Method and system for evaluating particle size distribution of rock ballast in double-mode shield construction of complex stratum

Abstract

The invention discloses a complex stratum dual-mode shield construction rock ballast particle size distribution evaluation method and system, and the method comprises the steps: obtaining a rock ballast image, carrying out the preprocessing and denoising, and obtaining a first image; utilizing a deep learning algorithm to identify the edge of the rock ballast, and separating overlapped rock ballast to obtain a second image; analyzing the shape of the rock ballast through morphological parameters, and performing three-dimensional reconstruction correction on the non-spherical particles to obtain a third image; performing data fusion on the third image in combination with laser particle size analysis data to obtain initial particle size distribution; aiming at a fine particle adhesion phenomenon, correcting particle size distribution by adopting a detection algorithm; dynamically updating particle size distribution according to the change of a crushing mechanism during shield mode switching; non-linear mutation in the complex stratum is analyzed through a mutation detection algorithm, and data smoothing is carried out; and finally, analyzing the error by using an error accumulation model, and if the error exceeds a threshold value, reprocessing until the error meets the requirement. According to the method, the precision and reliability of rock ballast particle size distribution evaluation are remarkably improved.

Description

Technical Field

[0001] The present invention belongs to the field of information technology, and particularly relates to a method and system for evaluating the particle size distribution of rock debris in the construction of a dual-mode shield tunneling machine in complex strata. Background Art

[0002] During the construction of a dual-mode shield tunneling machine in complex strata, the accurate evaluation of the particle size distribution of rock debris faces multiple technical challenges. First, the formation conditions are complex and variable, and the fragmentation characteristics of different lithologies vary significantly, resulting in irregular and unevenly distributed rock debris. Traditional single measurement methods are difficult to comprehensively capture the true particle size characteristics of rock debris. Image recognition technology has misjudgments when dealing with rock debris with blurred edges or overlaps. Laser particle size analysis has insufficient measurement accuracy for non-spherical particles, and mechanical screening cannot effectively handle the adhesion of fine particles. Second, when the dual-mode shield switches the excavation mode, the fragmentation mechanism of the rock debris changes, and a single measurement technology is difficult to dynamically adapt to this change. In addition, the cumulative measurement error of irregularly shaped rock debris will further affect the overall evaluation accuracy of the particle size distribution. Even more complex is that at the interface of different geological units in complex strata, the fragmentation mode of rock debris has significant non-linear characteristics, and existing technologies are difficult to accurately capture this sudden change. These problems together constitute the technical obstacles to the accurate evaluation of the particle size distribution of rock debris in the construction of a dual-mode shield tunneling machine in complex strata. Summary of the Invention

[0003] To solve the above technical problems, the present invention provides a method and system for evaluating the particle size distribution of rock debris in the construction of a dual-mode shield tunneling machine in complex strata. Among them, a method for evaluating the particle size distribution of rock debris in the construction of a dual-mode shield tunneling machine in complex strata includes:

[0004] Obtain the rock debris image data collected during the shield tunneling construction, preprocess to remove noise interference, and obtain the first image;

[0005] Adopt an image segmentation algorithm based on deep learning to identify the edges of the rock debris in the first image, judge whether the rock debris overlaps, and if it overlaps, perform separation processing to obtain the second image;

[0006] Perform shape analysis on the rock debris in the second image through preset morphological parameters, judge whether it is a non-spherical particle, and if it is non-spherical, use three-dimensional reconstruction technology for shape correction to obtain the third image;

[0007] According to the shape characteristics of the rock debris in the third image, combine with the particle size data obtained by a laser particle size analyzer to perform data fusion to obtain a preliminary particle size distribution result;

[0008] For the adhesion phenomenon of fine particles, adopt a preset adhesion particle detection algorithm to correct the preliminary particle size distribution result to obtain corrected particle size distribution data;

[0009] Obtain the data on the change of the rock fragmentation mechanism during the shield mode switching period. Combine it with the corrected particle size distribution data, and use a dynamic adaptive algorithm to update the particle size distribution in real time to obtain the dynamic particle size distribution result;

[0010] According to the non-linear mutation characteristics of the interfaces between different geological units in the complex stratum, use a mutation detection algorithm to analyze the dynamic particle size distribution result, judge whether there is a non-linear mutation, and if so, perform data smoothing processing to obtain the final particle size distribution evaluation result;

[0011] Through a preset error accumulation model, conduct error analysis on the final particle size distribution evaluation result, judge whether the error is within the preset threshold, and if it exceeds the threshold, reprocess the data until the error meets the requirements.

[0012] Preferably, the process of obtaining the rock debris image data collected during the shield construction, preprocessing to remove noise interference, and obtaining the first image includes:

[0013] Collect the rock debris images during the shield construction, and use image processing technology to remove the noise interference in the images;

[0014] Smooth the rock debris images through a filtering algorithm, and use an edge detection algorithm to extract the contour information of the rock debris images;

[0015] According to the preset clarity threshold, judge whether the image reaches the clear standard;

[0016] If the image does not reach the clear standard, adjust the filtering parameters and re-perform the noise removal process;

[0017] If the image reaches the clear standard, save it as the first image.

[0018] Preferably, the process of using an image segmentation algorithm based on deep learning to identify the edges of the rock debris in the first image, judge whether the rock debris overlaps, and if it overlaps, perform separation processing to obtain the second image includes:

[0019] Use a convolutional neural network to extract features from the first image. According to the extracted features, use a semantic segmentation algorithm to classify the image pixels and divide the rock debris area and the background area;

[0020] Judge whether there is pixel overlap in the rock debris area. If there is overlap, use a morphological processing method for area separation;

[0021] Smooth the separated rock debris area through a post-processing algorithm;

[0022] According to the preset accuracy threshold, judge whether the segmentation result meets the requirements. If it does not meet the requirements, adjust the model parameters and re-perform the segmentation;

[0023] If the segmentation result meets the requirements, a second image is generated, and the information of the segmented muck area is saved;

[0024] The muck area in the second image is refined using the region growing algorithm to obtain the final muck contour information.

[0025] Preferably, the process of analyzing the shape of the muck in the second image through preset morphological parameters to determine whether it is a non-spherical particle, and if it is non-spherical, using three-dimensional reconstruction technology for shape correction to obtain a third image includes:

[0026] Feature extraction is performed on the muck area using morphological parameters to determine the shape characteristic values of the muck;

[0027] Based on the shape characteristic values, the sphericity of the muck is judged. If the sphericity is lower than the preset threshold, it is marked as a non-spherical particle;

[0028] For non-spherical particles, three-dimensional reconstruction technology is used to correct the spatial shape of the muck to generate three-dimensional model data;

[0029] Through the three-dimensional model data, the corrected muck contour information is extracted to determine the geometric characteristics of the muck;

[0030] Based on the geometric characteristics, morphological optimization processing is performed on the muck, and a smoothing algorithm is used for post-processing of the optimized muck shape to obtain the initial data of the third image;

[0031] According to the preset shape accuracy threshold, it is judged whether the third image meets the requirements. If it meets, the corrected muck information is saved.

