Three-section surrounding rock structure plane sample construction method and system of physical adaptive soft boundary

Abstract

A physical self-adaptive soft boundary three-section surrounding rock structure surface sample construction method and system, obtains while drilling parameters and structure surface label data, and delimits structure surface event center interval; calculates multidimensional mechanical response characteristics in drilling depth dimension, front and rear influence radius; divides front stable, structure surface influence, and rear recovery three sections; extracts various statistical and differential characteristics in sections, and encapsulates into three-section sample units; generates boundary probability label and probability mask to obtain soft boundary supervision label; gives primary and secondary weights to multiple structure surfaces, and completes sample quality reliability evaluation. Based on data input module, event center interval determination module, mechanical response characteristic calculation module, influence radius calculation module, three-section division module, differential response characteristic calculation module, sample encapsulation module, self-adaptive soft boundary label generation module, primary and secondary label and sample quality evaluation module, sample library output module cooperates to form a construction system. The present application can improve the quality of training samples.

Description

Technical Field

[0001] This invention belongs to the technical fields of underground engineering surrounding rock detection, drilling measurement data processing, borehole viewing image analysis, training sample construction and intelligent recognition dataset generation, specifically involving a method and system for constructing a three-segment surrounding rock structure surface sample with physical adaptive soft boundary. Background Technology

[0002] In underground engineering projects such as coal mine roadways and tunnels, the surrounding rock typically contains structural surfaces such as fissures, delamination, and rock strata interfaces. Their spatial distribution and extent directly affect the integrity and bearing capacity of the surrounding rock, serving as crucial evidence for evaluating roadway surrounding rock stability and designing support parameters. Therefore, rapidly and accurately identifying the types and extent of these structural surfaces in the surrounding rock is of great significance for ensuring safe construction of underground engineering projects.

[0003] Currently, drilling parameters and borehole inspection data are widely used in in-situ detection and intelligent identification in underground engineering. Drilling detection technology continuously reflects the interaction between the drill bit and the surrounding rock by collecting drilling response parameters such as rotational speed, drilling rate, thrust, and torque during the drilling process. When the drill bit crosses surrounding rock structural surfaces such as fractures and delamination, drilling parameters often exhibit abrupt changes or abnormal fluctuations. Therefore, the development of surrounding rock structural surfaces can be determined by analyzing the drilling data. Borehole inspection data can visually represent the starting and ending positions, visible width, opening state, and degree of boundary ambiguity of structural surfaces, providing visual evidence for structural surface annotation. These two technologies provide information from the dimensions of mechanical response and visual representation, respectively. Their combination can provide a multi-source training data foundation for structural surface identification models.

[0004] However, existing drilling surface identification samples are typically constructed using fixed-length sliding windows or manual truncation, making it difficult to accurately represent the differences in response before, during, and after the drill bit enters the surface. For surfaces with a certain thickness and transition zone, using only a single hard boundary or fixed window as a label can easily lead to overly discrete boundary labels, insufficient deep learning of the start and end points of the surface by the model, and difficulty in controlling sample quality.

[0005] Furthermore, downhole drilling conditions are complex, and the effectiveness of structural surface boundary recognition is often affected by factors such as drill bit size, drilling speed and torque fluctuations, borehole wall imaging quality, and the ambiguity of structural surface boundaries. When different structural surfaces overlap within the same window, it can also cause difficulties in distinguishing between primary and secondary structural surfaces. Therefore, there is an urgent need for a method that can construct three-segment samples around the event center interval of a structural surface, adaptively generate soft boundary labels based on physical response and visual attributes, and quantitatively evaluate the sample quality. Summary of the Invention

[0006] To address the problems existing in the prior art, this invention provides a method and system for constructing three-segment surrounding rock structural surface samples with physically adaptive soft boundaries. This method can improve the boundary representation accuracy, sample quality controllability, and engineering adaptability of training samples for surrounding rock structural surface recognition models. It is suitable for constructing datasets for intelligent identification of surrounding rock structural surfaces in mine roadways, tunnels, and underground engineering projects during drilling. This system effectively solves the problems of imprecise structural surface boundary representation, overlapping labels of multiple structural surfaces, and uncontrollable sample quality in existing technologies, providing a reliable system-level solution for intelligent identification of surrounding rock structural surfaces in underground engineering projects.

[0007] To achieve the above objectives, this invention provides a method for constructing a three-segment surrounding rock structural surface sample with physically adaptive soft boundaries, comprising the following steps: Step 1: Data Acquisition and Determination of Event Center Interval on Structural Surface; Obtain the drilling parameter sequence and structural surface annotation results under a unified coordinate system, and use the interval defined by the start and end depths of the structural surface as the event center interval; Step 2: Calculation of mechanical response feature sequence; Based on the drilling parameter sequence, calculate the multi-dimensional mechanical response features distributed along the borehole depth direction; Step 3: Physical adaptive influence radius calculation; calculate the disturbance and recovery intensity in the neighborhood of the starting and ending depths of the structural surface respectively, and integrate multi-source parameters including drill bit diameter, drilling speed, fluctuation amplitude and visual attributes to adaptively determine the front influence radius and the back influence radius; Step 4: Three-segment spatial division based on influence radius; using the central interval of the structural surface event as the anchor point, adaptively divide the space into a pre-stabilized segment, a structural surface influence segment, and a post-recovery segment using the pre- and post-influence radii, and perform special processing for the case of composite structural surfaces; Step 5: Three-segment feature extraction and sample unit encapsulation; extract the drilling statistical features and mechanical response features in the pre-stabilization segment, the structural surface influence segment, and the post-recovery segment respectively, calculate the inbound and outbound differential response features, and encapsulate them into event-centered three-segment sample units according to the preset template; Step 6: Adaptive soft boundary label generation; Based on physical parameters such as the influence radius, determine the depth extension radius of the probability label, generate the starting and ending boundary probability labels, and construct the structural surface existence probability mask to form a soft boundary supervision label that conforms to the physical nature of the structural surface transition; Step 7: Generation of primary and secondary labels for multiple structural surfaces and evaluation of sample quality; For cases where multiple structural surfaces exist within a single sample unit, the primary and secondary label weights of each structural surface are calculated based on multi-source information to distinguish between primary and secondary structural surfaces; At the same time, the reliability of the sample is evaluated by comprehensively considering information such as image quality, parameter completeness rate, and annotation confidence.

[0008] As a preferred embodiment, the process of data acquisition and determination of the central interval of structural surface events in step 1 is as follows: S11: Obtain the drilling parameter sequence and structural surface annotation results under a unified borehole depth coordinate system; the drilling parameter sequence includes one or more of the following: drilling depth, drilling time, drilling speed, rotational speed, thrust, torque, drill pipe displacement, drilling vibration, and drilling rig power; the structural surface annotation results include at least the structural surface type, the structural surface starting depth, and the structural surface ending depth. S12: The depth range defined by the starting depth and ending depth of the structural surface is used as the central interval of the structural surface event. S13: Obtain the borehole inspection visual attribute fields, which include one or more of the following: visible width of the structural surface, ambiguity of the structural surface boundary, continuity of the structural surface, and open state of the structural surface.

[0009] As a preferred embodiment, the mechanical response characteristic sequence calculation process in step 2 is as follows: The mechanical response characteristic sequence distributed along the borehole depth direction is calculated based on the drilling parameter sequence; the mechanical response characteristics include multiple of the following: energy per unit footage, mechanical specific energy, torque-to-thrust ratio, drilling speed residual, energy mutation coefficient, drilling stiffness change rate, thrust fluctuation intensity, and torque fluctuation intensity.

[0010] As a preferred embodiment, the calculation process for the physical adaptive influence radius in step 3 is as follows: S31: Calculate the boundary disturbance intensity and the boundary recovery intensity in the neighborhood of the starting depth and the neighborhood of the ending depth of the structural surface, respectively. S32: Based on at least three parameters among drill bit diameter, current drilling speed, sampling time interval, thrust fluctuation amplitude, torque fluctuation amplitude, visible width of structural surface and ambiguity of structural surface boundary, and combined with the boundary disturbance intensity or the boundary recovery intensity, determine the pre-influence radius and post-influence radius of the event center interval of the structural surface, respectively.

[0011] As a preferred option, in step 4, the three-segment spatial partitioning process based on the radius of influence is as follows: Using the central interval of the structural surface event as the anchor point, and combining the pre-influence radius and the post-influence radius, the borehole depth direction is divided into the pre-stabilization section, the structural surface influence section, and the post-recovery section. The endpoint of the pre-stabilization segment does not exceed the initial depth of the structural surface, and its starting point is determined by subtracting the pre-influence radius from the initial depth of the structural surface. The structural surface influence segment covers the area from the initial depth of the structural surface to the final depth of the structural surface and its boundary transition range. Its starting point is the initial depth of the structural surface minus the pre-influence radius, and its ending point is the final depth of the structural surface plus the subsequent influence radius. The starting point of the subsequent recovery segment is not less than the final depth of the structural surface, and its ending point is determined by adding the subsequent influence radius to the final depth of the structural surface. The actual depth boundaries of each segment are adaptively adjusted based on whether there are other structural surface annotation intervals in the neighborhood. If there are other structural surface annotation intervals within the current stable segment or the subsequent recovery segment, the following processing is performed: If the distance between the other structural surface annotation intervals and the event center interval of the current structural surface is less than the corresponding influence radius, then the preceding influence radius or the subsequent influence radius of the current structural surface is shortened so that the boundary of the preceding stable segment or the subsequent recovery segment ends at the boundary of the other structural surface annotation intervals; or the three-segment sample unit of the current event center is marked as a composite structural surface sample to indicate that the sample contains complex response information of multiple structural surfaces.

