A method for diagnosing instability risk of unstable block in cavern under severe environment

Abstract

The application discloses a kind of unstable block instability risk diagnosis methods of cavern in harsh environment, belong to geological disaster monitoring technical field.The method is through laser radar, infrared thermal imaging, microseism and environmental sensor and is collected in multiple sources cooperation, based on environmental parameter adaptive fusion feature vector;Adopt the deep learning model of geometric feature constraint to identify structure surface and group;Based on real trace length distribution, three-dimensional discrete fracture network is constructed, and potential block is searched using improved mobility criterion;Time-varying stability coefficient and instability probability model are constructed, considering mechanical parameter degradation, microseismic cumulative damage and deformation acceleration, dynamic risk diagnosis and graded early warning are realized, and closed-loop feedback is formed.The application overcomes the defects of difficult data collection, low recognition accuracy, static analysis and overprediction, improves the accuracy, timeliness and automation level of diagnosis.

Description

Technical Field

[0001] This invention relates to the field of geological disaster monitoring, and in particular to a method for diagnosing the instability risk of unstable blocks in caverns under harsh environments. Background Technology

[0002] Structural planes developed in the surrounding rock of underground caverns cut into free-faced structures, forming movable blocks. Block instability is one of the main forms of surrounding rock failure. Accurately identifying unstable blocks and promptly diagnosing their instability risks is of great significance for engineering safety.

[0003] Currently, the following types of techniques are mainly used for stability analysis of surrounding rock blocks in caverns: (a) Structural Surface Recognition Technology The patent "An Automatic Recognition Method and System for Linear Structural Surfaces in Complex Rock Outcrops" (Application No.: CN202510590685.8) proposes an automatic recognition method for linear structural surfaces in complex rock outcrops. The method includes: acquiring images of the rock outcrops; using a DeepLabV3+ model to perform semantic segmentation on the input images; outputting a binary mask image of the structural surface trace pixels and performing topological reconstruction; establishing a mapping relationship between the image coordinate system and the three-dimensional point cloud to realize the conversion of the two-dimensional structural surface traces to three-dimensional space. The advantage of this method is that it achieves automated recognition of structural surfaces, reducing manual interaction. However, its limitations are: using a conventional semantic segmentation loss function without introducing geometric feature constraints, the identified structural surface boundaries may be relatively rough, and the accuracy of the attitude needs improvement; reconstruction is performed through the mapping from two-dimensional images to three-dimensional point clouds, resulting in information loss during the process; and it relies on optical images, making it difficult to guarantee the quality of data acquisition in harsh environments such as low visibility and strong winds.

[0004] (II) Block Stability Analysis Techniques The patent "A Stability Analysis Method for Underground Cavern Blocks Integrating Rigid Body Limit Equilibrium Method and Finite Element Method" (application number: CN202111409221.0) proposes a stability analysis method for underground cavern blocks, including: establishing a three-dimensional geometric model of the block and calculating a first safety factor using the rigid body limit equilibrium method; establishing a finite element model and calculating a second safety factor considering various loads; and taking the smaller value as the block stability safety factor. The advantage of this method is that it integrates two calculation methods, improving the reliability of the safety factor. However, its limitations are: the assumption of infinite extension of structural surfaces is still used when establishing the block geometric model, without considering the influence of the actual trace length distribution of structural surfaces on the block formation; and the calculated safety factor is a static one, without considering the influence of time effects such as the deterioration of structural surface mechanical parameters and damage accumulation on stability.

[0005] (III) Block monitoring and early warning technology The patent "Method and System for Monitoring Tunnel Surrounding Rock Block Collapse Based on Natural Vibration Frequency" (application number: CN202111044213.0) proposes a method for monitoring tunnel surrounding rock block collapse based on natural vibration frequency. This method assesses the risk of block collapse by monitoring changes in the natural vibration frequency of the surrounding rock blocks. The advantage of this method is that it achieves non-contact monitoring and the equipment deployment is relatively simple. However, its limitations are: it can only monitor blocks that have already loosened or deformed, making it difficult to identify potentially unstable blocks in advance; the natural vibration frequency is affected by environmental factors such as temperature and humidity, and the correspondence between characteristic signals and block damage needs further clarification; and the early warning threshold relies on empirical setting.

[0006] In summary, existing technologies suffer from drawbacks such as poor adaptability to harsh environments, fragmented identification and diagnosis, strong reliance on manual intervention, insufficient dynamic diagnosis, and limited applicability of formulas. Therefore, there is an urgent need to develop a fully automated, integrated, and dynamic method for monitoring and diagnosing unstable blocks in caverns that can adapt to harsh environments. Summary of the Invention

[0007] The purpose of this invention is to propose a method for diagnosing the instability risk of unstable blocks in caverns under harsh environments, thereby solving the technical problems of existing methods, such as poor adaptability to harsh environments, disconnect between identification and diagnosis, strong reliance on manual intervention, insufficient dynamic diagnosis, and limited applicability of formulas.