[0032] Preferably, the process of performing data fusion based on the muck shape characteristics in the third image and the particle size data obtained by a laser particle size analyzer to obtain a preliminary particle size distribution result includes:

[0033] Geometric feature extraction is performed on the muck area in the third image using morphological parameters to determine the spatial shape information of the muck;

[0034] Particle size data is obtained by a laser particle size analyzer. Combining the spatial shape information of the muck, a preliminary particle size distribution result is generated using a data fusion algorithm;

[0035] If there are outliers in the particle size distribution result, a smoothing algorithm is used to optimize the distribution curve to eliminate noise interference;

[0036] According to the preset accuracy threshold, it is judged whether the optimized particle size distribution meets the requirements. If it meets, the distribution result is saved;

[0037] For the saved particle size distribution results, a clustering algorithm is used to classify the particles to determine the proportion of rock debris in different particle size ranges;

[0038] Through the three-dimensional model data, the volume information of the rock debris is extracted, and combined with the particle size distribution results, a volume-particle size distribution curve of the rock debris is generated;

[0039] According to the volume-particle size distribution curve, a regression analysis algorithm is used to fit the distribution law to determine the characteristic parameters of the particle size distribution of the rock debris.

[0040] Preferably, the process of correcting the preliminary particle size distribution results by using a preset adhesion particle detection algorithm to obtain the corrected particle size distribution data includes:

[0041] Judge whether there are adhesion particles in the preliminary particle size distribution results. If there are adhesion particles, use the adhesion particle detection algorithm to identify them;

[0042] For the identified adhesion particles, correct the particle size distribution data through a removal algorithm;

[0043] According to the corrected particle size distribution data, judge whether it meets the preset threshold conditions;

[0044] If the threshold conditions are met, save the final particle size distribution data.

[0045] Preferably, the process of using a dynamic adaptive algorithm to update the particle size distribution in real time to obtain the dynamic particle size distribution results includes:

[0046] Obtain the data during the shield mode switching period and extract the change characteristics of the rock debris crushing mechanism;

[0047] Analyze the correlation between the change characteristics of the rock debris crushing mechanism and the corrected particle size distribution data;

[0048] Use the dynamic adaptive algorithm, combined with the corrected particle size distribution data, to update the particle size distribution in real time;

[0049] Judge whether the particle size distribution updated by the dynamic adaptive algorithm meets the preset accuracy conditions;

[0050] If the accuracy conditions are met, generate and store the dynamic particle size distribution results;

[0051] If the accuracy conditions are not met, adjust the parameters of the dynamic adaptive algorithm and update the particle size distribution again;

[0052] According to the finally generated dynamic particle size distribution results, output the change data of the rock debris crushing mechanism.

[0053] Preferably, the process of obtaining the final particle size distribution evaluation results includes:

[0054] Obtain the rock formation characteristic data of the interface between different geological units in complex strata, and extract the non-linear mutation characteristic parameters of the interface;

[0055] Use the mutation detection algorithm to analyze the dynamic particle size distribution result, and judge whether the non-linear mutation characteristic parameter exceeds the preset threshold;

[0056] If the non-linear mutation characteristic parameter exceeds the preset threshold, use the data smoothing algorithm to smooth the dynamic particle size distribution result;

[0057] Calculate the statistical characteristic values of the particle size distribution according to the smoothed dynamic particle size distribution result;

[0058] Use the clustering algorithm to classify the statistical characteristic values to determine the class label of the particle size distribution;

[0059] Match the class label with the preset particle size distribution evaluation criteria to generate the final particle size distribution evaluation result;

[0060] Store the final particle size distribution evaluation result and conduct correlation analysis in combination with the non-linear mutation characteristics of the geological unit interface.

[0061] The present invention also provides a rock debris particle size distribution evaluation system for double-mode shield construction in complex strata, including:

[0062] An image acquisition and preprocessing module, which is used to obtain the rock debris image data collected during the shield construction process, preprocess to remove noise interference, and obtain the first image;

[0063] A rock debris edge recognition and separation module, which is used to use an image segmentation algorithm based on deep learning to recognize the edge of the rock debris in the first image, judge whether the rock debris overlaps, and if it overlaps, perform separation processing to obtain the second image;

[0064] A morphological analysis and 3D reconstruction module, which is used to analyze the shape of the rock debris in the second image through preset morphological parameters, judge whether it is a non-spherical particle, and if it is non-spherical, use 3D reconstruction technology to correct the shape to obtain the third image;

[0065] A data fusion module, which is used to perform data fusion according to the rock debris shape characteristics in the third image and the particle size data obtained by a laser particle size analyzer to obtain a preliminary particle size distribution result;

[0066] An adhesion particle detection and correction module, which is used to correct the preliminary particle size distribution result by using a preset adhesion particle detection algorithm for the fine particle adhesion phenomenon to obtain the corrected particle size distribution data;

[0067] A dynamic particle size distribution update module, which is used to obtain the data on the change of the rock fragmentation mechanism during the shield mode switching period, combine the corrected particle size distribution data, and use a dynamic adaptive algorithm to update the particle size distribution in real time to obtain the dynamic particle size distribution result;

[0068] A mutation detection and data smoothing module, which is used to analyze the dynamic particle size distribution result by using a mutation detection algorithm according to the non-linear mutation characteristics of the interfaces between different geological units in complex strata, determine whether there is a non-linear mutation, and if so, perform data smoothing processing to obtain the final particle size distribution evaluation result;

[0069] An error analysis and correction module, which is used to analyze the error of the final particle size distribution evaluation result through a preset error accumulation model, determine whether the error is within the preset threshold, and if it exceeds the threshold, reprocess the data until the error meets the requirements.

[0070] Compared with the prior art, the present invention has the following advantages and technical effects:

[0071] The present invention obtains the rock debris image data, and through steps such as preprocessing, deep learning segmentation, shape analysis and three-dimensional reconstruction, combines the data of the laser particle size analyzer to obtain the preliminary particle size distribution result. Corrects for the fine particle adhesion phenomenon, and considering the change of the fragmentation mechanism during shield mode switching, uses a dynamic adaptive algorithm to update the particle size distribution in real time. Finally, according to the non-linear mutation characteristics of the interfaces between geological units in complex strata, performs mutation detection and data smoothing processing to obtain the final accurate particle size distribution evaluation result. The present invention effectively solves the accuracy problem of the evaluation of the rock debris particle size distribution in shield construction through multi-step data processing and analysis, and provides a reliable basis for engineering decision-making. Description of the Drawings

[0072] The drawings forming a part of this application are used to provide a further understanding of this application. The schematic embodiments of this application and their descriptions are used to explain this application and do not constitute an improper limitation to this application. In the drawings:

[0073] Figure 1 is a schematic flowchart of the method of the embodiment of the present invention;

[0074] Figure 2 is a schematic structural diagram of the system of the embodiment of the present invention. Detailed Embodiments

[0075] It should be noted that, without conflict, the embodiments in this application and the features in the embodiments can be combined with each other. The following will refer to the drawings and combine the embodiments to detail this application.

