Coal rock fatigue deformation crack propagation intelligent prediction method

Abstract

The invention provides a coal rock fatigue deformation crack propagation intelligent prediction method. The method comprises the following steps of sample preparation, rock mechanics test, real-time CT scanning, construction of image recognition time series data set, three-dimensional reconstruction of CT image, three-dimensional time series feature extraction, coal rock pore crack time series regression model training and verification, device system packaging and coal rock strain formula self-adaptive correction. The method proposes a method for selecting effective identification image data according to the time importance coefficient, realizes accurate data screening according to the crack propagation and deformation characteristics of coal rock itself, and effectively eliminates redundant data. The method proposes a way of extracting coal rock picture time series features from three dimensions of deformation increment, deformation direction and coal rock type, and constructs a time series regression prediction model by fusing the space-time distribution characteristics of coal rock failure, which provides a reliable method and idea for coal rock fatigue instability deformation monitoring and theoretical research.

Description

Technical Field

[0001] This invention relates to the field of rock mechanics, and in particular to an intelligent prediction method for crack propagation in coal and rock fatigue deformation. Background Technology

[0002] In underground mining engineering, coal and rock masses are frequently subjected to cyclic disturbances such as mining activities and vibrations. Under cyclic loading, the coal and rock mass undergoes both plastic and elastic deformation. The elastic deformation recovers after the load is removed, while the plastic deformation remains. Under cyclic disturbance, the initiation, propagation, and penetration of cracks in the coal and rock mass, as well as the magnitude of irreversible plastic deformation, are directly related to the degree of damage to the coal and rock mass, reflecting its fatigue performance under cyclic loading. Therefore, studying the crack propagation and deformation damage characteristics of coal and rock under cyclic loading and unloading is of great significance.

[0003] Domestic and international scholars have conducted extensive research on the mechanical deformation characteristics and damage evolution of coal and rock under cyclic loading and unloading. Many researchers have conducted uniaxial and triaxial compression tests and analyzed their stress-strain curves. Studies have revealed that under cyclic loading, the loading and unloading paths of coal and rock do not completely coincide; each loading and unloading cycle forms a hysteresis loop, leaving a permanent deformation. Currently, research on the stress-strain "hysteresis loop" phenomenon and irreversible strain under cyclic loading and unloading stress paths is mostly indirect and qualitative. Meanwhile, technological advancements and the development of artificial intelligence algorithms have provided new ideas and methods for rock mechanics research. High-precision industrial CT and other technologies enable real-time non-destructive monitoring, providing new methods and approaches for studying internal crack propagation and deformation damage in coal and rock. Based on this, this invention utilizes CT scanning technology and intelligent algorithms to predict coal and rock fatigue deformation and crack propagation based on visualized time-series images, providing a reliable concept, method, and technology for intelligent research on coal and rock safety. Summary of the Invention

[0004] The purpose of this invention is to provide an intelligent prediction method for crack propagation in coal and rock fatigue deformation, so as to solve the problems existing in the prior art.

[0005] The technical solution adopted to achieve the purpose of this invention is as follows: a smart prediction method for coal and rock fatigue deformation and crack propagation, comprising the following steps:

[0006] 1) Prepare coal and rock samples and construct a spatial coordinate system for the samples.

[0007] 2) The coal and rock samples were subjected to cyclic loading and unloading mechanical tests until the samples were destroyed. During the cyclic loading and unloading process, the coal and rock samples were subjected to high-precision full-range real-time CT scans.

[0008] 3) Perform real-time CT image preprocessing on coal and rock samples to construct an image recognition time-series dataset. Then perform 3D reconstruction of the CT images. Step 3) specifically includes the following sub-steps, which may be performed simultaneously or in any order:

[0009] 3.1) CT images at specific times are selected as time-series datasets based on the time importance coefficient φ. The time importance coefficient φ reflects the time effect of coal and rock pore fractures and volume changes. The time-series datasets include time-series radial two-dimensional slice datasets and axial two-dimensional slice datasets. Primary fractures in coal and rock samples are used as speckle structure features to locate pore fractures in coal and rock. Individual pore fracture time-series datasets are established for each located pore fracture. These individual pore fracture location time-series datasets characterize the process of a single pore fracture in coal and rock from initiation, development, propagation, connection, and eventual failure.

[0010] 3.2) After filtering, denoising, and enhancing the CT data using digital core software, the three-dimensional volume of pores and fractures was reconstructed.

[0011] 4) The time-series dataset constructed in step 3.1) is divided into a training set and a validation set. Time-series features are extracted from the training set data from the dimensions of deformation increment, deformation direction, and coal and rock type, respectively. Then, the time-series features are substituted into different time-series algorithms and fused to construct a coal and rock fatigue fracture propagation regression model to predict the development and propagation of pore fractures and the deformation of coal and rock in real time. The validation set and the pore fracture 3D reconstruction model obtained in step 3.2) are used to improve and validate the coal and rock fatigue fracture propagation regression model.

