Intelligent Classification Method and System for Surrounding Rock of Tunnel Face Based on Drilling Parameter Images

Abstract

The present invention discloses an intelligent classification method and system for surrounding rock of tunnel face based on drilling parameters images, specifically including the steps of: establishing a drilling parameter database and performing data cleaning using the box plot method; based on the cleaned data, generating a standardized data grid by establishing a grid and performing data interpolation and normalization on the grid; based on the data grid, converting the drilling parameters into pixel values to generate a drilling parameter cloud map of the tunnel face; establishing a convolutional neural network model to identify the drilling parameter cloud map of the tunnel face to obtain the surrounding rock level. The tunnel surrounding rock classification method of the present invention generates a drilling parameter cloud map of the tunnel face based on drilling parameters, realizes intelligent classification of surrounding rock through image recognition, fully considers the geological heterogeneity of the tunnel face, and has good accuracy and generalization performance in surrounding rock classification.

Description

Technical Field

[0001] The present invention relates to the technical field of tunnel engineering, and particularly relates to an intelligent classification method and system for surrounding rock of a tunnel face based on drilling parameters images. Background Art

[0002] Before the start of a tunnel project, on-site investigation needs to be carried out to determine the surrounding rock conditions. In China, the BQ value method is usually combined with geological exploration data to divide the surrounding rock quality from good to bad into grades I - VI, and the surrounding rock quality information is used to guide the tunnel design and construction plan. Drilling parameter measurement is an effective data acquisition technology, which can be used to describe the mechanical and structural characteristics of the surrounding rock, and can be used to identify the characteristics of the surrounding rock, such as the joints and fissures of the surrounding rock, the hardness of the surrounding rock. Further, it can also realize the identification of the surrounding rock grade.

[0003] In the existing intelligent classification model for surrounding rock based on drilling parameters, only the average value of drilling parameter data is used as the input during the model training process, without fully considering the influence of geological heterogeneity on the evaluation of the surrounding rock quality of the tunnel face. However, the geological non - uniformity of the tunnel face is widespread and will affect the stability of the tunnel. Therefore, there is an urgent need for a method for evaluating the quality of the surrounding rock of the tunnel face that takes into account the influence of geological heterogeneity on the evaluation of the surrounding rock quality of the tunnel face. Summary of the Invention

[0004] To solve the problems existing in the prior art, the present invention provides an intelligent classification method and system for surrounding rock of a tunnel face based on drilling parameters images, which consider the influence of geological heterogeneity on the evaluation of the surrounding rock quality of the tunnel face by generating drilling parameters images, and automatically obtain the surrounding rock grade by using image recognition methods, providing guidance for tunnel construction and solving the problems mentioned in the above background art.

[0005] To achieve the above object, the present invention provides the following technical solution: An intelligent classification method for surrounding rock of a tunnel face based on drilling parameters images, comprising the following steps:

[0006] S1. Establish a drilling parameter database and perform data cleaning using the box - plot method;

[0007] S2. Based on the cleaned data, generate a normalized data grid by establishing a grid and performing data interpolation and normalization on the grid;

[0008] S3. Based on the data grid, convert the drilling parameters into pixel values to generate a drilling parameter cloud map of the tunnel face;

[0009] S4. Establish a convolutional neural network model to identify the surrounding rock grade from the drilling parameter cloud map of the tunnel face.

[0010] Preferably, the drilling parameters in step S1 include four items: feed rate, propulsion pressure, impact pressure, and rotary pressure.

[0011] Preferably, the box plot method in step S1 specifically determines the non-outlier interval according to quartiles, and the interval calculation formula is:

[0012] [Q1 - 1.5×IQR, Q3 - 1.5×IQR];

[0013] In the formula: Q1 is the first quartile, Q3 is the third quartile, and IQR is equal to Q3 - Q1.

