A test-based rock fracture strength prediction method and system

Abstract

The application discloses a kind of rock fissure strength prediction method and system based on test, it is related to rock strength test technical field.A kind of rock fissure strength prediction system based on test, including have: rock sample acquisition module and rock strength prediction module.The present application can be from the local effect of microcrack tip and bifurcation, to the interaction of single fissure and its neighborhood, again to the overall connectivity behavior of multi-fissure network, form the continuous observation system of cross-scale by setting different scale samples in the same scene;Blind area detection can be introduced in the region of fissured rock sample, the key area that is difficult to characterize by conventional geometric parameters can be captured, so as to significantly improve the accuracy and interpretability of strength prediction, avoid the prediction deviation caused by ignoring local sensitive site.

Description

Technical Field

[0001] This invention relates to the field of rock strength testing technology, and in particular to a test-based method and system for predicting rock fracture strength. Background Technology

[0002] In existing rock mechanics research and engineering applications, methods for predicting the strength of fractured rocks mainly rely on traditional mechanical tests and geometric parameter statistics. Most methods obtain conventional characteristics such as fracture length, dip angle, spacing, and surface roughness, and combine them with uniaxial or triaxial loading test results to establish empirical formulas or numerical models. However, since the sample size is usually fixed, the differences in fracture geometric effects and network connectivity at different scales are not fully reflected, making it difficult to generalize the experimental results to engineering practice. Furthermore, there is a lack of effective detection of sensitive areas in fractures that are easily overlooked, such as fracture tip curvature, fracture bifurcation, insufficient interfacial fit, and microfracture clusters. These local geometric features often directly determine the failure initiation point, but they are not included in conventional prediction parameters. At the same time, existing methods are mostly carried out in standard laboratory environments, lacking systematic simulation of environmental conditions such as water, temperature, chemical effects, and time effects, making it difficult for the prediction results to reflect the strength evolution process under complex site conditions.

[0003] Therefore, there is a need for a rock fracture strength prediction method that combines cross-scale sample setup, fracture geometry blind zone detection, and multimodal multi-time-effect environmental loading to achieve high accuracy, interpretability, and confidence interval output. Summary of the Invention

[0004] The present invention aims to provide a test-based method and system for predicting rock fracture strength, which can achieve high-precision, interpretable strength prediction with a strength confidence interval under multiple environmental conditions.

[0005] A test-based method for predicting rock fracture strength includes the following steps:

[0006] Based on the actual rock scene, obtain the rock sample to be optimized, which contains rock fissures; obtain the geometric feature parameters of the rock fissures based on the rock sample to be optimized; divide the rock sample to be optimized into rock samples of different scales according to the geometric feature parameters of the rock fissures and the actual rock scene.

[0007] The following steps are used to test rock samples of any scale under standard conditions: 1. Acquire rock sample images; 2. Divide the rock sample into a basic rock sample region and a fractured rock sample region based on the images; 3. Extract component data from the basic rock sample region and the fractured rock sample region respectively to obtain the component parameter characteristics of the basic rock and the fractured rock; 4. Perform blind zone detection on the fractured rock sample region to obtain the geometric blind zone characteristics of the fractured rock; 5. Establish a stress test combination containing different axial parameters; 6. Use the stress test combination to perform graded loading monitoring on the rock sample and collect the stress fingerprint characteristics of the rock sample; 7. Predict the strength of the rock sample to obtain the predicted value of the standard rock strength.

[0008] A multimodal, multi-time-dependent environment is constructed, and the above specific steps are repeated to test rock samples of different scales to obtain multidimensional rock strength prediction values. The standard rock strength prediction values ​​and multidimensional rock strength prediction values ​​are compared to predict the strength, and the fracture strength prediction results and corresponding strength confidence intervals of the rock samples are output.

[0009] As a preferred embodiment of the present invention, the specific steps for dividing rock samples into rock samples of different scales based on the geometric characteristic parameters of rock fissures and the actual rock scene include:

[0010] A fracture geometry-scene feature set is constructed based on the actual rock scene and the geometric feature parameters of the rock fracture; microscale, mesoscale and macroscale specimens are set according to the fracture geometry-scene feature set; the minimum specimen size at each scale is determined according to the geometric feature parameters of the rock fracture.

[0011] As a preferred embodiment of the present invention, the specific steps for dividing a rock sample into a basic rock sample region and a fractured rock sample region based on a rock sample image include:

[0012] Image segmentation algorithms are used to process rock sample images to obtain fracture probability maps and fracture uncertainty maps. Based on the fracture probability maps and fracture uncertainty maps, an initial skeleton of the rock fracture region is extracted. The signed distance field and local fracture width of the initial skeleton of the rock fracture region are calculated. Based on the signed distance field and local fracture width, the inner and outer boundaries of the rock fracture are constructed.

[0013] Boundary buffer zones are identified based on the inner and outer boundaries of rock fissures; several sets of candidate boundaries are generated within the boundary buffer zones using an optimization function; the objectives of the optimization function include boundary smoothness constraints and fissure texture gradient information.

[0014] The candidate boundaries are offset and corrected to obtain the corrected candidate boundaries; the comprehensive similarity between the corrected candidate boundaries and the inner and outer boundaries of the rock fractures is calculated, and the optimal candidate boundaries are selected as the final fracture region boundaries; the rock samples are divided into basic rock sample regions and fractured rock sample regions according to the final fracture region boundaries.

[0015] As a preferred embodiment of the present invention, the specific steps for blind zone detection of fractured rock sample regions include:

[0016] A candidate blind zone seed set is generated in the fractured rock sample region. The candidate blind zone seed set includes geometric seeds, fitting seeds, and precursor seeds. For each type of candidate blind zone seed, a candidate blind zone ring is divided according to the final fracture region boundary. A preliminary blind zone block is generated in the candidate blind zone ring according to the maximum coverage-minimum overlap criterion. In the preliminary blind zone block, the selection weight of each preliminary blind zone block is calculated according to the fracture direction concentration, area density, and scene label.

[0017] Prioritize all the initial blind blocks according to their selection weights and output the final filtered blind blocks; perform feature detection on each filtered blind block to obtain the geometric features of the blind block;

[0018] By fusing the geometric features of the blind blocks on the rock sample, the geometric blind features of the fractured rock are obtained.