[0076] It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. And although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in a different order than here.

[0077] Embodiment 1

[0078] As Figure 1 shown, in this embodiment, a method for evaluating the particle size distribution of rock debris in double-mode shield tunneling in complex strata is provided, including:

[0079] Obtain the rock debris image data collected during the shield tunneling process, preprocess to remove noise interference, and obtain the first image;

[0080] Adopt an image segmentation algorithm based on deep learning to identify the edges of the rock debris in the first image, judge whether the rock debris overlaps, and if it overlaps, perform separation processing to obtain the second image;

[0081] Analyze the shape of the rock debris in the second image through preset morphological parameters, judge whether it is a non-spherical particle, and if it is non-spherical, use three-dimensional reconstruction technology for shape correction to obtain the third image;

[0082] According to the shape characteristics of the rock debris in the third image, combine with the particle size data obtained by the laser particle size analyzer, and perform data fusion to obtain the preliminary particle size distribution result;

[0083] Aiming at the fine particle adhesion phenomenon, adopt a preset adhesion particle detection algorithm to correct the preliminary particle size distribution result, remove the influence of the adhesion particles, and obtain the corrected particle size distribution data;

[0084] Obtain the change data of the rock debris crushing mechanism during the shield mode switching period, combine with the corrected particle size distribution data, and use a dynamic adaptive algorithm to update the particle size distribution in real time to obtain the dynamic particle size distribution result;

[0085] According to the non-linear mutation characteristics of the interfaces of different geological units in the complex strata, adopt a mutation detection algorithm to analyze the dynamic particle size distribution result, judge whether there is a non-linear mutation, and if there is, perform data smoothing processing to obtain the final particle size distribution evaluation result;

[0086] Through a preset error accumulation model, perform error analysis on the final particle size distribution evaluation result, judge whether the error is within the preset threshold, and if it exceeds the threshold, reprocess the data until the error meets the requirements.

[0087] Furthermore, the process of obtaining the rock debris image data collected during the shield tunneling process, preprocessing to remove noise interference, and obtaining the first image includes:

[0088] Collect the images of rock debris during the shield tunneling construction process, and use image processing technology to remove the noise interference in the images;

[0089] Perform smoothing processing on the rock debris images through a filtering algorithm to reduce the high-frequency noise in the images;

[0090] Adopt an edge detection algorithm to extract the contour information of the rock debris images and enhance the image details;

[0091] Judge whether the image meets the clarity standard according to the preset clarity threshold;

[0092] If the image does not meet the clarity standard, adjust the filtering parameters and re-perform the noise removal process;

[0093] If the image meets the clarity standard, save it as the first image for subsequent analysis.

[0094] Specifically, in this embodiment during the shield tunneling construction, a high-speed imaging device is used to collect the rock debris images, and a special waterproof and dustproof camera is installed at the tunneling head to collect the state of rock debris particles in real time. Such a collection environment is complex, and the images are easily affected by factors such as dust, vibration, and uneven light, resulting in noise interference. For image noise processing, this embodiment adopts methods such as Gaussian filtering or median filtering. For example, during the construction of a slurry shield, due to the image particle noise caused by slurry mixing, using a Gaussian filter kernel with a radius of five pixels can effectively reduce the high-frequency noise while maintaining the edge characteristics of the rock debris. In the construction of an earth pressure balance shield, for the salt-and-pepper noise caused by mechanical vibration, median filtering with a three-by-three window can be used to eliminate it.

[0095] In terms of edge detection, using the Sobel operator can better extract the contours of rock debris particles. For example, during the construction in a sandy cobble stratum, the shapes of the rock debris are irregular, and using the double-threshold Sobel edge detection can highlight the particle contours, which is beneficial for subsequent size analysis. In the soft soil stratum, the edges of the rock debris are blurred, and the Laplacian operator can be combined to enhance the edge features. The image clarity judgment can be based on the gradient value statistics. Taking the shield tunneling construction in a certain sandstone stratum as an example, set the average gradient threshold of the image to eighty. When the average gradient value of the processed image is lower than the threshold, it is necessary to increase the filtering intensity and reprocess. If it still does not meet the standard after three consecutive adjustments, it may be due to problems with the collection equipment, and it is necessary to check the camera.

[0096] Furthermore, the process of using an image segmentation algorithm based on deep learning to identify the edges of the rock debris in the first image, judge whether the rock debris overlaps, and perform separation processing if there is overlap to obtain the second image includes:

[0097] Use a convolutional neural network to extract features from the first image. According to the extracted features, use a semantic segmentation algorithm to classify the image pixels and divide the rock debris area and the background area;

[0098] Determine whether there are pixel overlaps in the muck area. If there are overlaps, use morphological processing methods to separate the areas;

[0099] Perform smoothing processing on the separated muck area through a post-processing algorithm to eliminate the noise and burrs generated during the segmentation process;

[0100] According to the preset precision threshold, determine whether the segmentation result meets the requirements. If it does not meet the requirements, adjust the model parameters and perform segmentation again;

[0101] If the segmentation result meets the requirements, generate a second image and save the information of the segmented muck area;

[0102] Use the region growing algorithm to refine the muck area in the second image to obtain the final muck contour information.

[0103] Specifically, in this embodiment, image features are extracted through multi-layer convolution and pooling operations of a convolutional neural network. Taking the muck image as an example, the first-layer convolution kernels are set to eight, with a size of three by three, for extracting low-level features such as edges and textures. The deep network extracts high-level features such as the shape and size of the muck, and the number of feature maps can reach one hundred and twenty-eight, effectively identifying the muck area features.

[0104] Semantic segmentation adopts a fully convolutional network architecture to classify each pixel. The image resolution is restored through upsampling to achieve pixel-level muck area division. The background class and the muck class can be set, and the output probability map shows the confidence of each pixel belonging to each class. The confidence threshold is set to 0.8, and the pixels with a confidence higher than the threshold are classified as the muck area.

[0105] For the overlapping area after segmentation, morphological processing uses opening and closing operations. First, erode with a circular structuring element with a radius of five pixels to remove small connections. Then perform dilation operations to fill the holes and achieve area separation. By adjusting the size of the structuring element, the separation degree can be controlled.

