Method for measuring and calculating earth pressure balance shield muck volume based on three-dimensional muck bucket scanning

Abstract

The present application relates to the technical field of shield construction, and provides a soil pressure balance shield muck volume measuring and calculating method based on three-dimensional muck bucket scanning, which comprises the following steps: step 1, preprocessing: filtering out other parts except the muck bucket, wherein the other parts include a railcar head and a pipe piece; step 2, muck bucket separation: dividing the muck bucket into independent point cloud clusters; step 3, coarse segmentation: firstly, removing the point clouds of the left and right side walls of the muck bucket; secondly, removing the point clouds of the front and back side walls of the muck; step 4, fine segmentation: removing the residual point clouds of the front and back side walls again; then, filtering the scattered points left in the above steps, that is, removing noise points; step 5, volume calculation: completing the missing point clouds in the middle of the empty muck bucket, dividing the obtained pure muck surface points into triangular meshes, and realizing muck bucket loading volume calculation; and obtaining the single tunneling ring muck volume by subtracting the full-load muck bucket volume result from the empty-load muck bucket volume result. The present application can preferably measure and calculate the muck volume.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of shield construction, in particular to a method for measuring and calculating the volume of earth pressure balance shield muck based on three-dimensional muck bucket scanning. BACKGROUND

[0002] Shield method has become the core technology of urban rail transit due to its advantages of safe operation, high construction efficiency, and small environmental impact. However, in complex and special geological conditions such as water-rich sand and gravel stratum, earth pressure balance shield often occurs over-excavation due to poor muck improvement effect, large stratum disturbance, and poor soil self-stability. If over-excavation is ignored or the over-excavation amount is not correctly estimated, it will lead to the formation of a cavity behind the lining after shield construction. The stratum cavity will cause ground subsidence under the action of long-term ground vehicle load or underground water. Ground subsidence directly affects urban traffic and may even damage existing building structures, which has a very adverse impact on urban construction.

[0003] Earth pressure balance shield (EPB) often uses large muck buckets carried by track cars to transport muck between the tunnel portal and the shield. The existing construction technology estimates the muck volume by weighing the muck bucket and manually observing the muck loading amount, which has the problems of strong subjectivity, large error, poor timeliness, etc., making it difficult to correctly estimate the over-excavation degree. The length, width, and height of the muck bucket are about 6m, 1.4m, and 2m, respectively, and the loading surface of the muck in the bucket has irregular undulations, which have certain randomness. The shield track car can usually carry 3-4 muck buckets, and the total volume of multiple muck buckets is usually greater than the theoretical muck discharge amount. When the muck bucket is not completely filled, it is considered that there is no over-excavation. During construction, an observer issues a command to determine whether the muck bucket is full. Due to the large volume of the muck bucket, there is a significant error in estimating the muck loading amount by the human eye. In addition, the muck bucket is weighed by a gantry crane during dumping to verify the muck volume. However, factors such as gantry crane vibration, weight sensor unbalanced load, and insufficient sensor accuracy can affect the measurement accuracy, and the weighing measurement reliability is insufficient.

[0004] To improve the accuracy of EPB residue volume measurement, many scholars have developed non-contact volume measurement systems according to the shield residue process. Some scholars have proposed an innovative method to realize automatic measurement of soil volume on the horizontal conveyor belt using three-dimensional laser scanning technology. Some scholars have developed a high-efficiency and accurate tunnel mud volume measurement system based on multi-line structured light technology. Some scholars have developed a real-time mud analysis system based on machine vision and deep learning, which realizes the measurement of mud quality and volume flow by image segmentation and feature extraction technology combined with conveyor belt scales and scanners, thereby providing support for intelligent control of shield tunneling. However, the residue volume measurement system based on conveyor belt scanning still faces many problems and cannot be effectively applied to engineering practice. For example, the movement of the conveyor belt (speed of 3 m / s) will produce obvious vibration, resulting in significant increase in measurement system error; the movement of the conveyor belt causes mud and water to splash and block LiDAR, resulting in missing of key point clouds and decrease in measurement accuracy; the movement speed of the conveyor belt is too fast, and the point cloud density is insufficient.

[0005] The prior art CN120747138A discloses a shield residue three-dimensional point cloud segmentation and volume calculation method based on deep learning, which adopts a deep learning method to construct a residue surface point cloud three-dimensional segmentation model. The deep learning method relies on a large number of samples for training, and the point cloud scale and quantity of the real shield construction scene are huge, and it takes a lot of time to make accurate samples. The physical shape and size of the residue hopper will change according to the differences in tunnel diameter and construction environment, and the huge training cost makes it difficult for the deep learning method to quickly adapt to three-dimensional entities of different scales. In addition, the deep learning algorithm requires a powerful GPU server, and the construction site usually does not have such a computer, which also limits the large-scale deployment and application of the deep learning method in the construction site.

[0006] In summary, there is an urgent need for a soil pressure balance shield residue volume measurement method based on three-dimensional residue hopper scanning to solve the above problems. SUMMARY

[0007] The content of the present application is to provide a soil pressure balance shield residue volume measurement method based on three-dimensional residue hopper scanning, which can overcome some or some of the defects of the prior art.

[0008] The soil pressure balance shield residue volume measurement method based on three-dimensional residue hopper scanning according to the present application comprises the following steps:

[0009] Step 1, pretreatment:

[0010] Filter out other parts other than the residue hopper, including the head of the railcar and the segment;

[0011] Step 2, residue hopper separation:

[0012] The slag bin is divided into independent point cloud clusters;

[0013] Step 3, coarse segmentation:

[0014] First, remove the dust particles from the left and right side walls of the slag hopper; then remove the dust particles from the front and back side walls of the slag hopper.

[0015] Step 4, fine segmentation:

[0016] Remove the remaining point cloud from the front and back sidewalls again; then filter the scattered points remaining from the above steps, i.e., remove noise.

[0017] Step 5, Volume Calculation:

[0018] Complete the missing point cloud in the middle of the empty slag hopper, divide the obtained pure slag surface points into triangular meshes to realize the calculation of slag hopper loading volume; and obtain the single tunneling ring slag discharge volume by subtracting the full-load slag hopper volume result from the empty slag hopper volume result.