[0012] As a preferred embodiment, the process of three-segment feature extraction and sample unit encapsulation in step 5 is as follows: S51: Extract the drilling statistical features and mechanical response features in the pre-stabilization section, the structural plane influence section, and the post-recovery section respectively; the drilling statistical features include the mean, variance, maximum and minimum values ​​of drilling speed, rotational speed, thrust and torque in each section; the mechanical response features are the statistical values ​​of each mechanical response feature in step 2 in the corresponding section, including the mean, variance and gradient; S52: Calculate the inbound differential response characteristics and the outbound differential response characteristics respectively; the inbound differential response characteristics satisfy the formula: The out-of-bounds difference response characteristics satisfy: ;in, The feature vector of the inbound differential response. The out-of-bounds difference response feature vector , and These are the feature vectors for the pre-stabilization segment, the structural plane-affected segment, and the post-recovery segment, respectively. Each feature vector is composed of the drilling-while-drilling statistical features and mechanical response features within the corresponding segment. This indicates normalization processing; S53: According to the preset sample field template, the depth boundary, intra-segment feature vector, inbound differential response feature, outbound differential response feature, structural surface type, inbound influence radius, outbound influence radius, and borehole observation visual attribute fields of the pre-stabilized segment, structural surface influence segment, and post-recovery segment are encapsulated into an event center three-segment sample unit. S54: Bind the structural surface type label, structural surface start depth information, structural surface end depth information and sample interval information to the three-segment sample unit of the event center to form the basic data sample for training, verification or sample library management of the surrounding rock structural surface recognition model.

[0013] As a preferred embodiment, the adaptive soft boundary label generation process in step 6 is as follows: S61: Obtain the three-segment sample unit of the event center and its corresponding structural surface start depth, structural surface end depth, preceding influence radius and subsequent influence radius; S62: Determine the depth extension radius of the starting boundary probability label based on at least three parameters: the preceding influence radius, the following influence radius, the drill bit diameter, the current drilling speed, the thrust fluctuation amplitude, the torque fluctuation amplitude, the visually visible width of the structure surface, and the ambiguity of the structure surface boundary. and the depth expansion radius of the endpoint boundary probability label ; S63: Based on the initial depth of the structural surface Depth expansion radius of the starting boundary probability label Generate starting point boundary probability labels Based on the structural plane termination depth and the depth expansion radius of the endpoint boundary probability label Generate endpoint boundary probability labels ; When the development length of the structural plane is less than and When summed, the depth expansion radii of the starting and ending boundary probability labels are proportionally compressed according to the development length of the structural surface. The compressed expansion radii are respectively... and ; S64: Generate a probability mask for the existence of structural surfaces based on the depth range of the structural surfaces, the probability labels of the starting and ending boundaries. The probability mask for the existence of the structural surface satisfies the following conditions: the existence probability of the central region of the structural surface is not less than 0.8, the existence probability of the transition region of the structural surface boundary varies continuously between 0.2 and 0.8, and the existence probability of the stable surrounding rock region outside the structural surface is not higher than 0.2; when the probability labels of the starting boundary and the ending boundary overlap, the existence probability of the overlapping region is limited so that the existence probability of the structural surface does not exceed 1. S65: Bind the starting boundary probability label, the ending boundary probability label, and the structural surface existence probability mask to the three-segment sample unit of the event center to form a soft boundary supervision label for learning the structural surface boundary of the borehole depth sequence.

[0014] As a preferred option, in step 7, the generation of primary and secondary labels for multiple structural surfaces and the evaluation of sample quality are performed. S71: Obtain the depth range of multiple structural surfaces contained in the three-segment sample unit of the event center, as well as their corresponding visually visible width, mechanical response anomaly intensity, depth overlap ratio with the sample unit, and annotation confidence. S72: Calculate the normalized visual width for each structural plane depth interval. Normalized value of abnormal mechanical response intensity The depth overlap ratio with the sample cell and labeled confidence normalized value The depth overlap ratio is the ratio of the overlap length between the depth interval of the structural surface and the depth interval of the current sample unit to the total length of the depth interval of the structural surface, and the value range is [0,1]. S73: Calculate the primary and secondary label weights for each structural surface based on visually visible width, mechanical response anomaly intensity, depth overlap ratio, and label confidence level; Primary and secondary label weights of each structural surface satisfy: ,in For the first Normalized value of the visually visible width of each structural surface; For the first Normalized values ​​of the mechanical response anomaly intensity corresponding to each structural surface; For the first The depth overlap ratio between each structural surface and the sample unit; For the first Normalized confidence values ​​of the labeled structural surfaces; to Preset weighting coefficients; This indicates all samples within the same sample unit. Summing of all structural surfaces; S74: Determine the primary and secondary structural surface labels based on the primary and secondary label weights; determine the structural surface with the largest primary and secondary label weights as the primary structural surface label, and determine other structural surfaces with primary and secondary label weights not lower than the preset secondary label threshold as secondary structural surface labels; when the primary and secondary label weights of all secondary structural surfaces are lower than the preset secondary label threshold, only the primary structural surface label is retained; bind the primary and secondary structural surface labels to the three-segment sample unit of the event center; S75: Obtain the three-segment sample unit of the event center and its corresponding borehole peep image clarity, drilling parameter integrity rate, structural surface boundary label confidence and depth correction residual; S76: Calculate sample reliability based on borehole observation image clarity, drilling parameter integrity rate, structural boundary label confidence level, and depth correction residual. ,in This represents the normalized value for the sharpness of the borehole peep view image; The percentage of complete drilling parameters; Label the confidence normalization values ​​for the structural surface boundaries; For depth correction, the normalized value of the residual; to The preset weighting coefficients satisfy the following conditions: ; S77: Determine the sample weight or verification status of the event center three-segment sample unit based on the sample reliability; when the sample reliability is lower than the preset reliability threshold, reduce the sample weight of the event center three-segment sample unit by multiplying the original weight by the ratio of the sample reliability to the reliability threshold, or mark the event center three-segment sample unit as a sample to be verified; when the sample reliability is higher than the preset high reliability threshold, mark the event center three-segment sample unit as a high-confidence training sample; bind the sample reliability, sample weight or verification status to the event center three-segment sample unit, and finally output a complete event center three-segment sample unit bound with structure surface type label, soft boundary supervision label, primary and secondary label weight, sample reliability and sample weight.

[0015] Compared with the prior art, the present invention has the following beneficial effects: 1. This invention uses the central interval of a structural surface event as an anchor point, adaptively dividing it into a pre-stabilization segment, a structural surface influence segment, and a post-recovery segment. It also introduces inbound and outbound differential response features, enabling a complete characterization of the dynamic mechanical response changes throughout the entire process of the drill bit entering, traversing, and exiting the structural surface. The three-segment sample unit not only retains complete drilling state information before and after the structural surface influence but also enhances the response abrupt change signal at the structural surface boundary through differential features, significantly improving the training samples' ability to express structural surface events.

[0016] 2. This invention determines the spatial boundary of a three-segment sample by using a physically adaptive influence radius. It integrates multiple parameters, including drill bit diameter, drilling speed, sampling time interval, thrust fluctuation amplitude, torque fluctuation amplitude, visually visible width of the structural surface, and structural surface boundary ambiguity, into quantitative calculation formulas for the pre-influence radius and post-influence radius. This adaptive partitioning mechanism allows the sample boundary to be dynamically adjusted according to the actual physical characteristics of the structural surface and drilling conditions, overcoming the problem of insufficient adaptability of fixed window or fixed radius partitioning to different types of structural surfaces. This makes the spatial partitioning of the three-segment sample more closely match the actual influence range when the drill bit traverses the structural surface.

[0017] 3. By constructing the boundary differential response features and the boundary differential response features, this invention quantifies the characteristic change patterns of the drill bit when it enters the structural plane from the stable surrounding rock and returns to the stable surrounding rock from the structural plane into a computable differential vector. This provides the identification model with structural plane boundary discrimination features with clear physical meaning, effectively enhancing the model's positioning accuracy of the structural plane's start and end depths.

[0018] 4. This invention replaces traditional hard boundary labels with a soft boundary supervision label system consisting of a starting boundary probability label, an ending boundary probability label, and a structural surface existence probability mask. The depth extension radius of the soft boundary label is adaptively determined by physical parameters such as the preceding influence radius, the following influence radius, the drill bit diameter, the drilling speed, and the fluctuation amplitude. This transforms the structural surface boundary from discrete depth points into a continuous probability distribution, which better conforms to the fuzzy transition characteristics of geological structural surface boundaries caused by drill bit size effects, gradual rock mass characteristics, and drilling disturbance transmission. This effectively reduces the boundary positioning errors caused by fixed windows and hard boundary labels.