[0008] To achieve the above objectives, the present invention adopts the following technical solution: A method and equipment for diagnosing the instability risk of unstable blocks in caverns under harsh environments, the method comprising the following steps: S1. Conduct multi-source data collaborative acquisition in harsh environments: Deploy lidar, infrared thermal imager, microseismic sensor and environmental sensor in key areas of the cavern to acquire environmental parameters in real time, and adaptively adjust the acquisition weight of each sensor according to the environmental parameters. The acquired laser point cloud, thermal imaging and microseismic data are fused into a unified multi-source feature vector F through a multi-source data fusion model. S2. Perform intelligent identification of rock mass structural surfaces: preprocess the point cloud data corresponding to the multi-source feature vector F output in step S1, use a deep learning semantic segmentation model with geometric feature constraint loss function to identify rock mass structural surfaces, obtain high-precision structural surface boundary and attitude information, and use an improved fuzzy mean clustering algorithm to automatically group the identified structural surfaces and calculate the geometric parameters of each group of structural surfaces. S3. Perform dynamic search of key blocks based on the true trace length distribution: Based on the structural surface geometric parameters obtained in step S2, construct a three-dimensional discrete fracture network (DFN) model. On the basis of traditional block theory, adopt an improved mobility discrimination criterion that considers the finite size of the structural surface. Through a recursive search algorithm, traverse all possible combinations of structural surfaces, identify and output the geometric shape, spatial location and actual volume of potential unstable blocks. S4. Perform dynamic diagnosis of instability risk considering time effects: For each potentially unstable block searched in step S3, calculate its initial stability coefficient F0, and introduce time effects and external disturbances to construct a time-varying stability coefficient model and an instability probability model. Taking into account the deterioration of structural surface mechanical parameters over time, cumulative damage from microseismic events, and block deformation acceleration, calculate the dynamic stability coefficient and instability probability, and then classify the risk level. S5. Provide early warning and dynamic feedback: Trigger differentiated early warning based on the risk level determined in step S4, and dynamically feed back real-time monitoring data to step S4, update the parameters of the time-varying stability coefficient model and the instability probability model, recalculate the dynamic stability coefficient and the instability probability, and form a closed-loop dynamic diagnosis.

[0009] A storage medium storing instructions and data for implementing a method for diagnosing the instability risk of unstable blocks in caverns under harsh environments.

[0010] A device for diagnosing the instability risk of unstable blocks in caverns under harsh environments includes: a processor and a storage medium; the processor loads and executes instructions and data in the storage medium to implement a method for diagnosing the instability risk of unstable blocks in caverns under harsh environments.

[0011] The beneficial effects of this invention are as follows: 1. Strong adaptability to harsh environments: This invention deploys lidar, infrared thermal imager, micro-vibration sensor and environmental sensor, and adaptively adjusts the weight of each sensor based on environmental parameters (visibility, temperature gradient, background noise), and fuses multi-source data into a unified feature vector F. This effectively solves the problem of poor data quality or failure of single sensor data in harsh environments such as low visibility, high cold, and strong winds, and significantly improves the success rate and reliability of data acquisition. 2. High accuracy in structural surface recognition: This invention introduces a geometric feature constraint loss function into the deep learning semantic segmentation model. By constraining the normal vector divergence and dip angle error, it forces the model to learn the true geometric properties of the structural surface, eliminating the problems of rough boundaries and attitude deviation caused by traditional methods that rely solely on pixel classification. This makes the structural surface boundary recognition more continuous, controls the attitude error within 2°, and improves the recognition accuracy by about 30%. 3. Reliable block search: This invention abandons the oversimplified assumption of "infinite extension of structural surfaces" in traditional block theory. It constructs a three-dimensional discrete fracture network (DFN) based on the probability distribution of measured trace lengths and adopts an improved mobility discrimination criterion that considers the finite size of structural surfaces. By correcting the block formation conditions and actual volume through the exponential relationship between actual trace length, average trace length, and maximum trace length, false blocks are effectively eliminated, and the searched potential unstable blocks are highly consistent with the on-site geological sketches. 4. Dynamic diagnosis and time-varying risk assessment: This invention constructs a time-varying stability coefficient model that includes three factors: the rate of deterioration of mechanical parameters (considering temperature, time exponent, and activation energy), the cumulative damage of microseismic events (considering event energy and distance attenuation), and the block deformation acceleration. Based on reliability theory, it establishes an instability probability model that considers parameter uncertainties, overcoming the shortcomings of traditional static safety factors that cannot reflect the deterioration of surrounding rock stability over time, and realizing a leap from "static judgment" to "dynamic prediction". 5. Closed-Loop Early Warning and Intelligent Feedback: This invention classifies risks into four levels based on a comprehensive risk index (integrating instability probability, block volume, and exposure degree), triggering differentiated early warnings (encrypted monitoring, yellow warning, red warning, and engineering intervention). Simultaneously, real-time monitoring data is dynamically fed back to step S4, continuously updating model parameters to form a closed-loop system of "collection-identification-search-diagnosis-early warning-feedback," significantly improving the timeliness and accuracy of early warnings and reducing the risk of false alarms and missed alarms. 6. High degree of integration and automation: This invention requires no manual intervention throughout the entire process from data acquisition, structural surface identification, block search to risk diagnosis and early warning. It is suitable for harsh environments such as high altitude, deep burial, high ground stress, strong unloading, high cold, and low visibility. It can be widely used in the monitoring of unstable blocks and disaster prevention in geotechnical engineering such as underground caverns, tunnels, and slopes. Attached Figure Description

