Intelligent identification method for three-dimensional crack nucleation of rock

Abstract

The application relates to an intelligent identification method for rock three-dimensional crack nucleation, comprising the following steps: S1, collecting acoustic emission signals of rock damage under compression loading; S2, three-dimensional positioning and coordinate normalization; S3, constructing a point set according to an intelligent time window or a strain window; S4, selecting a candidate clustering scale set Dk; S5, outputting optimal DBSCAN parameters; S6, obtaining a density clustering identification candidate nucleation cluster; S7, judging whether each cluster Ck is a crack nucleation cluster and evaluating the nucleation cluster confidence; S8, obtaining a three-dimensional nucleation body grid Bk,j; S9, outputting each nucleation body and quantitatively representing; S10, matching; and S11, providing a warning or control interface called by a monitoring system. The method combines real-time monitoring of acoustic emission and microseismic events, adopts adaptive intelligent parameter optimization, density clustering analysis, three-dimensional reconstruction and other technologies, and automatically identifies a rock crack nucleation area.

Description

Technical Field

[0001] This invention relates to the field of rock mass engineering technology, and in particular to an intelligent identification method for three-dimensional crack nucleation in rocks. Background Technology

[0002] Intelligent identification of three-dimensional crack nucleation in rocks is an important method for studying rock damage structures. In engineering practice, under external loads, stress concentration and energy accumulation induce crack nucleation, propagation, and penetration, leading to internal damage and loss of load-bearing capacity, ultimately resulting in fracture. Current technologies have limited research conditions for analyzing internal rock damage, making it difficult to achieve intelligent identification and quantitative analysis of internal crack damage. Crack nucleation and its interactions are key factors in the microscopic to macroscopic fracture of rocks. Current algorithms can only select parameters manually, lacking adaptive parameter optimization; the crack nucleation body cannot be accurately drawn in 3D, hindering online dynamic identification and tracking to form a "nucleation body trajectory" and "stability score." This impedes the timeliness and intelligence of the crack nucleation process. How to intelligently identify and quantitatively analyze rock crack nucleation is a pressing technical problem that needs to be solved in the field of intelligent identification of three-dimensional rock crack nucleation.

[0003] Currently, the main methods for identifying three-dimensional crack nucleation include CT-based "direct imaging of internal three-dimensional cracks" and 3D dynamic identification of surface cracks based on digital image correlation (3D-DIC). X-ray CT has segmentation processes and algorithms for rock CT and in-situ loaded micro-CT for observing crack evolution. It offers the most intuitive geometry, high resolution, and can quantitatively analyze connectivity / pore structure, making it suitable for microcrack research. However, the equipment is expensive, sample density is limited, it is sensitive to noise and artifacts, segmentation depends on parameters / training data, and it is difficult to apply on a large scale in the field. 3D-DIC provides strong visualization of the surface crack propagation process and quantitative displacement field, making it suitable for indoor experimental mechanism research. However, it mainly focuses on visible surfaces, is sensitive to dust, lighting, and speckle quality, and requires fusion with AE / ultrasound / CT for internal cracks.

[0004] Acoustic emission data from within the rock during loading was acquired using acoustic emission monitoring. 3D reconstruction of crack nucleation bodies was achieved using both DBSCAN and AlphaComplexShape algorithms. The acoustic emission data from within the rock was processed and 3D modeled to realize an intelligent identification method for 3D crack nucleation in rocks. Compared to traditional methods, this method allows for near real-time "tracking-expansion-penetration" evolution in experiments / downholes, is effective for internally invisible cracks, and is suitable for disaster gestation detection. Summary of the Invention

[0005] This invention provides an intelligent identification method for rock crack nucleation. This method combines acoustic emission (AE) and real-time monitoring of microseismic events, and adopts adaptive intelligent parameter optimization, density clustering analysis, three-dimensional reconstruction and other technologies to automatically identify rock crack nucleation regions and output their location, volume, quantity, morphology, evolution trajectory and identification confidence level.

[0006] Specifically, this invention provides an intelligent identification method for three-dimensional crack nucleation in rocks, the improvement of which is that the method includes: Step S1: Collect acoustic emission signals of rocks damaged under compression loading conditions, extract the amplitude, energy, frequency band characteristics and duration parameters of each triggering event from AE or microseismic point events, and calculate SNR and waveform quality indicators for saturation detection and crosstalk detection. Step S2: Use a time-inversion algorithm or a localization algorithm to convert the event into a three-dimensional point, map the coordinates to a unified reference system, and perform three-dimensional localization and coordinate normalization. Step S3: Construct a point set by scrolling through the smart time window or strain window; Step S4: For the point set Pk of each window, use the Monte Carlo method to generate a completely random spatial envelope, and calculate the three-dimensional Ripley K function K(d) to convert it into the L function L(d); automatically identify significant clustering distance intervals, and take the interval with the largest or stable deviation as the candidate clustering scale set D. k ; Step S5: Use Bayesian optimization, genetic algorithm or particle swarm optimization for fast optimization, and output the optimal density clustering algorithm parameters through adaptive intelligent optimization of density clustering algorithm parameters. Step S6: Introduce clustering quality indicators and combine them with noise rate and physical constraint penalty terms to obtain density clustering to identify candidate nucleation clusters; Step S7: Calculate the evidence features for each cluster Ck and output them using a rule engine or lightweight learning model to determine whether it is a crack nucleation cluster and evaluate the confidence of the nucleation cluster. Step S8: For clusters identified as nucleating, perform 3D reconstruction using Delaunay tetrahedralization and α-complex shape, adaptively determine α-alpha, construct the α-shape, and perform surface smoothing to obtain a 3D nucleating body mesh. ; Step S9: Output each nucleation body and quantify its characterization; Step S10, for and Matching is performed, including centroid distance, volume overlap, or morphological similarity matching, to obtain the nucleation body trajectory and achieve cross-window association and evolution tracking; : Indicates the j-th nucleation site identified within the current time window k; This indicates that the i-th nucleating body has been identified and recorded within the previous time window k-1; Step S11: Output the 3D visualization and recognition result file or recognition result database of nucleation body mesh, AE point cloud and energy mapping, and provide an early warning or control interface for the monitoring system to call.