[0019] As a preferred embodiment of the present invention, the specific steps for using a stress testing combination to perform graded loading monitoring on rock samples and collect the stress fingerprint characteristics of the rock samples include:

[0020] The axial parameters in the stress parameter combination include the number of axial forces, the axial direction, and the axial load.

[0021] The rock samples were subjected to strength tests by selecting different combinations of axial stress parameters in a stepwise manner to obtain the stress level characteristics of rock fractures. At the same time, at the end of each strength test, the rock samples were subjected to perturbation loading using small-amplitude perturbation cycles to extract the evolution characteristics of normal stiffness and shear stiffness.

[0022] After all stress parameter combination strength tests are completed, the collected rock fracture stress level characteristics, normal stiffness evolution characteristics, and shear stiffness evolution characteristics are horizontally aligned and feature extracted to obtain the stress fingerprint characteristics of the rock sample.

[0023] As a preferred embodiment of the present invention, the specific steps for predicting the strength of rock samples include:

[0024] Strength prediction is performed based on the stress fingerprint characteristics of rock samples, the geometric blind zone characteristics of fractured rocks, the basic rock composition parameter characteristics, and the composition parameter characteristics of fractured rocks to obtain the predicted value of standard rock strength.

[0025] By aligning the predicted values ​​of multidimensional rock strength and standard rock strength of rock samples at different scales over time, a set of strength comparison results is obtained.

[0026] The deviation amplitude between each predicted value in the strength comparison result set is calculated; when the deviation amplitude exceeds the preset deviation threshold, the corresponding predicted value is weighted and corrected to obtain the corrected rock strength prediction value; all predicted values ​​in the strength comparison result set are traversed and merged to obtain the fracture strength prediction result; the uncertainty of the fracture strength prediction result is estimated using the Bayesian inference algorithm, and the corresponding strength confidence interval is output.

[0027] A test-based rock fracture strength prediction system includes:

[0028] The rock sample acquisition module includes a sample construction unit; the sample construction unit is used to acquire rock samples containing rock fissures based on the rock field scene; to acquire the rock fissure geometric feature parameters based on the rock samples to be optimized; and to divide the rock samples to be optimized into rock samples of different scales according to the rock fissure geometric feature parameters and the rock field scene.

[0029] The rock strength prediction module includes a standard strength prediction unit and a comprehensive strength prediction unit. The standard strength prediction unit is used to test rock samples of any scale under standard conditions. The specific steps are: acquiring rock sample images; dividing the rock sample into a basic rock sample region and a fractured rock sample region based on the rock sample images; extracting component data from the basic rock sample region and the fractured rock sample region respectively to obtain the basic rock component parameter characteristics and fractured rock component parameter characteristics; performing blind zone detection on the fractured rock sample region to obtain the geometric blind zone characteristics of the fractured rock; establishing a stress test combination containing different axial parameters; using the stress test combination to perform graded loading monitoring on the rock sample and collecting the stress fingerprint characteristics of the rock sample; predicting the strength of the rock sample to obtain the standard rock strength prediction value. The comprehensive strength prediction unit is used to construct a multimodal, multi-time-effect environment, repeating the above specific steps to test rock samples of different scales to obtain multidimensional rock strength prediction values; comparing the standard rock strength prediction value and the multidimensional rock strength prediction value to predict the strength, and outputting the fracture strength prediction result and the corresponding strength confidence interval of the rock sample.

[0030] The present invention has the following advantages:

[0031] 1. This invention, by testing rock samples of different scales, can highlight both local geometric details and the overall fracture network effect. Small-scale samples can capture detailed features such as fracture tip curvature, bifurcation, and interfacial fit, while large-scale samples can cover the connectivity and possible penetration paths of multiple fractures. The combination of the two not only reveals the scale effect law of rock strength, but also provides a quantifiable theoretical basis for extrapolating from the laboratory to the engineering scale, thus unifying the prediction results between refinement and engineering applicability.

[0032] 2. This invention can compensate for the prediction bias caused by the lack of conventional geometric parameters by identifying the blind zone of rock fracture testing. This allows subsequent strength prediction to capture potential failure initiation points and strength attenuation sources in advance, thereby significantly improving the accuracy and stability of the prediction and enhancing the interpretability of the prediction results. It can clearly explain which characteristic factors have led to the decrease in rock mass strength.

[0033] 3. By setting up repeated tests under modal multi-time-effect environments, this invention can introduce external conditions such as water content, temperature, freeze-thaw cycles, chemical erosion, and loading cycles, allowing the real evolution of rock fractures under complex working conditions to be recorded. Furthermore, by acquiring comprehensive acoustic, optical, thermal, and mechanical response signals through multi-source monitoring, compared to a single standard environment, this setup can not only reveal the long-term influence of environmental factors on strength but also construct a generalized model across working conditions, outputting prediction results including confidence intervals, and providing a more reliable safety margin for engineering design. Attached Figure Description

[0034] Figure 1 This is a schematic diagram of a test-based rock fracture strength prediction system used in an embodiment of the present invention. Detailed Implementation

[0035] To enable those skilled in the art to better understand the technical solutions of this invention, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of this invention.

[0036] Example 1: A test-based method for predicting rock fracture strength, comprising the following steps:

[0037] Based on the actual rock scene, obtain the rock sample to be optimized, which contains rock fissures; obtain the geometric feature parameters of the rock fissures based on the rock sample to be optimized; divide the rock sample to be optimized into rock samples of different scales according to the geometric feature parameters of the rock fissures and the actual rock scene.

[0038] In the rock fracture strength prediction method, the actual rock scenario refers to the original state and external conditions of the rock in its natural or engineering environment, that is, the real rock mass environment from which the rock sample originates. A fracture geological survey is conducted at the proposed project site or outcrop location. By setting up survey lines and window samples, the strike, dip angle, length, spacing, filling condition, and development density of the target rock fractures are recorded. Simultaneously, environmental information such as bedding, faults, weathering zone range, water content, and temperature is obtained. Based on this, sampling points representing different fracture combinations and connectivity are selected. Samples containing natural fractures are obtained by core drilling or cutting blocks. The sampling coordinates, sampling depth, and structural plane orientation are immediately marked on-site. The samples are then sealed, moisturized, and vibration-proofed to reduce secondary disturbance to the fracture state during transportation and processing.