[0019] As a preferred option, step 1 specifically includes:

[0020] First, obtain the Z coordinate of each scan contour, and then calculate the sum of the X coordinates of all points within a single contour. As shown in formula (1); then the point cloud Z coordinates and their When a scatter plot is drawn, spatial information is mapped to planar information, and the curves... The maximum value represents the bottom surface of the tunnel, and they are distributed between the various carriages;

[0021] To extract the slag hopper, points larger than the bottom surface of the slag hopper are first removed, i.e., the tunnel bottom outline is removed, as described in formula (2); at this time, point cloud gaps appear between each carriage, and these gaps are separated by intervals. The calculation method is as shown in formula (3);

[0022] (1)

[0023] (2)

[0024] (3)

[0025] In the formula L For each point cloud contour to be processed, the set of Z coordinates is provided. The point set is , This represents the X-coordinate of the i-th point in the point set; This represents the contours that need to be retained, and each contour has a unique Z-coordinate; The contour position threshold is set to ensure it is greater than the contour of the empty slag hopper. ; Distance threshold; Scan the Z coordinate of the kth point of the profile.

[0026] As preferred, in step 2, The four hoppers are distributed between the first to fifth discontinuities from the vehicle head direction, and each hopper point cloud is extracted independently according to the Z coordinate distribution in each discontinuity.

[0027] As preferred, in step 3, the left and right side walls and the middle partition According to The boundary points are extracted according to the rate of change of the value, and the front and rear side walls of the hopper are processed by operating each profile one by one, and the boundary points are determined by calculating the tangent slope change trend of each point.

[0028] Specifically:

[0029] For the calculation of the boundary points of the left and right side walls and the middle partition, it is necessary to first divide the appropriate calculation range; for the empty hopper, the middle has a partition, and the partition is The distribution of the scatter plot is similar to that of the left and right side walls, and the point cloud is divided into two parts from the middle, and the boundary points are calculated respectively, and the calculation range of the boundary points is inwardly recessed by 20 points from both ends, including part of the soil surface points and the hopper side wall points; each point is analyzed one by one from the soil surface point to the side wall direction, and the boundary point The determination conditions and the point segmentation conditions are shown in equations (4) and (5):

[0030] (4)

[0031] , (5)

[0032] In the formula, represents the point set in the calculation range, is the ith point in the R is the X coordinate of the ith point, , represents the excluded point set and the effective point set, respectively; represents the X coordinate of the boundary point; The calculation of the front and rear side wall boundary points is different from that of the left and right side walls, considering the shape characteristics of the hopper, the division of the boundary point calculation range is shown in equations (6) and (7):

[0033] Let

[0034]

[0035] (6)

[0036] (7)​​

[0037] where represents a single set of contour points, , represents the maximum and minimum Y coordinates within represents the width of the calculation range; represents the Y coordinate of the i-th point within

[0038] The trend of the tangent slope from the slag surface to the side wall point is still obvious, and the split boundary point within is calculated accordingly;

[0039] The difference vector of each point is calculated as follows:

[0040] (8)

[0041] For the starting point and the ending point

[0042] (9)

[0043] The tangent slope and the angle are calculated as follows:

[0044] (10)

[0045] (11)

[0046] The calculation method of the boundary point is as follows:

[0047] (12)

[0048] , (13)

[0049] , respectively represent the removed point set and the effective point set.

[0050] As a preferred, in step 4, the fine segmentation method is divided into two parts, the first part uses the front and rear side wall segmentation method again to remove the slag boundary residual points, and the slag bucket top residual points are removed; the positioning method of the slag bucket top residual point is as follows:

[0051] (14)

[0052] (15)

[0053] wherein represents the top end residual point segmentation threshold, represents the bias distance, determined by the actual size of the slag pot;

[0054] The second part of the fine segmentation method is to use DBSCAN to remove the noise points that obviously deviate from the slag surface points.

[0055] As preferred, in step 5, specifically:

[0056] The segmented slag surface point cloud is projected to XY the plane, and the triangular mesh is generated according to the Delaunay method, the area of each triangular mesh is calculated, the volume of the three-prism is calculated according to the Z coordinate of the mesh vertex, and then all the three-prism volumes are added to obtain the loading volume of the slag pot.

[0057] As preferred, in step 5, the influence of the point cloud segmentation error on the volume calculation is reduced by the construction of the boundary, and the method of point cloud completion and boundary construction is as follows:

[0058] For the middle part missing of the empty load point cloud, the point cloud on the left and right sides of the missing part is copied to fill in, and the calculation method is as follows:

[0059] (16)

[0060] (17)

[0061] wherein , represents the missing boundary X coordinate, represents the missing part distance, represents the boundary point serial number; is the X coordinate of the i-th point in the segmented point cloud ;

[0062] (18)

[0063] (19)

[0064] (20)

[0065] wherein , is the point area to be used to fill in the missing part, , is the point set to fill in the missing part, is the new point set after filling in the missing part;

[0066] Let the arbitrary side of the constructed frame be a point set , the starting point of which is​ The destination is The starting and ending coordinates are determined based on the maximum and minimum values ​​of the actual X and Y coordinates of the point cloud, combined with the actual dimensions of the slag hopper. for Mean Z-coordinate of all points; side length for:

[0067] (twenty one)

[0068] according to In XY Row and column point density of a plane , structure The X and Y coordinates of the point are:

[0069] (twenty two)

[0070] For each border point Its Z-coordinate correction reference exist XY Projection of a plane ;set up for exist The nearest neighbor index;

[0071] (twenty three)

[0072] final The coordinates of any point inside are .

[0073] Preferably, in step 5, after obtaining a good triangular mesh, the volume of the three-dimensional solid formed by the slag surface and the slag hopper is calculated. The detailed calculation process is as follows:

[0074] Let the triangular mesh be in XY The coordinates of the three vertices on the plane are ( x 0, y 0),( x 1, y 1),( x 2, y 2) Area of ​​a single triangular grid for:

[0075] (twenty four)

[0076] Let the Z-axis coordinates of the three vertices of the triangle be... z 0 , z 1 , z 2The volume of a single triangular mesh is:

[0077] (25)

[0078] wherein, is the plane constructed by the coordinates of the points in the point cloud; Z is the axis coordinate;

[0079] In the calculation, a virtual bottom surface of the slag ladle is constructed for the point cloud according to the volume of the slag ladle, and the plane is constructed.The calculation method of the plane

[0080] (26)

[0081] wherein, is the actual volume of the slag ladle; and are the actual length and width of the slag ladle, respectively; is the original point cloud coordinate, is the plane coordinate constructed;

[0082] The loading volume of the slag ladle calculated based on the triangulation method is:

[0083] (27)

[0084] wherein, n represents the total number of grids into which the point cloud is divided; , represent the area and volume of the i-th grid, respectively; i

[0085] Finally, the actual discharging volume of each slag ladle is the loading volume of the slag ladle in the full load state minus the loading volume of the slag ladle in the empty load state :

[0086] (28).