[0007] Preferably, step S4 includes: Step S4-1, Coordinate Acquisition and Distance Calculation: Extract the three-dimensional spatial coordinates of all discrete event points within the crack nucleation region, and iterate through and calculate the three-dimensional Euclidean distance between any two points; Step S4-2, L(d) function calculation: Set the spatial observation scale, and calculate the three-dimensional Ripley K function K(d) based on the spatial distance between each point to obtain the expected number of point pairs within the distance d range; Step S4-3, K(d) normalization and clustering assessment: The variance of the K(d) function is stabilized and normalized using the L(d) function. By comparing the actual L(d) value with the theoretical value under a completely spatial random distribution, the clustering, randomness or discreteness of the crack nucleation region under different spatial scales d is quantitatively assessed.

[0008] Preferably, step S5 includes: Step S5-1 defines the optimization decision space, as shown in the following equation:

[0009] In the formula, H represents the joint search boundary space of the intelligent optimization algorithm; D K : This is the set of candidate clustering scales extracted within the k-th time window based on spatial point pattern statistical priors; The minimum statistical distance in the candidate set; The maximum statistical distance in the candidate set; This is the lower limit of the pre-defined minimum number of effective nucleation point frequencies; The upper limit of the maximum effective nucleation point frequency is set in advance; Step S5-2: Agent model initialization: For each set of parameters, an initial sampling set is used as the observation data to construct a Gaussian process surrogate model to fit the unknown distribution of the objective function; Step S5-3: Constructing the acquisition function and generating candidate solutions: Constructing the confidence upper limit set function And by maximizing the acquisition function, the optimal candidate solution for the probe point in the t-th iteration is output. : Step S5-4: Closed-loop evaluation of the objective function and model update: Extracted candidate solutions Substitute the data into the density clustering algorithm to perform trial clustering of the point cloud dataset for the current time window; Based on the clustering output, calculate the actual overall fitness score for the current parameters. ; to acquire new observation pairs Add to observation dataset In addition, the hyperparameters of the Gaussian process surrogate model are updated using Bayesian posterior probabilities; Step S5-5: Convergence judgment and optimal parameter output. Repeat steps S5-3 and S5-4 iteratively until the preset convergence condition is met or the maximum number of iterations is reached. After stopping the iteration, output the parameter combination with the highest score of the corresponding objective function in the observed dataset as the optimal density clustering parameters for the current time window. .

[0010] Preferably, step S10 includes: Step S10-1, perform centroid distance matching: calculate the spatial relative position of the centroids of the two nuclei; if the centroids do not undergo instantaneous shift, they are determined to be the same fracture source; Step S10-2, perform volume overlap matching: place the nucleation body 3D mesh from the previous stage and the mesh from the current stage in the same coordinate system, and determine the size of the spatial relative nesting and overlapping of the meshes; Step S10-3: Perform morphological similarity matching: compare the geometry of the two nuclei, including aspect ratio and spatial principal axis direction.

[0011] Preferably, step S10 includes: Construct a comprehensive evaluation function for evaluation and The degree of correlation is shown in the following formula: Let the matching degree be M. ; In the formula: Score based on Euclidean distance from the center of mass; The score is the intersection-union ratio of the three-dimensional volume. ; Scoring is based on morphological similarity. Distance weights; This is the overlap weight; This is the shape weight.

[0012] Preferably, step S10 includes: If the previous window is independent Independent of the current window When a one-to-one match is observed, the nucleated body is determined to be in a stable expansion state. If multiple independent nuclei in the previous window are the same nuclei in the current window... During matching, the nuclei are determined to be in a connected and merged state; If the current window If no matching ancestor is found in the previous window, the nucleosome is determined to be in a new germination state. If the previous window If no successor is found in the current window, the nucleus is determined to be in a dormant or closed state.

[0013] Compared with the prior art, the beneficial effects of the present invention are as follows: This application proposes an intelligent identification method for three-dimensional crack nucleation in rocks. Based on acoustic emission and microseismic point event data, and combined with techniques such as three-dimensional localization, spatial statistical analysis, adaptive parameter optimization, density clustering identification, three-dimensional reconstruction, and cross-window evolution tracking, it can automatically identify, visualize, and dynamically analyze the nucleation region of internal cracks in rocks. This overcomes the shortcomings of existing technologies, such as difficulty in real-time perception of internal cracks, reliance on manual setting of clustering parameters, difficulty in accurately reconstructing the outline of nucleated bodies, and difficulty in continuously tracking the crack evolution process. Furthermore, this invention can adaptively identify the point cloud distribution characteristics within different time windows. By acquiring clustering parameters, the objectivity, accuracy, and robustness of crack nucleation identification results are improved. Furthermore, by introducing physical constraints and confidence assessment mechanisms, the authenticity, reliability, and interpretability of the identification results are effectively enhanced. Simultaneously, this invention can quantitatively characterize the location, volume, morphology, quantity, and evolution trajectory of crack nuclei, providing more comprehensive data support for the analysis of rock damage evolution, the study of instability mechanisms, and the identification of hazardous areas. It is particularly suitable for dynamic disaster monitoring and early warning in mines, tunnels, and underground rock engineering, exhibiting high timeliness, intelligence, and engineering application value. Attached Figure Description