[0012] 5) Based on the coal and rock fatigue fracture propagation regression model obtained in step 4), the deformation and fracture propagation characteristics of coal and rock under cyclic loading and unloading can be predicted and characterized in real time.

[0013] 6) Based on the results of step 4), the time-series strain regression parameter set K can be obtained. Combining the "hysteresis loop" characteristic of the stress-strain curve, the strain formula and constitutive equation are modified from three dimensions: strain time series, coal and rock type, and strain direction. Simultaneously, the model from step 4) is encapsulated to form a device and system that can be widely applied. Further, in step 1), the large coal samples selected from the working face are processed into cylindrical specimens through cutting, grinding, and polishing. Alternatively, coal and rock powder can be mixed with water and pressed into cylindrical specimens using a mold.

[0014] Further, in step 1), the sample is arranged in a spatial rectangular coordinate system O-XYZ. The origin of the spatial rectangular coordinate system O-XYZ is located at the center of the bottom surface of the sample. The Z-axis of the spatial rectangular coordinate system O-XYZ is parallel to the length direction of the sample, and the X-axis and Y-axis are parallel to the radial direction of the sample.

[0015] Furthermore, initial CT images and all The CT images at time 1 and the times before and after that time are used as the time series dataset. Time importance coefficients. The method for determining is shown in equation (1). When the rate of change is greater than the threshold, is given Assign 1 if the threshold is not specified, otherwise assign 0. The threshold is determined through expert experience or by image observation.

[0016] (1)

[0017] In the formula, Let t be the number of effective pores and fractures in the coal and rock at time t. Let be the volume of the positioning hole fracture in the coal and rock at time t. N is the minimum increment of the number of fractures considered to have significant damage and deformation corresponding to this type of coal and rock. This represents the volume change of the positioning hole fissure. Let t be the volume of the coal and rock at time t. This represents the volume change of coal and rock.

[0018] Furthermore, in step 4), the ratio of the training set to the validation set of the time series dataset is 7:3, and can be adjusted according to specific experiments.

[0019] Furthermore, in step 4), the deformation increment temporal feature extraction method is to... Time and Calculate the absolute deformation increment of the pore after cyclic loading and unloading, taking the maximum overlap of the same pore at any given time:

[0020] Under cyclic loading over a period of time, the pores changed from spherical to ellipsoidal. Region a represents the reduced area within the original pore range, defined as the deformation reduction. hour The area of ​​a plane is denoted as The volume of the pore or fissure where it is located is denoted as The region shown in b represents the increased area within the original pore size, defined as the deformation increment. hour The area of ​​a plane is denoted as The volume of the pore or fissure where it is located is denoted as Furthermore, let point E and point F be arbitrary points on the boundary of the studied void fracture, and... Let E be the distance between points E and F. and These are the distances before and after the cyclic load is applied, respectively.

[0021] For a two-dimensional image of a pore fracture, the two-dimensional deformation increment function is defined as follows:

[0022] (2)

[0023] (3)

[0024] (4)

[0025] (5)

[0026] when At that time, the pores and fissures did not produce irreversible linear strain and area strain; when When the deformation increment of the pores and cracks equals the deformation decrement, irreversible linear strain is generated, but no area strain is generated; when When this occurs, it indicates that the deformation reduction of the pores and cracks is greater than the deformation increment, resulting in linear strain and area strain, and the pores and cracks are... of Contraction occurred within the plane. When this occurs, it indicates that the deformation reduction of the pores and cracks is less than the deformation increment, resulting in linear strain and area strain, and the pores and cracks are... of Expansion occurred within the plane. (5) This represents the absolute deformation increment.

[0027] Furthermore, in step 4), the method for extracting the temporal features of the deformation direction is to use the original crack as a positioning reference point and statistically analyze... Time and The direction of crack propagation at any given time is used to extract the spatial distribution features of crack angles and directions on the sliced ​​image. Furthermore, values ​​can be assigned to each angle and direction when... If a crack grows in a certain direction during a certain period, a corresponding value is assigned to that direction for that period, thereby quantifying the crack direction.

[0028] Furthermore, in step 4), the method for extracting the temporal features of coal and rock types is to statistically analyze the CT values ​​of CT image slices at different times. The CT values ​​are closely related to the density of coal and rock, and the magnitude of the CT values ​​can reflect different coal and rock types.

[0029] Furthermore, in step 4), the temporal algorithm utilizes full-scale skip links.

[0030] Furthermore, in step 4), the temporal algorithm employs full-scale deep supervision.

[0031] Furthermore, in step 5), the deformation features include deformation increment features and deformation direction features.

[0032] Furthermore, in step 6), the regression parameter set K is the parameter set of the regression model. After mapping and transformation, different correction coefficients can be formed, including time-series correction coefficients that consider the propagation of coal and rock pores and fractures. Correction coefficients for coal and rock types considering different coal and rock materials. Orientation correction factor due to different material loading methods The coal and rock strain formula is adaptively modified as follows:

[0033] (6)

[0034] (7)

[0035] (8)

[0036] In the formula, It is an equivalent change. It is volumetric strain. , , These are the linear strains in the x, y, and z directions, respectively.