[0106] Post - processing adopts a combination of median filtering and Gaussian filtering. The median filtering window is set to five by five to remove discrete noise points. The standard deviation of Gaussian filtering is taken as two to smooth the region boundary. Through repeated iteration, the region contour becomes smoother and more natural. The accuracy threshold can be evaluated from the change in contour curvature. When the curvature change rate is less than 5%, it is considered to meet the requirements. Region growing starts from the center point of the segmented region and expands according to pixel similarity. The similarity threshold is set such that the pixel difference is less than twenty and the gradient magnitude is less than ten. Through the eight - neighborhood expansion method, a complete rock debris contour is gradually grown. If over - growth occurs, it can be restricted by limiting the maximum growth area. In practical applications, taking a certain engineering site as an example, the original image contains a large number of rock debris fragments. After feature extraction by the convolutional network, rock debris regions of different sizes and shapes can be identified. Semantic segmentation divides the image into distinct rock debris regions and the background. Through morphological processing, the adhered rock debris blocks are separated. Post - processing smooths the region boundary, and finally a clear rock debris contour is obtained. These processing results can be used for subsequent analysis of rock debris particle size and statistical analysis of shape characteristics, providing data support for engineering construction.

[0107] Further, the process of performing shape analysis on the rock debris in the second image through preset morphological parameters to determine whether it is a non - spherical particle and, if non - spherical, using three - dimensional reconstruction technology for shape correction to obtain the third image includes:

[0108] Extract the feature of the rock debris region using morphological parameters to determine the shape characteristic value of the rock debris;

[0109] According to the shape characteristic value, judge the sphericity of the rock debris. If the sphericity is lower than the preset threshold, it is marked as a non - spherical particle;

[0110] For non - spherical particles, use three - dimensional reconstruction technology to perform spatial shape correction on the rock debris to generate three - dimensional model data;

[0111] Through the three - dimensional model data, extract the corrected rock debris contour information to determine the geometric characteristics of the rock debris;

[0112] According to the geometric characteristics, perform morphological optimization processing on the rock debris to eliminate shape errors;

[0113] Use a smoothing algorithm to post - process the optimized rock debris shape to obtain the initial data of the third image;

[0114] According to the preset shape accuracy threshold, judge whether the third image meets the requirements. If it meets, save the corrected rock debris information.

[0115] Specifically, the extraction of morphological parameters is the basis for analyzing the characteristics of rock debris, usually including perimeter, area, aspect ratio of major and minor axes, etc. Taking an actual scenario as an example, when obtaining an image of a piece of rock debris, its perimeter value can be obtained as 300 pixels, area as 5000 square pixels, and aspect ratio of major and minor axes as 2.5 through edge detection. These parameters directly reflect the shape characteristics of the rock debris.

[0116] Sphericity judgment is an important indicator for evaluating the regularity of rock debris. Assuming that the sphericity threshold is set to 0.85 and the actual sphericity of a certain piece of rock debris is 0.72, then shape correction is required. Sphericity calculation is based on the relationship between the projected area and perimeter of the rock debris. A lower sphericity often indicates the presence of edges or irregular protrusions on the rock debris.

[0117] 3D reconstruction technology constructs a three-dimensional model of the rock debris through multi-view images. In practice, this embodiment adopts a multi-camera arrangement scheme to collect rock debris images from different angles. For example, three cameras are arranged vertically with an angle of 120 degrees between them to ensure complete surface information collection. Through image registration and depth estimation, point cloud data is obtained, and then a three-dimensional model is reconstructed.

[0118] The extraction of geometric features needs to consider three-dimensional parameters such as the volume and surface area of the rock debris. Taking a certain piece of rock debris as an example, its volume is 8000 cubic millimeters, surface area is 2000 square millimeters, and surface-to-volume ratio is 0.25. These parameters can comprehensively reflect the spatial characteristics of the rock debris and provide a basis for subsequent shape optimization.

[0119] Morphological optimization processing mainly targets irregular protrusions and depressions. In practical applications, a smoothness threshold is set, and local correction is performed on the areas exceeding the threshold. For example, there is an acute angle on the surface of a certain rock debris with an angle of 35 degrees, and its transition angle is adjusted to 60 degrees through surface fitting to make the shape more regular. The application of the smoothing algorithm needs to balance shape correction and detail retention. The smoothing window size is set to 15 pixels, and the corrected contour is iteratively processed. The local curvature is calculated for each iteration. When the curvature change is less than the preset threshold, it is considered that the ideal smoothing effect is achieved. Shape accuracy threshold judgment is the key to ensuring the correction quality. Assuming that the area change before and after correction is required not to exceed 5%, and the perimeter change not to exceed 8%. For a certain piece of rock debris, the area change after correction is 3.2%, and the perimeter change is 6.5%, meeting the accuracy requirements, and the corrected result can be saved. The preservation of rock debris information needs to record complete shape parameters and the correction process. It includes original feature values, correction parameters, three-dimensional model data, etc. These information is of great value for subsequent analysis and quality control, can trace the entire processing process, and is convenient for optimization and improvement.

[0120] Furthermore, according to the shape characteristics of the rock debris in the third image, combined with the particle size data obtained by the laser particle size analyzer, the process of data fusion to obtain the preliminary particle size distribution result includes:

[0121] Morphological parameters are used to extract the geometric features of the muck area in the third image, and the spatial shape information of the muck is determined;

[0122] Particle size data are obtained by a laser particle size analyzer. Combining with the spatial shape information of the muck, a preliminary particle size distribution result is generated by using a data fusion algorithm;

[0123] If there are outliers in the particle size distribution result, a smoothing algorithm is used to optimize the distribution curve to eliminate noise interference;

[0124] According to a preset accuracy threshold, it is judged whether the optimized particle size distribution meets the requirements. If it meets, the distribution result is saved;

[0125] For the saved particle size distribution result, a clustering algorithm is used to classify the particles to determine the proportion of muck in different particle size ranges;

[0126] Through the three-dimensional model data, the volume information of the muck is extracted. Combining with the particle size distribution result, a volume-particle size distribution curve of the muck is generated;

[0127] According to the volume-particle size distribution curve, a regression analysis algorithm is used to fit the distribution law to determine the characteristic parameters of the particle size distribution of the muck.

[0128] Specifically, morphological parameters are important indicators reflecting the geometric features of particles, mainly including roundness, elongation, wedge degree, etc. Through the analysis of morphological parameters, the spatial shape characteristics of the muck can be quantitatively described. For example, for an irregular muck, its roundness is 0.65, elongation is 1.8, and wedge degree is 0.4, which indicates that the muck presents an irregular flat and long shape.