[0014] Figure 1 This is a flowchart of an intelligent identification method for three-dimensional crack nucleation in rocks, as described in this application. Figure 2 This is a diagram showing the three-dimensional spatial distribution of acoustic emission localization energy levels during rock damage in the method described in this application. Figure 3 This diagram illustrates the phased evolution, trajectory tracking, and topology merging process of acoustic emission nuclei involved in the method of this application. Figure 4 The image shows the AlphaComplexShape 3D reconstruction mesh of the crack nucleation body involved in the method of this application. Figure 5 This is a schematic diagram of the panoramic report on the phased evolution identification of rock crack nucleation involved in the method of this application; Figure 6 This is a schematic diagram of the acoustic emission characteristic hierarchical evolution early warning report involved in the method of this application. Detailed Implementation

[0015] To better understand this invention, the following description, in conjunction with the accompanying drawings and examples, will further illustrate the invention.

[0016] like Figure 1 As shown, the intelligent identification method for three-dimensional crack nucleation in rock damage includes the following steps: Step S1: Collect acoustic emission signals of rocks damaged under compression loading conditions. Extract parameters such as amplitude, energy, frequency band characteristics, and duration from each triggering event from point events such as AE / microseismic events, and calculate waveform quality indicators such as SNR and saturation / crosstalk detection.

[0017] Specifically, step S1, acoustic emission signal acquisition and multidimensional feature extraction, includes acquiring acoustic emission (AE) signals of rocks damaged under compression loading conditions. From AE / microseismic events, parameters such as amplitude, energy, frequency band characteristics, and duration are extracted for each triggering event. Specifically, the frequency band characteristics are quantified and divided into three specific energy levels—low-frequency band (LDZ), mid-frequency band (MDZ), and high-frequency band (HDZ)—through spectral analysis of the acoustic emission signals. The LDZ band is typically defined as 0-100kHz and is used to characterize the penetration and merging of large-scale cracks within the rock; the MDZ band is 100-300kHz and is used to characterize the stable development of fractures; the HDZ band is above 300kHz and is used to characterize the initiation of micro-fractures.

[0018] Specifically, for each triggered event, not only is the basic waveform recorded, but multi-dimensional spatiotemporal and energy parameters are also extracted, and waveform quality indicators are calculated to provide a precise data foundation for subsequent analysis. Step S1-1, Obtain ArrivalTime: Record the time using precise time synchronization technology and accurately calculate the signal propagation delay using a cross-correlation algorithm. As shown in the following formula: ; In the formula: and These represent signals received by different sensors. For signal correlation coefficient Step S1-2, Obtain the amplitude: The intensity of the event is characterized by extracting the maximum value of the signal time series, as shown in the following formula: ; In the formula, s(t) represents the time series of the signal; Step S1-3, Obtain Energy Characteristics: Integrate the signal amplitude to obtain energy characteristics. The energy level directly reflects the degree of rock damage, as shown in the following formula: ; In the formula, and The start and end time interval of the event Steps S1-4: Obtain frequency band features: Perform frequency domain analysis using Fast Fourier Transform (FFT) to extract the energy distribution in the spectral density function P(f).

[0019] In the formula: The power spectral density function represents the distribution of energy at different frequencies f. Time-series waveform of acoustic emission signal; The start and end times of the event signal; Frequency variable; The complex exponential basis functions of the Fourier transform are used to convert signals from the time domain to the frequency domain. Then, by calculating the energy or quantity proportion of each frequency band signal in the current time window point set Pk, a characteristic hierarchical evolution index is constructed. This is achieved by monitoring the waxing and waning relationships of the three proportion curves: LDZ, MDZ, and HDZ (e.g., ...). Figure 4 As shown in the figure, the transition of rock fracturing mechanism from microscopic HDZ dominance to macroscopic LDZ dominance is identified. When the proportion of LDZ shows an explosive increase and the slope of the curve changes abruptly, it is determined as a critical signal of physical merging of nuclei.

[0020] Steps S1-5 involve waveform quality assessment: additionally extracting the event duration (measuring event persistence), calculating the signal-to-noise ratio (SNR), and performing saturation and crosstalk detection, as shown in the following formula:

[0021] In the formula: SNR: the core indicator for measuring the quality of the received signal; E: represents the "Expected Value" in mathematics, which in engineering calculations usually refers to the average value of the signal over a certain period of time; SignalPower: effective signal power, representing the energy of the actual acoustic emission / microseismic signal generated when microcracks inside the rock fracture; NoisePower: noise power, representing the energy of environmental background interference (such as drilling rig vibration, ambient noise) or useless electrical noise generated by the sensor itself.

[0022] Step S2: Use a time-inversion algorithm or a positioning algorithm to convert the event into a three-dimensional point, map the coordinates to a unified reference system, and perform three-dimensional positioning and coordinate normalization.