[0014] Preferably, step S2 specifically includes:

[0015] S21. Based on the cleaned data, average the drilling parameters of all drill holes to obtain the representative characteristic value of each drill hole;

[0016] S22. Determine the grid size according to the tunnel design profile and grid the sample data;

[0017] S23. Use interpolation technology to interpolate the grid data.

[0018] Preferably, the interpolation technology includes nearest neighbor interpolation, linear interpolation, quadratic interpolation, cubic interpolation, and Lagrange interpolation.

[0019] Preferably, step S3 specifically includes:

[0020] S31. Normalize the sample data using the min-max method, and the formula is as follows:

[0021]

[0022] In the formula: x is the value before normalization, x max is the maximum value of the sample database before normalization, x min is the minimum value of the sample database before normalization, and x' is the value after normalization.

[0023] When performing normalization processing, in order to provide better uniformity, x max and x min respectively take the upper and lower limits of the non-outlier interval calculated in step S1; take the feed rate and its non-outlier interval as negative values, and then perform normalization.

[0024] S32. Use the drilling parameter values normalized to [0, 255] as the pixel values of the four channels, and generate a cloud map of drilling parameters using the four-channel (CMYK) mode.

[0025] Preferably, step S4 specifically includes:

[0026] S41. Expand the sample set by using the method of image data augmentation;

[0027] S42. Based on the sample set obtained by data augmentation, train a convolutional neural network model, input the cloud map of drilling parameters into the convolutional neural network model, and automatically obtain the corresponding surrounding rock level.

[0028] On the other hand, to achieve the above object, the present invention also provides the following technical solution: An intelligent grading system for tunnel face surrounding rock based on drilling parameter images, the system includes:

[0029] A data cleaning module, which establishes a drilling parameter database and performs data cleaning by using the box plot method;

[0030] A data grid module, which, based on the cleaned data, generates a standardized data grid by establishing a grid and performing data interpolation and normalization on the grid;

[0031] A drilling parameter cloud map generation module, which, based on the data grid, converts the drilling parameters into pixel values to generate a cloud map of drilling parameters for the tunnel face;

[0032] A surrounding rock level acquisition module, which establishes a convolutional neural network model and identifies the surrounding rock level by recognizing the cloud map of drilling parameters for the tunnel face.

[0033] The beneficial effects of the present invention are as follows: The intelligent grading method for tunnel face surrounding rock provided by the present invention is based on the drilling parameters collected during the drilling operation in the construction project. By generating the drilling parameter images, it takes into account the influence of geological inhomogeneity on the surrounding rock grading, and uses the image recognition method to construct an intelligent grading convolutional neural network model for the surrounding rock, realizing the intelligent grading of tunnel face surrounding rock considering geological inhomogeneity, providing guidance for tunnel construction, and improving the reliability of intelligent grading of tunnel surrounding rock. Description of the Drawings

[0034] Figure 1 is a flowchart of the steps of the intelligent grading method for tunnel face surrounding rock based on drilling parameter images provided in Embodiment 1 of the present invention;

[0035] Figure 2 is a schematic diagram of abnormal values of drilling parameters provided in Embodiment 1 of the present invention;

[0036] Figure 3 is a process diagram of sample data meshing using feed rate data provided in Embodiment 1 of the present invention;

[0037] Figure 4 is a schematic diagram of the relationship between each channel of drilling parameters and the CMYK cloud map provided in Embodiment 1 of the present invention;

[0038] Figure 5It is the CMYK cloud map of the tunneling face drilling parameters of typical sections with different surrounding rock grades provided in Embodiment 1 of the present invention. Figure 5 (a) is grade II surrounding rock, Figure 5 (b) is grade III surrounding rock, Figure 5 (c) is grade IV surrounding rock;

[0039] Figure 6 It is the CMYK cloud map of the drilling parameters of the typical section before and after horizontal flipping provided in Embodiment 1 of the present invention. Figure 6 (a) is before flipping, Figure 6 (b) is after flipping;

[0040] Figure 7 It is the intelligent surrounding rock grading transfer learning model based on the Inceptionv3 convolutional neural network provided in Embodiment 1 of the present invention.