[0019] 5. This invention uses a primary and secondary label weighting calculation mechanism to comprehensively consider information from four dimensions: visually visible width, mechanical response anomaly intensity, depth overlap ratio, and label confidence. This allows for the quantitative ranking of the importance of multiple structural surfaces within the same sample unit, effectively distinguishing between primary and secondary structural surfaces. This solves the problem of sample label confusion when multiple structural surfaces coexist, and improves the targeting of recognition model training in multi-structural-surface scenarios.

[0020] 6. This invention employs a sample reliability evaluation mechanism that integrates information from four dimensions: borehole observation image clarity, drilling parameter integrity rate, structural surface boundary label confidence level, and depth correction residual. This mechanism quantifies and scores the reliability of each sample unit, assigning sample weights or marking verification status accordingly. This mechanism can identify low-quality samples, samples requiring verification, and high-confidence training samples, enabling the model training process to prioritize high-quality samples and suppress the negative impact of low-quality samples. This is beneficial for improving the training data quality of the surrounding rock structural surface recognition model.

[0021] 7. This invention provides special handling for composite structural surfaces within the pre-stabilization segment or post-recovery segment that have other structural surface annotation intervals. By shortening the corresponding influence radius or marking them as composite structural surface samples, it avoids boundary conflicts of multiple structural surface events in the sample space, ensuring the effectiveness and stability of the sample construction method in densely developed multi-structural-surface areas.

[0022] 8. This invention binds soft boundary supervision labels, primary and secondary label weights, sample reliability and sample weights to event center three-segment sample units, outputting high-quality training samples with complete structure, rich annotations and controllable quality. These samples can be directly used for training, verification and sample library management of surrounding rock structure surface recognition models, providing a reliable data foundation for intelligent recognition of surrounding rock structure surfaces in underground engineering.

[0023] This method uses the central interval of structural surface events as anchor points, divides the sample space into three segments through physical adaptive influence radius, and introduces soft boundary probability labels, primary and secondary label weights, and sample reliability evaluation. It can improve the boundary representation accuracy, sample quality controllability, and engineering adaptability of training samples for surrounding rock structural surface recognition models. It effectively solves the problems of imprecise structural surface boundary representation, insufficient representation of structural surface entry, crossing, and recovery processes in training samples, lack of physical adaptive basis for soft boundary labels, overlapping of multiple structural surface labels, and difficulty in quantifying sample reliability in existing technologies. It is suitable for the construction of datasets for intelligent identification of surrounding rock structural surfaces during drilling in mine roadways, tunnels, and underground engineering.

[0024] This invention also provides a three-segment surrounding rock structural surface sample construction system for physically adaptive soft boundaries, used to implement a method for constructing three-segment surrounding rock structural surface samples with physically adaptive soft boundaries, including: The data input module is used to obtain the drilling parameter sequence and structural surface annotation results under a unified borehole depth coordinate system; The event center interval determination module is used to define the depth interval of the structural surface as the event center interval of the structural surface, which is defined by the starting depth and ending depth of the structural surface. The mechanical response characteristic calculation module is used to calculate the mechanical response characteristic sequence distributed along the borehole depth direction based on the drilling parameter sequence; The influence radius calculation module is used to determine the pre-influence radius and post-influence radius of the event center interval of the structural surface based on at least three parameters among the drill bit diameter, current drilling speed, sampling time interval, thrust fluctuation amplitude, torque fluctuation amplitude, visible width of the structural surface and ambiguity of the structural surface boundary, combined with the boundary disturbance intensity or the boundary recovery intensity. The three-segment division module is used to divide the borehole depth direction into a pre-stabilized segment, a structural surface influence segment, and a post-recovery segment, with the structural surface event center interval as the anchor point and in combination with the pre-influence radius and the post-influence radius. The differential response characteristic calculation module is used to calculate the inbound differential response characteristics and the outbound differential response characteristics, respectively. The sample encapsulation module is used to encapsulate the depth boundaries, intra-segment feature vectors, inbound differential response features, outbound differential response features, structural surface type, inbound influence radius, outbound influence radius, and borehole observation visual attribute fields of the pre-stabilized segment, structural surface influence segment, and post-recovery segment into event-centered three-segment sample units according to the preset sample field template, and bind them with structural surface type labels, structural surface start depth information, structural surface end depth information, and sample interval information; An adaptive soft boundary label generation module is used to determine the depth extension radius of the starting boundary probability label and the depth extension radius of the ending boundary probability label based on at least three parameters: the preceding influence radius, the following influence radius, the drill bit diameter, the current drilling speed, the thrust fluctuation amplitude, the torque fluctuation amplitude, the visually visible width of the structural surface, and the structural surface boundary ambiguity. It generates a starting boundary probability label based on the structural surface starting depth and the depth extension radius of the starting boundary probability label, and generates an ending boundary probability label based on the structural surface ending depth and the depth extension radius of the ending boundary probability label. It generates a structural surface existence probability mask based on the structural surface depth range, the starting boundary probability label, and the ending boundary probability label. Finally, it binds the starting boundary probability label, the ending boundary probability label, and the structural surface existence probability mask to the event center three-segment sample unit to form a soft boundary supervision label. The primary and secondary labeling and sample quality evaluation module is used to calculate the primary and secondary label weights and determine the primary and secondary structural surface labels when there are multiple structural surface depth intervals in a three-segment sample unit at the event center. This is done based on the visually visible width of each structural surface, the intensity of mechanical response anomalies, the depth overlap ratio with the sample unit, and the label confidence level. Simultaneously, it calculates the sample reliability based on the borehole peep image clarity, drilling parameter integrity rate, structural surface boundary label confidence level, and depth correction residual, and determines the sample weight or verification status of the three-segment sample unit at the event center based on the sample reliability. The sample library output module is used to output complete event-centric three-segment sample units bound with structural surface type labels, soft boundary supervision labels, primary and secondary label weights, sample reliability, and sample weights.

[0025] In this invention, the data input module ensures spatial consistency in subsequent sample construction from the data source. The event center interval determination module allows sample templates to be directly established around the structural surface occurrence interval, enhancing the correspondence between samples and structural surface events. The mechanical response feature calculation module constructs a mechanical state characterization sequence distributed along the drilling depth direction, providing a rich mechanical input foundation for subsequent intra-segment feature extraction and differential response feature calculation of the three-segment samples. The influence radius calculation module enables the physical adaptive determination of the spatial boundary of the three-segment samples, allowing the influence radius to be dynamically adjusted according to the actual physical characteristics of the structural surface and drilling conditions, and reflecting the asymmetry of disturbance levels at the structural surface inlet and outlet. The three-segment division module fully depicts the evolution of drilling state throughout the entire process of the drill bit entering, traversing, and exiting the structural surface. Simultaneously, it can detect complex structural surface situations and perform adaptive boundary adjustment or marking, avoiding boundary conflicts between multiple structural surface events in the sample space. The differential response feature calculation module provides the identification model with structural surface boundary discrimination features with clear physical meaning. Based on the sample encapsulation module, complete and directly usable foundational data samples for model training or sample library management can be generated. The adaptive soft boundary label generation module transforms structural surface boundaries from discrete depth points to continuous probability distributions, allowing the ambiguity range of labels to be adaptively determined based on the actual physical characteristics of the structural surface. Simultaneously, it provides soft boundary supervision signals with clear physical basis for the identification model. The primary and secondary label and sample quality evaluation modules quantify and rank the importance of multiple structural surfaces within the same sample unit, effectively solving the problem of sample label confusion when multiple structural surfaces coexist. Furthermore, it provides differentiated quality weights or verification labels for each sample unit, enabling the model training process to focus on high-quality samples and suppress the negative impact of low-quality samples. The sample library output module provides structurally complete, richly labeled, and quality-controllable data support for the training, validation, and sample library management of the surrounding rock structural surface identification model.

[0026] This system, through a modular collaborative architecture, enables the complete construction of training samples for the identification of surrounding rock structural surfaces, from physical features to high-quality data samples. It effectively solves the problems of imprecise structural surface boundary representation, overlapping labels of multiple structural surfaces, and uncontrollable sample quality in existing technologies, providing a reliable system-level solution for intelligent identification of surrounding rock structural surfaces in underground engineering. Attached Figure Description

[0027] Figure 1 This is a flowchart of the construction method in this invention; Figure 2 This is a schematic diagram of the physical adaptive influence radius calculation and the three-segment spatial partitioning based on the influence radius in this invention; Figure 3 This is a schematic diagram of the adaptive soft boundary label generation in this invention. Detailed Implementation

[0028] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0029] Example 1:

[0030] like Figure 1 As shown, this embodiment provides a method for constructing a three-segment surrounding rock structural surface sample with physically adaptive soft boundaries, including the following steps: Step 1: Data Acquisition and Determination of Event Center Interval on Structural Surface; Obtain the drilling parameter sequence and structural surface annotation results under a unified coordinate system, and use the interval defined by the start and end depths of the structural surface as the event center interval to establish physical anchor points for subsequent processing; As a preferred approach, the process of data acquisition and determination of the central interval of structural surface events is as follows: S11: Obtain the drilling parameter sequence and structural surface annotation results under a unified borehole depth coordinate system; the drilling parameter sequence includes one or more of the following: drilling depth, drilling time, drilling speed, rotational speed, thrust, torque, drill pipe displacement, drilling vibration, and drilling rig power; the structural surface annotation results include at least the structural surface type, structural surface starting depth, and structural surface ending depth, and are derived from one or more of the following: borehole inspection calibration results, borehole wall unfolding chart annotation results, or manual verification results; S12: The depth interval defined by the starting depth and ending depth of the structural surface is used as the central interval of the structural surface event; the central interval of the event directly builds a sample template around the occurrence interval of the structural surface, so that the sample construction process corresponds to the structural surface event itself, which is different from the conventional fixed sliding window method. S13: Obtain the borehole inspection visual attribute fields. The borehole inspection visual attribute fields include one or more of the following: visible width of the structural surface, ambiguity of the structural surface boundary, continuity of the structural surface, and open state of the structural surface. These are used for subsequent influence radius calculation and soft boundary label generation. This technical solution establishes a sample construction foundation anchored by the structural surface events themselves, by acquiring the drilling parameter sequence and structural surface annotation results under a unified borehole depth coordinate system and using the interval defined by the start and end depths of the structural surface as the event center interval. This differs from the conventional fixed sliding window method. Simultaneously, by acquiring visual attribute fields such as the visible width of the structural surface, boundary ambiguity, continuity, and opening state, crucial multi-source input information is provided for subsequent calculation of the physically adaptive influence radius and soft boundary label generation. This enables the sample construction process to integrate borehole inspection visual priors and drilling mechanical response data.

[0031] Step 2: Calculation of mechanical response feature sequence; Based on the drilling parameter sequence, calculate the multi-dimensional mechanical response features distributed along the drilling depth direction to characterize the mechanical state changes when the drill bit passes through the structural surface; As a preferred method, the process for calculating the mechanical response characteristic sequence is as follows: The mechanical response feature sequence distributed along the borehole depth direction is calculated based on the drilling parameter sequence. This mechanical response feature reflects the changes in the mechanical properties of the surrounding rock during drilling and forms the basis for subsequent feature extraction and boundary / outbound differential response feature calculation. The mechanical response feature includes multiple parameters such as energy per unit footage, mechanical specific energy, torque-to-thrust ratio, drilling speed residual, energy mutation coefficient, drilling stiffness variation rate, thrust fluctuation intensity, and torque fluctuation intensity. Specifically, energy per unit footage is the ratio of drilling energy to the corresponding footage; mechanical specific energy is the energy required to break a unit volume of rock; torque-to-thrust ratio is the ratio of torque to thrust; drilling speed residual is the deviation between the actual drilling speed and the theoretical drilling speed; the energy mutation coefficient is used to detect instantaneous changes in energy; the drilling stiffness variation rate reflects the instantaneous change in the relationship between thrust and drilling speed; and thrust fluctuation intensity and torque fluctuation intensity represent the degree of variation of the corresponding parameters within a local window. In this technical solution, multi-dimensional mechanical response features, including energy per unit footage, mechanical specific energy, torque-to-thrust ratio, drilling rate residual, energy mutation coefficient, drilling stiffness change rate, thrust fluctuation intensity, and torque fluctuation intensity, are extracted from the drilling parameter sequence to construct a mechanical state characterization sequence distributed along the borehole depth direction. Each feature has a clear physical meaning—energy features reflect rock breaking efficiency and energy mutation trends, force-torsion features reflect the interaction between the drill bit and the surrounding rock, and fluctuation features reflect the stability of the drilling process. The synergistic effect of these multi-dimensional features provides a rich and reliable mechanical input foundation for subsequent intra-segment feature extraction and boundary / outbound differential response feature calculation of the three-segment samples, enabling the samples to characterize the dynamic mechanical response changes of the drill bit throughout the entire process of traversing the structural surface from multiple perspectives.

[0032] Step 3: Physical adaptive influence radius calculation; calculate the disturbance and recovery intensity in the neighborhood of the starting and ending depths of the structural surface respectively, and integrate multi-source parameters including drill bit diameter, drilling speed, fluctuation amplitude and visual attributes to adaptively determine the front influence radius and the back influence radius; As a preferred method, the process for calculating the physically adaptive influence radius is as follows: S31: Calculate the boundary disturbance intensity and boundary recovery intensity in the neighborhood of the starting depth and ending depth of the structural plane, respectively; wherein, the boundary disturbance intensity reflects the degree of change in the mechanical response characteristics when the drill bit enters the structural plane from the stable surrounding rock, and the boundary recovery intensity reflects the degree of recovery of the mechanical response characteristics when the drill bit exits the structural plane back to the stable surrounding rock; the boundary disturbance intensity and boundary recovery intensity are calculated as follows: in the local neighborhood of the starting depth or ending depth of the structural plane, calculate the mean change rate or gradient amplitude of the preset mechanical response characteristics; S32: Based on at least three parameters among drill bit diameter, current drilling speed, sampling time interval, thrust fluctuation amplitude, torque fluctuation amplitude, visible width of structural surface and ambiguity of structural surface boundary, and combined with the boundary disturbance intensity or the boundary recovery intensity, determine the pre-influence radius and post-influence radius of the event center interval of the structural surface, respectively. As a preferred approach, within the neighborhood of the initial depth of the structural surface, the thrust fluctuation amplitude and torque fluctuation amplitude are obtained, and the preceding influence radius is calculated according to the following formula in combination with the boundary disturbance intensity; within the neighborhood of the final depth of the structural surface, the thrust fluctuation amplitude and torque fluctuation amplitude are obtained, and the following influence radius is calculated according to the following formula in combination with the boundary recovery intensity. ; In the formula, Indicates the preceding or following direction. This refers to the radius of influence before or after the initial impact, expressed in meters (m). The drill bit diameter is in meters (m). This is the current drilling speed, in m / s; The sampling time interval is expressed in seconds (s). This is the normalized value of the thrust fluctuation amplitude in the corresponding direction, which is dimensionless; This is the normalized value of the torque fluctuation amplitude in the corresponding direction, which is dimensionless. The normalized value of the visually visible width of the structural surface is dimensionless. This is the normalized value of the fuzziness at the boundary of the structural surface, which is dimensionless. to Preset weighting coefficients; and These are the preset minimum influence radius and the preset maximum influence radius, respectively, in meters; This represents the limiting function, which constrains the calculation results to... to between; The leading radius of influence is used to define the range of influence extending from the initial depth of the structural surface towards the stable surrounding rock, while the trailing radius of influence is used to define the recovery range extending from the final depth of the structural surface towards the stable surrounding rock. When the ingress disturbance intensity of the leading neighborhood is greater than the outgress recovery intensity of the trailing neighborhood, the leading radius of influence is greater than the trailing radius of influence, reflecting that the mechanical response disturbance at the entrance of the structural surface is more intense and the range of influence is wider. When the recovery fluctuation of the trailing neighborhood is stronger than the ingress fluctuation of the leading neighborhood, the trailing radius of influence is greater than the leading radius of influence, reflecting that the recovery process after the drill bit exits the structural surface is longer. In this technical solution, the ingress disturbance intensity and outgress recovery intensity are calculated separately in the neighborhood of the starting and ending depths of the structural surface. Multiple physical parameters, including drill bit diameter, drilling speed, sampling time interval, thrust fluctuation amplitude, torque fluctuation amplitude, visually visible width of the structural surface, and boundary ambiguity, are integrated into quantitative calculation formulas for the pre-influence radius and post-influence radius. This achieves the physically adaptive determination of the three-segment sample spatial boundary. The determination of the influence radius not only considers the objective differences in drilling conditions and mechanical response but also reflects the asymmetry of disturbance levels at the structural surface inlet and outlet—a larger pre-influence radius when the ingress disturbance is strong, and a larger post-influence radius when the recovery process is long. This makes the spatial division of the three-segment sample more closely match the actual physical influence range when the drill bit traverses the structural surface, providing a spatial framework with clear physical basis for subsequent intra-segment feature extraction and soft boundary label generation.