[0012] Figure 1 This is a flowchart illustrating the technical route of the method of the present invention; Figure 1 In the diagram, steps S1 to S5 are represented by rectangles, the flow direction is represented by solid arrows, and the dynamic feedback loop from step S5 to step S4 is represented by dashed arrows. Real-time monitoring data feedback is marked on the feedback loop. Figure 2 This is a schematic diagram of multi-source data collaborative acquisition and adaptive fusion in step S1 of the present invention; Figure 2 The attached diagrams are numbered as follows: LiDAR-110, Infrared Thermal Imager-120, Microseismic Sensor-130, Environmental Sensor-140, Multi-Source Data Fusion Module-200; Figure 3 This is a schematic diagram of the geometric feature constraints of the structural surface recognition model in step S2 of the present invention; Figure 4 This is a comparison chart of the block search results considering the limited size of the structural surface in step S3 of the present invention; Figure 5 This is a schematic diagram of the dynamic change curves of the time-varying stability coefficient and the instability probability in step S4 of the present invention. Figure 6 This is a schematic flowchart illustrating the monitoring setup for a group of deeply buried underground caverns in Example 2. Figure 7 This is a schematic flowchart illustrating the application of traffic tunnel monitoring in a high-altitude, cold region in Example 3; Figure 8 This is a schematic diagram of the hardware device operation according to an embodiment of the present invention. Detailed Implementation

[0013] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific examples described herein are merely illustrative and not intended to limit the scope of the invention.

[0014] Before formally describing the present invention, a general description of the solution of the present invention will be given first to facilitate understanding.

[0015] Example 1:

[0016] Please refer to Figure 1 The present invention provides a method for diagnosing the instability risk of unstable blocks in caverns under harsh environments, comprising the following steps: S1. Conduct multi-source data collaborative acquisition in harsh environments: Deploy lidar, infrared thermal imager, microseismic sensor and environmental sensor in key areas of the cavern to acquire environmental parameters in real time, and adaptively adjust the acquisition weight of each sensor according to the environmental parameters. The acquired laser point cloud, thermal imaging and microseismic data are fused into a unified multi-source feature vector F through a multi-source data fusion model. Please refer to Figure 2 Specifically, a lidar 110, an infrared thermal imager 120, a microseismic sensor 130, and an environmental sensor 140 are deployed in key areas of the cavern. The lidar is used to acquire high-precision three-dimensional point cloud data of the rock surface, the infrared thermal imager is used to acquire the temperature field to identify hidden fractures and water-bearing zones, the microseismic sensor is used to collect micro-fracture signals inside the rock mass, and the environmental sensor is used to monitor environmental parameters such as visibility, temperature gradient, and background noise. The system adaptively adjusts the acquisition weights of each sensor according to real-time environmental parameters, and fuses the point cloud, thermal imaging, and microseismic data into a unified feature vector F through the multi-source data fusion model proposed in this invention, which serves as the input for subsequent structural surface identification.

[0017] It should be noted that the multi-source data fusion model mentioned in step S1 is as follows:

[0018] Where F is the fused feature vector; f1, f2, and f3 are the feature vectors of point cloud, thermal imaging, and microseismic data, respectively; wᵢ(ε) is the adaptive weight of the i-th sensor based on the environmental parameter ε; ε1 is the reciprocal of visibility; ε2 is the temperature gradient; ε3 is the background noise level; and αᵢ is the environmental sensitivity coefficient.

[0019] As one embodiment, this step aims to address the problem of degraded data quality from a single sensor in harsh environments (such as low visibility, extreme cold, and strong winds). Specifically, one or more sensor networks are deployed in key areas of the cavern (such as the roof arch, intersections, and near fault zones). This network includes at least: a lidar 110 for acquiring high-precision three-dimensional spatial information of the rock surface; an infrared thermal imager 120 for acquiring the temperature field of the rock surface to identify hidden fractures and water-bearing zones; a microseismic sensor 130 for collecting micro-fracture signals inside the rock mass; and an environmental sensor 140 for real-time monitoring of environmental parameters such as visibility, temperature gradient, and background noise. Additionally, a multi-source data fusion module 200 is included to achieve multi-source data fusion.

[0020] As one example, during the construction of an underground powerhouse at a hydroelectric power station, the system detected a sudden increase in dust concentration (a sharp increase in the reciprocal of visibility ε1) in the current area due to blasting operations, detected by environmental sensors. At this time, the system automatically executes an adaptive data acquisition strategy: according to the formula... Recalculate the weights of each sensor.

[0021] Due to a severe deterioration in optical imaging quality, the weight of the infrared thermal imager 120... The wavelength of the LiDAR 110 is dynamically lowered; however, the wavelength of the LiDAR 110 has a certain penetrating power through smoke, and its weight... The weight of the micro-vibration sensor 130 was adjusted accordingly; at the same time, the weight of the micro-vibration sensor 130 was adjusted accordingly. The system remains unchanged and continues to monitor the aftershock effects following the blast. The system then assigns the weighted feature vector... By fusing the features, a unified multi-source feature vector is obtained. The fused feature vector F contains the optimal information about the rock mass surface and interior under the current environment, laying a data foundation for subsequent high-precision identification.

[0022] For example, during winter construction of a high-altitude tunnel, environmental sensors detected extremely low temperatures inside the tunnel (leading to a decrease in the contrast of the thermal imager). Larger), and the use of diesel machinery resulted in extremely low visibility ( Very large), microseismic background noise is at a moderate level ( (Medium). Assuming a preset environmental sensitivity coefficient. The values ​​are 0.5, 0.3, and 0.1 respectively. Then, calculate the weight numerator corresponding to visibility: Thermal imaging molecules: Micro-vibration molecules: After normalization, the lidar weights Approximately 0.01, thermal imaging weight Approximately 0.05, microseismic weight Approximately 0.94. The fused eigenvector F is primarily dominated by microseismic features, which is entirely consistent with the physical intuition that acoustic emission monitoring should be the primary method under harsh optical conditions. Conversely, under good lighting and temperature conditions, When the weights approach zero, the weights of the three factors will become equal, maximizing the utilization of information.