[0037] The present invention also discloses a computer-readable storage medium storing a computer program for implementing any of the above-described intelligent prediction methods for coal and rock fatigue deformation and crack propagation.

[0038] This invention also discloses an intelligent prediction device for coal and rock fatigue deformation and crack propagation, comprising a processor, an input unit, an output unit, and a storage unit. The input unit can receive CT image data from an external CT scanner. The output unit can reflect information about external manifestations. The storage unit is used to store digital core software, instruction sequences executed by the processor for implementing any of the aforementioned intelligent prediction methods for coal and rock fatigue deformation and crack propagation, and data during operation.

[0039] The technical effects of this invention are beyond doubt:

[0040] A. A method for selecting effective image data based on time importance coefficient was proposed, which enabled accurate data screening based on the crack propagation and deformation characteristics of coal and rock itself, effectively eliminating redundant data;

[0041] B. A method for extracting temporal features from coal and rock images from three dimensions: deformation increment, deformation direction, and coal and rock type is proposed. By integrating the spatiotemporal distribution characteristics of coal and rock failure, a temporal regression prediction model is constructed, providing a reliable method and approach for monitoring and theoretical research on coal and rock fatigue instability deformation. Attached Figure Description

[0042] Figure 1 Here is a flowchart of the intelligent prediction method;

[0043] Figure 2 This is a schematic diagram of the time-series characterization of pore and fracture increments. Detailed Implementation

[0044] The present invention will be further described below with reference to embodiments, but it should not be construed that the scope of the present invention is limited to the following embodiments. Various substitutions and modifications made based on ordinary technical knowledge and common practices in the art without departing from the above-described technical concept of the present invention should be included within the scope of protection of the present invention.

[0045] Example 1:

[0046] This embodiment uses laboratory methods based on CT image data to study the characteristics of irreversible plastic deformation and damage evolution of coal and rock. It proposes an intelligent prediction method for crack propagation in coal and rock fatigue deformation, including the following steps:

[0047] 1) Prepare coal and rock samples and construct a spatial coordinate system for the samples.

[0048] 2) The coal and rock samples were subjected to cyclic loading and unloading mechanical tests until the samples were destroyed. During the cyclic loading and unloading process, the coal and rock samples were subjected to high-precision full-range real-time CT scans.

[0049] 3) Perform real-time CT image preprocessing on coal and rock samples to construct an image recognition time-series dataset. Then perform 3D reconstruction of the CT images. Step 3) specifically includes the following sub-steps, which may be performed simultaneously or in any order:

[0050] 3.1) CT images at specific times are selected as time-series datasets based on the time importance coefficient φ. The time importance coefficient φ reflects the time effect of coal and rock pore fractures and volume changes. The time-series datasets include time-series radial two-dimensional slice datasets and axial two-dimensional slice datasets. Primary fractures in coal and rock samples are used as speckle structure features to locate pore fractures in coal and rock. Individual pore fracture time-series datasets are established for each located pore fracture. These individual pore fracture location time-series datasets characterize the process of a single pore fracture in coal and rock from initiation, development, propagation, connection, and eventual failure.

[0051] 3.2) After filtering, denoising, and enhancing the CT data using digital core software, the three-dimensional volume of pores and fractures was reconstructed.

[0052] 4) The time-series dataset constructed in step 3.1) is divided into a training set and a validation set. Time-series features are extracted from the training set data from the dimensions of deformation increment, deformation direction, and coal and rock type, respectively. Then, the time-series features are substituted into different time-series algorithms and fused to construct a coal and rock fatigue fracture propagation regression model to predict the development and propagation of pore fractures and the deformation of coal and rock in real time. The validation set and the pore fracture 3D reconstruction model obtained in step 3.2) are used to improve and validate the coal and rock fatigue fracture propagation regression model.

[0053] 5) Based on the time-series regression model of coal and rock fatigue fracture propagation obtained in step 4), the deformation and fracture propagation characteristics of coal and rock mass under cyclic loading and unloading can be predicted in real time and the characteristics can be characterized.

[0054] 6) Based on the results of step 45), the time-series strain regression parameter set K can be obtained. Combining the "hysteresis loop" characteristic of the stress-strain curve, the strain formula and constitutive equation are modified from three dimensions: strain time series, coal and rock type, and strain direction. At the same time, the model in step 4) is encapsulated to form a device and system that can be widely used to adaptively modify and adjust the coal and rock mass strain formula as time, coal and rock type, and pore and fracture direction change.

[0055] Example 2:

[0056] The main content of this embodiment is the same as that of Embodiment 1. In step 1), the large coal samples selected from the working face are processed into cylindrical specimens by cutting, grinding and polishing. Alternatively, coal and rock powder can be mixed with water and pressed into cylindrical specimens using a mold.