[0033] Step 4: Three-segment spatial division based on the radius of influence; such as Figure 2 As shown, the event center interval of the structural surface is used as the anchor point. The preceding and following influence radii are used to adaptively divide the preceding stable segment, the structural surface influence segment, and the following recovery segment. Special treatment is given to the case of the composite structural surface. As a preferred approach, the three-segment spatial partitioning process based on the radius of influence is as follows: Using the center interval of the structural surface event as the anchor point, and combining the pre-influence radius and post-influence radius, the borehole depth direction is divided into a pre-stabilization segment, a structural surface influence segment, and a post-recovery segment. The pre-stabilization segment is used to characterize the stable drilling state before the drill bit enters the structural surface, the structural surface influence segment is used to characterize the disturbance state during the process of the drill bit entering, crossing, or exiting the structural surface, and the post-recovery segment is used to characterize the recovery state after the drill bit crosses the structural surface. The endpoint of the pre-stabilization segment does not exceed the initial depth of the structural surface, and its starting point is determined by subtracting the pre-influence radius from the initial depth of the structural surface. The structural surface influence segment covers the area from the initial depth of the structural surface to the final depth of the structural surface and its boundary transition range. Its starting point is the initial depth of the structural surface minus the pre-influence radius, and its ending point is the final depth of the structural surface plus the subsequent influence radius. The starting point of the subsequent recovery segment is not less than the final depth of the structural surface, and its ending point is determined by adding the subsequent influence radius to the final depth of the structural surface. The actual depth boundaries of each segment are adaptively adjusted based on whether there are other structural surface annotation intervals in the neighborhood. If there are other structural surface annotation intervals within the current stable segment or the subsequent recovery segment, the following processing is performed: If the distance between the other structural surface annotation intervals and the event center interval of the current structural surface is less than the corresponding influence radius, then the preceding influence radius or the subsequent influence radius of the current structural surface is shortened so that the boundary of the preceding stable segment or the subsequent recovery segment ends at the boundary of the other structural surface annotation intervals; or the three-segment sample unit of the current event center is marked as a composite structural surface sample to indicate that the sample contains complex response information of multiple structural surfaces; In this technical solution, the central interval of the structural surface event is used as the anchor point. Combining the pre-influence radius and post-influence radius, the system adaptively divides the borehole along the depth direction into a pre-stabilization segment, a structural surface influence segment, and a post-recovery segment. This comprehensively depicts the evolution of the drilling state throughout the entire process of the drill bit entering, traversing, and exiting the structural surface. The boundaries of the three-segment division are determined by the physical influence radius rather than a fixed window. This allows for dynamic adjustment of the depth range of each segment based on actual working conditions such as drill bit diameter, drilling speed, and fluctuation amplitude, ensuring a precise correspondence between the spatial structure of the sample and the physical influence range of the structural surface event. Simultaneously, by detecting the presence of other structural surface labeled intervals within the pre-stabilization segment or post-recovery segment, adaptive boundary adjustments or labeling are performed for complex structural surface cases. This avoids boundary conflicts between multiple structural surface events in the sample space, ensuring the effectiveness of the sample construction method in densely developed multi-structural-surface areas.

[0034] Step 5: Three-segment feature extraction and sample unit encapsulation; extract the drilling statistical features and mechanical response features in the pre-stabilization segment, the structural surface influence segment, and the post-recovery segment respectively, calculate the inbound and outbound differential response features, and encapsulate them into event-centered three-segment sample units according to the preset template; As a preferred approach, the process of three-stage feature extraction and sample unit encapsulation is as follows: S51: Extract the drilling statistical features and mechanical response features in the pre-stabilization section, the structural plane influence section, and the post-recovery section respectively; the drilling statistical features include the mean, variance, maximum and minimum values ​​of drilling speed, rotational speed, thrust and torque in each section; the mechanical response features are the statistical values ​​of each mechanical response feature in step 2 in the corresponding section, including the mean, variance and gradient; S52: Calculate the boundary differential response characteristics and the outbound differential response characteristics respectively; the boundary differential response characteristics are used to characterize the characteristic change pattern of the drill bit when it enters the structural plane from the stable surrounding rock, and the outbound differential response characteristics are used to characterize the characteristic change pattern of the drill bit when it returns to the stable surrounding rock from the structural plane; the boundary differential response characteristics satisfy the formula: The out-of-bounds difference response characteristics satisfy: ;in, The feature vector of the inbound differential response. The out-of-bounds difference response feature vector , and These are the feature vectors for the pre-stabilization segment, the structural plane-affected segment, and the post-recovery segment, respectively. Each feature vector is composed of the drilling-while-drilling statistical features and mechanical response features within the corresponding segment. This indicates normalization processing, using L2 normalization or maximum / minimum value normalization; S53: According to the preset sample field template, the depth boundary, intra-segment feature vector, inbound differential response feature, outbound differential response feature, structural surface type, inbound influence radius, outbound influence radius, and optional borehole peeping visual attribute fields of the pre-stabilized segment, structural surface influence segment, and post-recovery segment are encapsulated into an event-centered three-segment sample unit; wherein, the borehole peeping visual attribute fields include one or more of the following: structural surface visually visible width, structural surface boundary ambiguity, structural surface continuity, and structural surface opening state. S54: Bind the structural surface type label, structural surface start depth information, structural surface end depth information and sample interval information to the three-segment sample unit of the event center to form the basic data sample for training, verification or sample library management of the surrounding rock structural surface recognition model. In this technical solution, drilling statistical features and mechanical response features are extracted from the pre-stabilization section, the structural plane influence section, and the post-recovery section, respectively. Incoming and outgoing differential response features are then calculated to quantify the dynamic mechanical response changes of the drill bit throughout its traversal of the structural plane into structured feature vectors. The incoming differential response features characterize the abrupt change pattern of the drill bit as it enters the structural plane from stable surrounding rock, while the outgoing differential response features characterize the recovery pattern as the drill bit recovers from the structural plane back to stable surrounding rock. This allows the samples to fully represent the bidirectional mechanical response changes at the structural plane boundary. By unifying the three depth boundaries, intra-segment feature vectors, differential response features, structural plane type, influence radius, and visual attribute fields into a unified event-centric three-segment sample unit, and binding it with structural plane type labels and start / end depth information, a complete set of basic data samples is formed that can be directly used for training, validating, or managing surrounding rock structural plane recognition models.

[0035] Step 6: Adaptive soft boundary label generation; such as Figure 3As shown, the depth extension radius of the probability label is determined based on physical parameters such as the influence radius, the start and end boundary probability labels are generated, and a structural surface existence probability mask is constructed to form a soft boundary supervision label that conforms to the physical nature of the structural surface transition. As a preferred option, the process of generating adaptive soft boundary labels is as follows: S61: Obtain the three-segment sample unit of the event center and its corresponding structural surface start depth, structural surface end depth, preceding influence radius and subsequent influence radius; S62: Determine the depth extension radius of the starting boundary probability label based on at least three parameters: the preceding influence radius, the following influence radius, the drill bit diameter, the current drilling speed, the thrust fluctuation amplitude, the torque fluctuation amplitude, the visually visible width of the structure surface, and the ambiguity of the structure surface boundary. and the depth expansion radius of the endpoint boundary probability label , and The units are all in meters; as a preferred method, and The determination method is as follows: the forward influence radius is used as the base value of the starting boundary probability label depth extension radius, the backward influence radius is used as the base value of the ending boundary probability label depth extension radius, and the ratio is adjusted according to the thrust fluctuation amplitude and torque fluctuation amplitude. S63: Based on the initial depth of the structural surface Depth expansion radius of the starting boundary probability label Generate starting point boundary probability labels Based on the structural plane termination depth and the depth expansion radius of the endpoint boundary probability label Generate endpoint boundary probability labels The probability labels for the starting and ending boundaries are probability distributions centered on the starting and ending depths of the structural surface, and their distribution forms are Gaussian, triangular, or trapezoidal probability distributions. When a Gaussian probability distribution is used, the starting boundary probability label satisfies the formula: The endpoint boundary probability labels satisfy the following formulas: ;in, The drilling depth is expressed in meters (m). The initial depth of the structural surface, in meters; The structural surface termination depth, in meters (m). The depth expansion radius of the starting boundary probability label, in meters; The depth expansion radius of the endpoint boundary probability label, in meters; and All are dimensionless probability values, with a range of (0,1]. When the length of the structural plane development (i.e. (less than) and When summed, the depth expansion radii of the starting and ending boundary probability labels are proportionally compressed according to the development length of the structural surface. The compressed expansion radii are respectively... and To avoid the start-point boundary probability label and the end-point boundary probability label completely overlapping throughout the entire structural surface depth range; S64: Generate a probability mask for the existence of structural surfaces based on the depth range of the structural surfaces, the probability labels of the starting and ending boundaries. As a preferred structural surface, a probability mask exists. The generation method is as follows: at the drilling depth At that point, take the starting point boundary probability label. and endpoint boundary probability labels The larger value in, i.e. The probability mask for the existence of the structural surface satisfies the following conditions: the existence probability of the central region of the structural surface is not less than 0.8, the existence probability of the transition region of the structural surface boundary varies continuously between 0.2 and 0.8, and the existence probability of the stable surrounding rock region outside the structural surface is not higher than 0.2; when the probability labels of the starting boundary and the ending boundary overlap, the existence probability of the overlapping region is limited so that the existence probability of the structural surface does not exceed 1. S65: Bind the starting boundary probability label, the ending boundary probability label, and the structural surface existence probability mask to the three-segment sample unit of the event center to form a soft boundary supervision label for learning the structural surface boundary of the borehole depth sequence; In this technical solution, physical parameters such as the pre-influence radius and post-influence radius are mapped to the depth extension radius of the starting and ending boundary probability labels. A Gaussian equiprobability distribution is used to generate soft boundary supervision labels, thus transforming the structural surface boundary from discrete depth points to a continuous probability distribution. The extension radius of the soft boundary label is constrained by multiple sources of parameters, including drill bit diameter, drilling speed, thrust fluctuation amplitude, torque fluctuation amplitude, the visually visible width of the structural surface, and boundary ambiguity. This allows the ambiguity range of the label to be adaptively determined based on the actual physical characteristics of the structural surface and drilling conditions, better reflecting the ambiguity transition characteristics of geological structural surface boundaries caused by drill bit size effects, gradual rock mass changes, and drilling disturbance transmission. Simultaneously, by proportionally compressing the extension radius for short structural surfaces and generating a structural surface existence probability mask that satisfies the characteristics of high probability at the center, continuous transition at the edges, and low probability on the outer sides, a soft boundary supervision signal with clear physical basis and effective reduction of boundary positioning errors is provided for the surrounding rock structural surface identification model.