[0041] Figure 8 It is the performance comparison result of the intelligent surrounding rock grading model provided in Embodiment 1 of the present invention.

[0042] Figure 9 It is the schematic diagram of the modules of the intelligent surrounding rock grading system for the tunneling face based on the drilling parameter images provided in Embodiment 2 of the present invention.

[0043] In the figure, 110 - data cleaning module; 120 - data grid module; 130 - drilling parameter cloud map generation module; 140 - surrounding rock grade acquisition module. Detailed implementation manners

[0044] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0045] Embodiment 1

[0046] Before the start of the tunnel project, on-site investigation shall be carried out to determine the surrounding rock conditions. In China, the BQ value method is usually combined with geological exploration data to divide the surrounding rock quality from good to bad into grades I - VI, and the surrounding rock quality information is used to guide the tunnel design and construction plan. Drilling parameter measurement is an effective data acquisition technology, which can be used to describe the mechanical and structural characteristics of the surrounding rock, and can be used to identify the characteristics of the surrounding rock, such as the joints and fractures of the surrounding rock, the hardness of the surrounding rock. Further, it can also realize the identification of the surrounding rock grade.

[0047] In the existing intelligent grading model for surrounding rock of drilling parameters, only the average value of drilling parameter data is used as the input during model training, without fully considering the influence of geological heterogeneity on the evaluation of the surrounding rock quality of the tunnel face. However, the geological non-uniformity of the tunnel face is widespread and will affect the stability of the tunnel.

[0048] Therefore, through long-term research, the present inventor provides an intelligent grading method for the surrounding rock of the tunnel face based on drilling parameter images, aiming to consider the influence of geological heterogeneity on the evaluation of the surrounding rock quality of the tunnel face by generating drilling parameter images, and automatically obtaining the tunnel surrounding rock level using image recognition methods to provide guidance for tunnel construction.

[0049] The present invention provides a technical solution: an intelligent grading method for the surrounding rock of the tunnel face based on drilling parameter images, and the step process is as Figure 1 shown, specifically including the following steps:

[0050] S1. Establish a drilling parameter database and perform data cleaning using the box plot method;

[0051] In a specific implementation, the drilling parameters include four items: feed rate, propulsion pressure, percussion pressure, and rotary pressure;

[0052] In a specific implementation, the drilling parameters may contain two types of outliers along the borehole depth. One is at the beginning of the borehole (about 0.3 - 0.5 m). To ensure that the borehole through the surrounding rock is straight, the hole opening is usually carried out very carefully, that is, before these parameters gradually increase to the normal level, the drilling pressure and speed are reduced ( Figure 2 , marked as "Type I"). The drilling parameters recorded at the beginning of these boreholes are not related to the characteristics of the surrounding rock. Sensor errors or the machine safety system cause the drilling parameters to occasionally be too large or too small ( Figure 2 , marked as "Type II"). The schematic diagram of the drilling parameter outliers is as Figure 2 shown.

[0053] In a specific implementation, for Type I drilling parameter outliers, they are directly deleted. For Type II drilling parameters, the box plot method is used to determine the non-outlier interval according to the quartiles, and the interval calculation formula is:

[0054] [Q1 - 1.5×IQR, Q3 - 1.5×IQR];

[0055] In the formula: Q1 is the first quartile, Q3 is the third quartile, and IQR is equal to Q3 - Q1.

[0056] In a specific implementation, the collected drilling parameter samples contain approximately 9.8 million data points, and their detailed information is shown in Table 1.

[0057] Statistical Feature Details of Drilling Parameters in Table 1

[0058]

[0059] In a specific implementation, the lower limit of the non-outlier interval of the feed rate is negative, which is unreasonable. This may be due to the drill bit retracting or sensor errors and has nothing to do with the surrounding rock quality. Therefore, the non-outlier interval range of the feed rate in Table 1 is changed from [-0.51, 5.50] to [0.00, 5.50]. The non-outlier intervals for sample data cleaning are shown in Table 2.