[0057] Example 3:

[0058] The main content of this embodiment is the same as that of Embodiment 1 or 2, except that in step 1), the sample is arranged in a spatial rectangular coordinate system O-XYZ. The origin of the spatial rectangular coordinate system O-XYZ is located at the center of the bottom surface of the sample. The Z-axis of the spatial rectangular coordinate system O-XYZ is parallel to the length direction of the sample, and the X-axis and Y-axis are parallel to the radial direction of the sample.

[0059] Example 4:

[0060] The main content of this embodiment is the same as that of embodiments 1 to 3, except that the initial CT images and all... The CT images at time 1 and the times before and after that time are used as the time series dataset. Time importance coefficients. The method for determining is shown in equation (1). When the rate of change is greater than the threshold, is given Assign 1 if the threshold is not specified, otherwise assign 0. The threshold is determined through expert experience or by image observation.

[0061] (1)

[0062] In the formula, Let t be the number of effective pores and fractures in the coal and rock at time t; Let t be the volume of the positioning hole fracture in the coal and rock at time t; N is the minimum increment of the number of fractures considered to have significant damage and deformation corresponding to this type of coal and rock; This represents the volume change of the positioning hole fissure; Let be the volume of coal and rock at time t; This represents the volume change of coal and rock.

[0063] Example 5:

[0064] The main content of this embodiment is the same as that of embodiments 1 to 4. In step 4), the ratio of the training set and the validation set of the time series dataset is 7:3, which can be adjusted according to the specific experiment.

[0065] Example 6:

[0066] The main content of this embodiment is the same as that of embodiments 1 to 5. However, in step 4), the method for extracting the deformation increment temporal features is to... Time and Calculate the absolute deformation increment of the pore after cyclic loading and unloading, taking the maximum overlap of the same pore at any given time:

[0067] See Figure 2 , Under cyclic loading over a period of time, the pores changed from spherical to ellipsoidal. Region a represents the reduced area within the original pore range, defined as the deformation reduction. hour The area of ​​a plane is denoted as The volume of the pore or fissure where it is located is denoted as The region shown in b represents the increased area within the original pore size, defined as the deformation increment. hour The area of ​​a plane is denoted as The volume of the pore or fissure where it is located is denoted as Furthermore, let point E and point F be arbitrary points on the boundary of the studied void fracture, and... Let E be the distance between points E and F. and These are the distances before and after the cyclic load is applied, respectively.

[0068] For a two-dimensional image of a pore fracture, the two-dimensional deformation increment function is defined as follows:

[0069] (2)

[0070] (3)

[0071] (4)

[0072] (5)

[0073] when At that time, the pores and fissures did not produce irreversible linear strain and area strain; when When the deformation increment of the pores and cracks equals the deformation decrement, irreversible linear strain is generated, but no area strain is generated; when When this occurs, it indicates that the deformation reduction of the pores and cracks is greater than the deformation increment, resulting in linear strain and area strain, and the pores and cracks are... of Contraction occurred within the plane. When this occurs, it indicates that the deformation reduction of the pores and cracks is less than the deformation increment, resulting in linear strain and area strain, and the pores and cracks are... of Expansion occurred within the plane. (5) This represents the absolute deformation increment.

[0074] Example 7:

[0075] The main content of this embodiment is the same as that of embodiments 1 to 6. However, in step 4), the method for extracting the temporal features of the deformation direction is to use the original crack as a positioning reference point and statistically analyze... Time and The direction of crack propagation at any given time is used to extract the spatial distribution features of crack angles and directions on the sliced ​​image. Furthermore, values ​​can be assigned to each angle and direction when... If a crack grows in a certain direction during a certain period, a corresponding value is assigned to that direction for that period, thereby quantifying the crack direction.

[0076] Example 8:

[0077] The main content of this embodiment is the same as that of embodiments 1 to 7. In step 4), the method for extracting the temporal features of coal and rock types is to statistically analyze the CT values ​​of CT image slices at different times. The CT value is closely related to the density of coal and rock, and the magnitude of the CT value can reflect different coal and rock types.

[0078] Example 9:

[0079] The main content of this embodiment is the same as that of embodiments 1 to 8, except that in step 4), the temporal algorithm utilizes full-scale skip links. The temporal algorithm employs full-scale depth supervision.

[0080] Example 10:

[0081] The main content of this embodiment is the same as that of embodiments 1 to 8. In step 5), the deformation features include deformation increment features and deformation direction features.

[0082] Example 11:

[0083] The main content of this embodiment is the same as that of embodiments 1 to 8, except that in step 6), the time-series correction coefficient for the development of coal and rock pores and fractures is considered. Correction coefficients for coal and rock types considering different coal and rock materials. Orientation correction factor due to different material loading methods The coal and rock strain formula is adaptively modified as follows:

[0084] (6) (7)

[0085] (8)

[0086] In the formula, It is an equivalent change. It is volumetric strain. , , These are the linear strains in the x, y, and z directions, respectively.

[0087] Example 12:

[0088] See Figure 1 This embodiment provides an intelligent prediction method for coal and rock fatigue deformation and crack propagation, including the following steps:

[0089] 1) Study sample preparation. Based on the actual needs of the research experiment, prepare rock samples and construct a sample spatial coordinate system.