[0036] Step 7: Generation of primary and secondary labels for multiple structural surfaces and evaluation of sample quality; For cases where multiple structural surfaces exist within a single sample unit, the primary and secondary label weights of each structural surface are calculated based on multi-source information to distinguish between primary and secondary structural surfaces; At the same time, the reliability of the sample is evaluated by comprehensively considering information such as image quality, parameter completeness rate, and annotation confidence, and weights are assigned to the samples or the verification status is marked. As a preferred approach, multi-structured surface primary and secondary label generation and sample quality evaluation S71: Obtain the depth intervals of multiple structural surfaces contained in the three-segment sample unit of the event center, as well as their corresponding visually visible width, mechanical response anomaly intensity, depth overlap ratio with the sample unit, and label confidence level; the mechanical response anomaly intensity is the degree of deviation of the preset mechanical response features within the structural surface depth interval from the reference value of stable surrounding rock, and is obtained by normalizing the absolute value of the difference between the mean of the mechanical response features within the interval and the reference value of stable surrounding rock. S72: Calculate the normalized visual width for each structural plane depth interval. Normalized value of abnormal mechanical response intensity The depth overlap ratio with the sample cell and labeled confidence normalized value The depth overlap ratio is the ratio of the overlap length between the depth interval of the structural surface and the depth interval of the current sample unit to the total length of the depth interval of the structural surface, and the value range is [0,1]. S73: Calculate the primary and secondary label weights for each structural surface based on visually visible width, mechanical response anomaly intensity, depth overlap ratio, and label confidence level; Primary and secondary label weights of each structural surface satisfy: ,in For the first The normalized, dimensionless visual width of each structural surface; For the first The normalized value of the mechanical response anomaly intensity corresponding to each structural surface is dimensionless. For the first The depth overlap ratio between each structural surface and the sample unit is dimensionless. For the first The dimensionless, dimensionless, normalized confidence scores of the labeled structural surfaces. to Preset weighting coefficients; This indicates all samples within the same sample unit. Summing of all structural surfaces; S74: Determine the primary and secondary structural surface labels based on the primary and secondary label weights; determine the structural surface with the largest primary and secondary label weights as the primary structural surface label, and determine other structural surfaces with primary and secondary label weights not lower than the preset secondary label threshold as secondary structural surface labels; when the primary and secondary label weights of all secondary structural surfaces are lower than the preset secondary label threshold, only the primary structural surface label is retained; bind the primary and secondary structural surface labels to the three-segment sample unit of the event center; S75: Obtain the three-segment sample unit of the event center and its corresponding borehole peep image sharpness, drilling parameter completeness rate, structural surface boundary label confidence level, and depth correction residual; borehole peep image sharpness is the image quality evaluation index, drilling parameter completeness rate is the ratio of the number of actual drilling parameters to the number of parameters that should be obtained, structural surface boundary label confidence level is the reliability score of the structural surface boundary position label, and depth correction residual is the residual error after correction of the deviation between different depth measurement systems; S76: Calculate sample reliability based on borehole observation image clarity, drilling parameter integrity rate, structural boundary label confidence level, and depth correction residual. ,in The normalized value for the image sharpness of borehole peep view is dimensionless. The percentage of complete drilling parameters is dimensionless. The confidence level normalization value is labeled for the structural surface boundary, and is dimensionless; The normalized value of the depth correction residual is dimensionless; to The preset weighting coefficients satisfy the following conditions: ; S77: Determine the sample weight or verification status of the event center three-segment sample unit based on the sample reliability; when the sample reliability is lower than the preset reliability threshold, reduce the sample weight of the event center three-segment sample unit by multiplying the original weight by the ratio of the sample reliability to the reliability threshold, or mark the event center three-segment sample unit as a sample to be verified; when the sample reliability is higher than the preset high reliability threshold, mark the event center three-segment sample unit as a high-confidence training sample; bind the sample reliability, sample weight or verification status to the event center three-segment sample unit, and finally output a complete event center three-segment sample unit bound with structure surface type label, soft boundary supervision label, primary and secondary label weight, sample reliability and sample weight.

[0037] This technical solution calculates primary and secondary label weights by integrating information from four dimensions: visually visible width, mechanical response anomaly intensity, depth overlap ratio, and label confidence. This enables the quantitative ranking of the importance of multiple structural surfaces within the same sample unit, effectively distinguishing between primary and secondary structural surfaces and resolving the label confusion problem when multiple structural surfaces coexist. Simultaneously, a sample reliability evaluation model is constructed by fusing borehole observation image clarity, drilling parameter integrity rate, structural surface boundary label confidence, and depth correction residuals. This model quantifies the reliability of each sample unit and assigns differentiated sample weights or labels to verify the status, allowing the model training process to prioritize high-quality samples and suppress the negative impact of low-quality samples. The final output is a complete event-centric three-segment sample unit bound with structural surface type labels, soft boundary supervision labels, primary and secondary label weights, sample reliability, and sample weights. This provides high-quality training data with complete structure, rich annotations, and controllable quality for the surrounding rock structural surface identification model.

[0038] Example 2:

[0039] This embodiment provides a three-segment surrounding rock structural surface sample construction system for physically adaptive soft boundaries, used to implement a method for constructing three-segment surrounding rock structural surface samples with physically adaptive soft boundaries, including: The data input module is used to acquire the drilling parameter sequence and structural surface annotation results under a unified borehole depth coordinate system. The drilling parameter sequence includes one or more of the following: drilling depth, drilling time, drilling speed, rotational speed, thrust, torque, drill pipe displacement, drilling vibration, and drilling rig power. The structural surface annotation results include at least the structural surface type, the structural surface starting depth, and the structural surface ending depth, and are derived from one or more of the following: borehole inspection calibration results, borehole wall unfolding chart annotation results, or manual verification results. The event center interval determination module is used to define the depth interval of the structural surface as the event center interval of the structural surface, which is defined by the starting depth and ending depth of the structural surface. The event center interval directly builds a sample template around the occurrence interval of the structural surface, so that the sample construction process corresponds to the structural surface event itself, which is different from the conventional fixed sliding window method. The mechanical response feature calculation module is used to calculate the mechanical response feature sequence distributed along the borehole depth direction based on the drilling parameter sequence. The mechanical response features are used to reflect the changes in the mechanical properties of the surrounding rock along the drilling process, including multiple parameters such as energy per unit footage, mechanical specific energy, torque-to-thrust ratio, drilling speed residual, energy mutation coefficient, drilling stiffness change rate, thrust fluctuation intensity, and torque fluctuation intensity. These features are the basis for subsequent feature extraction within the segment and calculation of the boundary / outbound differential response features. The influence radius calculation module is used to determine the pre-influence radius and post-influence radius of the event center interval of the structural surface based on at least three parameters, including drill bit diameter, current drilling speed, sampling time interval, thrust fluctuation amplitude, torque fluctuation amplitude, visible width of the structural surface, and structural surface boundary ambiguity, combined with the boundary disturbance intensity or the boundary recovery intensity. The pre-influence radius is used to limit the influence range extending from the starting depth of the structural surface towards the stable surrounding rock, and the post-influence radius is used to limit the recovery range extending from the ending depth of the structural surface towards the stable surrounding rock. When the boundary disturbance intensity of the pre-neighborhood is greater than the boundary recovery intensity of the post-neighborhood, the pre-influence radius is greater than the post-influence radius. When the recovery fluctuation of the post-neighborhood is stronger than the boundary fluctuation of the pre-neighborhood, the post-influence radius is greater than the pre-influence radius. This module is also used to determine the depth extension radius of the starting boundary probability label and the depth extension radius of the ending boundary probability label based on the pre-influence radius and post-influence radius, respectively. The three-segment division module is used to divide the borehole along the depth direction into a pre-stabilized segment, a structural surface-affected segment, and a post-recovery segment, using the structural surface event center interval as the anchor point and combining the pre-affected and post-affected radius of influence. The pre-stabilized segment represents the stable drilling state before the drill bit enters the structural surface, the structural surface-affected segment represents the disturbance state during the drill bit's entry, crossing, or exit from the structural surface, and the post-recovery segment represents the recovery state after the drill bit crosses the structural surface. When there are other structural surface-labeled intervals within the pre-stabilized segment or the post-recovery segment, this module adaptively adjusts the radius of influence or marks the sample as a composite structural surface sample. The differential response feature calculation module is used to calculate the boundary differential response features and the out-of-bounds differential response features respectively. The boundary differential response features are used to characterize the characteristic change pattern of the drill bit when it enters the structural plane from the stable surrounding rock, and the out-of-bounds differential response features are used to characterize the characteristic change pattern of the drill bit when it returns to the stable surrounding rock from the structural plane. The boundary differential response features are obtained by normalizing the difference between the feature vector of the structural plane influence segment and the feature vector of the preceding stable segment, and the out-of-bounds differential response features are obtained by normalizing the difference between the feature vector of the subsequent recovery segment and the feature vector of the structural plane influence segment. The sample encapsulation module is used to encapsulate the depth boundaries, intra-segment feature vectors, inbound differential response features, outbound differential response features, structural surface type, inbound influence radius, outbound influence radius, and optional borehole observation visual attribute fields of the pre-stabilized segment, structural surface influence segment, and post-recovery segment into event-centered three-segment sample units according to a preset sample field template, and binds them with structural surface type labels, structural surface start depth information, structural surface end depth information, and sample interval information; the borehole observation visual attribute fields include one or more of the following: structural surface visually visible width, structural surface boundary ambiguity, structural surface continuity, and structural surface open state; An adaptive soft boundary label generation module is used to determine the depth extension radius of the starting boundary probability label and the depth extension radius of the ending boundary probability label based on at least three parameters: the preceding influence radius, the following influence radius, the drill bit diameter, the current drilling speed, the thrust fluctuation amplitude, the torque fluctuation amplitude, the visually visible width of the structural surface, and the structural surface boundary ambiguity. It generates a starting boundary probability label based on the structural surface starting depth and the depth extension radius of the starting boundary probability label, and generates an ending boundary probability label based on the structural surface ending depth and the depth extension radius of the ending boundary probability label. It generates a structural surface existence probability mask based on the structural surface depth range, the starting boundary probability label, and the ending boundary probability label. Finally, it binds the starting boundary probability label, the ending boundary probability label, and the structural surface existence probability mask to the event center three-segment sample unit to form a soft boundary supervision label. The primary and secondary labeling and sample quality evaluation module is used to calculate the primary and secondary label weights and determine the primary and secondary structural surface labels when there are multiple structural surface depth intervals in a three-segment sample unit at the event center. This is done based on the visually visible width of each structural surface, the intensity of mechanical response anomalies, the depth overlap ratio with the sample unit, and the label confidence level. Simultaneously, it calculates the sample reliability based on the borehole peep image clarity, drilling parameter integrity rate, structural surface boundary label confidence level, and depth correction residual, and determines the sample weight or verification status of the three-segment sample unit at the event center based on the sample reliability. The sample library output module is used to output complete event center three-segment sample units bound with structural surface type labels, soft boundary supervision labels, primary and secondary label weights, sample reliability, and sample weights, providing data support for the training, verification, and sample library management of the surrounding rock structural surface recognition model; This system, through a modular collaborative architecture, enables the complete construction of training samples for the identification of surrounding rock structural surfaces, from physical features to high-quality data samples. It effectively solves the problems of imprecise structural surface boundary representation, overlapping labels of multiple structural surfaces, and uncontrollable sample quality in existing technologies, providing a reliable system-level solution for intelligent identification of surrounding rock structural surfaces in underground engineering.