[0023] S2. Perform intelligent identification of rock mass structural surfaces: preprocess the point cloud data corresponding to the multi-source feature vector F output in step S1, use a deep learning semantic segmentation model with geometric feature constraint loss function to identify rock mass structural surfaces, obtain high-precision structural surface boundary and attitude information, and use an improved fuzzy mean clustering algorithm to automatically group the identified structural surfaces and calculate the geometric parameters of each group of structural surfaces. Specifically, the point cloud data output from step S1 is preprocessed, including statistical filtering for noise reduction and moving least squares surface fitting, to eliminate noise interference from harsh environments. A deep learning-based semantic segmentation model is used to identify rock mass structural surfaces. During training, a geometric feature constraint loss function is introduced to force the model to learn the geometric properties of the structural surfaces by constraining the normal vector divergence and dip angle error, thereby improving the identification accuracy of structural surface boundaries. After identification, an improved fuzzy mean clustering algorithm is used to automatically group the structural surfaces, and the dominant attitude, mean trace length, and distribution characteristics of each group of structural surfaces are calculated.

[0024] It should be noted that the geometric feature constraints described in step S2 are implemented through a loss function: (2) in, For cross-entropy loss, For geometric constraint loss:

[0025] Where N is the total number of point clouds, and nᵢ is the normal vector of the structure surface. Let β and λ be the reference normal vector, β and λ be the equilibrium coefficients, and θⱼ be the tilt angle of the structural surface. The reference structural plane inclination angle is M, and the number of structural planes is M.

[0026] Please refer to Figure 3 , Figure 3The image shows a structural surface on a rock mass block, with its normal vector labeled. ,inclination and normal vector divergence The schematic curve.

[0027] For example, when training a deep learning model to identify fractured rock mass surface features, only cross-entropy loss is used. This can cause the model to tend to identify each individual rock block as a structural plane, outputting broken, discontinuous boundaries. Introducing this during training... and set In each iteration, the model not only calculates the difference between the predicted structural surface regions and the true labels (…), but also… It also calculates the normal vector divergence at each prediction point. divergence with the normal vector of the true structure surface The mean squared error. For a set of true structural surfaces with an attitude of 30°, if the model predicts a "structural surface" with an attitude of 150°, then... This introduces a large penalty term. Through backpropagation, the model is forced to learn the spatial continuity and consistency of attitude of structural surfaces. Ultimately, the trained model can clearly delineate the boundaries of continuous structural surfaces while ignoring the edges of isolated rock blocks, improving recognition accuracy by approximately 30% compared to traditional methods.

[0028] As one embodiment, this step aims to automatically and accurately extract the boundary and attitude information of structural surfaces (such as joints, fissures, and faults) from the fused 3D point cloud data.

[0029] First, the point cloud data output from step S1 is preprocessed, including using statistical filtering algorithms to remove outliers and employing moving least squares (MLS) to smooth the surface, thus partially eliminating data noise caused by harsh environments. Next, a deep learning-based semantic segmentation model (such as an improved U-Net or PointNet++) is used, and an innovative geometric feature constraint loss function is introduced during its training process. This forces the model to learn the geometric properties of the structural surfaces.

[0030] As one example, when identifying structural surfaces at the entrance of a high-altitude tunnel, the initial deep learning model resulted in jagged boundaries of the identified structural surfaces due to the presence of glacial till on the rock wall surface, leading to inaccurate attitude calculations. This was addressed by adding a geometric constraint term to the loss function. The model, during training, not only focuses on whether the pixel classification is correct ( Simultaneously constrain the predicted divergence of the structural surface normal vectors. and tilt angle The model maintains consistency with the actual geometric morphology. After this optimization, the structural surface boundaries output by the model are smooth and continuous, with the attitude error controlled within 2°. After identification, the system uses an improved fuzzy C-means clustering algorithm to automatically group structural surfaces with similar attitudes into the same group, and calculates the probability distribution characteristics of the dominant attitude (dip, dip angle) and trace length for each group.

[0031] S3. Perform dynamic search of key blocks based on the true trace length distribution: Based on the structural surface geometric parameters obtained in step S2, construct a three-dimensional discrete fracture network (DFN) model. On the basis of traditional block theory, adopt an improved mobility discrimination criterion that considers the finite size of the structural surface. Through a recursive search algorithm, traverse all possible combinations of structural surfaces, identify and output the geometric shape, spatial location and actual volume of potential unstable blocks. It should be noted that the improved mobility criterion described in step S3 is: block formation conditions: (3) Volume correction:

[0032] Where K is the number of structural facet groups that participate in forming the block, and Lᵢ is the actual trace length of the i-th structural facet group on the block boundary. Let be the average trace length, and δ(v, sᵢ) be the direction discrimination function. To determine the threshold, For the ideal block volume, This is the maximum trace length.

[0033] For example, in a block search case consisting of three sets of structural surfaces (J1, J2, J3) and one free surface, the traditional method directly determines that a block with a volume of [missing information] is formed because J1, J2, and J3 all extend infinitely. The movable block. When applying the criteria of this invention, the system first obtains the actual trace lengths of three sets of structural surfaces: and their average trace length: And the maximum trace length: Calculate the product of the first term: , , The product of the three is approximately 0.22. If the direction discrimination function... And assume Since 0.22 < 0.3, this combination of structural surfaces is deemed unable to form a stable, movable block. Simultaneously, the actual volume is calculated: The result is much smaller than the ideal volume, which is more consistent with the actual situation on site that "the structural planes of this group are not fully connected, so they cannot form large blocks and can only form small-sized dangerous rocks."