[0023] Specifically, such as Figure 2 As shown, 3D localization and coordinate normalization are performed, including: converting events into 3D points using time-lapse inversion or localization algorithms, mapping the coordinates to a unified reference frame, and performing 3D localization and coordinate normalization. The point cloud dataset after SNR filtering and coordinate normalization can effectively reduce the interference of non-fragmented noise on the spatial clustering evaluation index L(d). Specifically, time-lapse inversion or localization algorithms are used to convert one-dimensional time-domain events into 3D spatial points. Using time difference data from at least three sensors, the 3D position coordinates r are calculated using nonlinear least squares or Bayesian optimization. The localization formula is:

[0024] In the formula, c is the wave velocity inside the rock. Let be the Euclidean distance from the event point to sensor i. After localization, the results are mapped to a unified engineering coordinate system, and outliers are removed.

[0025] Step S3: Construct point sets by rolling according to intelligent time windows or strain windows, including: automatically adjusting the size of time windows or strain windows based on event timestamps and energy distributions, and outputting the point set P within each window. k This is for subsequent spatial clustering analysis.

[0026] Specifically, step S3 involves constructing a point set using either a smart time window or a strain window, which constitutes the smart time window / strain window partitioning. Specifically, to accurately capture the spatiotemporal evolution of crack nucleation at different stages, a dynamic partitioning strategy combining fixed and adaptive windows is adopted. In the basic embodiment, a cumulative time window is used to observe the global trajectory; in the enhanced embodiment, the system can automatically trigger slice window partitioning based on energy surges to capture transient merging behavior. During the stationary phase, a fixed time window or a fixed strain window is used. When an event rate or energy surge is detected, the window width is automatically reduced to improve temporal resolution.

[0027] Step S4: For the point set Pk of each window, generate the CSR envelope using Monte Carlo and calculate the 3D Ripley K(d) to L(d). That is, use Monte Carlo to generate a completely random spatial envelope and calculate the 3D Ripley K function K(d) to convert it to the L function L(d). Automatically identify significant clustering distance intervals and take the interval with the largest or stable deviation as the candidate clustering scale set D. k .

[0028] Specifically, step S4, spatial clustering intelligent evaluation (candidate scale generation), includes: for each pre-divided point set P k,A completely spatially random (CSR) envelope is generated based on Monte Carlo simulation. Subsequently, the point cloud clustering is measured by calculating the 3D Ripley K(d) function. To achieve variance stabilization, this function is converted to an L(d) function. By comparing the actual L(d) values ​​with the theoretical CSR envelope, significantly clustered distance intervals are automatically identified and used as the candidate clustering scale set D. K .

[0029] Preferably, in order to accurately analyze the distribution and evolution of crack nucleation regions in three-dimensional space, such as Figure 3 As shown, step S4, the spatial clustering analysis of the crack nucleation region, employs the three-dimensional Ripley K(d) and L(d) statistical methods, including: Step S4-1 Coordinate Acquisition and Distance Calculation: Extract the three-dimensional spatial coordinates of all discrete event points within the crack nucleation region, and iterate through and calculate the three-dimensional Euclidean distance between any two points.

[0030] Step S4-2(d) function calculation: Set the spatial observation scale (distance threshold d), and calculate the three-dimensional RipleyK(d) function based on the spatial distance between each point to obtain the expected number of point pairs within the distance d range.

[0031] Step S4-3, k(d) normalization and clustering assessment: The variance of the K(d) function is stabilized and normalized using the L(d) function; by comparing the actual L(d) value with the theoretical value under a completely spatially random distribution (CSR), the clustering, randomness or discreteness of the crack nucleation region under different spatial scales d is quantitatively assessed.

[0032] Step S5: Perform fast optimization using Bayesian optimization, genetic algorithm, or particle swarm optimization. Adaptive intelligent optimization of DBSCAN parameters is used to output the optimal DBSCAN parameters. Specifically, this includes: This step aims to avoid the enormous computational overhead of global grid search by employing a Gaussian Process (GP) surrogate model and a Bayesian optimization algorithm to quickly approximate the optimal solution of the comprehensive fitness objective function within the prior search space.

[0033] Step S5 is key to solving the problem of traditional density clustering relying on manual parameter tuning. In the candidate set... Within the framework, mechanisms such as Bayesian optimization, genetic algorithms, or particle swarm optimization are used for rapid optimization, outputting the optimal neighborhood radius (varepsilon) and core point threshold (MinPts). The optimization process is achieved by maximizing the following comprehensive fitness objective function: In the formula: Calinski–Harabasz index (CH): used to assess intra-cluster compactness and inter-cluster separation; Fragmentation penalty and Merge penalty: These respectively suppress non-physical distortions in the topology that excessively segment continuous fracture zones and anomalously connect independent fracture zones. The Frag index is not only related to the number of clusters, but also measures whether the identification results conform to the coherent physical characteristics of rock crack nucleation by calculating the coefficient of variation of cluster size distribution.

[0034] In the formula: A: represents intra-cluster distance / variance, measuring whether the microseismic points identified as a nucleated cluster are sufficiently compact. B: represents inter-cluster distance / variance, measuring whether the different nucleated clusters identified are sufficiently far apart. C: represents the total number of samples or features, a constant term related to the total number of microseismic events or the total spatial dispersion.

[0035] Noise Rate: Constrains the degree of loss of effective signal;

[0036] In the formula: The number of microseismic points that are judged as "noise" and removed by the DBSCAN algorithm during the clustering process because they are too isolated or too sparse.