[0129] The laser particle size analyzer measures the particle size through the principle of laser diffraction. Taking a group of muck samples as an example, the size data obtained by laser analysis show that the particle size range is 5 - 50 mm, and the median particle size is 25 mm. Combining these data with the morphological parameters and using the weighted average method for data fusion, a more accurate particle size distribution is obtained. In the particle size distribution result, outliers may occur. Assuming that there is an abnormal peak at a particle size of 40 mm, by using the Gaussian smoothing algorithm and setting the window width to 5 data points, such noise can be effectively eliminated. The smoothed distribution curve is smoother and more continuous, and can better reflect the true particle size distribution characteristics. Cluster analysis can divide the muck into different particle size grades. Using the K-means clustering method, the muck is divided into three categories: fine crushed material (5 - 15 mm), medium crushed material (15 - 30 mm), and coarse crushed material (30 - 50 mm). Statistics show that: the proportion of fine crushed material is 20%, the proportion of medium crushed material is 50%, and the proportion of coarse crushed material is 30%.

[0130] 3D reconstruction technology can obtain the volume information of rock debris. Through multi-view scanning and reconstruction, the volume of a piece of medium-crushed rock debris is 125 cubic centimeters. Combining the volume data of all rock debris with the particle size distribution, a volume-particle size distribution curve is plotted. The curve shows that as the particle size increases, the volume of a single particle shows an exponential growth trend. Using exponential regression analysis to fit the distribution law, the relationship expression between volume and particle size is obtained. Through this expression, the volume of rock debris with different particle sizes can be predicted, providing a basis for subsequent engineering applications. For example, the predicted volume of rock debris with a particle size of 40 mm is about 250 cubic centimeters, and the error from the measured value is within 5%. The volume-particle size distribution characteristics reflect the grading situation of rock debris. A good grading curve should present a continuous and smooth "S" shape, indicating that the rock debris of each particle size is evenly distributed. This distribution is beneficial to improving the stability and bearing capacity of the roadbed, reducing the void ratio, and extending the service life of the roadbed. By analyzing characteristic parameters such as the uniformity coefficient and curvature coefficient, the grading quality can be evaluated to guide the optimization of batching.

[0131] Further, the process of correcting the preliminary particle size distribution result by using a preset adhesion particle detection algorithm to obtain the corrected particle size distribution data includes:

[0132] Judge whether there are adhesion particles in the preliminary particle size distribution result. If there are adhesion particles, use the adhesion particle detection algorithm to identify them;

[0133] For the identified adhesion particles, correct the particle size distribution data through a removal algorithm;

[0134] According to the corrected particle size distribution data, judge whether it meets the preset threshold conditions;

[0135] If the threshold conditions are met, save the final particle size distribution data.

[0136] Specifically, for the problem of adhesion particles, the detection algorithm based on the contour concavity and convexity features can identify groups of adhered particles. For example, when it is detected that the contour curvature changes abnormally or the area ratio deviates from the normal range, it can be determined as an adhesion particle. In the correction stage, the watershed algorithm can effectively separate the adhesion particles. For example, for two closely connected rock debris particles, they are segmented by finding the position of the minimum cross-section of the contour, so as to obtain independent particle contours. When judging the threshold conditions, the allowable ranges of statistical parameters such as the coefficient of variation and skewness of the particle size distribution can be set. For example, if the coefficient of variation of the particle size distribution of a certain batch of rock debris samples exceeds the preset value of 0.3, or the absolute value of the skewness is greater than 1.5, it is considered that the distribution is abnormal and re-sampling or adjusting the processing parameters is required. This multiple threshold judgment mechanism can ensure that the finally obtained particle size distribution data has good statistical characteristics and practical reference value. Through these strict data processing and quality control processes, the accuracy and reliability of rock debris particle size analysis can be significantly improved.

[0137] Furthermore, the process of using a dynamic adaptive algorithm to update the particle size distribution in real time to obtain the dynamic particle size distribution result includes:

[0138] Obtain the data during the shield mode switching period and extract the change characteristics of the rock fragmentation mechanism;

[0139] Analyze the correlation between the change characteristics of the rock fragmentation mechanism and the corrected particle size distribution data;

[0140] Use a dynamic adaptive algorithm, combined with the corrected particle size distribution data, to update the particle size distribution in real time;

[0141] Judge whether the particle size distribution updated by the dynamic adaptive algorithm meets the preset accuracy conditions;

[0142] If the accuracy conditions are met, generate and store the dynamic particle size distribution result;

[0143] If the accuracy conditions are not met, adjust the parameters of the dynamic adaptive algorithm and re-update the particle size distribution;

[0144] According to the finally generated dynamic particle size distribution result, output the change data of the rock fragmentation mechanism.

[0145] Specifically, in the construction of shield tunnels, mode switching will significantly affect the rock fragmentation mechanism. The data during the switching period includes the change records of the tunneling parameters of the shield machine, cutter head rotation speed, and propulsion speed. For example, when the shield machine switches from the excavation mode to the slurry addition mode, the cutter head rotation speed drops from three revolutions per minute to two revolutions per minute, and the propulsion speed drops from two meters per hour to one point five meters per hour. These changes directly affect the fragmentation effect of the rock debris. The change characteristics of the rock fragmentation mechanism are mainly manifested in the morphology and size distribution of the fragmented particles. Under different tunneling modes, due to the different ways of the tool interacting with the rock, different fragmentation characteristics will be produced. For example, in the hard rock mode, the rock debris is angular and the particle size distribution is relatively uniform, and most particle sizes are between ten and twenty millimeters. In the soft rock mode, the rock debris is irregular and the particle size distribution range is wider, ranging from five millimeters to thirty millimeters.

[0146] For the correlation analysis between the fragmentation mechanism change and the particle size distribution, it can be achieved by establishing a correlation model. For example, in a certain tunnel project, when the cutter head rotation speed decreases by 10%, the average particle size increases by 15%, and at the same time, the proportion of particles larger than 25 millimeters increases by 20%. This kind of correlation data can be used to predict the fragmentation effect under different working conditions.

[0147] The dynamic adaptive algorithm continuously optimizes the particle size distribution calculation model through real-time data feedback. This algorithm dynamically adjusts the calculation parameters according to historical data and current working conditions. For example, in the muddy water shield mode, the algorithm automatically adjusts the weight coefficients of the particle size distribution according to the change of the muddy water pressure. When the muddy water pressure rises, the weight of the proportion of fine particles is increased; when the pressure drops, the weight of large particles is appropriately increased.

[0148] The preset accuracy conditions usually include indicators such as the fitting degree of the particle size distribution curve and the prediction error rate. For example, it is required that the determination coefficient of the particle size distribution curve is greater than 0.95, and the average relative error between the predicted value and the measured value is less than 8%. If these conditions are not met, the algorithm parameters need to be adjusted, such as increasing the sampling frequency or modifying the weight calculation method. The finally output data on the change mechanism of rock fragmentation not only includes the dynamic changes of the particle size distribution, but also comprehensive indicators such as fragmentation efficiency and energy consumption. These data can be used to guide the optimization of construction parameters. For example, in a certain project, it is found through analysis that when the cutter head rotation speed is controlled at 2.5 revolutions per minute, both the fragmentation effect can be ensured and the energy consumption can be reduced.