[0039] The specific steps for classifying rock samples into different scales based on the geometric characteristics of rock fractures and the actual rock scene include:

[0040] A fracture geometry-scene feature set is constructed based on the actual rock scene and the geometric feature parameters of the rock fracture; microscale, mesoscale and macroscale specimens are set according to the fracture geometry-scene feature set; the minimum specimen size at each scale is determined according to the geometric feature parameters of the rock fracture.

[0041] The geometric features of the acquired rock sample to be optimized are obtained. Specific steps include, but are not limited to, placing the sample into a 3D imaging device or computed tomography system to acquire 3D voxel images of the internal fractures, and combining this with a surface topography scanner or optical microscopy to acquire surface undulation information of the fracture surface. By performing threshold segmentation and connected component analysis on the 3D images, the overall morphology of the fractures can be identified, and their length, extension direction, dip angle, spacing, and connectivity within the sample can be calculated. At the fracture tip region, image skeletonization and local curvature analysis are used to extract the curvature and bifurcation features of the fracture tip to reflect stress concentration. Sensitive locations; in characterizing fracture surfaces, the difference in elevation between the two fracture surfaces is compared using a digital elevation model to obtain the roughness and degree of fit or mismatch of the fracture surface, thereby obtaining inter-surface matching parameters; at the macroscopic level, statistical analysis of the fracture network is used to calculate the distribution density, continuity probability, and directionality of fractures in three-dimensional space, forming a comprehensive index that can reflect the complexity of the fracture network; finally, these geometric feature data from points, lines, surfaces to the overall network are archived into a structured parameter set as geometric feature parameters of rock fractures; this preserves the detailed information of the fractures and ensures the integrity and repeatability of the data that can be used for quantitative analysis in subsequent strength prediction.

[0042] Based on the aforementioned geometric parameters and field information, the statistical data on fracture spacing, length distribution, probability density of strike and dip angle, fracture surface density or volume density, connectivity and penetration probability, spatial relationship between bedding and faults, water content and temperature conditions obtained in the field are aligned with the corresponding geometric parameters measured in the laboratory using a unified coordinate system and dimensions. Representative values, dispersion and confidence ranges are given for each type of feature to form a searchable dataset. The mapping relationship and weight between samples and field units are labeled to obtain a fracture geometry-scene feature set.

[0043] Based on the feature set, specific scale division rules are set. Generally speaking: when the field shows small fracture spacing, high connectivity, or potential through-paths, larger volumes are preferred to cover the interactions of multiple fracture groups; when the research objective emphasizes the effect of fracture tip, fracture surface adhesion, or infilling, smaller volumes are preferred to ensure the resolution and control of local details; at the same time, the long axis of the sample is required to maintain a correspondence with the direction of the principal stress or the orientation of the potential slip surface, and the directional distribution, density, and connectivity of the sample are verified using the scene fit index to determine whether additional sampling or scale replacement is needed; in addition, the specific size rules for rock sample setting are as follows: during implementation, microscale samples are used for detailed studies of fracture tips, bifurcations, and the adhesion of two fracture surfaces, used to capture local geometric details and inter-surface adhesion features, and are generally set to twenty to fifty times the minimum fracture aperture. To ensure the resolution and signal-to-noise ratio of morphological information and tomographic imaging, mesoscale specimens are used to study the coupling behavior of a single fracture and its neighboring microfractures. They reflect the interaction between the fracture body and surrounding secondary fractures. Generally, their side length is set to approximately 0.5 to 2 times the average fracture spacing in the field to simultaneously cover the target fracture and its influence range. Macroscale specimens are used to examine the connectivity of multiple fracture groups and the possible through-paths to reveal the overall effect of the fracture network. Generally, their side length is more than twice the average fracture spacing in the field and does not exceed a reasonable proportion of the engineering characteristic dimensions. They must also ensure that they contain at least two sets of fractures in different orientations or a complete and continuous long fracture segment. All specimens are processed and marked with an orientation consistent with the direction of the principal stress or the excavation direction in the field. The above dimensions only show one setting. The specific scale setting rules are set according to the conditions of the field specimens.

[0044] The following steps are used to test rock samples of any scale under standard conditions: 1. Acquire rock sample images; 2. Divide the rock sample into a basic rock sample region and a fractured rock sample region based on the images; 3. Extract component data from the basic rock sample region and the fractured rock sample region respectively to obtain the component parameter characteristics of the basic rock and the fractured rock; 4. Perform blind zone detection on the fractured rock sample region to obtain the geometric blind zone characteristics of the fractured rock; 5. Establish a stress test combination containing different axial parameters; 6. Use the stress test combination to perform graded loading monitoring on the rock sample and collect the stress fingerprint characteristics of the rock sample; 7. Predict the strength of the rock sample to obtain the predicted value of the standard rock strength.

[0045] The specific steps for dividing a rock sample into a basic rock sample region and a fractured rock sample region based on the rock sample image include:

[0046] Image segmentation algorithms are used to process rock sample images to obtain fracture probability maps and fracture uncertainty maps. Based on the fracture probability maps and fracture uncertainty maps, an initial skeleton of the rock fracture region is extracted. The signed distance field and local fracture width of the initial skeleton of the rock fracture region are calculated. Based on the signed distance field and local fracture width, the inner and outer boundaries of the rock fracture are constructed.

[0047] Boundary buffer zones are identified based on the inner and outer boundaries of rock fissures; several sets of candidate boundaries are generated within the boundary buffer zones using an optimization function; the objectives of the optimization function include boundary smoothness constraints and fissure texture gradient information.

[0048] The candidate boundaries are offset and corrected to obtain the corrected candidate boundaries; the comprehensive similarity between the corrected candidate boundaries and the inner and outer boundaries of the rock fractures is calculated, and the optimal candidate boundaries are selected as the final fracture region boundaries; the rock samples are divided into basic rock sample regions and fractured rock sample regions according to the final fracture region boundaries.

[0049] When testing rock samples of any scale under standard conditions, high-quality images of the rock samples are first acquired. The imaging method should be one that can simultaneously reflect the internal structure and surface morphology to ensure that the resolution and signal-to-noise ratio are sufficient to distinguish the smallest crack aperture and the subtle texture changes in its neighborhood. After imaging, the images are denoised, brightness equalized and geometrically corrected so that subsequent region division and feature calculation are based on a stable and consistent image baseline.