[0090] 2) Cyclic loading and unloading rock mechanics tests with high-precision, full-range real-time CT scanning. The stress path and real-time CT scanning for cyclic loading and unloading rock mechanics tests should be determined based on the problem being studied and the actual conditions of the laboratory. CT scanning should be performed in real time during the loading and unloading of the specimen. The resolution of the CT scanner used for CT scanning should reach the micrometer level to ensure the identification of deformation and damage in micropores and cracks, thereby improving the precision and accuracy of the test.

[0091] 3) Real-time CT image preprocessing of coal and rock samples, and construction of time-series image recognition datasets. Two-dimensional slices were extracted radially and axially using Matlab, and three-dimensional reconstruction of the CT images was performed using digital core software such as Avizo. The time-series dataset for locating single pore fractures characterizes the process of a single pore fracture in coal and rock from initiation, development, propagation, connection, and eventual failure. Since there are multiple pore fractures in coal and rock, time-series datasets for locating single fractures should be constructed for pore fractures that meet certain conditions. The pore fractures that meet these conditions are defined as those whose size is greater than a threshold value, which is obtained through expert experience.

[0092] The method for extracting radial and axial images using Matlab is as follows: based on the time importance coefficient. This determines the selection of CT images at different times. Specifically, it selects the CT images from the initial time and all... The CT images at time 1 and the times before and after that time are used as the time-series dataset. Because the development of coal and rock pores and fractures is a slow process in the early stages of cyclic loading and unloading, the CT image features at certain time points are very similar. The existence of these "redundant" image data will not have a positive effect on model training; it will only increase the computational load and reduce the model's efficiency. Therefore, it is necessary to ensure that the image features can still be fully extracted even with a large number of similar samples. This embodiment summarizes the laws governing the initiation, development, propagation, and connection of coal and rock pores and fractures, as well as volume changes, and proposes a time importance coefficient that reflects the time effect of coal and rock pores and fractures and volume changes. That is, selecting the initial CT images and all The CT images at time 1 and the times before and after that time are used as the time-series dataset. The time importance coefficients... The method for determining it is as follows:

[0093] (1)

[0094] In the formula, This indicates the number of effective pores and fractures in the coal and rock at time t. Effective pores and fractures refer to pores and fractures that are large enough to significantly affect the damage, deformation, and permeability of the coal and rock. The lower limit of the size of the effective fractures can be determined based on expert experience. This represents the volume of the locating hole fracture in the coal and rock at time t. This volume should include the volume of the extended fractures connected to the original hole fracture. This is the volume change of the fissure in the positioning hole. Only when this value exceeds a certain threshold is the fissure volume considered to have changed significantly. This represents the volume of coal and rock at time t. This represents the change in the volume of coal and rock. Only when this value exceeds a certain threshold is the volume of coal and rock considered to have changed significantly.

[0095] Specify each moment Take 1 or take 0, when A value of 1 indicates a significant change in the volume or number of pores and fractures, or in the volume of coal and rock, compared to the previous moment. This change can be determined by specifying a threshold value; when the rate of change exceeds this threshold, it is considered a significant change. Appendix 1: Otherwise, assign 0. This threshold should be determined based on expert experience or by observing a significant change in the image. A value of 0 indicates that at this moment, at least one of the following quantities—the volume of pores and fractures, the number of pores and fractures, or the volume of coal and rock—has not changed significantly compared to the previous moment. Initially... The value of is determined based on expert experience and rock mechanics test results. When the value is 1, at least one of the following parameters—the number of coal and rock pores and fractures, the volume of pores and fractures, and the volume of coal and rock—has changed significantly compared to the previous time point; when... At time 0, the volume of coal and rock pores and fractures, the number of pores and fractures, and the volume of coal and rock showed almost no change compared to the previous time point. In the same experiment, Once its value is fixed, it will not change.

[0096] The method for three-dimensional reconstruction of CT images uses digital core software such as Avizo to reconstruct the initial time. CT data at time 1 and times before and after it are filtered, denoised, and enhanced before 3D volume reconstruction. CT image processing software can be one of the following: VGStudio Max, Avizo / Amira, Mimics, ORS / Dragonfly, etc.

[0097] This embodiment uses the DVC method, which utilizes the primary fractures in the coal and rock samples as speckle structure features to locate the pore fractures in the coal and rock.

[0098] Constructing a time-series dataset involves creating a single time-series dataset for a single, located pore fracture. The steps are as follows:

[0099] a) Select M time-series CT images at different times from the single locating hole crack to be identified, including radial and axial images, and crop them into 88N square images with a side length of 50-256 pixels. The value of M should not be less than 50. The value of M should be large enough so that the trained model is more accurate.

[0100] b) The cropped image is labeled with the features corresponding to different times. This process is completed based on expert experience.

[0101] c) Divide images of a single positioning hole crack at different scales into training and test sets, and increase the training data by means of translation, rotation, etc., to avoid overfitting.

