The design speed of the tunnel which is a single-hole double-track tunnel is 350km/h. The geological conditions of some longitudinal sections are shown in Fig. 3.

(1) Topography and geomorphology

The landform of the tunnel site belongs to the low-middle mountain and peak cluster valley. The terrain undulates greatly, and the maximum relative elevation difference can reach 700m. The natural slope is generally 20 ~ 55°, and steep cliffs are formed locally. The vegetation is developed.

(2) Formation lithology

The stratum lithology along the tunnel is mainly soluble limestone, and the specific lithology is described in Table 3.

Table 3

The formation lithology crossed by the Chaoyang Tunnel and its characteristics.
Mileage section	Stratum lithologic	Lithological description	Age and system	Sign
DK159 + 802 ~ DK160 + 651, DK170 + 740 ~ DK172 + 536	Limestone, siliceous shale, shale with coal seam	The limestone is gray and medium-thick layered, with cryptocrystalline structure. The siliceous shale and shale are gray to grayish yellow interspersed with thin layer brownish yellow after weathering, and the joints are developed.	Upper Permian Changxing Formation and Wujiaping Formation	P₃w + c
DK160 + 324 ~ DK161 + 651, DK170 + 740 ~ DK171 + 750	Limestone	Gray and grayish white, thick to very thick layered, with cryptocrystalline structure and karst development, and hard rock.	Middle Permian Maokou Formation	P₂m
DK161 + 651 ~ DK162 + 180, DK170 + 070 ~ DK171 + 740	Limestone with shale	The limestone is gray and gray-black, medium-thick layered interspersed with thin layer, with cryptocrystalline structure, the rock is hard. The shale is dark-gray with argillaceous cementation and thin layered structure. The shale is soft and weakly weathering, which is easy to disintegrate when exposed to air, and soften when exposed to water.	Middle Permian Qixia Formation	P₂q
DK162 + 180 ~ DK162 + 275, DK170 + 030 ~ DK170 + 070	Quartz sandstone, sandstone, shale with coal seam	Gray, grayish white and grayish yellow with thin layers in the middle. The sandstone is hard and the shale with coal seam is soft.	Middle Permian Liangshan Formation	P₁l
DK162 + 275 ~ DK163 + 130, DK169 + 680 ~ DK170 + 030	Limestone	Gray and grayish white, thick to very thick layered with cryptocrystalline structure. The rock is hard and karst is developed.	Upper Carboniferous Maping Formation	C₃mp
DK163 + 240 ~ DK163 + 800, DK169 + 180 ~ DK169 + 680	Limestone with shale	Gray and dark gray, medium-thick layered with cryptocrystalline structure. The rock is hard and karst is developed.	Middle Carboniferous Huanglong Formation	C₂hn
DK163 + 800 ~ DK164 + 443, DK168 + 706 ~ DK169 + 180	Thick limestone	Gray and dark gray, medium-thick layered with cryptocrystalline structure. The rock is hard and karst is developed.	Lower Carboniferous Datang Stage Shangsi member	C₂d²
DK164 + 443 ~ DK164 + 678, DK168 + 513 ~ DK168 + 706	Shale with argillaceous limestone, carbonaceous shale	Variegated, mostly grayish white, grayish yellow, and grayish black with medium-thick layer with thin layer and argillaceous cementation. The shale is thin and flake, the carbonaceous shale is flake and scaly.	Lower Carboniferous Datang Stage upper Jiusi stage	C₁d¹
DK164 + 678 ~ DK168 + 513	Argillaceous limestone with shale, carbonaceous shale, sandstone	Variegated, mostly grayish white, grayish yellow, and grayish black with medium-thick layer with thin layer and argillaceous cementation. The sandstone and argillaceous limestone are medium-thick layered, the shale is thin and flake, the carbonaceous shale is flake and scaly.	Lower Carboniferous Datang Stage lower Jiusi stage	C₁d¹

(3) Geological structure

According to the site investigation, the regional structure is Shuili anticline. Under the influence of the regional structure, three secondary faults are developed in the tunnel crossing section, namely Die 1# reverse fault, Shuiyin normal fault and Chaoyang reverse fault.

a. Die 1# reverse fault

It intersects with the tunnel line near DK162 + 665. According to the comprehensive analysis of the terrain survey and geophysical results, the width of the fault fracture zone is about 60m, and the influencing length of the tunnel is about 70m. Moreover, the lower Permian Liangshan Formation was found to be staggered through field investigation.

b. Shuiyin normal fault

It intersects with the tunnel line near DK168 + 400. There is no obvious sign of the fault near the tunnel line, and the occurrence of rock strata on both sides of the fault changes greatly.

c. Chaoyang reverse fault

It intersects with the tunnel line near DK172 + 023. The influencing length of the tunnel is about 100m.