[0149] Furthermore, the process of obtaining the final evaluation result of the particle size distribution includes:

[0150] Obtain the rock layer characteristic data of the interfaces between different geological units in the complex stratum, and extract the non-linear mutation characteristic parameters of the interfaces;

[0151] Use the mutation detection algorithm to analyze the dynamic particle size distribution result, and judge whether the non-linear mutation characteristic parameters exceed the preset threshold;

[0152] If the non-linear mutation characteristic parameters exceed the preset threshold, use the data smoothing algorithm to smooth the dynamic particle size distribution result;

[0153] According to the smoothed dynamic particle size distribution result, calculate the statistical characteristic values of the particle size distribution;

[0154] Use the clustering algorithm to classify the statistical characteristic values and determine the class labels of the particle size distribution;

[0155] Match the class labels with the preset particle size distribution evaluation criteria to generate the final particle size distribution evaluation result;

[0156] Store the final particle size distribution evaluation result and conduct correlation analysis in combination with the non-linear mutation characteristics of the geological unit interfaces.

[0157] Specifically, in complex strata, the interfaces between different geological units often exhibit significant nonlinear mutation characteristics. For example, the interface between sandstone and mudstone may exhibit lithologic abrupt changes and differences in the degree of fracture development. These characteristics are captured through field exploration and drilling data, including parameters such as the integrity coefficient of the rock formation, joint spacing, and degree of weathering. For example, in a subway tunnel construction section, at the interface between sandstone and mudstone, the integrity coefficient suddenly changes from 0.85 to 0.4, and the joint spacing decreases from 50 cm to 5 cm, indicating significant nonlinear mutation at the interface.

[0158] Dynamic particle size distribution results are analyzed using mutation detection algorithms, such as wavelet analysis. By setting appropriate thresholds, for example, when the particle size variation between adjacent segments exceeds 30%, a mutation point is identified. In a real-world application, during tunnel excavation, after the shield machine switched modes, the rock slag particle size distribution experienced dramatic fluctuations, with the maximum particle size jumping from 80 mm to 150 mm, triggering a mutation detection alert.

[0159] Data smoothing uses a moving average or exponential smoothing method to eliminate abnormal fluctuations. For example, after applying a five-point moving average to a certain segment of sudden change data, the particle size distribution curve becomes flat, and the maximum particle size fluctuation range converges to the range of 90 to 110 mm. Statistical characteristic value calculations include average particle size, standard deviation, skewness coefficient, and other parameters. These parameters comprehensively reflect the characteristics of rock slag crushing. Cluster analysis uses fuzzy clustering methods to classify particle size distribution characteristics. For example, in a certain engineering practice, based on the two characteristic parameters of average particle size and standard deviation, rock slag samples were divided into three categories: fine crushing, medium crushing, and coarse crushing. The fine crushing type has an average particle size of less than 30 mm and a standard deviation of less than 10 mm; the medium crushing type has an average particle size between 30 and 70 mm and a standard deviation between 10 and 20 mm; and the coarse crushing type has an average particle size greater than 70 mm and a standard deviation greater than 20 mm.

[0160] The particle size distribution evaluation standard is usually formulated in combination with the project needs, such as the requirements of the conveying system and the subsequent processing technology. The evaluation indicators include grade matching rationality, particle size uniformity, etc. In a certain project, the content of fine aggregate is required to be no more than 20% and the maximum particle size is no more than 120 mm. Based on this, an evaluation system is established to divide the particle size distribution quality into three levels: excellent, qualified, and unqualified. Through correlation analysis, it is found that at the geological interface, the particle size distribution of rock slag often changes significantly. For example, in a transition zone from hard rock to soft rock, as the rock strength decreases, the content of fine aggregate increases significantly and the average particle size decreases. This change trend is highly consistent with the nonlinear mutation characteristics of the interface, providing an important basis for the optimization of construction parameters.

[0161] In view of the problem of evaluating the particle size distribution of rock debris in the construction of a dual-mode shield tunneling machine in complex strata, this embodiment proposes a comprehensive solution. Through multi-step data processing and analysis, the accuracy and reliability of the evaluation of the particle size distribution of rock debris are significantly improved. The specific technical effects are as follows:

[0162] 1. Multi-modal data fusion and generation of high-precision particle size distribution;

[0163] This embodiment combines image processing technology and data from a laser particle size analyzer. Through a deep learning algorithm, the rock debris images are segmented and recognized, and a three-dimensional reconstruction technology is used to correct the shapes of non-spherical particles. This multi-modal data fusion method can effectively solve the problem of insufficient accuracy of traditional single measurement methods (such as laser particle size analysis or image recognition) when dealing with complex rock debris shapes. For example, laser particle size analysis has a large measurement error for non-spherical particles, and image recognition is prone to misjudgment when dealing with overlapping rock debris. Through data fusion and three-dimensional reconstruction in this embodiment, more accurate particle size distribution results are generated, providing a reliable basis for subsequent engineering decisions.

[0164] 2. Dynamic adaptive update and real-time monitoring of particle size distribution;

[0165] During the shield tunneling construction process, the crushing mechanism of rock debris will change due to the switching of the shield mode, resulting in dynamic changes in the particle size distribution. This embodiment obtains data during the shield mode switching period and uses a dynamic adaptive algorithm to update the particle size distribution in real time. This method can quickly respond to changes in construction conditions, ensuring the timeliness and accuracy of the particle size distribution data. For example, at the interface between hard rock and soft rock, the crushing characteristics of rock debris will change significantly, and the dynamic adaptive algorithm can adjust the particle size distribution model in real time to avoid measurement lag caused by changes in the construction mode.

[0166] 3. Nonlinear mutation detection and data smoothing processing;

[0167] In complex strata, the rock debris crushing mode at the interface between different geological units has significant non-linear characteristics, and it is difficult for traditional methods to accurately capture such mutational changes. This embodiment analyzes the dynamic particle size distribution results through a mutation detection algorithm to determine whether there are non-linear mutations, and uses data smoothing processing technology to optimize the mutant data. This method can effectively reduce measurement errors caused by geological mutations and ensure the continuity and stability of the particle size distribution curve. For example, at the interface between sandstone and mudstone, the particle size distribution of rock debris may fluctuate violently. Through mutation detection and smoothing processing, more realistic particle size distribution evaluation results can be generated.