[0050] It should be noted that the standard environment refers to the uniform external conditions set to ensure the comparability and repeatability of results when testing rock samples. These conditions typically include maintaining a laboratory atmosphere at normal temperature and pressure, controlling the temperature at around 20 degrees Celsius with fluctuations not exceeding 2 degrees Celsius, maintaining the relative humidity at around 50%, ensuring there is no significant corrosiveness or high humidity interference in the air, ensuring the sample is in a stable dry or specified moisture content state before loading, and avoiding the influence of external factors such as strong vibration, light, or electromagnetic interference during the test, so that the mechanical response of the sample is controlled only by the loading path and is not affected by environmental fluctuations.

[0051] In the segmentation step of rock sample images, the image after denoising and geometric correction is first input into the segmentation model or regularized segmentation process, outputting a probability distribution map of each pixel belonging to a fracture and a corresponding uncertainty map. The probability map is used to characterize the differences between the fracture and the parent rock in terms of grayscale, texture, and edge response, while the uncertainty map is used to mark blurred areas caused by uneven illumination, imaging artifacts, low signal-to-noise ratio, or material property transitions. Both retain soft information in floating-point form without premature binarization, serving as the basis for subsequent geometric calculations and boundary safety distance adjustment. In the step of extracting the initial fracture skeleton based on the fracture probability map and fracture uncertainty map, the high-probability connected components are first refined and topology-preserving skeletonized to obtain a skeleton that reflects the main direction and distribution of the fracture. The centerline network of the cross relationship is used; to avoid noise traction, isolated small pieces or low-probability filaments are suppressed or broken without destroying the main connectivity structure. At the same time, in high uncertainty areas, the skeleton is allowed to be locally empty and the position labels are recorded so as not to be misled during subsequent boundary search; then, the shortest distance from the whole image pixels to the skeleton is established with the initial skeleton of the rock fracture region as a reference, and the skeleton side and the outside are distinguished to obtain a signed distance field. At the same time, intensity or texture profiles are generated along the direction approximately orthogonal to the skeleton normal. The local fracture width of each point is estimated using the full width at half maximum or the bilateral reach distance; when the fracture is anisotropic or multimodal, the minimum envelope width in multiple directions is preferred and scale consistency correction is performed to obtain a continuous, smooth and dimensionally clear local fracture width.

[0052] Specifically, the inner boundary of the rock fissure is defined as a set of distance contour lines, and its contour radius is a linear combination of local width and uncertainty weights, so that the inner boundary is moderately close to the fissure in fine fissures and high confidence regions without excessively adhering to the wall; the outer boundary of the rock fissure is defined as a larger contour radius and maintains a strict inclusion relationship with the inner boundary, so that it is appropriately moved outward in rough, poorly adhered, or high uncertainty regions to avoid intrusion into the parent rock. Both boundaries achieve continuity and workability through curvature regularization.

[0053] In the step of identifying boundary buffer zones based on inner and outer boundaries, the closed region between two boundaries is used as a buffer zone, and an adaptive thickness and cost field are calculated for each connected piece. When the local surface is rough, the fit is low, or the probability fluctuates drastically, the buffer zone is automatically thickened to provide a more ample search corridor, while it is appropriately narrowed in areas with uniform texture and clear signal. At the same time, it shields the abnormal areas caused by imaging artifacts or metallic marks, so that subsequent optimization is only performed within the reliable space. The candidate boundary is regarded as one or more closed curves, and the optimization objective function composed of curvature penalty, texture gradient consistency, equidistant preference with the skeleton, and cross-material property jump penalty is minimized in the boundary buffer zone domain. In order to cover different branch and end morphologies, initialization is performed from multiple starting points and multiple ending points, and multiple solution sets are solved to form several geometrically smooth candidate boundary curves that are compatible with lithological texture.

[0054] Specifically, in the optimization function, curvature penalty is used to constrain the smoothness of the boundary curve, preventing jagged edges or violent non-physical oscillations, and ensuring that the boundary morphology is geometrically continuous and manufacturable. Texture gradient consistency is used to guide the boundary to fit the real transition zone between the fracture and the parent rock in the image, that is, to give higher weights to areas with significant grayscale or texture gradients, making the boundary more consistent with the material property boundaries in the image. Equidistance preference with the skeleton reflects that the boundary should maintain a relatively balanced interval relationship with the fracture skeleton as a whole, avoiding both excessive narrowness due to close contact with the skeleton and excessive intrusion into the parent rock away from the skeleton, thereby maintaining the coordination between the geometric center of the fracture and the boundary. Cross-material property jump penalty is used to constrain the boundary from frequently crossing the parent rock area or from producing false bends due to noise. When the boundary crosses a continuous area that clearly belongs to the basic rock area, the energy cost is increased to suppress unreasonable extension. By minimizing the comprehensive results of these four indices, the obtained boundary has both geometric smoothness and fits the texture and geometric features of the real fracture, while avoiding over-reliance on local noise or artifacts, thereby ensuring the stability and physical rationality of the partitioning results.

[0055] In the boundary screening step, specifically: when offsetting the candidate boundary, the candidate curve is first projected onto the cost field within the boundary buffer zone. Taking into account the uncertainty distribution, local crack width estimation, main texture direction and intensity gradient, as well as physical clues such as strain localization zone and acoustic emission hotspots obtained from synchronous experiments, small and restricted normal displacement and tangential reparameterization are applied to each discrete point of the curve. The displacement amplitude is limited to a certain proportion of the buffer zone thickness at that location to avoid exceeding the boundary. The curvature change is restricted to a continuous range to prevent sharp corners or self-crossing. At the same time, outward correction is applied to locations with significantly smaller local widths, and inward correction is applied to locations with significantly larger local widths, so that the curve remains smooth and continuous globally, and the two types of error sources, namely, wall adhesion and redundancy, are avoided as much as possible locally. After completion, the consistency of the curve length, curvature and the distance distribution of the skeleton is checked to obtain the corrected candidate boundary.