In addition, the tunnel passes through five contact zones between the soluble rock and non-soluble rock.

(4) Hydrogeological characteristics

a. Surface water

The surface water system that the tunnel passes through belongs to the Dagou river system, which includes the mainstream of the Zhang river and the tributaries of the Fangcun river. The Zhang river is the largest in Libo county. Its main channel is about 50km long and the main tributary is the Lada river. The Fangcun river flows from north to south through Yanpai town and Heyong town of Sandu county, and then enters Libo county. The main flow direction of surface water is NE ~ SW along the trough valley. Meanwhile, due to the control of the transverse fracture, there are a series of transverse valleys cutting through the main structure line. Therefore, the surface catchment including groundwater within a certain range flows into the longitudinal main channel through the transverse valleys.

b. Groundwater

The groundwater is mainly pore water in the Quaternary soil, bedrock fissure water, and karst conduit water with medium water quantity. It is recharged by atmospheric precipitation and surface water and discharged in the form of runoff, springs, and underground rivers. The pore water is mainly stored in the Quaternary Holocene slope eluvium and the completely weathered bedrock of the entrance and exit in the tunnel area. Its water content is relatively poor, and it is mainly recharged by atmospheric precipitation and surface water. The bedrock fissure water is mainly stored in the non-soluble rock stratum with weak content, which is rich only in the zones of the geological structure-activity and fracture, and the structural joint fissure development. There is a large amount of groundwater in the joint intensive zone and fault fracture zone, near which strand water inrush may be encountered during the tunnel construction. The water in the soluble rock stratum belongs to the karst fissure water or karst conduit water. As the surface depressions, sinkholes, and karst caves are relatively developed, it was inferred that the degree of karst development in the tunnel area is medium to strong. However, the distribution of karst water is not uniform and irregular. The groundwater will increase exponentially during the rainy season.

The alternate distribution of soluble rock with good permeability and non-soluble rock with poor permeability in the hydrogeological unit determines that the overall flow direction of groundwater in the survey area is along the structural line. That is after the surface water infiltrates into the vertical fractures of the ground-exposed soluble rock or the karst caves, it flows longitudinally along the main structural line. When it is cut by a transverse valley, a spring point can be formed near the contact line between the soluble rock and non-soluble rock.

4.2 Water inrush situation

The Chaoyang tunnel was designed with 2 transverse galleries, 1 inclined shaft, and 2 parallel headings. A parallel heading with an elevation of 3m lower than the main tunnel was set at the tunnel exit for drainage during the construction and operation. Its design length is 2800m, and the distance between the heading and the main tunnel is 30m. When the parallel heading was constructed to the working face PDK170 + 671 whose buried depth is about 230m on June 10, 2018, a large-scale water inrush occurred. A large number of mechanical equipment was destroyed and three workers died, which caused significant economic losses and harmful social impact. Since May 30, 2018, the development process of tunnel water inrush was listed in Table 4.

Table 4

Water inrush process of the parallel heading PDK170 + 671 at the Chaoyang tunnel exit.
Date	Description
May 30, 2018	Muddy water gushed from the boreholes in the center of tunnel floor with about 10m spray distance. The water pressure gradually increased, and the maximum spray distance reached 20m. After a period of time, the gushing water became clear and the pressure remained constant.
May 30, 2018 to June 1, 2018	There was continuous moderate rain or heavy rain on the ground. The water flow in the tunnel increased from 570 m³/h to 1100m³/h, and then tended to be stable.
June 2, 2018 to June 3, 2018	Four advance horizontal drilling was carried out on the tunnel face, and the drill pipe had protrusion. In particular, when the 4th borehole was drilled to 8.5m, water began to flow. When the borehole reached 9m, it could not be drilled because of excessive water output.
June 4, 2018 to June 9, 2018	A solution crack with 2m long and 0.5m high was exposed on the right side of the tunnel floor, and there was a stream of water flowing out. The geological radar, advance horizontal drilling and drilling around the boreholes were used to further identify the geological conditions.
June 10, 2018	A large-scale water and mud inrush suddenly occurred. The submerged depth was about 2.5 m above the tunnel floor. It returned to the original water inflow after an hour. The water inflow was about 5.7×10⁴ m³.