[0034] Please refer to Figure 4 , Figure 4 The left side of the image shows the search results for the traditional method (assuming infinite extension of the structural plane), while the right side shows the search results for the method of this invention. As one embodiment, this step aims to overcome the over-prediction problem of blocks caused by the assumption of "infinitely extended structural surfaces" in traditional block theory. Based on the geometric parameters (attitude, trace length distribution) of the structural surface groups obtained in step S2, a three-dimensional discrete fracture network (DFN) model is constructed using the Monte Carlo method. In this model, the diameter (trace length) of each group of structural surfaces is no longer a fixed value, but follows the probability distribution obtained from the actual measurements in step S2. Based on this, an improved mobility criterion is used to search for truly movable blocks.

[0035] As one example, in the sidewall of an underground factory, the traditional method assumes a set of structural planes with an attitude of 240° dip and a dip angle of 70° running through the entire calculation area, thereby "predicting" a volume of 10m³. 3 The system identified a wedge-shaped block, but no such large block was actually found on-site. Using step S3 of this invention, the system first constructed a DFN model, where the trace lengths of the structural surfaces follow a log-normal distribution with a mean of 1.5m. An improved discriminant function was applied during the recursive search for combinations of structural surfaces. Calculations revealed that the actual trace length of a certain set of structural surfaces that contribute to the formation of the block is... Much smaller than its average trace length This led to the combination of Value below preset threshold Therefore, the block is determined to be immovable. Furthermore, the actual volume of the formed block is also determined according to the formula... After correction, the output results closely match the on-site geological sketches, effectively eliminating false blocks that were over-predicted.

[0036] S4. Perform dynamic diagnosis of instability risk considering time effects: For each potentially unstable block searched in step S3, calculate its initial stability coefficient F0, and introduce time effects and external disturbances to construct a time-varying stability coefficient model and an instability probability model. Taking into account the deterioration of structural surface mechanical parameters over time, cumulative damage from microseismic events, and block deformation acceleration, calculate the dynamic stability coefficient and instability probability, and then classify the risk level. Specifically, for each potentially unstable block identified in step S3, the initial safety factor for sliding failure and toppling failure is calculated separately, and the smaller one is taken as the initial stability factor F0. Introducing time effects and external disturbances, a time-varying stability coefficient model is constructed, comprehensively considering three factors: the deterioration of structural surface mechanical parameters over time, cumulative damage from microseismic events, and block deformation acceleration, to calculate the dynamic stability coefficient. Furthermore, based on catastrophe theory and reliability methods, an instability probability model is established, and a comprehensive risk index is calculated by combining the block volume and engineering exposure level, classifying it into four risk levels: low, medium, high, and extremely high.

[0037] It should be noted that the time-varying stability coefficient model and instability probability model mentioned in step S4 are as follows: Time-varying stability coefficient: (4) Degradation rate:

[0038] Cumulative damage variable:

[0039] Probability of instability:

[0040] Where F0 is the initial stability coefficient, γ(t) is the degradation rate, and m is the time exponent. R is the activation energy, T(t) is the gas constant, and D(t) is the rock mass temperature; D(t) is the cumulative damage, N(t) is the total number of microseismic events, Eⱼ is the energy of the j-th event, E0 is the reference energy, rⱼ is the distance of the event from the reference block, and r0 is the reference distance. Let η be the critical damage value, η be the damage coupling coefficient, and A(t) be the deformation acceleration. To maximize the impact on acceleration, ξ is the acceleration coupling coefficient; Let κ be the probability of instability, and κ be the sensitivity coefficient. The critical safety factor is μ, where μ is the uncertainty amplification factor. , The standard deviation and mean of the safety factor.

[0041] For example, for a key block in a high-stress soft rock tunnel, its initial... The system monitored the following data: ① Based on the on-site temperature and indoor degradation test parameters ( ), calculate the cumulative degradation items for the first 30 days. ② The microseismic system recorded multiple events, resulting in cumulative damage. Let the critical damage be... Damage coupling coefficient The damage term is ③ The displacement gauge measures the acceleration of the block. , , Then the acceleration term is .therefore, The stability coefficient is already far below the critical value. Substituting this value into the instability probability model, we take... Then the logical function part However, due to the uncertainties in various parameters, a safety factor standard deviation is assumed. mean Uncertainty amplification factor Then the final This example illustrates that although the calculated stability coefficient is extremely low, the system conservatively presents a low probability of instability because the various degradation parameters have not yet converged, thus avoiding false alarms. As time progresses, if the parameters stabilize, If the value is reduced, the probability of instability will increase rapidly.

[0042] Please refer to Figure 5 , Figure 5 The horizontal axis represents time t (days), and the vertical axis represents the stability coefficient. (Range 0.5~2.0), the right vertical axis represents the probability of instability. (Range 0~1.0). The figure shows a descending solid line curve. ) and the rising dashed curve ( The critical safety factor is marked with a horizontal dashed line. = 1.3; As one embodiment, this step aims to address the problem that traditional static safety factors cannot reflect the deterioration of surrounding rock stability over time. First, for each potentially unstable block identified in step S3, the initial safety factor for sliding and toppling failure is calculated using the rigid body limit equilibrium method. Then, a time-varying stability coefficient model is constructed, which dynamically evaluates stability from three dimensions: mechanical parameter deterioration, microseismic cumulative damage, and deformation acceleration.

