Rock mass high-energy event early warning method based on acoustic emission probability compensation and fracture reconstruction

Abstract

The application discloses a rock mass high-energy event early warning method of acoustic emission probability compensation and fissure reconstruction, and belongs to the technical field of rock mass engineering safety monitoring. The method is characterized in that acoustic emission sensors are arranged to obtain event waveforms, time of arrival, coordinates, energy and frequency spectrum, a forward sliding time window is used to inversely calculate a seismic source moment tensor, a rupture plane occurrence and a seismic source radius from a historical modeling window, a seismic source is taken as a node, edges are connected according to a double criterion that a center distance between two seismic sources is not greater than a sum of the seismic source radius and a geometric intersection of the rupture plane exists, a three-dimensional fissure network is reconstructed by graph theory and fissure energy density is calculated, a three-dimensional detection probability field is calculated based on a single sensor energy amplitude level-seismic source distance pickup probability and at least four sensor joint pickup conditions, the fissure energy density is compensated for probability, a rolling threshold is determined according to historical modeling window data, a high-risk area is identified and early warning is output. The application can improve the three-dimensional space early warning capability of rock mass high-energy events under the condition that there is a directional missing detection and spatial incompleteness in monitoring data.

Description

Technical Field

[0001] This invention relates to the fields of rock mass engineering safety monitoring and acoustic emission monitoring technology, and in particular to a method for early warning of high-energy events in rock masses based on acoustic emission probability compensation and fracture reconstruction. This method can be used for three-dimensional spatial early warning of high-energy acoustic emission events during the fracturing process of coal mine overburden, underground engineering surrounding rock, and other loaded rock masses. Background Technology

[0002] In deep coal mines and underground rock mass engineering, rock masses undergo several stages under high stress and disturbances such as mining, excavation, or other disturbances, including stable microcrack initiation, interaction and accumulation, and unstable connection. During the unstable connection stage, local fracture networks may rapidly expand with significant energy release, leading to high-energy acoustic emission events or mine tremor events. These high-energy events are closely related to dynamic disasters such as rockbursts and are important monitoring targets for assessing rock mass instability risks.

[0003] Acoustic emission monitoring can continuously record elastic wave signals generated by rock mass fracturing, and invert parameters such as event location, energy, source moment tensor, and fracture surface attitude from waveform, time, amplitude, spectrum, and multi-sensor recording information. Existing high-energy event early warning methods mostly use indicators such as event energy, frequency, b-value, stress, or wave velocity anomalies, which can reflect the overall activity level of the rock mass, but still have limitations in spatial positioning. When risk areas are close in planar projection but far apart vertically, relying solely on two-dimensional projection or regional zoning can easily lead to significant spatial early warning errors.

[0004] On the other hand, acoustic emission or microseismic monitoring data are usually not complete. Influenced by sensor placement, monitoring boundaries, propagation attenuation, and the radiation directionality of the seismic source, the probability of events at different locations, with different energies, and different rupture mechanisms being picked up by sensors varies. With a fixed sensor array, the radiation directionality of shear-type seismic sources makes events in strong radiation directions easier to record, while events in unfavorable directions may be missed. This results in an incomplete fracture network constructed based on recorded events and a systematic underestimation of energy accumulation indicators in regions with low detection probability.

[0005] Existing three-dimensional early warning methods mostly focus on regional hazard imaging or stress field inversion, making it difficult to directly characterize the geometric connectivity between potential fracture surfaces. Furthermore, most existing early warning indicators assume complete monitoring data and rarely use the spatial detection probability of the monitoring system directly for risk indicator compensation. Therefore, there is an urgent need for an early warning method that can unify source mechanisms, fracture geometric connectivity, energy accumulation characteristics, and monitoring detection probabilities into a single three-dimensional computational framework to improve the spatial early warning accuracy and stability of high-energy rock mass events even with incomplete monitoring data. Summary of the Invention

[0006] The purpose of this invention is to provide a method for early warning of high-energy events in rock masses based on acoustic emission probability compensation and fracture reconstruction, so as to solve the problems of insufficient spatial dimension, inadequate characterization of potential fracture surface connectivity, and underestimation of risk indicators caused by directional omissions in monitoring data in the existing technology.

[0007] To achieve the above objectives, the technical solution adopted by this invention is: a method for early warning of high-energy events in rock masses based on acoustic emission probability compensation and fracture reconstruction, comprising the following steps: S1. Establish an acoustic emission monitoring system to collect acoustic emission waveforms and event data generated during rock loading, mining, or excavation, and establish a forward sliding time window according to the time axis. The forward sliding time window includes non-overlapping historical modeling windows and future prediction windows; data after the current moment is not included in the current early warning model construction.