4.3 Results and analysis of visual monitoring for water inrush

The surrounding rock of section PDK170 + 394ཞPDK170 + 814 of the parallel heading at the exit of Chaoyang Tunnel is mainly Permian limestone intercalated with shale (P₁q), and its grade is Level III. The buried depth of the face PDK170 + 671 is about 230m. According to the design data, the Class III bolting shotcrete and full-section construction were adopted in the section. Before the large-scale water and mud inrush occurs, the following measures were taken: the one was to strengthen the monitoring measurement of the surrounding rock behind the face. Once an abnormality occurs, the construction on the tunnel face will stop immediately. If necessary, the additional umbrella arch was set to ensure safety. The second was that the geological radar and advanced horizontal drilling were used to further explore the geological conditions near the face after clearing the mud in the tunnel. To observe the stability of the tunnel surrounding rock, the monitoring equipment was arranged at a distance of about 30m from the tunnel face. The water inrush was monitored by video. A video was selected to analyze the visual recognition of water inrush, and red was used to identify the water inrush flow. The analysis results were shown in Fig. 4. It can be seen from Fig. 4 that the location of the water inrush is in the right spandrel and vault of the tunnel face. The initial water flow and velocity of the water inrush were small. The red marking area was mainly concentrated in the right spandrel and vault of the tunnel face. Then the water inflow and velocity suddenly increased and the water quickly broke into the tunnel. The red area also increased accordingly. At the same time, the traditional inter-frame difference method was also used for the visual analysis of the water inrush monitoring video, and the results were shown in Fig. 5. It can be seen from Fig. 5 that the traditional method is greatly affected by the noise and the water inrush can not be effectively identified. However, the visual monitoring result of the proposed method also produced some noise in addition to the red marking. The reasons were analyzed as follows: (1) the monitoring equipment was far away from the tunnel face. The sharpness was affected. (2) the water inrush caused the shaking of the monitoring and lighting equipment.

(1) The water-bearing or water-conducting disaster-causing structure is one of the necessary conditions for the occurrence of water and mud inrush in tunnels. When the tunnel face during the excavation is close to or through the disaster-causing structures, there is often precursory information such as changes in lithology, high-pressure water jet, and water turbidity. These are the basis for visual monitoring of water and mud inrush.

(2) Given the difficulty of constructing the background model in the background subtraction method, easily generating the “hole” in the inter-frame difference method, and the complex calculation in the optical flow method, an improved inter-frame difference method for visual monitoring of water and mud inrush was proposed. Then the monitoring video of borehole water inrush in the construction process of a tunnel was used to verify the monitoring effect of the proposed method. By comparing with the traditional inter-frame difference method, it is proved that the proposed method has better monitoring accuracy and early warning effect.

(3) Relying on the Chaoyang Tunnel from the Guiyang to Nanning Railway, the proposed method in this paper has been successfully applied to the very large-scale water and mud inrush at the parallel heading PDK170 + 671 of the Chaoyang Tunnel. It was found that the location of water inrush is in the right spandrel and vault of the tunnel face, and the water inflow and velocity have a significant increasing trend. Due to the long monitoring distance and the shaking of the monitoring equipment, some noise was generated in the recognition images of the visual monitoring. However, its effect is still significantly better than that of the traditional inter-frame difference method.

Acknowledgments

This work was supported by the National Science Foundation for Distinguished Young Scholars of China (Grant No.52025091), National Natural Science Foundation of China (NO.51679131), Natural Science Foundation of Chongqing (cstc2021jcyj-msxmX0133), Natural Science Foundation of Gansu Province (20JR10RA472), Project of Science and Technology Research Program of Chongqing Education Commission of China (KJQN201901423, KJQN202300714).

Data availability

Data will be available by the corresponding author on reasonable request.