[0037] The total number of microseismic event points collected within the current time window.

[0038] The specific implementation steps are as follows: Step S5-1: Define the optimization decision space: Define two-dimensional optimization decision variables consisting of density clustering parameters. Based on the candidate clustering scale set D extracted in step S4 by spatial statistical analysis. K The joint search boundary space H is constructed using the preset core point threshold set M: ; In the formula: H represents the joint search boundary space (or two-dimensional decision variable space) of intelligent optimization algorithms (such as Bayesian optimization). It defines the set of all legal solutions that the algorithm is allowed to explore when performing parameter adaptive optimization. D K : The set of candidate clustering scales within the k-th time window extracted based on spatial point pattern statistics (such as 3DRipley's K function). and : These are the minimum and maximum statistical distances in the candidate set, serving as the upper and lower bounds of the search interval. and The minimum and maximum limits of the effective nucleation point frequency are not pre-defined.

[0039] Specifically, this invention proposes a multi-objective collaborative optimization criterion based on matching spatial topology with physical representation. This criterion is achieved by constructing a comprehensive fitness objective function that includes clustering quality, noise rate, and topological distortion penalty terms. Specifically: The algorithm uses a candidate set D generated by spatial statistical scaling. K The goal is to find the optimal combination of parameters to overcome two non-physically realistic clustering topological distortions: one is the fragmentation distortion, where continuous microcrack nucleation regions are incorrectly fragmented into a large number of solitary subclusters; the other is the over-merging distortion, where multiple spatially independent nucleation regions are abnormally connected into a single giant cluster with chaotic physical meaning.

[0040] Therefore, the comprehensive fitness objective function constructed in this invention The expression is as follows: ; In the formula: A clustering quality index characterizing the significance of spatial clustering; The noise rate is used to characterize the degree of loss of effective signal. The fragmentation penalty term, constructed for the fragmented distortion morphology, is positively correlated with the total number of clusters and is used to suppress excessive fragmentation of the target. The over-merging penalty term for the over-merging distortion pattern is quantified by evaluating the spatial variance of the cluster size (such as the number of event points or the envelope volume) to suppress the target anomalous connectivity. These are the preset weighting coefficients for noise rate, fragmentation penalty, and over-merging penalty, respectively, with values ​​ranging from (0,1).

[0041] By maximizing the overall fitness objective function, the solution formula is shown below: ; In the formula, : Represents the neighborhood radius parameter (optimization independent variable) in density clustering algorithms (such as DBSCAN), used to define the local clustering and density reachability range of rock microseismic events in three-dimensional space; : Represents the core point threshold parameter (optimization independent variable) in the density clustering algorithm, which represents the minimum microseismic point frequency required to determine the effective nucleation of microcracks in a certain spatial region.

[0042] Step S5-2: Agent model initialization: Within the joint search boundary space H, randomization is performed using Latin hypercube sampling (LHS). Initial parameter combinations are used for each set of parameters, with the initial sample set serving as the observation data. A Gaussian process surrogate model gp is constructed to fit the unknown distribution of the objective function: In the formula, For predicting the mean function; The covariance kernel function is preferred (Matern5 / 2 kernel function is preferred to accommodate non-smooth abrupt changes in the parameter space).

[0043] Step S5-3: Construction of Acquisition Function and Generation of Candidate Solutions To balance the "exploration" (searching for regions of high uncertainty) and "exploration" (searching for regions of current extreme values) in the optimization process, an Upper Confidence Bound (UCB) acquisition function is constructed. :

[0044] In the formula, and These are the mean and standard deviation predicted by the Gaussian process surrogate model, respectively. To explore the development tradeoff coefficients, the optimal probe point candidate solution is output in the t-th iteration by maximizing the acquisition function. : ; Step S5-4: Closed-loop evaluation of the objective function and model update: Before calculating the actual comprehensive fitness score, discrete points identified as noise need to be removed. Only the intra-cluster compactness and inter-cluster separation of the identified candidate clusters are evaluated. The extracted candidate solutions... Substitute the DBSCAN algorithm to perform trial clustering of the point cloud dataset for the current time window.

[0045] Based on the clustering output, calculate the actual overall fitness score for the current parameters. The newly acquired observation pairs Add to observation dataset In this study, the hyperparameters of the Gaussian process surrogate model are updated using Bayesian posterior probabilities.

[0046] Step S5-5 convergence determination and optimal parameter output, repeat steps S5-3 and S5-4 iteratively until the preset convergence condition is met (continuous convergence). If the improvement in the objective function value is less than the decision threshold in the next iteration, or if the maximum number of iterations is reached, the iteration stops. After stopping the iteration, the parameter combination with the highest objective function score in the output observation dataset is used as the optimal density clustering parameter for the current time window. .

[0047] Step S6: Introduce clustering quality indicators and combine them with noise rate and physical constraint penalty terms that only apply to "non-physical real connectivity" to obtain density clustering identification candidate nucleation clusters.

[0048] Specifically, density clustering identifies candidate nucleation clusters. The optimal parameters obtained through optimization drive the DBSCAN algorithm to perform clustering, dividing the acoustic emission point cloud into several clusters with consistent density, and automatically identifying core points, boundary points, and noise points to form candidate nucleation clusters.

[0049] Step S7: Calculate the evidence features for each cluster Ck and output them using a rule engine or lightweight learning model to determine whether it is a crack nucleation cluster and evaluate the confidence level of the nucleation cluster.