[0008] S2. Perform waveform preprocessing, time-based picking, and 3D positioning on acoustic emission events within the historical modeling window to obtain event spatial coordinates, event energy, and spectral parameters; perform source moment tensor inversion on the events and determine the true rupture surface to obtain the azimuth, dip angle, normal vector, slip vector, and source radius of the rupture surface.

[0009] S3. Using the epicenter of the event as a node, determine the fracture connectivity of any two events sequentially. When the Euclidean distance between the two epicenters is not greater than the sum of the radii of the two event epicenters, and the corresponding rupture surfaces of the two events have a geometric intersection in three-dimensional space, determine that the fractures corresponding to the two events are connected and establish an edge between the two nodes; traverse all event pairs in the historical modeling window and reconstruct the three-dimensional fracture network using graph theory.

[0010] S4. Calculate the fracture energy density of each node based on a three-dimensional fracture network. The fracture energy density is the ratio of the total acoustic emission energy of the events corresponding to the fractures connected to that node to the sum of the equivalent aperture parameters of the connected fractures. It is used to characterize the degree of energy accumulation on a unit fracture expansion scale of the local fracture network. A fracture energy density spatial field is formed by three-dimensional kernel density estimation.

[0011] S5. Based on the waveform acquisition results of each acoustic emission sensor, project the events within the historical modeling window onto an energy amplitude-source distance space with event energy amplitude as the abscissa and source distance as the ordinate. Statistically calculate the acquisition probability of a single sensor under different energy amplitudes and source distances. Based on the condition that an event must be acquired by at least four sensors for 3D localization, calculate the 3D detection probability field of the acoustic emission monitoring system at different spatial locations and different event energy amplitudes.

[0012] S6. Based on the three-dimensional detection probability field, perform probability compensation on the fracture energy density to obtain the probability-compensated fracture energy density. Probability compensation involves dividing the original fracture energy density by the larger of the detection probability at the corresponding spatial location and energy level and the preset lower limit of the detection probability, in order to avoid numerical instability in the compensation calculation in low detection probability areas.

[0013] S7. The warning threshold is determined solely by the probability-compensated fracture energy density distribution within the current historical modeling window. Three-dimensional regions where the probability-compensated fracture energy density exceeds the warning threshold are classified as high-risk areas. Spatial warning results for high-energy rock mass events within the future prediction window are output. Subsequently, the historical modeling window and the future prediction window are updated along the time axis to achieve continuous forward warning.

[0014] Further, in step S1, the length of the historical modeling window is determined based on the autocorrelation function of the acoustic emission event sequence. The autocorrelation function of the event count or event energy sequence is calculated, and the time lag corresponding to when the autocorrelation function first enters the Bartlett confidence threshold is used as the fluctuation scale. This time scale represents the duration during which preceding micro-fracture events still have a significant correlation with subsequent fracture activity. This fluctuation scale is used as the baseline length of the fixed-time historical modeling window. When the number of events within the fixed-time window is insufficient to stably reconstruct the fracture network, an event window containing a preset number of consecutive events is used instead of the fixed-time historical modeling window. The length of the future prediction window is no greater than the length of the historical modeling window, and the sliding step size is one future prediction window length, ensuring that the current prediction only uses previously occurring acoustic emission events.

[0015] Further, the source moment tensor inversion in step S2 includes waveform preprocessing, frequency filtering, P-wave first arrival picking, source location, low-frequency level calculation, initial moment tensor inversion, moment tensor iterative correction based on event clustering, and determination of the true rupture surface. First, events with similar waveforms and spatiotemporal characteristics are clustered. Within each cluster, the theoretical amplitude is calculated back and compared with the observed amplitude. The input amplitude and moment tensor results are iteratively updated until the normalized root mean square error meets a preset condition, at which point the source moment tensor, source type, and rupture surface attitude are output. The source radius is calculated using the Brune source model, and the source radius r satisfies... Where c is the wave velocity in the source region, k is the Brune source model coefficient, and f c This is the longitudinal wave corner frequency.

[0016] Furthermore, in step S3, the existence of a geometric intersection line between the corresponding rupture surfaces of the two events in three-dimensional space means that the finite rupture surfaces, determined by the azimuth, dip angle, and spatial position of the two rupture surfaces, have an intersection line within the effective radius of the two seismic sources. In the graph structure, a fracture network skeleton for three-dimensional visualization can be generated along the shortest propagation path between connected nodes.