Li, S.C., Xu, Z.H., Huang, X., Lin, P., Zhao, X.C., Zhang, Q.S., Yang, L., Zhang, X., Sun, H.F. & Pan, D.D. Classification, geological identification, hazard mode and typical case studies of hazard-causing structures for water and mud inrush in tunnels. Chin J Rock Mech Rock Eng. 37(5), 1041–1069. DOI: 10.13722/j.cnki.jrme.2017.1332 (2018).
Wang, J., Li, S.C., Li, L.P., Lin, P., Xu, Z.H. & Gao, C.L. Attribute recognition model for risk assessment of water inrush. Bull Eng Geol Environ. 78 (2), 1057–1071. https://doi.org/10.1007/s10064-017-1159-4 (2019).
Beard, A.N. Tunnel safety, risk assessment and decision-making, Tunn Undergr Space Tech. 25(1), 91–94. https://doi.org/10.1016/j.tust.2009.07.006 (2010).
Alija, S., Torrijo, F.J. & Quinta-Ferreira, M. Geological engineering problems associated with tunnel construction in karst rock masses: The case of Gavarres tunnel (Spain). Eng Geol. 157, 103–111. https://doi.org/10.1016/j.enggeo.2013.02.010 (2013).
Golian, M., Teshnizi, E.S., Parise, M., Terzić, J., Milanović, S., RistićVakanjac, V., Mahdad, M., Abbasi, M., Taghikhani, H. & Saadat, H. A new analytical method for determination of discharge duration in tunnels subjected to groundwater inrush. Bull Eng Geol Environ. 80, 3293–3313. https://doi.org/10.1007/s10064-021-02140-6 (2021).
Li, S.C., Liu, C., Zhou, Z.Q., Li, L.P., Shi, S.S. & Yuan, Y.C. Multi-sources information fusion analysis of water inrush disaster in tunnels based on improved theory of evidence. Tunn. Undergr. Space Technol. 113, 103948. https://doi.org/10.1016/j.tust.2021.103948 (2021).
Li, X.Z., Zhang, P.X., He, Z.C., Cheng, M.L. & Guo, L. Identification of geological structure which induced heavy water and mud inrush in tunnel excavation: A case study on Lingjiao tunnel. Tunn Undergr Space Technol. 69, 203–208. https://doi.org/10.1016/j.tust.2017.06.014 (2017).
Song, K.I., Cho, G.C. & Chang, S.B. Identification, remediation, and analysis of karst sinkholes in the longest railroad tunnel in South Korea. Eng Geol. 135–136, 92–105. https://doi.org/10.1016/j.enggeo.2012.02.018 (2012).
Li, S.C., Gao, C.L., Zhou, Z.Q., Li, L.P., Wang, M.X., Yuan, Y.C. & Wang, J. Analysis on the Precursor Information of Water Inrush in Karst Tunnels: A True Triaxial Model Test Study. Rock Mech. Rock Eng. 52, 373–384. https://doi.org/10.1007/s00603-018-1582-2 (2019).
Wang, Y.C., Chen, F., Sui, W.H., Meng, F.S. & Geng, F. Large-scale model test for studying the water inrush during tunnel excavation in fault. Bull Eng Geol Environ. 81, 238. https://doi.org/10.1007/s10064-022-02733-9 (2022).
Jiang, F.X., Ye, G.X., Wang, C.W., Zhang, D.Y. & Guan, Y.Q. Application of high-precision microseismic monitoring technique to water inrush monitoring in coal mine. Chin J Rock Mech Eng. 27(9), 1932–1938 (2008).
Li, S.C., Zhou, Z.Q., Li, L.P., Xu, Z.H., Zhang, Q.Q. & Shi, S.S. Risk assessment of water inrush in karst tunnels based on attribute synthetic evaluation system. Tunn. Undergr. Space Technol. 38, 50–58. https://doi.org/10.1016/j.tust.2013.05.001 (2013).
Raposo, J.R., Molinero, J. & Dafonte, J. Quantitative evaluation of hydrogeological impact produced by tunnel construction using water balance models. Eng Geol. 116 (3–4), 323–332. https://doi.org/10.1016/j.enggeo.2010.09.014 (2010).
Li, S.C., Li, X.Z., Jing, H.W., Yang, X.L., Rong, X.L. & Chen, W.Z. Research Development of Catastrophe Mechanism and Forecast Controlling Theory of Water Inrush and Mud Gushing in Deep Long Tunnel. Chin Basic Sci. 19(03), 27–43. DOI: 10.3969/j.issn.1009-2412.2017.03.005 (2017).
Liu, J.Q., Sun, Y.K., Li, C.J., Yuan, H.L., Chen, W.Z., Liu, X.L. & Zhou, X.S. Field monitoring and numerical analysis of tunnel water inrush and the environmental changes. Tunn Undergr Space Technol. 122, 104360. https://doi.org/10.1016/j.tust.2021.104360 (2022).
Wang, T.T., Zhan, S.S., Chen, C.H. & Su, W.C. Characterizing fractures to mitigate inrush of water into a shaft using hydrogeological approaches. Tunn Undergr Space Technol. 61, 205–220. https://doi.org/10.1016/j.tust.2016.10.011 (2017).