[0087] The beneficial effects of the present application are as follows:

[0088] The point cloud segmentation method provided by the application realizes the extraction of the muck surface point cloud by using the three-dimensional space characteristics of the muck bucket. This method does not need to be trained, consumes less computing resources, and can realize accurate segmentation by only adjusting the model parameters for muck buckets of different scales. The method provided by the application does not need large-scale parallel matrix operations, and a general CPU computer can run the method and has high computing efficiency. The performance of the proposed segmentation method is verified by comparison with existing deep learning point cloud segmentation methods. The error rate of the volume calculation method proposed by the application is much smaller than that of existing weighing or human eye volume estimation methods. The method can accurately calculate the shield muck volume, and has significant engineering application value. The application can improve the control accuracy of shield muck and reduce shield overbreak. BRIEF DESCRIPTION OF DRAWINGS

[0089] Figure 1 The flowchart of a soil pressure balance shield muck volume calculation method based on three-dimensional muck bucket scanning in embodiment 1;

[0090] Figure 2 The implementation process of muck bucket separation in embodiment 1 (different colors of the muck bucket represent that each muck bucket is independent);

[0091] Figure 3 (a) is the implementation process of muck bucket side wall rough segmentation. Rough segmentation is divided into two stages, rough segmentation of the left and right side walls of the muck bucket and rough segmentation of the front and back side walls of the muck bucket.

[0092] Figure 3 (b) is the implementation method of the left and right side wall rough segmentation;

[0093] Figure 3 (c) is the implementation method of the front and back side wall rough segmentation;

[0094] Figure 4 The schematic diagram of the fine segmentation effect in embodiment 1;

[0095] Figure 5 The optimization effect diagram of the triangular mesh in embodiment 1;

[0096] Figure 6 (a) is a schematic diagram of the test volume result distribution in embodiment 2;

[0097] Figure 6 (b) is a schematic diagram of the average error change trend with the load level in embodiment 2;

[0098] Figure 7 The schematic diagram of the test sample coordinate offset result in embodiment 2;

[0099] Figure 8 (a) is a schematic diagram of the Z coordinate offset relative error in embodiment 2;

[0100] Figure 8 (b) is a schematic diagram of the Y coordinate offset relative error in embodiment 2;

[0101] Fig. 9(a) is a schematic diagram of random down-sampling of the profile of the test sample in Example 2;

[0102] Fig. 9(b) is a schematic diagram of random down-sampling of the inner point of the profile of the test sample in Example 2;

[0103] Fig. 10(a) is a schematic diagram of the influence of the point density of the profile sampling on the volume calculation in Example 2;

[0104] Fig. 10(b) is a schematic diagram of the influence of the point density of the profile sampling on the volume calculation in Example 2;

[0105] Figure 11 Fig. 11 is a schematic diagram of the principle that the unit grid error and the point cloud density error in Example 2 are equivalent to the Z coordinate error. DETAILED DESCRIPTION

[0106] In order to further understand the content of the present application, the present application will be described in detail in conjunction with the drawings and examples. It should be understood that the examples are only used to explain the present application but not to limit the present application.

[0107] Example 1

[0108] The point cloud scanning of the slag bucket car will generate background points such as the profile of the slag bucket and the tunnel lining, which are irrelevant to the volume calculation. In order to measure the loading condition of the slag bucket with high precision, it is necessary to segment the point cloud of the transport train and extract the pure slag surface points. Therefore, the present embodiment proposes a soil pressure balance shield slag volume calculation method based on three-dimensional slag scanning according to the shape characteristics of the track car point cloud, as shown in Figure 1 The proposed method can adapt to both empty and full slag buckets.

[0109] Since the slag bucket is a single-sided open shell, the slag soil usually cannot be stacked to the slag bucket, and the outer surface of the slag bucket has various edges, and deformation and slag soil adhesion may occur, so that the volume calculation by three-dimensional reconstruction to construct a solid will produce a large error. Therefore, the idea of volume calculation is to extract the pure slag surface by point cloud segmentation and then divide the unit grid for numerical integration. In actual engineering, the size and shape of the slag bucket will change with the size of the shield and the difference of the engineering project. In order to improve the adaptability of the method to the actual engineering and reduce the calculation consumption, the method is studied according to the space characteristics of the slag bucket, and a high-precision slag soil point cloud segmentation is realized by superimposing various non-deep learning methods. The method includes the following steps:

[0110] Step 1, pretreatment:

[0111] Filter out the other parts except the four slag buckets, including the track car head and the segment.

[0112] The collected railcar point cloud contains slag hopper, locomotive, mud hopper, pipe segment to be installed, tunnel and other parts. Fine point cloud segmentation first needs to extract four slag hoppers from the complete point cloud. Each component of the railcar is distributed along the tunnel longitudinally one by one, that is, the line scanning point cloud is formed by stacking multiple plane profiles in the Z axis, that is, the Z coordinates of each profile are the same. The distribution characteristics of the point cloud can be used to quickly filter other components of the railcar.

[0113] The implementation process is as shown in Figure 2 . Since the scanning range is limited in the tunnel, the Y coordinate distribution of all profiles is consistent, and the morphological difference of the point cloud is mainly in the X direction. Therefore, the Z coordinates of each scanning profile are obtained first, and the X coordinate sum of all points in a single profile is calculated , as shown in formula (1); then the Z coordinates of the point cloud and their are plotted into a scatter plot. It can be seen that the trend of the railcar point cloud is consistent with the spatial characteristics of the three-dimensional point cloud. At this time, the spatial information is mapped into plane information, making data processing more convenient. The maximum value in the curve represents the tunnel bottom surface (the radar is located at the top of the tunnel), and they are distributed between each carriage.