[0050] Specifically, for each cluster C k The evidence features are calculated and output using a rule engine or lightweight learning model to determine whether it is a "crack nucleation cluster" and to evaluate the confidence of the nucleation cluster. In this embodiment, the confidence evaluation adopts a nonlinear mapping function, which quantifies the possibility of nucleation by evaluating the point cloud space duty cycle (i.e., the ratio of event frequency to envelope volume) of the target cluster.

[0051] Step S8: For clusters identified as nucleating, perform 3D reconstruction using Delaunay tetrahedronization and alpha complex shape, adaptively determine α-alpha, construct the α-shape, and perform surface smoothing to obtain a 3D nucleating body mesh. .

[0052] Specifically, such as Figure 4 As shown, step S8 performs AlphaComplexShape 3D reconstruction. High-confidence nucleating clusters are selected, and Delaunay tetrahedralization and AlphaComplexShape reconstruction are implemented to finely characterize the true 3D contours of the nucleating region. The rolling sphere radius parameter is adaptively determined by minimizing the reconstruction error objective function, and surface smoothing is performed after constructing the alpha-shape, ultimately outputting a high-quality 3D mesh.

[0053] When extracting the three-dimensional spatial geometric features of each target cluster, the rolling sphere radius parameter in the adaptive determination algorithm is used. Its adaptive determination criteria include the quantile criterion of the circumscribed sphere radius distribution or the minimum error criterion for reconstructing geometry; in the preliminary identification stage, The parameters can be obtained by linearly mapping the neighborhood radius Eps obtained through optimization; in the fine reconstruction stage, dynamic fine-tuning is performed by extracting the quantiles of the radius distribution of the circumscribed sphere of the Delaunay tetrahedron. Specifically, for any AE event cluster point set... The specific logic of employing the circumsphere radius distribution quantile criterion includes performing Delaunay tetrahedral partitioning on the discrete point cloud of the target cluster, extracting the set of circumsphere radii for all tetrahedral elements, constructing an empirical cumulative distribution function for the circumsphere radii, and selecting the radius value corresponding to a preset cumulative probability quantile as the rolling sphere radius parameter. The specific logic of adopting the reconstruction geometric minimum error criterion is as follows: A reconstruction error objective function is constructed, including a volume change rate penalty term and a surface topological hole penalty term. Within the value space formed by the set of circumscribed sphere radii, the radius value that minimizes the reconstruction error objective function is iteratively solved, and this value is determined as the rolling sphere radius parameter. .

[0054] Step S9: Output and quantify each nucleation body: Based on the generated 3D reconstructed mesh, quantitatively calculate the core geometric and mechanical properties of the nucleation body.

[0055] Specifically, the spatial geometric feature parameters and microseismic physical feature parameters of the reconstructed three-dimensional nucleating body mesh are extracted to complete the multi-dimensional quantitative characterization of each microcrack nucleating body within the current time window. The spatial geometric feature parameters include at least the three-dimensional coordinates of the centroid and the envelope volume, and the microseismic physical feature parameters include at least the cluster accumulated energy and the event frequency.

[0056] Step S10, for and Matching is performed, including centroid distance, volume overlap, or morphological similarity matching, to obtain nucleation body trajectories and achieve cross-window association and evolution tracking.

[0057] Above: k and k-1: represent two adjacent time windows (or monitoring phases). k is the latest time window currently being analyzed, and k-1 is its previous phase; B: Represents an independent three-dimensional nucleation body extracted through density clustering and network reconstruction; : Indicates the j-th nucleation site identified within the current time window k; : Indicates the i-th nucleating body that has been identified and recorded within the previous time window k-1; In order to determine and To determine whether they belong to the same nucleogen, the algorithm needs to compare features from three physical dimensions. The specific methods for constructing the match are as follows: 1. Centroid Distance Matching: Calculate the spatial relative position of the three-dimensional geometric centers (centroids) of two nuclei. If they are extensions of the same fracture origin, the centroids should not undergo instantaneous "translocation". 2: Volume overlap matching: Place the nucleation body 3D mesh from the previous stage and the mesh from the current stage in the same coordinate system, and observe the relative spatial nesting and overlapping volume sizes of them; 3. Morphological similarity matching: Compare the geometric shapes of the two. For example, aspect ratio, spatial principal axis direction (dominant direction of the fracture), etc.

[0058] "Matching degree" is a quantitative indicator (usually between 0 and 1) used for scoring and evaluation. and The degree of correlation. This is typically calculated by constructing a comprehensive evaluation function: Let the matching degree be M. ; Where: Euclidean distance score of the centroid ( (The closer the distance, the higher the score. A decay function is typically used to convert distance into a score between 0 and 1.) 3D volume intersection-union ratio score ( ): That is, the intersecting volume divided by the total merged volume. The formula is: The greater the overlap, the closer the score is to 1; Morphological similarity score It can be calculated by obtaining the cosine of the angle between the eigenvectors through principal component analysis (PCA).

[0059] Distance weights; This is the overlap weight; For shape weights; after calculating the matching degree between all nuclei in the current window and all nuclei in the previous window, a matching degree matrix will be formed.

[0060] Step S10, nucleogene cross-window association and evolution tracking: such as Figure 4 As shown, based on the centroid Euclidean distance, three-dimensional volume intersection-union ratio, and geometric similarity of the three-dimensional nuclei, the cross-window spatiotemporal correlation matching degree of the nuclei within adjacent time windows is calculated; based on the spatiotemporal correlation matching degree, the temporal evolution state of the nuclei is determined, a cross-window correlation sequence is constructed, and the spatiotemporal evolution trajectory of the microcrack nuclei is output.