[0102] 4) Training and Validation of the Coal and Rock Pore Fracture Temporal Regression Model. The temporal dataset constructed in step 3) above is divided into a training set and a validation set. Temporal features are extracted from the training set data from the dimensions of deformation increment, deformation direction, and coal and rock type, respectively. Then, the temporal features are substituted into different temporal algorithms and fused to construct a coal and rock fatigue fracture propagation regression model, which can predict the development and propagation of pore fractures and the deformation of coal and rock in real time. The model is trained and learned based on two-dimensional CT images processed by Matlab. The temporal regression model should be improved and validated by combining the three-dimensional reconstruction model of pore fractures from CT processing software and the validation set.

[0103] In this time-series dataset, the ratio of the training set to the validation set is 7:3.

[0104] To fully capture the semantic information of coal and rock pore and fracture development and expansion images under cyclic loading, Unet3+ was selected as the network model for segmenting the main pore and fracture region. By fusing deep and shallow features through full-scale skip connections and deep supervision, it supervises features at each scale while reducing network parameters and improving computational efficiency. The Unet3+ algorithm effectively combines high-level and low-level semantics from pore and fracture deformation increment features and directional feature maps at different feature scales using full-scale skip connections.

[0105] (9)

[0106] Where i represents the number of the encoder / decoder layer along the encoding direction, N represents the total number of encoders, C represents the convolution operation, the main convolution operation here is the identification and processing of the increase and decrease features and direction features of coal and rock pore fracture deformation, U and D represent upsampling and downsampling operations respectively, H represents the "feature fusion" mechanism (i.e., 1 convolutional layer + 1 batch regularization layer + 1 ReLU function layer), and [ ] represents the channel dimension splicing and fusion operation.

[0107] The Unet3+ algorithm employs full-scale deep supervision, learns hierarchical representations from comprehensive aggregated feature maps, and optimizes the hybrid loss function to enhance the boundaries of coal and rock pore fractures.

[0108] To further highlight the boundary characteristics of coal and rock pores and fractures, the multi-scale structural similarity index loss function MS-SSIM was used to assign different weights to different boundaries. The calculation formula is as follows:

[0109] (10)

[0110] In the formula, M represents the number of scales in the coal and rock CT image. , and , These are the mean and variance of x and y, respectively. Let x and y represent the covariance of two different coal and rock CT images of the same size. and The weight ratio of the two parts in each scale. and These are two small constants used to avoid the abnormal situation where the denominator is zero.

[0111] The temporal feature extraction algorithms used are Temporal 3D ConvNets (T3D) and SE-ResNet. T3D is primarily an improvement on the densely connected network of the DenseNet algorithm. Its main feature is a TTL (Temporal Transition Layer) convolutional layer containing 3D convolutional kernels with variable temporal ranges, capable of capturing temporal features of varying lengths. T3D takes a set of images as input; the algorithm only predicts features when the number of input images exceeds the set limit. Therefore, when the number of input images is less than the set limit, the SE-ResNet structure is used to capture subtle differences between coal and rock image categories and to address the issue of unclear temporal features when there are only a few coal and rock images as input.

[0112] 5) Extraction of deformation increment and deformation direction characteristics. Related studies indicate that coal and rock undergo irreversible plastic deformation under cyclic loading and unloading, forming a "hysteresis loop" in the stress-strain curve. Based on the aforementioned time-series regression model, the deformation characteristics of coal and rock under cyclic unloading can be characterized. The method for extracting the deformation increment is to... Time and The maximum overlap of the same pore at any given time, such as Figure 2 As shown, after cyclic loading, the pores inside the coal and rock will undergo compressive deformation. Due to the different hardness and expansion and ductility properties of coal and rock materials, the volume of the pores and cracks inside the coal and rock will increase or decrease.

[0113] Under cyclic loading over a period of time, the pores changed from spherical to ellipsoidal. Region a represents the reduced area within the original pore range, defined as the deformation reduction. hour The area of ​​a plane is denoted as The volume of the pore or fissure where it is located is denoted as The region shown in b represents the increased area within the original pore size, defined as the deformation increment. hour The area of ​​a plane is denoted as The volume of the pore or fissure where it is located is denoted as Furthermore, let point E and point F be arbitrary points on the boundary of the studied void fracture, and... Let E be the distance between points E and F. and These are the distances before and after the cyclic load is applied, respectively.

[0114] When considering only the two-dimensional image of the pore fracture, the two-dimensional deformation increment function is defined as follows:

[0115] (2)

[0116] (3)

[0117] (4)

[0118] (5)

[0119] when At that time, the pores and fissures did not produce irreversible linear strain and area strain; when When the deformation increment of the pores and cracks equals the deformation decrement, irreversible linear strain is generated, but no area strain is generated; when When this occurs, it indicates that the deformation reduction of the pores and cracks is greater than the deformation increment, resulting in linear strain and area strain, and the pores and cracks are... of Contraction occurred within the plane. When this occurs, it indicates that the deformation reduction of the pores and cracks is less than the deformation increment, resulting in linear strain and area strain, and the pores and cracks are... of Expansion occurred within the plane. (5) This represents the absolute deformation increment.