[0114] To extract the slag hopper, first remove the points greater than the bottom surface of the slag hopper, that is, the tunnel bottom surface profile is removed, as described in formula (2); at this time, point cloud is missing between each carriage, which is represented by an interval , and the calculation method is as shown in formula (3);

[0115] (1)

[0116] (2)

[0117] (3)

[0118] In the formula, L is the Z coordinate set of all point cloud profiles to be processed (each Z layer corresponds to a profile); for each , the point set is , , which represents the X coordinate of the i-th point in the point set; represents the profile to be retained, each profile has a unique Z coordinate; is the profile position threshold, which is greater than the of the empty slag hopper profile; is the distance threshold; is the Z coordinate of the k-th point of the scanning profile.

[0119] Step 2, slag hopper separation:

[0120] The four slag hoppers were divided into independent point cloud clusters to facilitate subsequent segmentation operations.

[0121] In step 2, If sorted from the front of the vehicle, the four slag hoppers are distributed between the first and fifth intervals. The point cloud of each slag hopper is extracted independently based on the Z coordinate distribution in each interval.

[0122] Step 3, coarse segmentation:

[0123] First, remove the dust particles from the left and right side walls of the slag hopper; then remove the dust particles from the front and back side walls of the slag.

[0124] The challenge in segmenting the point cloud on the surface of construction waste lies in the fact that it is spatially continuous with the sidewalls of the hopper, and the segmentation boundary dynamically evolves with loading conditions. To accurately extract the surface of the construction waste, various methods are used to segment the point cloud. In the coarse segmentation step, two steps are used to remove the sidewalls of the hopper. As shown in Figure 3(a), the coarse segmentation operates on a single hopper, from the hopper's... As can be seen from the planar curve diagram, the left and right side walls and the middle partition... Significantly larger than the surface of the slag. And it has a significant trend of change, which can be based on The rate of change of values ​​is used to extract boundary points. For the treatment of the front and rear sidewalls of the slag hopper, each contour needs to be processed individually. From the contour coordinate scatter plot, it can be seen that the difference in X-coordinates between the front and rear sidewall points and the slag surface points decreases, reducing the reliability of determining boundary points based on the trend of X-value changes. Boundary points are determined by calculating the trend of the tangent slope at each point.

[0125] For the calculation of the boundary points of the left and right sidewalls and the middle partition, it is first necessary to define an appropriate calculation range. An excessively large calculation range will lead to a decrease in the reliability of the algorithm. The method proposed in this embodiment is also applicable to fully loaded point clouds. The case of an empty cloud is more complex, so an empty cloud is used as an example.

[0126] For empty slag hoppers, there is a partition in the middle, and the partition is in... The distribution on the scatter plot is similar to that of the left and right sidewalls. The point cloud can be divided into two parts from the middle and calculated separately. The calculation range of the boundary points is indented by 20 points from both ends, including some points on the surface of the slag and points on the sidewalls of the slag hopper. The light green rectangles in Figures 3(b) and 3(c) mark the calculation range of the boundary points. Each point is analyzed one by one from the slag surface points toward the sidewalls. The calculation direction of the points is shown by the green arrows in Figures 3(b) and 3(c). Boundary points The judgment conditions and the point segmentation conditions are shown in equations (4) and (5):

[0127] (4)

[0128] , (5)

[0129] wherein denotes the point set within the calculation range, is R the i-th point within the calculation range, is the X coordinate of the i-th point, , denotes the point set to be culled and the valid point set, respectively; denotes the X coordinate of the boundary point;

[0130] As shown in Fig. 3(c), the calculation of the front and rear sidewall boundary points is different from that of the left and right sidewalls. The coarse segmentation of the left and right sidewalls is performed by generating the coordinates of the slag pot profile and and calculating the boundary points, and the left and right sidewall points and the middle partition points are culled. The coarse segmentation of the front and rear sidewalls is performed by traversing the slag pot profile to calculate the boundary points and cull the front and rear sidewall points. Since the sidewall depth of the empty slag pot changes dynamically with the cross-sectional profile, the boundary point calculation range cannot be divided by the number of points. Considering the shape characteristics of the slag pot, the division of the boundary point calculation range is shown in the following equations (6) and (7):

[0131] Let

[0132] (6)

[0133] (7)

[0134] wherein denotes the single profile point set, , denotes the maximum and minimum Y coordinates within ; denotes the calculation range width; denotes the Y coordinate of the i-th point within .

[0135] Although the trend of the X coordinate of the points within is not obvious, the trend of the tangent slope from the slag surface to the sidewall point is still obvious, and the division boundary point within is calculated accordingly;

[0136] The difference vector of each point is calculated as follows:

[0137] (8)

[0138] For the starting point and the ending point

[0139] (9)

[0140] tangent slope and angle is calculated as follows:

[0141] (10)

[0142] (11)

[0143] boundary points The calculation method is shown as follows:

[0144] (12)

[0145] , (13)

[0146] , respectively represent the set of points to be removed and the set of valid points.

[0147] Step 4, fine segmentation:

[0148] The remaining point cloud of the front and rear side walls is removed again; then the scattered points remaining after the above step are filtered, that is, the noise points are removed.

[0149] The coarse segmentation method removes most of the side wall points of the slag, but due to the division of the boundary points, there may still be deviations, so using only one coarse segmentation method cannot completely remove the side wall points of the slag bucket, and there are still some side wall points remaining at the front and rear side wall boundaries of the slag surface. In addition, the slag bucket point cloud may leave some points in the processing process from pretreatment to fine segmentation, and these points become noise points of the slag surface point cloud. Therefore, a set of fine segmentation methods is used to remove these scattered points. The fine segmentation method is divided into two parts, the first part uses the front and rear side wall segmentation method again to remove the slag boundary residual points, and removes the residual points at the top end of the slag bucket. The positioning method of the residual points at the top end of the slag bucket is as follows:

[0150] (14)

[0151] (15)

[0152] In the formula, represents the top end residual point segmentation threshold, represents the offset distance, which is determined by the actual size of the slag bucket;

[0153] The second part of the fine segmentation method is to remove the noise points that obviously deviate from the slag surface points by using Density-Based Spatial Clustering of Applications with Noise (DBSCAN) on the entire slag surface point cloud. DBSCAN is a density-based spatial clustering algorithm that is good at finding clusters of arbitrary shape and automatically identifying noise points. DBSCAN determines point density by the number of domain points, and finally finds the maximum set of density-connected objects, while points outside the set are noise points. After a series of segmentations, the remaining noise points are distributed discretely, and there is a significant density difference between the slag surface point cloud. The DBSCAN algorithm based on density can efficiently filter the remaining noise points. The fine segmentation effect is shown in Figure 4 .