[0056] When calculating the comprehensive similarity between the candidate boundary and the inner and outer boundaries and selecting the optimal boundary, a unified score containing multiple evaluations is constructed: First, the distance balance with the inner and outer boundaries, requiring the curve to be close to the middle position of the two and avoiding large areas biased to either side; second, the conformity with the image texture gradient, requiring the curve to follow the transition zone of the real material properties; third, the geometric homology with the fracture skeleton, requiring reasonable topological correspondence in the neighborhood of branches, intersections and endpoints; fourth, the regularity of the curve itself, including closure, curvature smoothness and reasonable length; and fifth, the coverage of physical anomaly areas, so that the boundary preferentially passes through areas of strain concentration or energy accumulation. The above items are normalized and weighted according to the scene weight to obtain a comprehensive score. When there is branch conflict or multiple solutions coexist, the connection consistency and the minimum cost integral are used as criteria for pruning and merging. Finally, the curve with the highest comprehensive score and that passes the consistency check is selected as the final fracture region boundary.

[0057] After obtaining the final fracture region boundary, a spatial mask is generated between the fracture region and the basic rock region using this boundary as the dividing line. Extremely small isolated patches are then screened for connectivity issues and have small holes filled, resulting in smooth region boundaries, reasonable topology, and ease of subsequent processing. Subsequently, based on the mask, the subsequent component extraction, geometric feature calculation, and mechanical monitoring data are aligned with coordinates and labeled to form a traceable data record. The record includes quality indicators such as boundary origin, buffer zone thickness, mean uncertainty, distance uniformity, and comprehensive score. Ultimately, the sample is stably divided into a basic rock sample region and a fractured rock sample region, providing a reliable spatial benchmark for subsequent blind zone detection and stress fingerprint construction.

[0058] The specific steps for blind zone detection in fractured rock sample areas include:

[0059] A candidate blind zone seed set is generated in the fractured rock sample region. The candidate blind zone seed set includes geometric seeds, fitting seeds, and precursor seeds. For each type of candidate blind zone seed, a candidate blind zone ring is divided according to the final fracture region boundary. A preliminary blind zone block is generated in the candidate blind zone ring according to the maximum coverage-minimum overlap criterion. In the preliminary blind zone block, the selection weight of each preliminary blind zone block is calculated according to the fracture direction concentration, area density, and scene label.

[0060] When generating candidate blind zone seeds in the fractured rock sample region, the final fracture region boundary and fracture skeleton are used as spatial references. Sensitivity is evaluated pixel-by-pixel in the image domain by comprehensively considering three types of indicators: geometry, interface adhesion, and pre-stress indicators. Extreme or over-threshold positions are extracted as seed points or small blocks. Geometric seeds are obtained from areas where the curvature of the fracture tip increases significantly, the degree of the bifurcation node increases, and the distance between adjacent fracture tips decreases abnormally. Adhesion seeds are obtained from areas where the matching degree decreases after the elevation of the two sides of the fracture surface is aligned, the mismatch gradient increases, and the proportion of local contact area decreases. Pre-stress seeds are obtained from the corresponding projection positions in the image of strain localization zones, thermal anomaly accumulation zones, or dense acoustic event zones that accompany the loading process. Imaging uncertainty is used as the elimination weight to avoid noise-triggered pseudo-seeds.

[0061] For each type of candidate blind zone seed, when constructing candidate blind zone rings based on the final fracture region boundary, two safety distance contour lines are determined using signed distance and local fracture width. The inner line restricts the boundary from excessively approaching the fracture surface, while the outer line restricts the blind zone from intruding into the parent rock body. The ring thickness is adaptively adjusted according to local width, image uncertainty, and surface roughness. It automatically thickens in areas with rough fractures or poor fit, and moderately narrows in areas with clear texture and stable boundaries. Through connectivity constraints, the ring maintains a continuous corridor at bifurcation and intersection, facilitating subsequent blind zone search within a reliable range. The ring is discretized into multi-scale tiles, prioritizing combinations that can cover more seeds while having minimal overlap with selected tiles, optimizing for maximum coverage and minimum overlap. Penalties are applied to tiles that approach the sample boundary or cross obviously homogeneous parent rock areas to ensure that the initially selected blind zone blocks completely enclose sensitive areas while avoiding ineffective expansion. Adjacent small blocks are allowed to merge into more fitted irregular blocks at complex geometries such as bifurcation and confluence to improve representation capabilities.

[0062] It should be noted that the aforementioned tiles refer to several regularized small regions or sub-blocks artificially divided within the candidate blind zone annulus. This is equivalent to superimposing a set of multi-scale grid windows on a continuous annulus space to facilitate statistical and combined analysis. Each tile represents a local segment within the annulus, and its size and shape can be adaptively adjusted according to the local width, directional distribution, and bifurcation complexity of the fracture. This ensures that the potential blind zone seeds are fully covered without losing details due to overly coarse division or introducing excessive redundancy due to overly fine division. By optimizing the coverage and overlap among these tiles, the candidate blind zone blocks that best represent the sensitive parts of the fracture can be selected more effectively.

[0063] The consistency between the fracture orientation concentration and the dominant orientation within the block, the degree of fit between the fracture area density and the field statistics, and the degree of matching of scene labels (such as bedding dominance, cross joints, fault fracture zones, and infill control) are normalized and weighted. At the same time, the comprehensive scores of geometric sensitivity and fit sensitivity are superimposed, and uncertainty and imaging quality are used as correction factors to obtain the selection weight of the initial blind block, which provides a basis for subsequent ranking.

[0064] Prioritize all initially selected blind blocks according to their selection weights, and output the final filtered blind blocks. Perform feature detection on each filtered blind block to obtain its geometric features. Select blind blocks in descending order of weight, and merge and remove redundancy from blind blocks that are spatially highly overlapping or functionally equivalent. When adjacent high-weight blocks form a narrow and repetitive coverage, prioritize retaining the block shape that can more completely represent the continuous transition from crack tip to bifurcation to merging. For weak directions with insufficient coverage, appropriately insert the second-highest weight block to ensure the representativeness of direction and scale. Finally, obtain filtered blind blocks that are spatially balanced, consistent with scene statistics, and have sufficient coverage of intensity-sensitive areas.

[0065] When performing feature detection on each blind block, robust statistics of geometric details are calculated under the local coordinates of the blind block, including the quantile and gradient change of the crack tip curvature, the degree and angle distribution of the bifurcation nodes, the nearest distance between adjacent crack tips and their directionality, the matching degree and mismatch gradient of the two sides of the crack surface, the mean and dispersion of the contact area ratio and the local crack width, and the microcrack density and connectivity, etc. To reduce the impact of noise, all quantization results are corrected for scale consistency and checked for repeated measurements, and the quality label and uncertainty are recorded for weighted use in the subsequent fusion stage.