[0061] After obtaining the matching degree matrix, the algorithm sets a threshold and, based on the matching correspondence (1-to-1, many-to-1, 0-to-1, etc.), determines and classifies the evolutionary state of the nucleogene into the following physical processes: State A: Stable expansion: The judgment logic is: a certain value in the previous window With a certain window It exhibits a high degree of "one-to-one" matching (greater than the threshold).

[0062] State B: Connectivity Merging The determination logic is as follows: multiple independent nuclei in the previous window and the same giant nuclei in the current window. It presents a high degree of matching ("many to one").

[0063] State C: Newborn germination: Decision logic: The current window's No ancestors with a satisfactory match ("0 to 1") were found in the previous window.

[0064] State D: Dormant / Closed Decision logic: the previous window No successor can be found in the current window ("1 to 0").

[0065] Specifically, in step S10, the evolutionary trajectory of nuclei is tracked by calculating the centroid distance, volume overlap, and morphological similarity of nuclei, marking the processes of nuclei addition, extinction, merging, and splitting, and plotting the evolution curves of volume-time and number-time.

[0066] Specifically, for nuclei generated in adjacent time windows, their centroid distance, volume intersection-union ratio, and geometric similarity are calculated to carry out spatiotemporal correlation matching.

[0067]

[0068] Based on this, the evolutionary states of each nucleation region, such as addition, growth, division, or merging, are determined, and the evolutionary trajectory curve of the nucleogen is constructed and output. Where: (Center of mass): The position of the center of the nucleus in adjacent windows moves; (Volume): Compare the changes in the size of the nuclei; (Shape): Compare the geometric similarity of nuclei. S11. Output and Application Interface: Outputs 3D visualization and recognition result files or recognition result databases of nucleation body mesh, AE point cloud and energy mapping, and provides early warning or control interfaces for monitoring systems to call.

[0069] Step S11 outputs the grid, spatial microseismic point cloud and multi-source energy mapping features for spatial registration and fusion, outputs a three-dimensional visualization model and structured nucleation evolution identification results for storage; based on the structured nucleation evolution identification results, provides dynamic early warning and linkage control signals for rock mass instability and rupture to the external monitoring system through the configured early warning control interface.

[0070] Specifically, such as Figure 5 , Figure 6 As shown, the confidence score of the nucleated body is output and used to update the parameters online. Specifically, the online parameter update is to adaptively update the weight coefficients of the noise rate, fragmentation penalty term, and over-merging penalty term in the comprehensive fitness objective function in the next time window.

[0071] The specific steps include: Extract the multidimensional physical features of each target cluster after reconstruction, including the average signal-to-noise ratio, spatial compactness, and rupture energy density of microseismic events within the cluster; Based on a multi-source evidence fusion model, the nucleation confidence score of each target cluster is calculated; an online feedback adjustment mechanism based on the nucleation confidence score is constructed to calculate the global average confidence score of all target clusters within the current time window; based on the deviation between the global average confidence score and the preset expected threshold, the penalty weight coefficient in the comprehensive fitness objective function in the next time window is adaptively updated to realize the online closed-loop evolution of dynamic target recognition parameters.

[0072] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0073] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0074] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0075] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0076] The above are merely embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention are included within the scope of the claims of the present invention pending approval.

Claims

1. A smart identification method for three-dimensional crack nucleation in rocks, characterized in that, The method includes: Step S1: Collect acoustic emission signals of rocks damaged under compression loading conditions, extract the amplitude, energy, frequency band characteristics and duration parameters of each triggering event from AE or microseismic point events, and calculate SNR and waveform quality indicators for saturation detection and crosstalk detection. Step S2: Use a time-inversion algorithm or a localization algorithm to convert the event into a three-dimensional point, map the coordinates to a unified reference system, and perform three-dimensional localization and coordinate normalization. Step S3: Construct a point set by scrolling through the smart time window or strain window; Step S4: For the point set Pk of each window, use the Monte Carlo method to generate a completely random spatial envelope, and calculate the three-dimensional Ripley K function K(d) to convert it into the L function L(d); automatically identify significant clustering distance intervals, and take the interval with the largest or stable deviation as the candidate clustering scale set D. k ; Step S5: Use Bayesian optimization, genetic algorithm or particle swarm optimization for fast optimization, and output the optimal density clustering algorithm parameters through adaptive intelligent optimization of density clustering algorithm parameters. Step S6: Introduce clustering quality indicators and combine them with noise rate and physical constraint penalty terms to obtain density clustering to identify candidate nucleation clusters; Step S7: Calculate the evidence features for each cluster Ck and output them using a rule engine or lightweight learning model to determine whether it is a crack nucleation cluster and evaluate the confidence of the nucleation cluster. Step S8: For the nucleation clusters identified, perform 3D reconstruction using Delaunay tetrahedralization and α complex shape, adaptively determine α-alpha, construct the α-shape, and perform surface smoothing to obtain a 3D nucleation mesh. ; Step S9: Output each nucleation body and quantify its characterization; Step S10, for and Matching is performed, including centroid distance, volume overlap, or morphological similarity matching, to obtain nucleation body trajectories and achieve cross-window association and evolution tracking; : Indicates the j-th nucleation site identified within the current time window k; This indicates that the i-th nucleating body has been identified and recorded within the previous time window k-1; Step S11: Output the 3D visualization and recognition result file or recognition result database of nucleation body mesh, AE point cloud and energy mapping, and provide an early warning or control interface for the monitoring system to call; Step S5 includes: Step S5-1 defines the optimization decision space, as shown in the following equation: ； In the formula, H represents the joint search boundary space of the intelligent optimization algorithm; D K : This is the set of candidate clustering scales extracted within the k-th time window based on spatial point pattern statistical priors; The minimum statistical distance in the candidate set; The maximum statistical distance in the candidate set; This is the lower limit of the pre-defined minimum number of effective nucleation point frequencies; The upper limit of the maximum effective nucleation point frequency is set in advance; The optimization process is achieved by maximizing the following comprehensive fitness objective function, as shown in the following equation: ； In the formula: CH is the Calinski-Harabasz index, used to evaluate the compactness within clusters and the separation between clusters; Frag and Merge penalties respectively suppress the non-physical distortions that over-segment continuous fracture zones and abnormally connect independent fracture zones; the Frag index also measures whether the identification results conform to the coherent physical characteristics of rock crack nucleation by calculating the coefficient of variation of cluster size distribution, as shown in the following formula: ； In the formula: A represents intra-cluster distance / variance; B represents inter-cluster distance / variance; C represents the total number of samples or features; The noise rate (NoiseRate) is used to constrain the loss of effective signal, as shown in the following formula: ； In the formula: This indicates the number of microseismic points that were identified as noise and removed during the clustering process by the DBSCAN algorithm. This represents the total number of microseismic event points collected within the current time window. The preset weighting coefficient for the noise rate; Preset weighting coefficients for fragmentation penalty items; The preset weighting coefficient for the merging penalty term; , and The value range is (0,1).