[0154] Step 5, volume calculation:

[0155] Complete the missing point cloud in the empty slag bucket, divide the obtained pure slag surface points into triangular meshes, and realize the volume calculation of the slag bucket loading; subtract the full load slag bucket volume result from the empty load slag bucket volume result to obtain the single tunneling ring slag volume.

[0156] The triangular subdivision volume method is a numerical method for calculating the volume of three-dimensional shapes based on triangular mesh discretization, especially suitable for irregular surfaces or scattered point cloud data volume solving. This method has a significant precision advantage for point cloud data obtained on construction sites, where the size of the slag bucket is fixed and the slag surface changes. When calculating, project the segmented slag surface point cloud to a XY plane, generate triangular meshes according to the Delaunay method, calculate the area of each triangular mesh, calculate the volume of the triangular prism according to the Z coordinate of the mesh vertex, and then add up the volume of all triangular prisms to obtain the volume of the slag bucket loading. However, point cloud missing and abrupt points after point cloud segmentation will cause a large number of abnormal triangular mesh elements, increasing the error, as shown in Figure 5 . The grid size at the point cloud missing place is much larger than the other parts, and these large grids introduce significant errors. In addition, due to the irregularity of the point cloud edge, a large number of abnormal triangular meshes are generated under the Delaunay rule, and the area of these meshes is much larger than the average level. These abnormal meshes have uncertainty, which reduces the uncertainty of the calculation result. Since point cloud segmentation inevitably causes some points to be lost, boundary construction is an important step to reduce the impact of point cloud segmentation error on volume calculation. The implementation method of point cloud completion and boundary construction is as follows:

[0157] For the middle missing of the empty load point cloud, copy the point cloud on the left and right sides of the missing part to fill it in. The missing part is continuous with the point cloud on both sides, and the difference in Z-axis coordinates is minimal, so the introduced error is relatively small. The calculation method is as follows:

[0158] (16)

[0159] (17)

[0160] wherein , represents the missing boundary X coordinate, represents the missing part interval, represents the boundary point serial number; is the X coordinate of the i-th point in the segmented point cloud

[0161] (18)

[0162] (19)

[0163] (20)

[0164] wherein , is the point area to be used for filling in the missing part, , is the point set for filling in the missing part, is the new point set after filling in the missing part;

[0165] Let the arbitrary side of the constructed frame be the point set , the starting point of which is , and the ending point of which is , the starting point and the ending point coordinates being determined according to the maximum and minimum values of the actual X, Y coordinates of the point cloud in combination with the actual size of the slag pot; is the average value of the Z coordinates of all points in , and the side length is:

[0166] (21)

[0167] According to the row and column point density of in the X-Y plane, XY , construct , wherein the X, Y coordinates of the points are:

[0168] (22)

[0169] For each frame point , the correction of the Z coordinate thereof is referenced to the projection in the X-Y plane; XY ; let be the projection of in the X-Y plane​​​​The nearest point index of ;

[0170] (23)

[0171] Finally The coordinates of any point inside are .

[0172] After obtaining a good triangular mesh, the volume of the slag surface and the slag bucket forming a three-dimensional entity is calculated, and the detailed calculation process is as follows:

[0173] Let the coordinates of the three vertices of the triangular mesh on the XY plane be x 0, y 0),( x 1, y 1),( x 2, y 2), the area of a single triangular mesh is :

[0174] (24)

[0175] Let the Z-axis coordinates of the three vertices of the triangle be z 0 , z 1 , z 2 , then the volume of a single triangular mesh is

[0176] (25)

[0177] where is the Z-axis coordinate of the constructed plane; Z

[0178] Since the real slag bucket bottom surface is not a two-dimensional plane, each point needs to be spatially registered with the normal position of the slag bucket bottom surface, and then the Z-axis distance from the point to the bottom surface is calculated. This process significantly increases the calculation complexity and resource consumption. To simplify the calculation, a virtual slag bucket bottom surface is constructed for the point cloud according to the slag bucket volume when calculating, and the calculation method of the plane is as follows:

[0179] (26)

[0180] where is the actual volume of the slag bucket; and are the actual length and width of the slag bucket, respectively; is the original point cloud coordinate, is the constructed plane coordinate;

[0181] The loading volume of the slag pot calculated based on the triangulation method is:

[0182] (27)

[0183] wherein, n represents the total number of grids in which the point cloud is divided; , respectively represent the area and volume of the first i grid;

[0184] Finally, the actual slag volume of each slag pot is the loading volume of the slag pot in the full load state minus the loading volume of the slag pot in the empty load state (residual slag left after pouring):

[0185] (28).

[0186] Example 2

[0187] In this embodiment, the error analysis of point cloud volume calculation is carried out. Considering the limitations of construction progress and test site in the construction process, it is difficult to directly control and accurately measure the actual loading capacity of the slag pot, and it is impossible to analyze the proposed volume calculation method in detail. In order to improve the reliability of the proposed volume calculation method for actual slag point cloud, a slag line scanning test is carried out to obtain point cloud data under different loading capacities, and the effects of slag loading capacity, scanning accuracy, point cloud density and other factors on volume calculation are discussed to further reduce the measurement error on the construction site.

[0188] 1) Test method

[0189] In order to simulate the point cloud collection effect of the shield construction site and accurately measure the actual loading volume, a slag line scanning simulation system is designed in the laboratory in this embodiment. The system is composed of a scaled slag pot, a moving support, a laser radar (of the same type as used in the construction site), a large electronic scale and machine-made sand. The line scanning LiDAR is installed at the top of the moving support, and the point cloud scanning of the slag pot is completed by manually pushing the moving support. The moving support is designed to move only in one direction, ensuring that the support and the slag pot only move relative to each other in one direction. The slag pot is placed on the large electronic scale, so the loading volume of the slag pot can be accurately controlled by the total mass of the slag pot. The test is respectively carried out on 44.56dm 3 , 89.12dm 3 , 133.68dm 3 , 178.24dm 3 , 222.80dm 3The sand soil under the volume was scanned for 10 times, and a total of 50 test samples were obtained. In order to simulate the actual shape of the slag soil surface, the surface morphology of the sand soil was artificially adjusted before each point cloud scanning, and an irregular curved surface was randomly constructed to verify the measurement error of the radar on the irregular curved surface.