[0066] The geometric features of blind blocks on rock samples are fused to obtain the geometric blind zone features of fractured rocks. During the feature fusion, indices of different dimensions are first standardized and dimensionless. Then, fusion weights are assigned based on their correlation with the strength weakening mechanism and stability. Geometric sensitivity, interface fit, and hierarchical connectivity information are aggregated into a unified blind zone feature vector at the sample level. When multiple blocks exhibit strong correlation redundancy, correlation reduction or principal component extraction is used to retain the main information axis. The source block number, spatial location, and uncertainty are also archived, forming geometric blind zone features of fractured rocks that can be directly used for prediction modeling and have traceable interpretability.

[0067] Geometric blind zone features in fractured rocks are used to compensate for the shortcomings of traditional geometric parameters (such as length, dip angle, spacing, and roughness) in characterizing the impact of fractures. They highlight key local areas that are most sensitive to strength prediction but are easily overlooked, such as high-stress zones with concentrated fracture tip curvature, potential instability sources at bifurcation and intersection, weak contact zones with low fracture surface fit, and hidden pathways formed by microfracture clouds. By introducing these blind zone features, not only can the accuracy of strength prediction be significantly improved, but the interpretability of the prediction results can also be enhanced. For example, it can explain why certain fractures are more likely to become failure initiation points. At the same time, this feature can reflect the actual failure mechanism of rock masses under different scales and environmental conditions, providing a more scientific basis for stability analysis, safety margin design, and fracture control in engineering.

[0068] The specific steps for using a stress testing assembly to monitor graded loading of rock samples and collect the stress fingerprint characteristics of the rock samples include:

[0069] The axial parameters in the stress parameter combination include the number of axial forces, the axial direction, and the axial load.

[0070] Specifically: the number of axial axes is used to distinguish between single-axis, biaxial, and multiaxial combined loading conditions. The axial direction is set according to the relative relationship between the crack orientation and dip angle in the specimen, so that a representative angular distribution is formed between the main loading axis and the potential shear surface, bedding plane, or joint group. The axial load gives the target peak value, loading rate, and holding time for each axis, and clarifies whether the confining pressure or lateral constraint participates in its amplitude range. To ensure the comparability of results between different combinations, all parameters are uniformly normalized to the same benchmark after the specimen geometry, end face flatness, and sensor calibration are completed. At the same time, safe shutdown criteria and data quality indicators are predefined to ensure that subsequent graded loading is carried out within traceable and controllable boundaries.

[0071] The rock samples were subjected to strength tests by selecting different combinations of axial stress parameters in a stepwise manner to obtain the stress level characteristics of rock fractures. At the same time, at the end of each strength test, the rock samples were subjected to perturbation loading using small-amplitude perturbation cycles to extract the evolution characteristics of normal stiffness and shear stiffness.

[0072] By selecting and applying combinations of different axial parameters in a stepwise manner according to a proportional hierarchy, strength tests were conducted on the specimens to obtain the response trajectory of the fracture at each stress level. Each loading stage included a stable loading segment and a constant load holding segment to fully capture the stage characteristics of fracture closure, initiation, and propagation. At the end of each stage, a small-amplitude perturbation cycle was superimposed to perform high-resolution sampling of the force-displacement or stress-strain curves. The instantaneous and asymptotic changes of normal stiffness and shear stiffness were inverted using the tangent slope and hysteresis loop slope, respectively. To reduce the influence of historical damage and noise on the estimation, the perturbation cycle adopted symmetrical amplitude and consistent frequency, and a brief unloading and drift correction were performed before entering the next loading stage. This resulted in the formation of rock fracture stress level characteristics, normal stiffness evolution characteristics, and shear stiffness evolution characteristics jointly characterized by stress level, stiffness degradation amplitude, and energy hysteresis loop morphology.

[0073] After all stress parameter combination strength tests are completed, the collected rock fracture stress level characteristics, normal stiffness evolution characteristics and shear stiffness evolution characteristics are horizontally aligned and feature extracted to obtain the stress fingerprint characteristics of the rock sample.

[0074] After all stress parameters were combined, the fracture stress level characteristics, normal stiffness evolution characteristics, and shear stiffness evolution characteristics obtained from different combinations were aligned with a unified time axis and stress level. Interpolation resampling and baseline drift correction were used to map the multi-source data to the same reference coordinate system, and noise filtering and robust statistics were used to suppress occasional outliers. Subsequently, discriminative indices were extracted from the aligned sequence, including characteristic inflection points of each stress level, staged degradation rate of stiffness, hysteresis area and shape parameters, coupling terms between different axes, and anisotropic quantification values ​​related to fracture orientation. Finally, these standardized and dimensionless quantities were combined into a structured vector, which served as the stress fingerprint characteristics of the rock sample under multiaxial and multi-level loading histories, providing a unified and stable model input for subsequent strength prediction and uncertainty assessment.

[0075] A multimodal, multi-time-dependent environment is constructed, and the above specific steps are repeated to test rock samples of different scales to obtain multidimensional rock strength prediction values. The standard rock strength prediction values ​​and multidimensional rock strength prediction values ​​are compared to predict the strength, and the fracture strength prediction results and corresponding strength confidence intervals of the rock samples are output.

[0076] Multimodal and multi-time-effect environments refer to a set of test scenarios that, in addition to standard laboratory conditions, systematically superimpose multiple external effects and monitoring methods, and set different durations and cycles in the time dimension. The external effects may include changes in water content, confining pressure and lateral pressure paths, temperature and freeze-thaw cycles, chemical immersion, loading rate and loading-unloading cycles, etc. The monitoring methods can simultaneously use acoustic emission, ultrasound, digital image correlation, infrared thermography and conventional mechanical sensors to record the response trajectory of rocks under different environments and time scales. Unlike standard environments, which only obtain one-time data under constant temperature and humidity, constant medium and single loading regime, multimodal and multi-time-effect environments aim to reproduce complex working conditions and evolution processes in reality, so that the same family of samples can generate comparable strength and degradation information under multiple working conditions and time periods, providing a basis for predicting rock fracture strength across scales and scenarios.

[0077] The specific steps for predicting the strength of rock samples include:

[0078] Strength prediction is performed based on the stress fingerprint characteristics of rock samples, the geometric blind zone characteristics of fractured rocks, the basic rock composition parameter characteristics, and the composition parameter characteristics of fractured rocks to obtain the predicted value of standard rock strength.