[0060] Table 2 Non-outlier Intervals for Sample Data Cleaning

[0061]

[0062] S2. Based on the cleaned data, by establishing a grid and performing data interpolation and normalization on the grid, a normalized data grid is generated;

[0063] 1) Based on the cleaned data, average the drilling parameters of all boreholes to obtain the representative characteristic values of each borehole;

[0064] 2) According to the tunnel design profile, determine the grid size and grid the sample data;

[0065] In a specific implementation, according to the tunnel design profile, the grid size is determined to be 15.15 m (width) × 12.80 m (height). The grid spacing is 0.05 m both horizontally and vertically. The grid consists of 303 × 256 squares with a side length of 0.05 m.

[0066] 3) Use interpolation techniques for grid data interpolation.

[0067] Interpolation techniques include nearest neighbor interpolation, linear interpolation, quadratic interpolation, cubic interpolation, and Lagrange interpolation.

[0068] In a specific implementation, the nearest neighbor interpolation method is used for grid data interpolation. This method first finds the actual borehole closest to the grid point and determines the drilling parameter of the grid point directly according to this drilling parameter. Therefore, this method ensures the allocation of all grid point data and can accurately reflect the characteristics of the local geological uniformity of the tunnel face. Please refer to Figure 3 , Figure 3 which shows the process of gridding sample data using feed rate data.

[0069] S3. Based on the data grid, convert the drilling parameters into pixel values to generate a drilling parameter cloud map of the tunnel face;

[0070] 1) The min-max method is used to normalize the sample data, and the formula is as follows:

[0071]

[0072] Where: x is the value before normalization, x max is the maximum value of the sample database before normalization, x min is the minimum value of the sample database before normalization, and x' is the value after normalization.

[0073] During the normalization process, in order to provide better uniformity, x max and x min respectively take the upper and lower limits of the non-outlier interval calculated in step S1; take the feed rate and its non-outlier interval as negative values, and then perform normalization.

[0074] 2) The drilling parameter values normalized to [0, 255] are used as the pixel values of the four channels, and a drilling parameter cloud map is generated using the four-channel (CMYK) mode.

[0075] The CMYK mode is a color matching mode used in color printing. It uses the principle of mixing three primary colors and black ink to form the so-called "full-color printing".

[0076] In a specific implementation, c is cyan in the CMYK mode, corresponding to the impact pressure value in the present invention. When the pixel value changes from 0 to 255, the cyan component changes from bright to dark, indicating that the impact pressure value changes from 64.50 bar to 196.50 bar. This corresponds to the change of the surrounding rock grade from high to low (i.e., the change of the surrounding rock quality from low to high). m is magenta in the CMYK mode, corresponding to the propulsion pressure in the present invention. When the pixel value changes from 0 to 255, the magenta changes from bright to dark, which means that the propulsion pressure changes from 4.50 bar to 112.50 bar. This indicates that the surrounding rock grade changes from high to low (i.e., the surrounding rock quality changes from low to high). y is yellow in the CMYK mode, corresponding to the rotary pressure in the present invention. When the pixel value changes from 0 to 255, the yellow changes from bright to dark, that is, the rotary pressure changes from 17.00 bar to 153.00 bar. This shows that the surrounding rock changes from high to low (i.e., the surrounding rock quality changes from low to high). k is black in the CMYK mode, corresponding to the feed rate in the present invention. When the pixel value changes from 0 to 255, the black changes from bright to dark, that is, the feed rate changes from 5.50 m / min to 0 m / min. This indicates that the surrounding rock grade changes from high to low (i.e., the surrounding rock quality changes from low to high). The relationship between the drilling parameters and each channel of the CMYK cloud map is as Figure 4 shown.