[0190] The scanning frequency of the test radar is 15-50Hz / Line, the angle resolution is 0.05-0.5°, the angle range is 0-276°, the working range is 0.05-120m, and the statistical error is ≤5mm (0.05-5m); ≤10mm (5-20m, @10klx).

[0191] 2) The influence of loading on error

[0192] According to the processing method of the field data, the background points such as the side wall of the slag bucket and the ground in the test sample are removed, the loading volume is calculated using the triangulation method for the sand soil surface, and the error rate (calculation method as shown in formula 29) is calculated. The test results of 50 samples are shown in FIG. 6(a). The results show that when the loading rate is low, the volume calculation error rate is large, and the distribution range is wide, and a significant negative deviation occurs. When the loading amount is 44.56dm 3 , the error is distributed between-35% and 35%. With the increase of the loading amount, the error value decreases, and the error range narrows down, and when the loading amount is 222.80dm 3 , the error distribution range converges to-3%~3%. The average error rate under each loading amount is calculated to analyze the change trend, and the results are shown in FIG. 6(b). The test results show that the difference of the slag bucket loading amount has a significant influence on the volume calculation result, and there is a significant inverse exponential decay relationship between the loading level (%) and the average error rate (%), that is, the error rate decreases with the increase of the loading level, but the reduction rate gradually decays and eventually converges. When the loading amount is 15.3%, the error rate is as high as 20.63%, and when the loading amount reaches 76.6%, the error rate is only 1.86%. According to the fitted decay curve, when the loading amount reaches 95% (the loading level of the actual full load of the slag bucket is usually 95%), the calculation error rate is only 1.38%.

[0193] For the test point cloud, the volume calculation error is mainly determined by the system error of the point cloud scanning, that is, the inaccuracy of the point cloud in describing the three-dimensional surface of the slag soil due to the insufficient accuracy of the equipment and scanning, which leads to the deviation of the calculation result. The scanning system error is determined by the equipment and scanning method, and will not change with the loading amount. However, with the increase of the loading amount, the influence of the scanning system error on the volume calculation result is gradually diluted, so there is an inverse exponential decay relationship between the loading level (%) and the average error rate (%).

[0194] (29)

[0195] In the formula This represents the calculation error rate. and This indicates the calculated volume and the actual volume.

[0196] For triangulation volume calculations, systematic errors generated by point cloud scanning cause deviations between the calculated volume of the triangular prism and the actual situation. These errors typically fall into two categories: one caused by the height error of the triangular prism, i.e., the Z-coordinate error of the point; the other is due to the point's position... XY The distribution deviation on the plane, i.e., the error in the X and Y coordinates of points. To test the impact of scanning error on the calculation results, the experiment applied offsets to the Z and Y coordinates of 50 sets of test samples. The offset effects are shown below. Figure 7 As shown. For the Z-coordinate offset, the experiment tested the error results when the overall Z-coordinate increased by 20, 10, -10, and -20 mm (the maximum measurement error for LiDAR is 10 mm). It should be noted that the Z-coordinate offset of line-scan radar is usually caused by the measurement error of the equipment, and the Z-coordinate error of each point is usually random. Therefore, the experiment tested cases where the Z-coordinate was randomly offset but the average Z-coordinate offset of the point cloud remained at 20, 10, -10, and -20 mm. The results show that, because the point cloud of the experimental sample is uniformly distributed, the difference between the calculated results of random offset and overall offset is negligible. For points in... XY The distribution error of the plane was tested by randomly offsetting the Y coordinate. Considering that the Y coordinate interval of the test sample was 6 mm, the offset scale was set to 3, 6, 9, and 12 mm. Figure 7 The data shows that as the offset scale increases, the distribution of points in the X and Y planes gradually becomes chaotic, and the inhomogeneity of the triangular mesh increases accordingly.

[0197] The effects of point cloud offset on volume are shown in Figures 8(a) and 8(b). To highlight the impact of offset on calculation error, the relative errors caused by point cloud offset are statistically analyzed. The results show that the calculation error increases linearly with the increase of the Z-coordinate, unaffected by the offset direction, and the rate of error increase decreases with the increase of the loading rate. The impact of Y-coordinate offset on calculation error is relatively smaller than that of Z-coordinate offset (at 44.56 dm). 3 89.12dm 3 133.68dm 3 178.24dm 3 222.80dm 3The relative error rates of the volume under the Z coordinate deviation of 10 mm were 16.0074, 8.32869, 5.5061, 4.16164, and 3.29, respectively, and the relative error rates of the Y coordinate deviation of 9 mm were 13.647, 6.83013, 2.9694, 2.121, and 1.0869, respectively. The error growth rates under each loading rate were also relatively small. The results show that, under the triangulation computer mechanism, the height deviation of the triangular prism is more fatal than the grid form deviation. Moreover, in actual point cloud collection, the occurrence probability of Y coordinate bias is relatively low.

[0198] In summary, under the condition of triangulation method, the slag hopper scanning has a significant advantage in volume measurement. Even if the systematic error of the point cloud in describing the slag surface is artificially increased, the influence of the calculation result on the volume error of the full load state is still small. The height of the actual slag hopper can usually reach 2 m, and the volume can exceed 20 m 3 . The scanning error of the slag surface will further reduce the relative volume of the slag hopper. The belt conveyor slag profile scanning does not have the inclusiveness of scanning system error, and the system error of LiDAR in the Z direction will directly reflect the cumulative result of the slag profile.

[0199] 3) Influence of point cloud density on volume result

[0200] In addition to the bias error of LiDAR measurement points, another potential error factor is the change in point cloud density. The collection frequency of line scanning LiDAR is constant, and then the speed change of the rail car will cause the density change of the line scanning point profile. In order to explore the influence of point cloud density on volume calculation, the study statistically analyzes the results of volume error rate under different downsampling levels. The test tests the profile sampling and profile point sampling (the same effect as global sampling), and the sampling method is shown in FIG. 9 (a) and FIG. 9 (b). Profile sampling randomly discards a certain proportion of profiles in the global point cloud. This sampling method can simulate the density change of the line profile when the rail is not moving at a constant speed; profile point sampling randomly discards a certain proportion of points in each profile. This way can simulate the change of the angle resolution of the line LiDAR. Since the test sample and the actual slag point cloud have certain differences, and the change of the actual situation point cloud density has certain randomness, the discarded points under each sampling method are random. Under this condition, the test calculates 50 times for each sampling level and method, and finally takes the average value of the calculation. This controls the uncertainty brought by the randomness of sampling, so that the final result reaches the target confidence level. Since the volume calculation is based on the surface point cloud, the projection density of the surface point cloud on the plane is used to measure the volume error. XY

[0201] ​The test results are shown in FIG. 10(a) and FIG. 10(b), and the values in the heat map grid represent the proportion of the relative original sample volume error growth. The results show that the volume error rate increases exponentially with the increase of the point cloud downsampling level, and the error rate and the growth rate are affected by the loading amount. At the same downsampling level, the higher the loading amount, the smaller the volume error rate.