[0043] As one example, the initial stability coefficient of a certain block This exceeds the design safety threshold of 1.3. However, the cavern is located in a high-stress zone with a large daily temperature difference. The system uses the formula... Calculate the degradation rate over time. migration and temperature The periodic changes cause the structural surface mechanical parameters to continuously deteriorate. Simultaneously, microseismic sensors detect multiple low-energy tensile events, accumulating damage variables. The acceleration gradually increased. Furthermore, the displacement gauge monitored the block deformation acceleration. From 0 to a positive value. Taking all the above factors into account, the time-varying stability coefficient model... Calculated dynamic coefficients It dropped to 1.25 after three months. Furthermore, through an instability probability model... The calculated instability probability is as high as 0.75, and the system classifies it as a high-risk level based on this.

[0044] S5. Provide early warning and dynamic feedback: Trigger differentiated early warning based on the risk level determined in step S4, and dynamically feed back real-time monitoring data to step S4, update the parameters of the time-varying stability coefficient model and the instability probability model, recalculate the dynamic stability coefficient and the instability probability, and form a closed-loop dynamic diagnosis.

[0045] Based on the risk level determined in step S4, differentiated early warnings are triggered: for medium risk, monitoring frequency is increased; for high risk, a yellow warning is issued and on-site inspections are strengthened; for extremely high risk, a red warning is issued and immediate engineering measures or personnel evacuation is recommended. Simultaneously, real-time monitoring data (displacement, microseismic activity, environmental parameters) is dynamically fed back to step S4 to update model parameters and recalculate time-varying stability coefficients and instability probabilities, forming a closed-loop dynamic diagnosis from data acquisition to early warning, ensuring the timeliness and accuracy of early warnings.

[0046] It should be noted that the risk levels mentioned in step S5 include low risk, medium risk, high risk, and extremely high risk, classified according to the comprehensive risk index R:

[0047] in, Let V be the probability of instability, V be the volume of the block, V0 be the reference volume, α be the volume impact index, and β(E) be the exposure coefficient. The risk level is determined based on R falling within different threshold ranges. Furthermore, the differentiated early warning described in step S5 specifically involves: increasing the monitoring frequency for medium risk; issuing a yellow alert and strengthening on-site inspections for high risk; and issuing a red alert and recommending immediate engineering measures or personnel evacuation for extremely high risk.

[0048] For example, the system identified three potentially unstable blocks.

[0049] Block A: Volume (Small), located at the bottom of the side wall (exposure level) ), probability of instability .Pick ,but .

[0050] Block B: Volume (Large), located at the top arch (exposure level) ), probability of instability .but .

[0051] Block C: Volume (Huge), located above the main passage (level of exposure) ), probability of instability .but System preset thresholds: Low risk Medium risk. High risk, The risk level is extremely high. Therefore, block A is low risk (small size, even if it falls, the impact is limited, only recorded); block B is high risk (large size and relatively high probability), triggering a yellow alert, and it is recommended to increase patrols; block C is medium risk (although huge in size, the current probability is low), triggering encrypted monitoring. This comprehensive index... It takes into account probability, consequences (volume), and exposure level, which is more scientific than using safety factor or probability alone.

[0052] As one embodiment, this step aims to achieve closed-loop control from diagnosis to treatment. Based on the risk level calculated in step S4, the system triggers corresponding early warning mechanisms and response measures. Simultaneously, the latest real-time monitoring data (displacement, microseismic activity, environmental parameters) is used as new input and dynamically fed back to step S4 to update the parameters in the time-varying model, enabling rolling calculation of the risk coefficient.

[0053] In one embodiment, if a block in step S4 has a dynamic stability coefficient that drops to 1.25, the system determines it as "high risk" and immediately issues a yellow alert. Upon receiving the alert, the on-site engineer increased the frequency of patrols in the area to once per shift and deployed prisms for twice-daily displacement monitoring. A week later, displacement data showed that the block's deformation rate accelerated. Feedback to the system showed that the model recalculated the instability probability to 0.88. The system automatically upgraded the alert level to "extremely high risk" and issued a red alert, while recommending: "Immediately reinforce the system with anchor bolts or seal with shotcrete; evacuate personnel from the area if necessary." The on-site team quickly completed the reinforcement as instructed, and subsequent monitoring data showed that the deformation tended to converge, successfully averting a potential collapse.

[0054] Example 2:

[0055] This embodiment uses a deeply buried underground powerhouse of a hydroelectric power station as an example to illustrate the specific implementation process of the present invention. See also... Figure 6 .

[0056] Multiple monitoring sections were installed on the roof arch, side walls, and end walls of the factory building (such as...). Figure 6As shown, sections 1 to 5 are each equipped with a data acquisition system. The lidar 110 uses a terrestrial 3D laser scanner; the scanning frequency can be set according to the deformation rate, and the scanning accuracy is better than ±1mm@100m. The infrared thermal imager 120 has a temperature resolution better than 0.05℃, and the acquisition frequency can be set to one frame every 30 minutes. The micro-seismic sensor 130 consists of multiple single-component accelerometers; the sampling frequency can be set to 10kHz, and the sensor spacing is determined according to site conditions. The environmental sensor 140 monitors parameters such as temperature, humidity, visibility, and wind speed; the sampling frequency can be set to once per minute.

[0057] The system automatically adjusts the acquisition mode based on real-time environmental parameters. During monitoring, when encountering heavy fog, the system automatically increases the weight of LiDAR 110 and decreases the weight of other sensors; when encountering strong winds, the scanning frequency of LiDAR 110 is increased to eliminate wind noise. The fused feature vector F is calculated using formula (1). After several months of continuous monitoring, point cloud data, thermal imaging data, and microseismic events were collected, and the data acquisition success rate was significantly higher than that of traditional single-sensor methods.