[0120] When considering the three-dimensional volume of pores and fissures, when When the deformation reduction of the pore is equal to the deformation increment, it indicates that no irreversible volumetric strain has occurred in the pore; when When the deformation reduction of the pores and fissures is greater than the deformation increment, the volume of the pores and fissures has shrunk; when When the deformation reduction of the pores and fissures is less than the deformation increment, the volume of the pores and fissures has expanded.

[0121] Calculate the area strain and volumetric strain of the pore fracture at a given moment:

[0122] (11)

[0123] (12)

[0124] Throughout the entire process of coal and rock cyclic loading and unloading, The area strain and volumetric strain at any given time are clearly time-dependent quantities. Calculations are performed on the same coal and rock sample separately. The strain value at time t is used to fit a strain-time curve based on the obtained data:

[0125] (13)

[0126] (14)

[0127] The optimal fitting method for the above function fitting should be selected based on the distribution characteristics of the acquired data. The parameter set K obtained from the function fitting will be used to incorporate time-series correction coefficients. The formula for strain theory is modified.

[0128] The deformation direction feature extraction method is to extract the deformation direction feature. Time and When the same pore and fracture overlap to the maximum extent, the pores within the coal and rock will undergo compressive deformation after cyclic loading. Due to the heterogeneity and anisotropy of coal and rock materials, the deformation of the internal pores and fractures will exhibit directional characteristics. This embodiment summarizes the laws of rock mechanics research and divides coal and rock deformation into three directional characteristics: radial deformation characteristics, axial deformation characteristics, and total deformation characteristics. The radial deformation characteristic is characterized by the deformation of the pore and fracture in the radial or near-radial direction being greater than the axial deformation; the axial deformation characteristic is characterized by the deformation of the pore and fracture in the axial or near-axial direction being greater than the radial deformation; the total deformation characteristic is characterized by the deformation of the pore and fracture in the axial direction being approximately the same as the deformation in the radial direction.

[0129] 6) Adaptive Correction of Coal and Rock Strain Formula. Existing rock mechanics strain formulas are derived based on the isotropic assumption. However, in practical engineering, coal and rock are often anisotropic, heterogeneous, and nonlinear materials. This embodiment proposes a time-series correction coefficient based on the results of an intelligent learning algorithm. Coal and rock type correction coefficient Direction correction factor This allows the strain calculation formula to be adaptively corrected and adjusted as time, coal and rock type, and pore and fracture direction change, making the theoretical formula for strain calculation more consistent with the actual situation on the engineering site.

[0130] (6)

[0131] (7)

[0132] (8)

[0133] In the formula, It is an equivalent change. It is volumetric strain. , , These are the linear strains in the x, y, and z directions, respectively.

[0134] Timing correction factor This study considers the temporal effect of coal and rock pore and fracture development. In the initial stage of loading, the strain change in coal and rock is relatively small, while in the plastic deformation stage, the strain change is larger. Based on the parameter set K fitted by the strain data function, a pseudo-temporal correction coefficient is constructed. .

[0135] (15)

[0136] Coal and rock type correction factor Considering the characteristics of different coal and rock materials, deep learning algorithms are used to identify the differences in strain among different types of coal and rock, and different material combinations are assigned different correction coefficients for coal and rock types. .

[0137] Direction correction factor Considering the directional characteristics caused by different material loading methods, different directional correction coefficients are determined based on intelligent algorithms.

[0138] Example 13:

[0139] This embodiment provides a computer-readable storage medium storing a computer program for implementing any one of the intelligent prediction methods for coal and rock fatigue deformation and crack propagation in embodiments 1 to 12.

[0140] Example 14:

[0141] This embodiment provides an intelligent prediction device for coal and rock fatigue deformation and crack propagation, including a processor, an input unit, an output unit, and a storage unit. The input unit can receive CT image data from an external CT scanner. The output unit can reflect information about external manifestations. The storage unit is used to store digital core software, instruction sequences executed by the processor for implementing any one of the intelligent prediction methods for coal and rock fatigue deformation and crack propagation in embodiments 1-12, and data during operation.