[0202] Comparing the contour sampling results and the contour point sampling results, it can be seen that the former has a greater impact on volume calculation, which is reflected in that the error growth rate of contour sampling is generally greater than that of contour point sampling at the same downsampling ratio. As can be seen from FIG. 10(a) and FIG. 10(b), contour point sampling can still better maintain the main shape of the contour at a high downsampling level, while contour sampling will lose spatial information faster at the same downsampling level. Therefore, during the process of slag hopper point cloud scanning, the matching degree of LiDAR scanning and the movement of the rail car should be improved, and the contour density of the slag surface should be increased.

[0203] In FIG. 10(a) and FIG. 10(b), the downsampling levels at which the error rate of each loading volume exceeds 1% are marked with black dashed squares, and the black square is used as a boundary. It is considered that the downsampling levels less than the black square have a relatively small impact on the volume error at each loading volume. In general, when the point cloud density is not less than 70% of the test sample, the impact of insufficient point density on the description error of the surface contour on the volume calculation result does not exceed 1%.

[0204] 4) Estimation of actual slag hopper volume error

[0205] It is difficult to accurately measure the slag volume on the construction site, and it is not convenient to carry out related work of actual volume error analysis. Therefore, the indoor test results provide theoretical support for estimating and optimizing the actual slag volume calculation error. The above research shows that the error of point cloud volume calculation based on the triangulation method can be divided into Z coordinate bias error, unit grid error (X, Y coordinate deviation), and point cloud density error (point cloud density deficiency causes distortion of point cloud description of spatial information). Considering that the slag surface is always limited in the given contour of the slag hopper, the unit grid error and the point cloud density error can also be essentially regarded as the Z coordinate bias error, as shown in the principle of Figure 11 The bottom area of the slag point cloud mapped to the XY plane is always equal to the bottom area of the slag hopper, regardless of how the grid is divided. The total area of the unit grid is constant, but the deviation of X, Y coordinates causes the volume of the triangular prism to deviate, and this error can be equivalent to the height bias of the triangular prism. For the point cloud density error, the lost point cloud reduces the description of the surface details, which is equivalent to interpolating between two real points, and the height of the interpolated point deviates from the real surface. Therefore, all errors can be represented by the Z coordinate bias error of the point. The loading volume of the slag hopper can be written as:

[0206] (30)

[0207] (31)

[0208] where , represents the average value of the Z coordinate of the triangular element mesh and its area in the XY XY plane. Since the area of the slag pot bottom is constant, it can be represented by where is the average value of the Z coordinate of the point cloud, and represents the area of the slag pot bottom. Then is divided into a standard part and a bias part Since is a constant, and are linearly related.

[0209] In order to estimate the error of the point cloud of the engineering slag pot, the equivalent Z coordinate bias of the radar measurement can be obtained by inversely calculating the volume error of the test sample (X, Y coordinate errors have been included). For the real slag point cloud, the density difference between it and the test sample needs to be considered. The average point cloud density of all 50 test samples is 45245.7 points / m 2 , and the actual slag surface point cloud density is about 14326.41 points / m 2 , so the introduced density error can be regarded as the case of 70% downsampling level. Therefore, is calculated as follows:

[0210] (32)

[0211] (33)

[0212] where , , respectively represent the average volume calculation error rate of the test sample under the i th loading capacity, the density error after downsampling of the point cloud, and the average loading height. represents the average equivalent Z coordinate bias of the line scan point cloud, represents the loading height of the actual slag. The calculated , the equivalent loading height of the 6.2 m diameter shield under empty load is about 0.357 m, and the equivalent loading height of the full load of slag is about 1.7 m. Finally, the system error of LiDAR scanning for the volume measurement of empty load and full load of slag is and about 2.44%, 0.514%, respectively.

[0213] The above description of the present application and its embodiments is illustrative and not restrictive, and the present application is not limited to the embodiments shown in the drawings. Therefore, if a person of ordinary skill in the art is inspired by the above description, and without departing from the spirit of the present application, similar structural modes and embodiments can be designed without creativity, and all of them should belong to the protection scope of the present application.