[0058] Statistical filtering was applied to the point cloud data for noise reduction, and then the local surface was fitted using the moving least squares method. A semantic segmentation model based on deep learning was constructed, and a geometric feature-constrained loss function (Equation 2) was introduced, with a balance coefficient set. The model was trained until convergence. Several structural surfaces were identified. Improved fuzzy C-means clustering was used for automatic grouping, with clustering features including attitude, trace length, and spacing. The number of clusters was automatically determined by the silhouette coefficient. The distribution characteristics of dominant attitude and trace length in each group were calculated. The structural surface identification accuracy was significantly better than that of conventional methods, and the attitude error was significantly reduced.

[0059] Based on the identified structural surface information, a three-dimensional DFN model is constructed using the Monte Carlo method, and the diameter of each set of structural surfaces follows the probability distribution of the measured trace length. Mobility discrimination is improved using formula (3), and a discrimination threshold is set. The method recursively searches for potentially unstable blocks and identifies key blocks with large volumes. Compared to the traditional infinite extension assumption, this method significantly reduces the number and volume of blocks, and the block locations show a high degree of agreement with the on-site geological sketches.

[0060] Stability analysis was performed on key blocks. Taking a typical block as an example, the initial stability coefficient was determined based on field test or empirical parameters. Deterioration parameters were obtained from indoor deterioration tests; microseismic monitoring showed the cumulative damage level; and displacement monitoring showed the deformation acceleration. The time-varying stability coefficient and instability probability were calculated using formula (4), and the risk level was classified.

[0061] Based on the risk level, high-risk blocks are subjected to intensive monitoring, and the monitoring data is dynamically fed back to step 4 to update the model parameters. Over time, the stability coefficient of some blocks decreases, and the probability of instability increases, triggering corresponding early warnings. On-site reinforcement measures are promptly implemented according to the early warning level to avoid potential collapse accidents. This embodiment verifies the effectiveness of the invention in deep-buried caverns under high ground stress.

[0062] Example 3:

[0063] This embodiment uses a traffic tunnel in a high-altitude, frigid region as an example to verify the adaptability of the present invention to harsh environments. See also Figure 7 .

[0064] Multiple monitoring sections were set up at the tunnel arch crown, left arch waist, and right arch waist (e.g., Figure 7 As shown in the diagram (sections 1, 2, and 3), each section is equipped with a data acquisition system (equipment is the same as in Example 1). For high-altitude, low-visibility environments, the system automatically adjusts its acquisition strategy: increasing the weight of the lidar 110 and decreasing the weight of other sensors. The infrared thermal imager 120 successfully identified multiple ice fissure zones (difficult to identify with conventional optical images). The environmental sensor 140 records temperature changes for subsequent degradation rate calculations. Continuous monitoring for several months during winter maintained a high data acquisition success rate.

[0065] Several structural planes were identified and grouped into several groups of dominant structural planes. Several potentially unstable blocks were identified, including several large critical blocks.

[0066] Considering the deterioration effect of freeze-thaw cycles on mechanical parameters, the deterioration rate was set to a value higher than that of normal temperature environments. The time-varying stability coefficients of each block were calculated, predicting a decrease in the stability coefficients of several blocks after the spring snowmelt season, indicating an increased probability of instability and thus a high risk. An early warning was issued. Drainage and localized anchor bolt reinforcement measures were implemented on-site. During the spring snowmelt season, on-site inspections revealed crack propagation in the warning area's unstable rock mass, but due to the prior reinforcement, no collapse occurred. The monitoring data showed good agreement with the actual on-site conditions. This embodiment verifies the data acquisition capabilities and dynamic diagnostic effectiveness of the present invention in harsh environments.

[0067] Example 4:

[0068] Please see Figure 8 , Figure 8 This is a schematic diagram of the hardware device in operation according to an embodiment of the present invention. The hardware device specifically includes: a device 401 for collecting information on unstable blocks in a cavern under harsh conditions, intelligent identification and instability risk diagnosis, a processor 402 and a storage medium 403.

[0069] A device 401 for diagnosing the instability risk of unstable blocks in caverns under harsh environments: The device 401 implements the method for diagnosing the instability risk of unstable blocks in caverns under harsh environments.

[0070] Processor 402: The processor 402 loads and executes the instructions and data in the storage medium 403 to implement the method for diagnosing the instability risk of unstable blocks in caverns under harsh environments.

[0071] Storage medium 403: The storage medium 403 stores instructions and data; the storage medium 403 is used to implement the method for diagnosing the instability risk of unstable blocks in caverns under harsh environments.

[0072] This invention is not limited to the specific embodiments described above. Those skilled in the art can implement this invention using various other specific embodiments based on the content disclosed herein. Therefore, any design that adopts the design structure and concept of this invention and makes some simple changes or modifications falls within the scope of protection of this invention.