[0168] 4. Adhesive particle correction and improvement of the ability to handle fine particles;

[0169] Fine particle adhesion is a significant factor affecting particle size distribution accuracy. This embodiment uses a pre-defined adhesion particle detection algorithm to correct the initial particle size distribution results, removing the influence of adhered particles. This method effectively addresses the error accumulation problem associated with traditional methods for processing fine particles, improving the accuracy of particle size distribution assessments. For example, in silt sand formations, fine particles tend to clump together, resulting in an overly large particle size distribution. Adhesion particle correction accurately separates fine particles, generating more realistic particle size distribution data.

[0170] 5. Error accumulation model and quality control;

[0171] This embodiment uses a preset error accumulation model to perform error analysis on the final particle size distribution evaluation results to determine whether the error is within a preset threshold. If the threshold is exceeded, data processing is performed again until the error meets the requirements. This quality control mechanism can effectively ensure the reliability and consistency of the particle size distribution evaluation results and avoid misjudgments caused by measurement errors or improper data processing. For example, during continuous construction, through real-time monitoring of the error accumulation model, deviations in data processing can be discovered and corrected in a timely manner, ensuring that the particle size distribution evaluation results are always at a high level of precision.

[0172] 6. System integration and automation processing capabilities;

[0173] This example not only proposes a method but also provides a complete system solution, encompassing multiple modules, including image acquisition and preprocessing, rock slag edge identification and separation, morphological analysis and 3D reconstruction, data fusion, dynamic updating, and mutation detection. This integrated system approach automates the evaluation of rock slag particle size distribution, reducing manual intervention and improving work efficiency. For example, through the automated system, construction personnel can obtain particle size distribution data in real time and adjust shield construction parameters based on the evaluation results, optimizing the construction process.

[0174] In summary, this embodiment comprehensively solves the accuracy and reliability issues of rock slag particle size distribution evaluation during dual-mode shield construction in complex strata through technical means such as multimodal data fusion, dynamic adaptive updating, nonlinear mutation detection, adhesion particle correction, and error accumulation model, providing strong support for engineering decision-making in shield construction.

[0175] Example 2

[0176] like Figure 2 As shown, based on the same inventive concept, this embodiment also provides a system for evaluating rock slag particle size distribution in dual-mode shield construction in complex strata, comprising:

[0177] The image acquisition and preprocessing module is used to obtain rock slag image data collected during shield construction, preprocess and remove noise interference to obtain a first image;

[0178] The rock debris edge recognition and separation module is used to recognize the edges of rock debris in the first image by using an image segmentation algorithm based on deep learning, determine whether the rock debris overlaps, and if so, perform separation processing to obtain the second image;

[0179] The morphological analysis and three-dimensional reconstruction module is used to analyze the shape of the rock debris in the second image through preset morphological parameters, determine whether it is a non-spherical particle, and if it is non-spherical, use three-dimensional reconstruction technology to correct the shape to obtain the third image;

[0180] The data fusion module is used to perform data fusion based on the shape characteristics of the rock debris in the third image and the particle size data obtained by a laser particle size analyzer to obtain a preliminary particle size distribution result;

[0181] The adhered particle detection and correction module is used to correct the preliminary particle size distribution result by using a preset adhered particle detection algorithm for the phenomenon of fine particle adhesion to obtain the corrected particle size distribution data;

[0182] The dynamic particle size distribution update module is used to obtain the change data of the rock debris crushing mechanism during the shield mode switching period, combine it with the corrected particle size distribution data, and use a dynamic adaptive algorithm to update the particle size distribution in real time to obtain the dynamic particle size distribution result;

[0183] The mutation detection and data smoothing module is used to analyze the dynamic particle size distribution result by using a mutation detection algorithm according to the non-linear mutation characteristics of the interfaces between different geological units in complex strata, determine whether there is a non-linear mutation, and if so, perform data smoothing processing to obtain the final particle size distribution evaluation result;

[0184] The error analysis and correction module is used to analyze the error of the final particle size distribution evaluation result through a preset error accumulation model, determine whether the error is within the preset threshold, and if it exceeds the threshold, reprocess the data until the error meets the requirements.

[0185] A rock debris particle size distribution evaluation system for double-mode shield tunneling in complex strata provided in this embodiment has all the advantages of the rock debris particle size distribution evaluation method for double-mode shield tunneling in complex strata provided in Embodiment 1.

[0186] Embodiment 3

[0187] This embodiment also discloses a computer device, including a memory, a processor, and a computer program stored on the memory, and the processor executes the computer program to implement the steps of the method described in Embodiment 1.

[0188] Embodiment 4

[0189] This embodiment also discloses a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the method described in Embodiment 1 are implemented.

[0190] Embodiment 5

[0191] This embodiment also discloses a computer program product, including a computer program. When the computer program is executed by a processor, the steps of the method described in Embodiment 1 are implemented.

[0192] The above are only the preferred specific embodiments of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by those skilled in the art within the technical scope disclosed in the present application should be covered by the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. An evaluation method for the particle size distribution of rock debris in the construction of a dual-mode shield tunneling machine in complex strata, characterized in that, Including: Obtain the image data of the muck collected during the shield tunneling construction, preprocess to remove noise interference, and obtain the first image; Adopt an image segmentation algorithm based on deep learning to identify the edges of the muck in the first image, judge whether the muck overlaps, and if it overlaps, perform separation processing to obtain the second image; Perform shape analysis on the muck in the second image through preset morphological parameters to judge whether it is a non-spherical particle. If it is non-spherical, use three-dimensional reconstruction technology to correct the shape and obtain the third image; According to the shape characteristics of the muck in the third image, combine with the particle size data obtained by the laser particle size analyzer, and perform data fusion to obtain the preliminary particle size distribution result; For the phenomenon of fine particle adhesion, adopt a preset adhesion particle detection algorithm to correct the preliminary particle size distribution result to obtain the corrected particle size distribution data; Obtain the data on the change of the muck fragmentation mechanism during the shield mode switching period, combine with the corrected particle size distribution data, and use a dynamic adaptive algorithm to update the particle size distribution in real time to obtain the dynamic particle size distribution result; According to the non-linear mutation characteristics of the interfaces between different geological units in the complex stratum, adopt a mutation detection algorithm to analyze the dynamic particle size distribution result, judge whether there is a non-linear mutation, and if there is, perform data smoothing processing to obtain the final particle size distribution evaluation result; Through a preset error accumulation model, perform error analysis on the final particle size distribution evaluation result, judge whether the error is within the preset threshold, and if it exceeds the threshold, reprocess the data until the error meets the requirements.

2. The method according to claim 1, wherein: The process of obtaining the image data of the muck collected during the shield tunneling construction, preprocessing to remove noise interference, and obtaining the first image includes: Collect the muck images during the shield tunneling construction, and use image processing technology to remove the noise interference in the images; Perform smoothing processing on the muck images through a filtering algorithm, and use an edge detection algorithm to extract the contour information of the muck images; According to the preset clarity threshold, judge whether the image meets the clarity standard; If the image does not meet the clarity standard, adjust the filtering parameters and re-perform the noise removal processing; If the image meets the clarity standard, save it as the first image.