Yang, T.H. Tang, C.A. Tan, H.Z. Zhu, W.C. & Feng, Q.Y. State of the art of inrush models in rock mass failure and developing trend for prediction and forecast of groundwater inrush. Chinese J. Rock Mech Eng. 6(02), 268–277. DOI: 10.1016/S1672-6529(07)60007-9 (2007).
Bai, J.W., Li, S.C., Liu, R.T., Zhang, Q.S., Zhang, H.J. & Sha, F. Multi-field information monitoring and warning of delayed water bursting in deep rock fault. Chin J Rock Mech Eng. 34(11), 2327–2335. DOI: 10.13722/j.cnki.jrme.2015.0994 (2015).
Wang, S., Li, L.P., Cheng, S., Yang, J.Y., Jin, H., Gao, S. & Wen, T. Study on an improved real-time monitoring and fusion prewarning method for water inrush in tunnels. Tunn Undergr Space Technol. 112, 103884. https://doi.org/10.1016/j.tust.2021.103884 (2021).
Shapiro, S.A. & Dinske, C. Fluid-induced seismicity: Pressure diffusion and hydraulic fracturing. Geophys Prospect. 57(2), 301–310. https://doi.org/10.1111/j.1365-2478.2008.00770.x (2009).
Dou, L.M., He, J., Gong, S.Y., Song, Y.F. & Liu, H. A case study of micro-seismic monitoring: Goaf water-inrush dynamic hazards. J China Univ Min Technol. 41(1), 20–25. DOI:10.1007/s11783-011-0280-z (2012).
Zhang, P.H., Yang, T.H., Yu, Q.L., Xu, T., Shi, W.H. & Li, S.C. Study of a seepage channel formation using the combination of microseismic monitoring technique and numerical method in Zhangmatun iron mine. Rock Mech Rock Eng. 49(9), 3699–3708. https://doi.org/10.1007/s00603-016-1007-z (2016).
Sun, D.D., Hou, X.G., Yang, T.H., Zhao, Y., Zhang, P.H., Yang, B., Liu, Y.L. & Ma, K. Calibration of Water Inrush Channel and Numerical Simulation of Water Inrush Process in Coal Mine Roof. Front Earth Sci. 10. https://doi.org/10.3389/feart.2022.931508 (2022).
Ma, K., Sun, X.Y., Tang, C.A., Yuan, F.Z., Wang, S.J. & Chen, T. Floor water inrush analysis based on mechanical failure characters and microseismic monitoring, Tunn Undergr Space Technol. 108, 103698. https://doi.org/10.1016/j.tust.2020.103698 (2021).
Xu, Z.M., Sun, Y.J., Gong, S.Y. & Zhu, Z.K. Monitoring and numerical simulation of formation of water inrush pathway caused by coal mining above confined water with high pressure. Chin J Rock Mech Eng. 31(08), 1698–1704. DOI: 10.3969/j.issn.1000-6915.2012.08.025 (2012).
Li, S.C., B. Liu, L.C. Nie, Z.Y. Liu, M.Z. Tian, S.R. Wang, M.X. Su, Q. Guo, Detecting and monitoring of water inrush in tunnels and coal mines using direct current resistivity method: A review, J. Rock Mech. Geotech. Eng. 7(4) (2015) 469–478, http://dx.doi.org/10.1016/j.jrmge.2015.06.004.
Liu, B., Liu, Z.Y., Fan, K.R., Nie, L.C. & Zhang, X.X. An improved Time-Lapse resistivity tomography to monitor and estimate the impact on the groundwater system induced by tunnel excavation. Tunn Undergr Space Technol. 66, 107–120. https://doi.org/10.1016/j.tust.2017.04.008 (2017).
Liu, Z.X., Xue, G.Q. & Zhang, X.K. The feasibility of monitoring lagging water inrush by using parameter of polarizability. Chin J Geophys. 65(8), 3186–3197. DOI: 10.6038/cjg2022P0280 (2022).
Ren, Z.H., Fang, F.Z., Yan, N. & Wu, Y. State of the Art in Defect Detection Based on Machine Vision. Int J Precis Eng Manuf.-Green Tech. 9, 661–691. https://doi.org/10.1007/s40684-021-00343-6 (2022).
Stent, S., Gherardi, R., Stenger, B., Soga, K. & Cipolla, R. Visual change detection on tunnel linings. Mach Vision Appl. 27, 319–330. https://doi.org/10.1007/s00138-014-0648-8 (2016).
Huang, H.W., Zhao, S., Zhang, D.M. & Chen, J.Y. Deep learning-based instance segmentation of cracks from shield tunnel lining images. Struct Infrastruct Eng. 18(2), 183–196. https://doi.org/10.1080/15732479.2020.1838559 (2020).
Zhao, S., Zhang, D.M., Xue, Y.D., Zhou, M.L. & Huang, H.W. A deep learning-based approach for refined crack evaluation from shield tunnel lining images. Automat Constr. 132, 103934. https://doi.org/10.1016/j.autcon.2021.103934 (2021).
Chen, J.Y., Yang, T.J., Zhang, D.M., Huang, H.W. & Tian, Y. Deep learning based classification of rock structure of tunnel face. Geosci Front. 12(1), 395–404. https://doi.org/10.1016/j.gsf.2020.04.003 (2021).
Chen, J.Y., Zhou, M.L., Zhang, D.M., Huang, H.W. & Zhang, F.S. Quantification of water inflow in rock tunnel faces via convolutional neural network approach. Automat Constr. 123, 103526. https://doi.org/10.1016/j.autcon.2020.103526 (2021).