[0040] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A method for constructing a three-segment surrounding rock structural surface sample with physically adaptive soft boundary, characterized in that, Includes the following steps: Step 1: Data Acquisition and Determination of Event Center Interval on Structural Surface; Obtain the drilling parameter sequence and structural surface annotation results under a unified coordinate system, and use the interval defined by the start and end depths of the structural surface as the event center interval; Step 2: Calculation of mechanical response characteristic sequence; Based on the drilling parameter sequence, the multi-dimensional mechanical response characteristics distributed along the borehole depth direction are calculated; Step 3: Physical adaptive influence radius calculation; calculate the disturbance and recovery intensity in the neighborhood of the starting and ending depths of the structural surface respectively, and integrate multi-source parameters including drill bit diameter, drilling speed, fluctuation amplitude and visual attributes to adaptively determine the front influence radius and the back influence radius; Step 4: Three-segment spatial division based on influence radius; using the central interval of the structural surface event as the anchor point, adaptively divide the space into a pre-stabilized segment, a structural surface influence segment, and a post-recovery segment using the pre- and post-influence radii, and perform special processing for the case of composite structural surfaces; Step 5: Three-segment feature extraction and sample unit encapsulation; extract the drilling statistical features and mechanical response features in the pre-stabilization segment, the structural surface influence segment, and the post-recovery segment respectively, calculate the inbound and outbound differential response features, and encapsulate them into event-centered three-segment sample units according to the preset template; Step 6: Adaptive soft boundary label generation; Based on physical parameters such as the influence radius, determine the depth extension radius of the probability label, generate the starting and ending boundary probability labels, and construct the structural surface existence probability mask to form a soft boundary supervision label that conforms to the physical nature of the structural surface transition; Step 7: Generation of primary and secondary labels for multiple structural surfaces and evaluation of sample quality; For cases where multiple structural surfaces exist within a single sample unit, the primary and secondary label weights of each structural surface are calculated based on multi-source information to distinguish between primary and secondary structural surfaces; At the same time, the reliability of the sample is evaluated by comprehensively considering information such as image quality, parameter completeness rate, and annotation confidence.

2. The method for constructing a three-segment surrounding rock structural surface sample with physically adaptive soft boundary as described in claim 1, characterized in that, In step 1, the process of data acquisition and determination of the central interval of structural surface events is as follows: S11: Obtain the drilling parameter sequence and structural surface annotation results under a unified borehole depth coordinate system; the drilling parameter sequence includes one or more of the following: drilling depth, drilling time, drilling speed, rotational speed, thrust, torque, drill pipe displacement, drilling vibration, and drilling rig power; the structural surface annotation results include at least the structural surface type, the structural surface starting depth, and the structural surface ending depth. S12: The depth range defined by the starting depth and ending depth of the structural surface is used as the central interval of the structural surface event. S13: Obtain the borehole inspection visual attribute fields, which include one or more of the following: visible width of the structural surface, ambiguity of the structural surface boundary, continuity of the structural surface, and open state of the structural surface.

3. The method for constructing a three-segment surrounding rock structural surface sample based on a physically adaptive soft boundary as described in claim 1, characterized in that, In step 2, the process of calculating the mechanical response characteristic sequence is as follows: The mechanical response characteristic sequence distributed along the borehole depth direction is calculated based on the drilling parameter sequence; the mechanical response characteristics include multiple of the following: energy per unit footage, mechanical specific energy, torque-to-thrust ratio, drilling speed residual, energy mutation coefficient, drilling stiffness change rate, thrust fluctuation intensity, and torque fluctuation intensity.

4. The method for constructing a three-segment surrounding rock structural surface sample based on a physically adaptive soft boundary as described in claim 1, characterized in that, In step 3, the calculation process for the physical adaptive influence radius is as follows: S31: Calculate the boundary disturbance intensity and the boundary recovery intensity in the neighborhood of the starting depth and the neighborhood of the ending depth of the structural surface, respectively. S32: Based on at least three parameters among drill bit diameter, current drilling speed, sampling time interval, thrust fluctuation amplitude, torque fluctuation amplitude, visible width of structural surface and ambiguity of structural surface boundary, and combined with the boundary disturbance intensity or the boundary recovery intensity, determine the pre-influence radius and post-influence radius of the event center interval of the structural surface, respectively.

5. The method for constructing a three-segment surrounding rock structural surface sample with physically adaptive soft boundary according to claim 1, characterized in that, In step 4, the three-segment spatial partitioning process based on the radius of influence is as follows: Using the central interval of the structural surface event as the anchor point, and combining the pre-influence radius and the post-influence radius, the borehole depth direction is divided into the pre-stabilization section, the structural surface influence section, and the post-recovery section. The endpoint of the pre-stabilization segment does not exceed the initial depth of the structural surface, and its starting point is determined by subtracting the pre-influence radius from the initial depth of the structural surface. The structural surface influence segment covers the area from the initial depth of the structural surface to the final depth of the structural surface and its boundary transition range. Its starting point is the initial depth of the structural surface minus the pre-influence radius, and its ending point is the final depth of the structural surface plus the subsequent influence radius. The starting point of the subsequent recovery segment is not less than the final depth of the structural surface, and its ending point is determined by adding the subsequent influence radius to the final depth of the structural surface. The actual depth boundaries of each segment are adaptively adjusted based on whether there are other structural surface annotation intervals in the neighborhood. If there are other structural surface annotation intervals within the current stable segment or the subsequent recovery segment, the following processing is performed: If the distance between the other structural surface annotation intervals and the event center interval of the current structural surface is less than the corresponding influence radius, then the preceding influence radius or the subsequent influence radius of the current structural surface is shortened so that the boundary of the preceding stable segment or the subsequent recovery segment ends at the boundary of the other structural surface annotation intervals; or the three-segment sample unit of the current event center is marked as a composite structural surface sample to indicate that the sample contains complex response information of multiple structural surfaces.