[0077] The CMYK cloud maps of the drilling parameters of typical sections with different surrounding rock grades are as Figure 5 shown, and are composed of Figure 5It can be seen that the CMYK cloud maps of the drilling parameters of the typical cross-sections of the surrounding rocks of Grade II, Grade III, and Grade IV (as shown in Figure 5 a, Figure 5 b, Figure 5 c) all show non-uniform characteristics, reflecting that the quality of the surrounding rocks on the same heading face is not the same; and as the grade of the surrounding rock increases (from Grade IV to Grade II), the quality of the surrounding rock improves, and the color of the corresponding CMYK cloud map of the drilling parameters becomes darker, which is consistent with the corresponding relationship between the drilling parameters, the quality of the surrounding rock, and each single channel of CMYK.

[0078] S4. Establish a convolutional neural network model to identify the grade of the surrounding rock by recognizing the cloud map of the drilling parameters of the tunnel heading face.

[0079] 1) Use the method of image data augmentation to expand the sample set.

[0080] In a specific implementation, considering that the tunnel contour is a left-right symmetric structure and the quality of the surrounding rock on the left or right side of the tunnel heading face does not affect the judgment of the overall surrounding rock quality, the sample data is augmented by randomly flipping it horizontally. Figure 6 The CMYK cloud maps of the drilling parameters before and after horizontal flipping are shown.

[0081] 2) Based on the sample set with data augmentation, train the convolutional neural network model, input the cloud map of the drilling parameters into the convolutional neural network model, and automatically obtain the corresponding grade of the surrounding rock.

[0082] In a specific implementation, based on the Inception-v3 model, a surrounding rock classification CNN model is established through transfer learning. The CMYK cloud map of the drilling parameters of the tunnel heading face is recognized, and the corresponding grade of the surrounding rock is output. Figure 7 The intelligent surrounding rock classification model based on Inception-v3 is shown.

[0083] In a specific implementation, the average value of the drilling parameters of the heading face is also used as the input, and six machine learning algorithms, namely Support Vector Machine (SVM for short), K-Nearest Neighbour (KNN for short), Random Forest (RF for short), Extra Trees (ET for short), BaggingClassifier (Bag for short), and Gradient Boosting (GB for short), are used to train six intelligent surrounding rock classification models for performance comparison. The comparison results are as shown in Figure 8 shown.

[0084] From Figure 8It can be seen that, compared with the surrounding rock intelligent grading model with the average value of the heading face drilling parameters as the input, the accuracy of the surrounding rock intelligent grading model (CNN model) with the CMYK cloud map of drilling parameters as the input is increased by 5% - 14%, which reflects the superiority of the method provided by this patent.

[0085] The intelligent surrounding rock grading method for tunnel heading face provided by the present invention is based on the drilling parameters collected during the drilling process in the construction, constructs an intelligent surrounding rock grading convolutional neural network model, realizes the intelligent grading of the surrounding rock of the tunnel heading face considering geological heterogeneity, provides timely and effective guidance for the on-site construction of the tunnel, and improves the reliability of the intelligent grading of the tunnel surrounding rock.

[0086] Embodiment 2

[0087] Based on the same inventive concept as the above method embodiment, the embodiment of the present application also provides an intelligent surrounding rock grading system for tunnel heading face based on the drilling parameter image, and this system can realize the functions provided by the above method embodiment, such as Figure 9 As shown, the system includes:

[0088] A data cleaning module 110, which establishes a drilling parameter database and performs data cleaning using the box plot method;

[0089] A data grid module 120, which based on the cleaned data, generates a standardized data grid by establishing a grid and performing data interpolation and normalization on the grid;

[0090] A drilling parameter cloud map generation module 130, which based on the data grid, converts the drilling parameters into pixel values to generate a CMYK cloud map of the tunnel heading face drilling parameters;

[0091] A surrounding rock level acquisition module 140, which establishes a convolutional neural network model and identifies the surrounding rock level from the CMYK cloud map of the tunnel heading face drilling parameters.