Claims

1. A method for intelligent prediction of crack propagation in coal and rock fatigue deformation, characterized in that, Includes the following steps: 1) Prepare coal and rock samples and construct a spatial coordinate system for the samples; 2) Conduct cyclic loading and unloading mechanical tests on the coal and rock samples until the samples are destroyed; during the cyclic loading and unloading process, perform high-precision full-range real-time CT scans on the coal and rock samples. 3) Perform real-time CT image preprocessing on coal and rock samples to construct an image recognition time-series dataset; and perform 3D reconstruction of the CT images; step 3) specifically includes the following sub-steps: 3.1) CT images at specific times are selected as time-series datasets based on the time importance coefficient φ; whereby the time importance coefficient φ reflects the time effect of coal and rock pore fractures and volume changes; the time-series datasets include time-series radial two-dimensional slice datasets and axial two-dimensional slice datasets; using the primary fractures in the coal and rock samples as speckle structure features, the pore fractures in the coal and rock are located; a single pore fracture time-series dataset is established for each located pore fracture; the single pore fracture location time-series dataset characterizes the process of a single pore fracture in the coal and rock mass from initiation, development, expansion, connection, and eventual failure; the initial time, all =1 time and CT images before and after time point 1 are used as a time-series dataset; time importance coefficients The method for determining is shown in equation (1); at the initial time t=0, give Assign 1; When t>0, if the rate of change of the number of effective pores and fractures, the volume of locating pores and fractures, or the volume of coal and rock in equation (1) compared to the previous moment is greater than the threshold, then assign 1; Assign 1 if the threshold is not specified, otherwise assign 0; the threshold is determined by expert experience or by image observation. (1) In the formula, Let t be the number of effective pores and fractures in the coal and rock at time t; Let be the volume of the positioning hole fracture in the coal and rock at time t; N represents the volume change of the positioning hole fracture; N is the minimum increment of the number of fractures considered to have significant damage and deformation corresponding to this type of coal and rock. Let be the volume of coal and rock at time t; This represents the volume change of coal and rock. 3.2) After filtering, denoising, and enhancing the CT data using digital core software, three-dimensional volume reconstruction of pores and fractures was performed; 4) The time-series dataset constructed in step 3.1) is divided into a training set and a validation set. Time-series features are extracted from the training set data from the dimensions of deformation increment, deformation direction, and coal and rock type, respectively. Then, the time-series features are substituted into different time-series algorithms and fused to construct a coal and rock fatigue fracture propagation regression model to predict the development and propagation of pore fractures and the deformation of coal and rock in real time. The validation set and the pore fracture 3D reconstruction model obtained in step 3.2) are used to improve and validate the coal and rock fatigue fracture propagation regression model. 5) Based on the coal and rock fatigue fracture propagation regression model obtained in step 4), the deformation and fracture propagation characteristics of coal and rock under cyclic loading and unloading are predicted and characterized in real time. 6) Based on the results of step 4), the time-series strain regression parameter set K is obtained. Combining the hysteresis loop characteristics of the stress-strain curve, the strain formula and constitutive equation are modified from three dimensions: strain time series, coal and rock type, and strain direction. The regression parameter set K is the parameter set of the regression model. After mapping and transformation, different correction coefficients are formed, including time-series correction coefficients that consider the propagation of coal and rock pores and fractures. Correction coefficients for coal and rock types considering different coal and rock materials. Orientation correction factor considering different material loading methods The coal and rock strain formula is adaptively modified as follows: （6） （7） （8） In the formula, It is an equivalent change. It is volumetric strain. , , These are the linear strains in the x, y, and z directions, respectively.

2. The intelligent prediction method for coal and rock fatigue deformation and crack propagation according to claim 1, characterized in that: In step 1), the large coal samples selected from the working face are processed into cylindrical raw coal specimens by cutting, grinding and polishing; or coal and rock powder are mixed with water and pressed into cylindrical coal specimens using a mold.

3. The intelligent prediction method for coal and rock fatigue deformation and crack propagation according to claim 2, characterized in that: In step 1), the sample is arranged in a spatial rectangular coordinate system O-XYZ; the origin of the spatial rectangular coordinate system O-XYZ is located at the center of the bottom surface of the sample; the Z-axis of the spatial rectangular coordinate system O-XYZ is parallel to the length direction of the sample, and the X-axis and Y-axis are parallel to the radial direction of the sample.

4. The intelligent prediction method for coal and rock fatigue deformation and crack propagation according to claim 1, characterized in that: In step 4), the deformation increment temporal feature extraction method is to... Time and The maximum overlap of the same pore at any given time is calculated, and the absolute deformation increment after cyclic loading and unloading of the pore fracture is calculated. The temporal feature extraction method for deformation direction uses the original fracture as a positioning reference point and statistically analyzes... Time and The direction of crack propagation at any given moment is used to extract the spatial distribution features of crack angle and direction on the slice image; The method for extracting the temporal features of coal and rock types is to statistically analyze the CT values ​​of CT image slices at different times.

5. The intelligent prediction method for coal and rock fatigue deformation and crack propagation according to claim 1, characterized in that: In step 4), the temporal algorithm utilizes full-scale skip links.

6. The intelligent prediction method for coal and rock fatigue deformation and crack propagation according to claim 1, characterized in that: In step 5), the deformation features include deformation increment features and deformation direction features.

7. A computer-readable storage medium, characterized in that: The system contains a computer program for implementing the intelligent prediction method for coal and rock fatigue deformation and crack propagation as described in any one of claims 1 to 6.

8. A smart prediction device for coal and rock fatigue deformation and crack propagation, characterized in that: It includes a processor, an input unit, an output unit, and a storage unit; the input unit receives CT image data from an external CT scanner; the output unit reflects information about the external appearance; the storage unit is used to store data of a digital core software, executed by the processor, of a coal and rock fatigue deformation crack propagation intelligent prediction method as described in any one of claims 1 to 6.

Images