3. The method according to claim 1, wherein: The process of adopting an image segmentation algorithm based on deep learning to identify the edges of the muck in the first image, judge whether the muck overlaps, and if it overlaps, perform separation processing to obtain the second image includes: Adopt a convolutional neural network to extract features from the first image. According to the extracted features, use a semantic segmentation algorithm to classify the image pixels and divide the muck area and the background area; Judge whether there is pixel overlap in the muck area. If there is overlap, use a morphological processing method to separate the areas; Perform smoothing processing on the separated muck areas through a post-processing algorithm; According to the preset accuracy threshold, judge whether the segmentation result meets the requirements. If it does not meet the requirements, adjust the model parameters and re-perform the segmentation; If the segmentation result meets the requirements, generate the second image and save the information of the segmented muck areas. The regional growth algorithm is used to refine the rock debris area in the second image to obtain the final rock debris contour information.

4. The method according to claim 1, wherein the process of analyzing the shape of the rock debris in the second image by preset morphological parameters to determine whether it is a non-spherical particle, and if it is non-spherical, using three-dimensional reconstruction technology for shape correction to obtain the third image includes: Using morphological parameters to extract features of the rock debris area to determine the shape characteristic values of the rock debris; According to the shape characteristic values, judging the sphericity of the rock debris. If the sphericity is lower than the preset threshold, it is marked as a non-spherical particle; For non-spherical particles, using three-dimensional reconstruction technology to correct the spatial shape of the rock debris to generate three-dimensional model data; Through the three-dimensional model data, extracting the corrected rock debris contour information to determine the geometric characteristics of the rock debris; According to the geometric characteristics, performing morphological optimization processing on the rock debris, and using a smoothing algorithm to post-process the optimized rock debris shape to obtain the initial data of the third image; According to the preset shape accuracy threshold, judging whether the third image meets the requirements. If it meets the requirements, the corrected rock debris information is saved.

5. The method according to claim 1, wherein the process of performing data fusion on the shape characteristic of the rock debris in the third image and the particle size data obtained by a laser particle size analyzer to obtain a preliminary particle size distribution result includes: Using morphological parameters to extract the geometric characteristics of the rock debris area in the third image to determine the spatial shape information of the rock debris; Obtaining particle size data by a laser particle size analyzer, combining the spatial shape information of the rock debris, and using a data fusion algorithm to generate a preliminary particle size distribution result; If there are outliers in the particle size distribution result, using a smoothing algorithm to optimize the distribution curve to eliminate noise interference; According to the preset accuracy threshold, judging whether the optimized particle size distribution meets the requirements. If it meets the requirements, the distribution result is saved; For the saved particle size distribution result, using a clustering algorithm to classify the particles to determine the proportion of rock debris in different particle size ranges; Through the three-dimensional model data, extracting the volume information of the rock debris, and combining the particle size distribution result to generate a volume-particle size distribution curve of the rock debris; According to the volume-particle size distribution curve, using a regression analysis algorithm to fit the distribution law to determine the particle size distribution characteristic parameters of the rock debris. ​ ​ ​ ​ ​ ​ ​ ​ Obtain data during shield mode switching to extract the changing characteristics of rock fragmentation mechanism; Analyze the correlation between the changing characteristics of the rock fragmentation mechanism and the corrected particle size distribution data; Adopt dynamic adaptive algorithm and combine with corrected particle size distribution data to update particle size distribution in real time; Determine whether the particle size distribution after the dynamic adaptive algorithm updates meets the preset accuracy conditions; If the accuracy conditions are met, the dynamic particle size distribution results are generated and stored; If the accuracy conditions are not met, adjust the dynamic adaptive algorithm parameters and re-update the particle size distribution; According to the final dynamic particle size distribution results, the rock slag crushing mechanism change data is output.

8. The method according to claim 1, characterized in that The process of obtaining the final particle size distribution evaluation results includes: Obtain rock formation characteristic data at the interface of different geological units in complex strata and extract nonlinear mutation characteristic parameters of the interface; The dynamic particle size distribution results are analyzed using a mutation detection algorithm to determine whether the nonlinear mutation characteristic parameters exceed the preset threshold; If the nonlinear mutation characteristic parameter exceeds a preset threshold, a data smoothing algorithm is used to smooth the dynamic particle size distribution result; Calculate the statistical characteristic value of the particle size distribution according to the dynamic particle size distribution result after smoothing; Using a clustering algorithm to classify the statistical characteristic values to determine a category label of the particle size distribution; Matching the category label with a preset particle size distribution evaluation standard to generate a final particle size distribution evaluation result; The final particle size distribution evaluation results are stored and correlated with the nonlinear mutation characteristics of the geological unit interface for analysis.

9. An evaluation system for the particle size distribution of rock debris in the construction of a dual-mode shield tunneling machine in complex strata, characterized in that, include: The image acquisition and preprocessing module is used to obtain rock slag image data collected during shield construction, preprocess and remove noise interference to obtain a first image; a rock slag edge recognition and separation module, configured to use a deep learning-based image segmentation algorithm to recognize rock slag edges in the first image, determine whether the rock slag overlaps, and if so, perform separation processing to obtain a second image; a morphological analysis and three-dimensional reconstruction module, configured to perform shape analysis on the rock ballast in the second image using preset morphological parameters to determine whether the rock ballast is non-spherical; if so, perform shape correction using three-dimensional reconstruction technology to obtain a third image; a data fusion module for fusing data based on the rock ballast shape features in the third image and the particle size data obtained by the laser particle size analyzer to obtain a preliminary particle size distribution result; Adhesion particle detection and correction module is used to correct the preliminary particle size distribution results based on the fine particle adhesion phenomenon using a preset adhesion particle detection algorithm to obtain the corrected particle size distribution data; The dynamic particle size distribution update module is used to obtain the change data of the rock fragmentation mechanism during the shield mode switching period, and combines the corrected particle size distribution data with a dynamic adaptive algorithm to update the particle size distribution in real time to obtain the dynamic particle size distribution results. Mutation detection and data smoothing module, which is used to analyze the dynamic particle size distribution results by using a mutation detection algorithm according to the non-linear mutation characteristics of the interfaces between different geological units in complex strata, determine whether there is a non-linear mutation, and if so, perform data smoothing processing to obtain the final particle size distribution evaluation result; Error analysis and correction module, which is used to analyze the error of the final particle size distribution evaluation result through a preset error accumulation model, determine whether the error is within a preset threshold, and if it exceeds the threshold, reprocess the data until the error meets the requirements.

Images