[0092] Although the present invention has been described in detail with reference to the foregoing embodiments, for those skilled in the art, they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements for some of the technical features. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. An intelligent classification method for surrounding rock of tunnel face based on drilling parameter images, characterized in that, It includes the following steps: S1. Establish a drilling parameter database and perform data cleaning using the box plot method; S2. Based on the cleaned data, generate a normalized data grid by creating a grid and performing data interpolation and normalization on the grid; S3. Based on the data grid, convert the drilling parameters into pixel values to generate a cloud map of the drilling parameters of the tunnel face; specifically, it includes: taking the drilling parameter values normalized to [0, 255] as the pixel values of four channels and generating the cloud map of the drilling parameters using the four-channel CMYK mode; S4. Establish a convolutional neural network model to identify the cloud map of the drilling parameters of the tunnel face to obtain the surrounding rock level.

2. The intelligent classification method for surrounding rock of tunnel heading face based on drilling parameter images according to claim 1, wherein: In step S1, the drilling parameters include four items: feed rate, propulsion pressure, percussion pressure, and rotary pressure.

3. The intelligent classification method for surrounding rock of tunnel heading face based on drilling parameter images according to claim 1, characterized in that: In step S1, the box plot method specifically determines the non-outlier interval according to the quartiles, and the interval calculation formula is: [Q1 - 1.5×IQR, Q3 - 1.5×IQR]; In the formula: Q1 is the first quartile, Q3 is the third quartile, and IQR is equal to Q3 - Q1.

4. The intelligent classification method for surrounding rock of tunnel heading face based on drilling parameter images according to claim 1, characterized in that: Step S2 specifically includes: S21. Based on the cleaned data, average the drilling parameters of all drill holes to obtain the representative characteristic value of each drill hole; S22. Determine the grid size according to the tunnel design profile and grid the sample data; S23. Use the interpolation technique to perform grid data interpolation to generate a normalized data grid.

5. The intelligent classification method for surrounding rock of tunnel heading face based on drilling parameter images according to claim 4, wherein: The interpolation techniques include nearest neighbor interpolation, linear interpolation, quadratic interpolation, cubic interpolation, and Lagrange interpolation.

6. The intelligent classification method for surrounding rock of tunnel heading face based on drilling parameter images according to claim 1, characterized in that: The normalization in step S2 specifically includes: Perform normalization processing on the sample data using the min-max method, and the formula is as follows: where: x is the value before normalization, x max is the maximum value of the sample database before normalization, x min is the minimum value of the sample database before normalization, and x′ is the value after normalization; When performing normalization processing, x max and x min respectively take the upper and lower limits of the non-outlier interval obtained by calculating in step S1; take the feed rate and its non-outlier interval as negative values, and then perform normalization.

7. The intelligent classification method for surrounding rock of tunnel heading face based on drilling parameter images according to claim 1, characterized in that: Step S4 specifically includes: S41. Use the method of image data augmentation to expand the sample set; S42. Based on the sample set after data augmentation, train the convolutional neural network model, input the cloud map of the drilling parameters into the convolutional neural network model, and automatically obtain the corresponding surrounding rock level.

8. An intelligent classification system for surrounding rock of tunnel face based on drilling parameter images, characterized in that: The system includes: A data cleaning module (110), which establishes a drilling parameter database and performs data cleaning using the box plot method; A data grid module (120), which generates a normalized data grid by creating a grid and performing data interpolation and normalization on the grid based on the cleaned data; A drilling parameter cloud map generation module (130), which converts the drilling parameters into pixel values based on the data grid to generate a cloud map of the drilling parameters of the tunnel face; specifically, it includes: taking the drilling parameter values normalized to [0, 255] as the pixel values of four channels and generating the cloud map of the drilling parameters using the four-channel CMYK mode; A surrounding rock level acquisition module (140), which establishes a convolutional neural network model to identify the cloud map of the drilling parameters of the tunnel face to obtain the surrounding rock level.

Images