Claims

1. A method for diagnosing the instability risk of unstable blocks in caverns under harsh environments, characterized in that: Includes the following steps: S1. Conduct multi-source data collaborative acquisition in harsh environments: Deploy lidar, infrared thermal imager, microseismic sensor and environmental sensor in key areas of the cavern to acquire environmental parameters in real time, and adaptively adjust the acquisition weight of each sensor according to the environmental parameters. The acquired laser point cloud, thermal imaging and microseismic data are fused into a unified multi-source feature vector F through a multi-source data fusion model. S2. Perform intelligent identification of rock mass structural surfaces: preprocess the point cloud data corresponding to the multi-source feature vector F output in step S1, use a deep learning semantic segmentation model with geometric feature constraint loss function to identify rock mass structural surfaces, obtain high-precision structural surface boundary and attitude information, and use an improved fuzzy mean clustering algorithm to automatically group the identified structural surfaces and calculate the geometric parameters of each group of structural surfaces. S3. Perform dynamic search of key blocks based on the true trace length distribution: Based on the structural surface geometric parameters obtained in step S2, construct a three-dimensional discrete fracture network (DFN) model. On the basis of traditional block theory, adopt an improved mobility discrimination criterion that considers the finite size of the structural surface. Through a recursive search algorithm, traverse all possible combinations of structural surfaces, identify and output the geometric shape, spatial location and actual volume of potential unstable blocks. S4. Perform dynamic diagnosis of instability risk considering time effects: For each potentially unstable block searched in step S3, calculate its initial stability coefficient F0, and introduce time effects and external disturbances to construct a time-varying stability coefficient model and an instability probability model. Taking into account the deterioration of structural surface mechanical parameters over time, cumulative damage from microseismic events, and block deformation acceleration, calculate the dynamic stability coefficient and instability probability, and then classify the risk level. S5. Provide early warning and dynamic feedback: Trigger differentiated early warning based on the risk level determined in step S4, and dynamically feed back real-time monitoring data to step S4, update the parameters of the time-varying stability coefficient model and the instability probability model, recalculate the dynamic stability coefficient and the instability probability, and form a closed-loop dynamic diagnosis.

2. The method for diagnosing the instability risk of unstable blocks in caverns under harsh environments according to claim 1, characterized in that: The multi-source data fusion model mentioned in step S1 is as follows: Where F is the fused feature vector; f1, f2, and f3 are the feature vectors of point cloud, thermal imaging, and microseismic data, respectively; wᵢ(ε) is the adaptive weight of the i-th sensor based on the environmental parameter ε; ε1 is the reciprocal of visibility; ε2 is the temperature gradient; ε3 is the background noise level; and αᵢ is the environmental sensitivity coefficient.

3. The method for diagnosing the instability risk of unstable blocks in caverns under harsh environments according to claim 1, characterized in that: The geometric feature constraints described in step S2 are implemented through a loss function: in, For cross-entropy loss, For geometric constraint loss: Where N is the total number of point clouds, and nᵢ is the normal vector of the structure surface. Let β and λ be the reference normal vector, β and λ be the equilibrium coefficients, and θⱼ be the tilt angle of the structural surface. The reference structural plane inclination angle is M, and the number of structural planes is M.

4. The method for diagnosing the instability risk of unstable blocks in a cavern under harsh conditions according to claim 1, characterized in that: The improved mobility criterion described in step S3 is: block formation conditions: Volume correction: Where K is the number of structural facet groups that participate in forming the block, and Lᵢ is the actual trace length of the i-th structural facet group on the block boundary. Let be the average trace length, and δ(v, sᵢ) be the direction discrimination function. To determine the threshold, For the ideal block volume, This is the maximum trace length.

5. The method for diagnosing the instability risk of unstable blocks in caverns under harsh environments according to claim 1, characterized in that: The time-varying stability coefficient model and instability probability model mentioned in step S4 are as follows: Time-varying stability coefficient: Degradation rate: Cumulative damage variable: Probability of instability: Where F0 is the initial stability coefficient, γ(t) is the degradation rate, and m is the time exponent. R is the activation energy, T(t) is the gas constant, and D(t) is the rock mass temperature; D(t) is the cumulative damage, N(t) is the total number of microseismic events, Eⱼ is the energy of the j-th event, E0 is the reference energy, rⱼ is the distance of the event from the reference block, and r0 is the reference distance. Let η be the critical damage value, η be the damage coupling coefficient, and A(t) be the deformation acceleration. To maximize the impact on acceleration, ξ is the acceleration coupling coefficient; Let κ be the probability of instability, and κ be the sensitivity coefficient. The critical safety factor is μ, where μ is the uncertainty amplification factor. , The standard deviation and mean of the safety factor.

6. The method for diagnosing the instability risk of unstable blocks in a cavern under harsh conditions according to claim 1, characterized in that, The risk levels mentioned in step S5 include low risk, medium risk, high risk, and extremely high risk.

7. The method for diagnosing the instability risk of unstable blocks in a cavern under harsh conditions according to claim 6, characterized in that, Risk levels are classified based on the comprehensive risk index R: in, V is the probability of instability, V is the volume of the block, V0 is the reference volume, α is the volume influence index, and β(E) is the exposure coefficient; the risk level is determined based on R falling within different threshold ranges.

8. The method for diagnosing the instability risk of unstable blocks in a cavern under harsh conditions according to claim 7, characterized in that, The differentiated early warning mentioned in step S5 is as follows: when the risk is medium, the monitoring frequency is increased; when the risk is high, a yellow warning is issued and on-site inspections are strengthened; when the risk is extremely high, a red warning is issued and it is recommended to take immediate engineering measures or evacuate personnel.

9. A storage medium, characterized in that: The storage medium stores instructions and data to implement the method for diagnosing the instability risk of unstable blocks in caverns under harsh environments as described in any one of claims 1 to 8.

10. A diagnostic device for the instability risk of unstable blocks in caverns under harsh environments, characterized in that: include: A processor and a storage medium; the processor loads and executes instructions and data in the storage medium to implement the method for diagnosing the instability risk of unstable blocks in caverns under harsh environments as described in any one of claims 1 to 8.

Images