Claims

1. A method for calculating the volume of earth pressure balance shield tunnel excavated soil based on three-dimensional slag hopper scanning, characterized in that: Includes the following steps: Step 1, Preprocessing: Filter out all parts except the slag hopper, which includes the railcar head and segments; Step 2, slag hopper separation: The slag bin is divided into independent point cloud clusters; Step 3, coarse segmentation: First, remove the dust particles from the left and right side walls of the slag hopper; then remove the dust particles from the front and back side walls of the slag hopper. In step 3, the left and right side walls and the middle partition... according to Extracting boundary points based on the rate of change of values. The sum of the X coordinates of all points within a single contour; for the treatment of the front and rear sidewalls of the slag hopper, each contour needs to be operated on one by one, and the boundary points are determined by calculating the trend of the change of the tangent slope of each point. Specifically: For the calculation of the boundary points of the left and right side walls and the middle partition, the calculation range needs to be defined first; for an empty slag hopper, there is a partition in the middle, and the partition is located at... The distribution on the scatter plot is similar to that of the left and right sidewalls. The point cloud is divided into two parts from the middle and calculated separately. The calculation range of the boundary points is indented by 20 points from both ends, including some points on the surface of the slag and points on the sidewalls of the slag hopper. Each point is analyzed one by one from the points on the slag surface towards the sidewalls, including the boundary points. The judgment conditions and the point segmentation conditions are shown in equations (4) and (5): （4）； ， （5）； In the formula Represents the set of points within the computational range. for The i-th point inside, for The X coordinate, , These represent the set of points to be removed and the set of valid points, respectively. Indicates the X coordinate of the boundary point; The calculation of the boundary points of the front and rear side walls is different from that of the left and right side walls. Considering the shape characteristics of the slag hopper, the division of the boundary point calculation range is shown in equations (6) and (7) as follows: set up ; （6）； （7）； In the formula Represents a single set of contour points. , express The maximum and minimum Y coordinates within the range; Indicates the width of the calculation range; express The y-coordinate of the i-th point within the range; The trend of the tangent slope from the surface of the slag to the sidewall point is still obvious, and this is used to calculate... The dividing boundary points within; Difference vectors of each point The calculation is as follows: （8）； Regarding the starting and ending points: 、 （9）； Tangent slope and angle The calculation is as follows: （10）； （11）； Boundary points The calculation method is as follows: （12）； ， （13）； , These represent the set of points to be removed and the set of valid points, respectively. Step 4, fine segmentation: Remove the remaining point cloud from the front and back sidewalls again; then filter the scattered points remaining from the above steps, i.e., remove noise. In step 4, the fine segmentation method is divided into two parts. The first part uses the front and rear sidewall segmentation method again to remove the residual points at the boundary of the slag and remove the residual points at the top of the slag hopper; the residual points at the top of the slag hopper... The positioning method is as follows: （14）； （15）； In the formula This indicates the threshold for segmentation of the residual points at the top. This indicates the offset distance, which is determined by the actual dimensions of the slag hopper. The second part of the fine segmentation method is to use DBSCAN to remove noise points that are significantly off-center from the surface of the slag and soil. Step 5, Volume Calculation: Complete the missing point cloud in the middle of the empty slag hopper, divide the obtained pure slag surface points into triangular meshes to realize the calculation of slag hopper loading volume; and obtain the single tunneling ring slag discharge volume by subtracting the full-load slag hopper volume result from the empty slag hopper volume result.

2. The method for calculating the volume of earth pressure balance shield tunnel excavated soil based on three-dimensional slag hopper scanning according to claim 1, characterized in that: Step 1 specifically involves: First, obtain the Z coordinate of each scan contour, and then calculate the sum of the X coordinates of all points within a single contour. As shown in formula (1); then the point cloud Z coordinates and their When a scatter plot is drawn, spatial information is mapped to planar information, and the curves... The maximum value represents the bottom surface of the tunnel, and they are distributed between the various carriages; To extract the slag hopper, points larger than the bottom surface of the slag hopper are first removed, i.e., the tunnel bottom outline is removed, as described in formula (2); at this time, point cloud gaps appear between each carriage, and these gaps are separated by intervals. The calculation method is as shown in formula (3); （1）； （2）； （3）； In the formula L For each point cloud contour to be processed, the set of Z coordinates is provided. The point set is , This represents the X-coordinate of the i-th point in the point set; This represents the contours that need to be retained, and each contour has a unique Z-coordinate; The contour position threshold is set to ensure it is greater than the contour of the empty slag hopper. ; Distance threshold; The Z coordinate of the k-th point of the scanned contour.

3. The method for calculating the volume of earth pressure balance shield tunnel excavated soil based on three-dimensional slag hopper scanning according to claim 2, characterized in that: In step 2, If sorted from the front of the vehicle, the four slag hoppers are distributed between the first and fifth intervals. The point cloud of each slag hopper is extracted independently based on the Z coordinate distribution in each interval.

4. The method for calculating the volume of earth pressure balance shield tunnel excavated soil based on three-dimensional slag hopper scanning according to claim 3, characterized in that: Step 5 specifically involves: Project the point cloud of the segmented waste soil surface onto... XY In the plane, triangular meshes are generated according to the Delaunay method, the area of ​​each triangular mesh is calculated, the volume of the triangular prism is calculated based on the Z coordinate of the mesh vertex, and then all triangular prisms are accumulated to obtain the slag hopper loading volume.

5. The method for calculating the volume of earth pressure balance shield tunnel excavated soil based on three-dimensional slag hopper scanning according to claim 4, characterized in that: In step 5, the impact of point cloud segmentation errors on volume calculation is reduced by constructing boundaries. The methods for point cloud completion and boundary construction are as follows: For the missing part in the center of the empty point cloud, the point clouds on the left and right sides of the missing part are copied to fill it in. The calculation method is as follows: （16）； （17）； In the formula , Indicates the missing boundary X coordinate. Indicates the spacing of the missing part. Indicates the boundary point number; To segment point clouds The i-th point The X coordinate; （18）； （19）； （20）； In the formula , The area to be filled with missing points. , To fill in the missing point set, The new set of points after filling in the gaps; Let the arbitrary edge of the constructed border be a set of points. Its starting point is The destination is The starting and ending coordinates are determined based on the maximum and minimum values ​​of the actual X and Y coordinates of the point cloud, combined with the actual dimensions of the slag hopper. for Mean Z-coordinate of all points; side length for: （21）； according to In XY Row and column point density of a plane , structure The X and Y coordinates of the point are: （22）； For each border point Its Z-coordinate correction reference exist XY Projection of a plane ;set up for exist The nearest neighbor index; （23）； final The coordinates of any point inside are .

6. The method for calculating the volume of earth pressure balance shield tunnel excavated soil based on three-dimensional slag hopper scanning according to claim 5, characterized in that: In step 5, after obtaining a good triangular mesh, the volume of the three-dimensional solid formed by the slag surface and the slag hopper is calculated. The detailed calculation process is as follows: Let the triangular mesh be in XY The coordinates of the three vertices on the plane are ( x 0, y 0),( x 1, y 1),( x 2, y 2) Area of ​​a single triangular grid for: （24）； Let the Z-axis coordinates of the three vertices of the triangle be... z 0 , z 1 , z 2 The volume of a single triangular mesh is: （25）； in, To construct a plane Z Axis coordinates; During the calculation, a virtual bottom surface of the slag hopper is constructed based on the point cloud of the slag hopper volume, and a plane is constructed. The calculation method is as follows: （26）； in, This refers to the actual volume of the slag hopper; and These are the actual length and width of the slag hopper, respectively. The original point cloud coordinates, To construct planar coordinates; The slag hopper loading volume calculated based on the triangulation method is: （27）； in, n This represents the total number of grids into which the point cloud is divided; , They represent the first i The area and volume of each grid cell; Ultimately, the actual slag discharge volume of each slag bucket The loading volume of the slag hopper under full load condition Excluding the slag hopper loading volume under empty conditions : 。

Images