[0079] First, data from different sources are standardized in terms of dimensions, value range, and sampling frequency. The key moments, stiffness degradation rate, and energy index in the stress fingerprint, along with the crack tip curvature, bifurcation degree, and inter-surface fit in the geometric blind zone, as well as the mineral composition and pore structure of the two types of regions, are fed into a prediction model with physical constraints. The model takes the mechanical failure strength under standard conditions as the learning target and outputs the standard rock strength prediction value under a single working condition by fitting the coupling relationship between geometric control, material properties, and load evolution.

[0080] The multidimensional rock strength prediction values ​​corresponding to rock samples of different scales and the standard rock strength prediction values ​​are time-aligned to obtain a strength comparison result set. When aligning the multidimensional rock strength prediction values ​​corresponding to rock samples of different scales and the standard rock strength prediction values, the stress level, cycle number and equivalent damage variable in the loading process are used as references to map the prediction sequences of each scale and each working condition to a unified reference coordinate system. If necessary, interpolation and resampling correction are performed to correct the time drift caused by different sampling step sizes and different loading rates. The same failure criterion (e.g., the stiffness before the peak value drops to a certain proportion or the energy index reaches an inflection point) is used as the alignment anchor point to form a strength comparison result set that can be compared item by item, so that the prediction values ​​from micro, meso, and macro scales and multiple working conditions are comparable on the same scale.

[0081] The deviation amplitude between each predicted value in the strength comparison result set is calculated; when the deviation amplitude exceeds the preset deviation threshold, the corresponding predicted value is weighted and corrected to obtain the corrected rock strength prediction value; all predicted values ​​in the strength comparison result set are traversed and merged to obtain the fracture strength prediction result; the uncertainty of the fracture strength prediction result is estimated using the Bayesian inference algorithm, and the corresponding strength confidence interval is output.

[0082] The deviation between each pair of predicted values ​​is calculated, and this deviation is evaluated in conjunction with their respective quality indicators. When the deviation exceeds a preset deviation threshold, the predicted values ​​with larger deviations are weighted and corrected based on factors such as data uncertainty, scene fit, blind zone intensity indicativeness, and monitoring signal integrity. During the multi-value fusion process, results with stable data sources, smaller errors, and higher consistency with the actual crack scene receive higher weights, while results with more noise interference, larger scene deviations, or missing signals are assigned lower weights. After this differential correction, an updated intensity prediction sequence is formed, providing input for subsequent traversal fusion. The preset deviation threshold is set manually by professional technicians.

[0083] After weight correction, the intensity comparison result set is traversed and fused. Weighted aggregation under consistency constraints is used to summarize the multi-scale, multi-condition predictions into a single fracture intensity prediction result for the current rock scenario. During the fusion process, outlier predictions are suppressed or re-evaluated, and highly correlated repetitive information from the same source is dimensionality-reduced to ensure that the fused result reflects the common trends from different evidence without excessive weighting, ultimately forming a comprehensive intensity estimate that considers both local details and overall network effects. After obtaining the comprehensive fracture intensity prediction result, Bayesian inference is used to estimate uncertainty. Starting with residual statistics or prior experience under standard conditions, and combining the consistency and uncertainty of multiple sources of evidence in this experiment, a probability distribution of the prediction result is established, and an intensity range is given at a given confidence level. This range simultaneously integrates three sources: measurement noise, model structural uncertainty, and environmental variability, making the output intensity confidence interval both statistically significant and reflecting the safety margin that should be reserved in engineering, thus providing directly usable interval conclusions for design and safety assessment.

[0084] Example 2, a test-based rock fracture strength prediction system, see [link to example]. Figure 1 As shown, it includes:

[0085] The rock sample acquisition module includes a sample construction unit; the sample construction unit is used to acquire rock samples containing rock fissures based on the rock field scene; to acquire the rock fissure geometric feature parameters based on the rock samples to be optimized; and to divide the rock samples to be optimized into rock samples of different scales according to the rock fissure geometric feature parameters and the rock field scene.

[0086] The rock strength prediction module includes a standard strength prediction unit and a comprehensive strength prediction unit. The standard strength prediction unit is used to test rock samples of any scale under standard conditions. The specific steps are: acquiring rock sample images; dividing the rock sample into a basic rock sample region and a fractured rock sample region based on the rock sample images; extracting component data from the basic rock sample region and the fractured rock sample region respectively to obtain the basic rock component parameter characteristics and fractured rock component parameter characteristics; performing blind zone detection on the fractured rock sample region to obtain the geometric blind zone characteristics of the fractured rock; establishing a stress test combination containing different axial parameters; using the stress test combination to perform graded loading monitoring on the rock sample and collecting the stress fingerprint characteristics of the rock sample; predicting the strength of the rock sample to obtain the standard rock strength prediction value. The comprehensive strength prediction unit is used to construct a multimodal, multi-time-effect environment, repeating the above specific steps to test rock samples of different scales to obtain multidimensional rock strength prediction values; comparing the standard rock strength prediction value and the multidimensional rock strength prediction value to predict the strength, and outputting the fracture strength prediction result and the corresponding strength confidence interval of the rock sample.

[0087] It should be understood that those skilled in the art can make improvements or modifications based on the above description, and all such improvements and modifications should fall within the protection scope of the appended claims. Parts not described in detail in this specification are prior art known to those skilled in the art.