No competing interests reported.

Download PDF

Version 1

posted

You are reading this latest preprint version

Study on an improved intelligent visual monitoring method of water inrush in complex construction environment of tunnels and engineering application

Status:

Version 1

Abstract

Figures

1. Introduction

2. Visual monitoring Basis of water and mud inrush

3. Visual monitoring method

3.1 Improved inter-frame difference method

3.2 Verification

4 Engineering application

4.1 Engineering background of Chaoyang tunnel

4.2 Water inrush situation

4.3 Results and analysis of visual monitoring for water inrush

5 Conclusion

Declarations

Acknowledgments

Data availability

References

Additional Declarations

Status:

Version 1

Algorithm	Definition	Formula	Advantage	Disadvantage
Background subtraction method	The differential image is obtained by subtract the input current frame image from the previously established continuously updated background model to extract the foreground moving target.	\(D\left( {x,y,t} \right)=\left\{ {\begin{array}{*{20}{c}} 1&{if\left\| {F\left( {x,y,t} \right) - F\left( {x,y,{b_t}} \right)} \right\|>T} \\ 0&{else} \end{array}} \right.\)	Its calculation is simple, target contour is complete, moving information is accurate and rich, and it can meet the real-time video detection.	The background model need to be constantly updated and the monitoring results are vulnerable to the interference from background, lighting and other factors
Inter-frame difference method	The differential image is obtained by subtracting the pixels of the consecutive two-frame or three-frame images in time, and then the moving target is extracted by comparing the absolute value of gray difference with the set threshold.	\(D\left( {x,y,t} \right)=\left\{ {\begin{array}{*{20}{c}} 1&{if\left\| {F\left( {x,y,t} \right) - F\left( {x,y,t+1} \right)} \right\|>T} \\ 0&{else} \end{array}} \right.\)	It is easy to implementation and sensitive to moving targets, and it has fast operation speed and strong ability to resist the interference from light, shadow and other environmental factors	The too fast moving speed or the large moving displacement of the target between adjacent frames easily generates the “hole” or “ghost”, so that the complete target image cannot be extracted
Optical flow method	The spatial moving object is projected onto a two-dimensional plane image, and then the displacement vector of each pixel is calculated by using the temporal variation and correlation of optical flow of consecutive frame sequences in the video image	\(\frac{{\partial F\left( {x,y,t} \right)}}{{\partial x}}\frac{{dx}}{{dt}}+\frac{{\partial F\left( {x,y,t} \right)}}{{\partial y}}\frac{{dy}}{{dt}}+\frac{{\partial F\left( {x,y,t} \right)}}{{\partial t}}=0\)	It has high detection accuracy and is applicable to camera fixed and moving scenes	The calculation of optical flow field is complicated, it is vulnerable to light and other factors