6. The method for constructing a three-segment surrounding rock structural surface sample with physically adaptive soft boundary according to claim 1, characterized in that, In step 5, the process of three-segment feature extraction and sample unit encapsulation is as follows: S51: Extract the drilling statistical features and mechanical response features in the pre-stabilization section, the structural plane influence section, and the post-recovery section respectively; the drilling statistical features include the mean, variance, maximum and minimum values ​​of drilling speed, rotational speed, thrust and torque in each section; the mechanical response features are the statistical values ​​of each mechanical response feature in step 2 in the corresponding section, including the mean, variance and gradient; S52: Calculate the inbound differential response characteristics and the outbound differential response characteristics respectively; The boundary difference response characteristics satisfy the formula: The out-of-bounds difference response characteristics satisfy: ;in, The feature vector of the inbound differential response. The out-of-bounds difference response feature vector , and These are the feature vectors for the pre-stabilization segment, the structural plane-affected segment, and the post-recovery segment, respectively. Each feature vector is composed of the drilling-while-drilling statistical features and mechanical response features within the corresponding segment. This indicates normalization processing; S53: According to the preset sample field template, the depth boundary, intra-segment feature vector, inbound differential response feature, outbound differential response feature, structural surface type, inbound influence radius, outbound influence radius, and borehole observation visual attribute fields of the pre-stabilized segment, structural surface influence segment, and post-recovery segment are encapsulated into an event center three-segment sample unit. S54: Bind the structural surface type label, structural surface start depth information, structural surface end depth information and sample interval information to the three-segment sample unit of the event center to form the basic data sample for training, verification or sample library management of the surrounding rock structural surface recognition model.

7. The method for constructing a three-segment surrounding rock structural surface sample with physically adaptive soft boundary according to claim 1, characterized in that, In step 6, the adaptive soft boundary label generation process is as follows: S61: Obtain the three-segment sample unit of the event center and its corresponding structural surface start depth, structural surface end depth, preceding influence radius and subsequent influence radius; S62: Determine the depth extension radius of the starting boundary probability label based on at least three parameters: the preceding influence radius, the following influence radius, the drill bit diameter, the current drilling speed, the thrust fluctuation amplitude, the torque fluctuation amplitude, the visually visible width of the structure surface, and the ambiguity of the structure surface boundary. and the depth expansion radius of the endpoint boundary probability label ; S63: Based on the initial depth of the structural surface Depth expansion radius of the starting boundary probability label Generate starting point boundary probability labels Based on the structural surface termination depth and the depth expansion radius of the endpoint boundary probability label Generate endpoint boundary probability labels ; When the development length of the structural plane is less than and When summed, the depth expansion radii of the starting and ending boundary probability labels are proportionally compressed according to the development length of the structural surface. The compressed expansion radii are respectively... and ; S64: Generate a probability mask for the existence of structural surfaces based on the depth range of the structural surfaces, the probability labels of the starting and ending boundaries. The probability mask for the existence of the structural surface satisfies the following conditions: the existence probability of the central region of the structural surface is not less than 0.8, the existence probability of the transition region of the structural surface boundary varies continuously between 0.2 and 0.8, and the existence probability of the stable surrounding rock region outside the structural surface is not higher than 0.2; when the probability labels of the starting boundary and the ending boundary overlap, the existence probability of the overlapping region is limited so that the existence probability of the structural surface does not exceed 1. S65: Bind the starting boundary probability label, the ending boundary probability label, and the structural surface existence probability mask to the three-segment sample unit of the event center to form a soft boundary supervision label for learning the structural surface boundary of the borehole depth sequence.

8. The method for constructing a three-segment surrounding rock structural surface sample with physically adaptive soft boundary according to claim 1, characterized in that, In step 7, the generation of primary and secondary labels for multiple structural surfaces and the evaluation of sample quality are performed. S71: Obtain the depth range of multiple structural surfaces contained in the three-segment sample unit of the event center, as well as their corresponding visually visible width, mechanical response anomaly intensity, depth overlap ratio with the sample unit, and annotation confidence. S72: Calculate the normalized visual width for each structural plane depth interval. Normalized value of abnormal mechanical response intensity The depth overlap ratio with the sample cell and labeled confidence normalized value The depth overlap ratio is the ratio of the overlap length between the depth interval of the structural surface and the depth interval of the current sample unit to the total length of the depth interval of the structural surface, and the value range is [0,1]. S73: Calculate the primary and secondary label weights for each structural surface based on visually visible width, mechanical response anomaly intensity, depth overlap ratio, and label confidence level; Primary and secondary label weights of each structural surface satisfy: ,in For the first Normalized value of the visually visible width of each structural surface; For the first Normalized values ​​of the abnormal mechanical response intensity corresponding to each structural surface; For the first The depth overlap ratio between each structural surface and the sample unit; For the first Normalized confidence values ​​of the labeled structural surfaces; to Preset weighting coefficients; This indicates all samples within the same sample unit. Summing of all structural surfaces; S74: Determine the primary and secondary structural surface labels based on the primary and secondary label weights; determine the structural surface with the largest primary and secondary label weights as the primary structural surface label, and determine other structural surfaces with primary and secondary label weights not lower than the preset secondary label threshold as secondary structural surface labels; when the primary and secondary label weights of all secondary structural surfaces are lower than the preset secondary label threshold, only the primary structural surface label is retained; bind the primary and secondary structural surface labels to the three-segment sample unit of the event center; S75: Obtain the three-segment sample unit of the event center and its corresponding borehole peep image clarity, drilling parameter integrity rate, structural surface boundary label confidence and depth correction residual; S76: Calculate sample reliability based on borehole observation image clarity, drilling parameter integrity rate, structural boundary label confidence level, and depth correction residual. ,in This represents the normalized value for the sharpness of the borehole peep view image; The percentage of complete drilling parameters; Label the confidence normalization values ​​for the structural surface boundaries; For depth correction, the normalized value of the residual; to The preset weighting coefficients are used, and they satisfy the following conditions: ; S77: Determine the sample weight or verification status of the event center three-segment sample unit based on the sample reliability; when the sample reliability is lower than the preset reliability threshold, reduce the sample weight of the event center three-segment sample unit by multiplying the original weight by the ratio of the sample reliability to the reliability threshold, or mark the event center three-segment sample unit as a sample to be verified; when the sample reliability is higher than the preset high reliability threshold, mark the event center three-segment sample unit as a high-confidence training sample. The sample reliability, sample weight, or verification status are bound to the event center three-segment sample unit, and the final output is a complete event center three-segment sample unit bound with structure surface type label, soft boundary supervision label, primary and secondary label weight, sample reliability, and sample weight.

9. A system for constructing three-segment surrounding rock structural surface samples with physically adaptive soft boundaries, used to implement the method for constructing three-segment surrounding rock structural surface samples with physically adaptive soft boundaries as described in any one of claims 1 to 8, characterized in that, include: The data input module is used to obtain the drilling parameter sequence and structural surface annotation results under a unified borehole depth coordinate system; The event center interval determination module is used to define the depth interval of the structural surface as the event center interval of the structural surface, which is defined by the starting depth and ending depth of the structural surface. The mechanical response characteristic calculation module is used to calculate the mechanical response characteristic sequence distributed along the borehole depth direction based on the drilling parameter sequence; The influence radius calculation module is used to determine the pre-influence radius and post-influence radius of the event center interval of the structural surface based on at least three parameters among the drill bit diameter, current drilling speed, sampling time interval, thrust fluctuation amplitude, torque fluctuation amplitude, visible width of the structural surface and ambiguity of the structural surface boundary, combined with the boundary disturbance intensity or the boundary recovery intensity. The three-segment division module is used to divide the borehole depth direction into a pre-stabilized segment, a structural surface influence segment, and a post-recovery segment, with the structural surface event center interval as the anchor point and in combination with the pre-influence radius and the post-influence radius. The differential response characteristic calculation module is used to calculate the inbound differential response characteristics and the outbound differential response characteristics, respectively. The sample encapsulation module is used to encapsulate the depth boundaries, intra-segment feature vectors, inbound differential response features, outbound differential response features, structural surface type, inbound influence radius, outbound influence radius, and borehole observation visual attribute fields of the pre-stabilized segment, structural surface influence segment, and post-recovery segment into event-centered three-segment sample units according to the preset sample field template, and bind them with structural surface type labels, structural surface start depth information, structural surface end depth information, and sample interval information; An adaptive soft boundary label generation module is used to determine the depth extension radius of the starting boundary probability label and the depth extension radius of the ending boundary probability label based on at least three parameters: the preceding influence radius, the following influence radius, the drill bit diameter, the current drilling speed, the thrust fluctuation amplitude, the torque fluctuation amplitude, the visually visible width of the structural surface, and the structural surface boundary ambiguity. It generates a starting boundary probability label based on the structural surface starting depth and the depth extension radius of the starting boundary probability label, and generates an ending boundary probability label based on the structural surface ending depth and the depth extension radius of the ending boundary probability label. It generates a structural surface existence probability mask based on the structural surface depth range, the starting boundary probability label, and the ending boundary probability label. Finally, it binds the starting boundary probability label, the ending boundary probability label, and the structural surface existence probability mask to the event center three-segment sample unit to form a soft boundary supervision label. The primary and secondary labeling and sample quality evaluation module is used to calculate the primary and secondary label weights and determine the primary and secondary structural surface labels when there are multiple structural surface depth intervals in a three-segment sample unit at the event center. This is done based on the visually visible width of each structural surface, the intensity of mechanical response anomalies, the depth overlap ratio with the sample unit, and the label confidence level. Simultaneously, it calculates the sample reliability based on the borehole peep image clarity, drilling parameter integrity rate, structural surface boundary label confidence level, and depth correction residual, and determines the sample weight or verification status of the three-segment sample unit at the event center based on the sample reliability. The sample library output module is used to output complete event-centric three-segment sample units bound with structural surface type labels, soft boundary supervision labels, primary and secondary label weights, sample reliability, and sample weights.

Images