2. The intelligent identification method for three-dimensional crack nucleation in rocks as described in claim 1, characterized in that, Step S4 includes: Step S4-1, Coordinate Acquisition and Distance Calculation: Extract the three-dimensional spatial coordinates of all discrete event points within the crack nucleation region, and iterate through and calculate the three-dimensional Euclidean distance between any two points; Step S4-2, L(d) function calculation: Set the spatial observation scale, and calculate the three-dimensional Ripley K function K(d) based on the spatial distance between each point to obtain the expected number of point pairs within the distance d range; Step S4-3, K(d) normalization and clustering assessment: The variance of the K(d) function is stabilized and normalized using the L(d) function. By comparing the actual L(d) value with the theoretical value under a completely spatial random distribution, the clustering, randomness or discreteness of the crack nucleation region under different spatial scales d is quantitatively assessed.

3. The intelligent identification method for three-dimensional crack nucleation in rocks as described in claim 1, characterized in that, Step S5 also includes: Step S5-2 Proxy model initialization: For each set of parameters, an initial sampling set is used as the observation data to construct a Gaussian process surrogate model to fit the unknown distribution of the objective function; Step S5-3: Constructing the acquisition function and generating candidate solutions: Constructing the confidence upper limit set function And by maximizing the acquisition function, the optimal candidate solution for the probe point in the t-th iteration is output. : Step S5-4: Closed-loop evaluation of the objective function and model update: Extracted candidate solutions Substitute the data into the density clustering algorithm to perform trial clustering of the point cloud dataset for the current time window; Based on the clustering output, calculate the actual overall fitness score for the current parameters. ; to acquire new observation pairs Add to observation dataset In addition, the hyperparameters of the Gaussian process surrogate model are updated using Bayesian posterior probabilities; Step S5-5: Convergence judgment and optimal parameter output. Repeat steps S5-3 and S5-4 iteratively until the preset convergence condition is met or the maximum number of iterations is reached. After stopping the iteration, output the parameter combination with the highest score of the corresponding objective function in the observed dataset as the optimal density clustering parameters for the current time window. .

4. The intelligent identification method for three-dimensional crack nucleation in rocks as described in claim 1, characterized in that, Step S10 includes: Step S10-1, perform centroid distance matching: calculate the spatial relative position of the centroids of the two nuclei; if the centroids do not undergo instantaneous shift, they are determined to be the same fracture source; Step S10-2, perform volume overlap matching: place the nucleation body 3D mesh from the previous stage and the mesh from the current stage in the same coordinate system, and determine the size of the spatial relative nesting and overlapping of the meshes; Step S10-3: Perform morphological similarity matching: compare the geometry of the two nuclei, including aspect ratio and spatial principal axis direction.

5. The intelligent identification method for three-dimensional crack nucleation in rocks as described in claim 1, characterized in that, Step S10 includes: Construct a comprehensive evaluation function for evaluation and The degree of correlation is shown in the following formula: Let the matching degree be M. ； In the formula: Score based on Euclidean distance from the center of mass; The score is the intersection-union ratio of the three-dimensional volume. ; Scoring is based on morphological similarity. Distance weights; This is the overlap weight; This is the shape weight.

6. The intelligent identification method for three-dimensional crack nucleation in rocks as described in claim 1, characterized in that, Step S10 includes: If the previous window is independent Independent of the current window When a one-to-one match is observed, the nucleated body is determined to be in a stable expansion state. If multiple independent nuclei in the previous window are the same nuclei in the current window... During matching, the nuclei are determined to be in a connected and merged state; If the current window If no matching ancestor is found in the previous window, the nucleosome is determined to be in a new germination state. If the previous window If no successor is found in the current window, the nucleus is determined to be in a dormant or closed state.

Images