Claims

1. A test-based method for predicting rock fracture strength, characterized in that, Includes the following steps: Obtain rock samples containing rock fissures from a real-world rock scene; obtain geometric feature parameters of the rock fissures from the rock samples to be optimized; Based on the geometric characteristics of rock fractures and the actual rock scene, the rock samples to be optimized are divided into rock samples of different scales. The specific steps for testing rock samples of any scale under standard conditions are as follows: acquiring rock sample images; dividing the rock sample into basic rock sample region and fractured rock sample region based on the rock sample images; extracting component data from the basic rock sample region and fractured rock sample region respectively to obtain basic rock component parameter characteristics and fractured rock component parameter characteristics. Blind zone detection was performed on the fractured rock sample area to obtain the geometric blind zone characteristics of the fractured rock; a stress test combination with different axial parameters was established; the stress test combination was used to perform graded loading monitoring on the rock sample to collect the stress fingerprint characteristics of the rock sample; the strength of the rock sample was predicted to obtain the predicted value of standard rock strength. Construct a multimodal and multi-time-dependent environment, repeat the above specific steps to test rock samples of different scales, and obtain multidimensional rock strength prediction values; compare the standard rock strength prediction values ​​and the multidimensional rock strength prediction values ​​to predict the strength, and output the fracture strength prediction results and corresponding strength confidence intervals of the rock samples; Image segmentation algorithms were used to process rock sample images to obtain fracture probability maps and fracture uncertainty maps; the initial skeleton of the rock fracture region was extracted based on the fracture probability maps and fracture uncertainty maps. Calculate the signed distance field and local fracture width of the initial framework of the rock fracture region; construct the inner and outer boundaries of the rock fracture based on the signed distance field and local fracture width; Boundary buffer zones are identified based on the inner and outer boundaries of rock fissures; several sets of candidate boundaries are generated within the boundary buffer zones using an optimization function; the objectives of the optimization function include boundary smoothness constraints and fissure texture gradient information. The specific steps for blind zone detection in fractured rock sample areas include: A candidate blind zone seed set is generated in the fractured rock sample region. The candidate blind zone seed set includes geometric seeds, fitting seeds, and precursor seeds. For each type of candidate blind zone seed, a candidate blind zone ring is divided according to the final fracture region boundary. A preliminary blind zone block is generated in the candidate blind zone ring according to the maximum coverage-minimum overlap criterion. In the preliminary blind zone block, the selection weight of each preliminary blind zone block is calculated according to the fracture direction concentration, area density, and scene label. Prioritize all the initial blind blocks according to their selection weights and output the final filtered blind blocks; perform feature detection on each filtered blind block to obtain the geometric features of the blind block; By fusing the geometric features of the blind blocks on the rock sample, the geometric blind features of the fractured rock are obtained.

2. The test-based method for predicting rock fracture strength according to claim 1, characterized in that, The specific steps for classifying rock samples into different scales based on the geometric characteristics of rock fractures and the actual rock scene include: A fracture geometry-scene feature set is constructed based on the actual rock scene and the geometric feature parameters of the rock fracture; microscale, mesoscale and macroscale specimens are set according to the fracture geometry-scene feature set; the minimum specimen size at each scale is determined according to the geometric feature parameters of the rock fracture.

3. The test-based method for predicting rock fracture strength according to claim 2, characterized in that, The specific steps for dividing a rock sample into a basic rock sample region and a fractured rock sample region based on the rock sample image include: The candidate boundaries are offset and corrected to obtain the corrected candidate boundaries; the comprehensive similarity between the corrected candidate boundaries and the inner and outer boundaries of the rock fractures is calculated, and the optimal candidate boundaries are selected as the final fracture region boundaries; the rock samples are divided into basic rock sample regions and fractured rock sample regions according to the final fracture region boundaries.

4. The test-based method for predicting rock fracture strength according to claim 3, characterized in that, The specific steps for using a stress testing assembly to monitor graded loading of rock samples and collect the stress fingerprint characteristics of the rock samples include: The axial parameters in the stress parameter combination include the number of axial forces, the axial direction, and the axial load. The rock samples were subjected to strength tests by selecting different combinations of axial stress parameters in a stepwise manner to obtain the stress level characteristics of rock fractures. At the same time, at the end of each strength test, the rock samples were subjected to perturbation loading using small-amplitude perturbation cycles to extract the evolution characteristics of normal stiffness and shear stiffness. After all stress parameter combination strength tests are completed, the collected rock fracture stress level characteristics, normal stiffness evolution characteristics, and shear stiffness evolution characteristics are horizontally aligned and feature extracted to obtain the stress fingerprint characteristics of the rock sample.

5. The test-based method for predicting rock fracture strength according to claim 4, characterized in that, The specific steps for predicting the strength of rock samples include: Strength prediction is performed based on the stress fingerprint characteristics of rock samples, the geometric blind zone characteristics of fractured rocks, the basic rock composition parameter characteristics, and the composition parameter characteristics of fractured rocks to obtain the predicted value of standard rock strength. By aligning the predicted values ​​of multidimensional rock strength and standard rock strength of rock samples at different scales over time, a set of strength comparison results is obtained. The deviation amplitude between each predicted value in the strength comparison result set is calculated; when the deviation amplitude exceeds the preset deviation threshold, the corresponding predicted value is weighted and corrected to obtain the corrected rock strength prediction value; all predicted values ​​in the strength comparison result set are traversed and merged to obtain the fracture strength prediction result; the uncertainty of the fracture strength prediction result is estimated using the Bayesian inference algorithm, and the corresponding strength confidence interval is output.

6. A test-based rock fracture strength prediction system, characterized in that, The system employs a test-based rock fracture strength prediction method according to any one of claims 1-5, comprising: The rock sample acquisition module includes a sample construction unit; the sample construction unit is used to acquire rock samples containing rock fissures based on the rock field scene; to acquire the rock fissure geometric feature parameters based on the rock samples to be optimized; and to divide the rock samples to be optimized into rock samples of different scales according to the rock fissure geometric feature parameters and the rock field scene. The rock strength prediction module includes a standard strength prediction unit and a comprehensive strength prediction unit. The standard strength prediction unit is used to test rock samples of any scale under standard conditions. The specific steps are: acquiring rock sample images; dividing the rock sample into a basic rock sample region and a fractured rock sample region based on the rock sample images; extracting component data from the basic rock sample region and the fractured rock sample region respectively to obtain the basic rock component parameter characteristics and fractured rock component parameter characteristics; performing blind zone detection on the fractured rock sample region to obtain the geometric blind zone characteristics of the fractured rock; establishing a stress test combination containing different axial parameters; using the stress test combination to perform graded loading monitoring on the rock sample and collecting the stress fingerprint characteristics of the rock sample; predicting the strength of the rock sample to obtain the standard rock strength prediction value. The comprehensive strength prediction unit is used to construct a multimodal, multi-time-effect environment, repeating the above specific steps to test rock samples of different scales to obtain multidimensional rock strength prediction values; comparing the standard rock strength prediction value and the multidimensional rock strength prediction value to predict the strength, and outputting the fracture strength prediction result and the corresponding strength confidence interval of the rock sample.

Images