[0017] Furthermore, in step S4, the equivalent aperture parameter of the fracture is calculated based on the trace of the source moment tensor, the P-wave corner frequency, the S-wave velocity in the source region, the Lamé constant, and the dot product of the unit vector normal to the fracture surface and the unit vector of the slip direction. The source radius, fracture aperture, and event energy used in the fracture energy density calculation are all obtained from acoustic emission monitoring data and source inversion results.

[0018] Furthermore, in step S5, the single sensor picking probability is determined as follows: at each grid node in the energy amplitude-source distance space, the number of picked events and the number of unpicked events within the search radius are counted. When the total number of events is not less than the minimum sample size, the picking probability is the ratio of the number of picked events to the total number of events. When the total number of events is less than the minimum sample size, corresponding events are added from the energy amplitude-source distance space to form a conservative estimate, and a monotonic physical constraint is applied that the picking probability must not decrease when the energy increases or the distance decreases.

[0019] Further, in step S5, the three-dimensional detection probability field is the probability that an event is picked up by four or more sensors, which is obtained by summing and complementing the probabilities that the event is picked up by exactly zero, one, two, and three sensors. The joint picking probability of each sensor combination is the product of the picking probabilities of the selected sensors and the product of the non-selected sensors' non-picking probabilities.

[0020] Furthermore, in step S6, the fracture energy density is probabilistically compensated. satisfy ,in The original fracture energy density, Let be the system detection probability at the spatial coordinates and energy level of the i-th acoustic emission event. This represents the lower limit of the detection probability.

[0021] Furthermore, in step S7, the warning threshold is the 75th percentile of the probability-compensated fracture energy density within the current historical modeling window, which corresponds to the warning threshold of the future prediction window, thus avoiding the use of future event information for ex-post threshold fitting.

[0022] Furthermore, in step S4, the three-dimensional kernel density estimation uses the fracture energy density value of each node and the three-dimensional coordinates of the node as inputs to perform spatial smoothing on the discrete node indices, thereby obtaining a continuous three-dimensional fracture energy density spatial field.

[0023] Compared with the prior art, the present invention has the following advantages: 1. This invention uses the event location, source radius, and rupture surface orientation to determine the geometric connectivity of the fracture, and directly reconstructs the three-dimensional fracture network in a graph theory manner. Compared with only performing two-dimensional projection or region clustering on the event coordinates, it can more directly characterize the spatial penetration process of the potential fracture surface.

[0024] 2. This invention constructs a fracture energy density index, which couples the acoustic emission energy involved in connecting fractures with the fracture aperture scale to characterize the local energy concentration on the potential fracture surface. The main parameters are objectively obtained by monitoring and source inversion, without relying on artificial weight allocation.

[0025] 3. This invention quantifies the detection probability of the sensor array under different location, energy and source mechanism conditions based on the detection probability method, and uses the detection probability to compensate for the crack energy density. This can correct the missed detection deviation caused by the radiation directionality of the source and the sensor layout, and improve the reliability of early warning under incomplete monitoring data conditions.

[0026] 4. This invention adopts a forward computation method that strictly isolates the historical modeling window and the future prediction window. The gap network, detection probability, risk index and early warning threshold all use data before the prediction time, which is suitable for building a continuous rolling real-time early warning process. Attached Figure Description

[0027] Figure 1 This is a flowchart of a rock mass high-energy event early warning method based on acoustic emission monitoring probability compensation and three-dimensional fracture network reconstruction in an embodiment of the present invention.

[0028] Figure 2 This is a schematic diagram illustrating the relationship between the historical modeling window and the future prediction window of the strict forward sliding time window in an embodiment of the present invention.

[0029] Figure 3 This is a schematic diagram of three-dimensional fracture network reconstruction based on source parameters and graph theory in an embodiment of the present invention.

[0030] Figure 4 This is a schematic diagram of the topological calculation unit for fracture energy density and three-dimensional kernel density imaging in an embodiment of the present invention.

[0031] Figure 5 This is a schematic diagram illustrating the process of calculating the single sensor pickup probability and the system three-dimensional detection probability field using the probability-based complete magnitude method in this embodiment of the invention.

[0032] Figure 6 This is a schematic diagram illustrating the calculation of probability-compensated fracture energy density and the identification of three-dimensional high-risk areas in an embodiment of the present invention.

[0033] Figure 7 This is a comparison diagram of the blind forward prediction results of high-energy events before and after compensation in an embodiment of the present invention. Detailed Implementation

[0034] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. The described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, other implementation methods obtained by those skilled in the art without creative effort all fall within the applicable scope of the technical solution of the present invention.

[0035] To make the objectives, features, and advantages of this invention clearer, the invention will be further described below in conjunction with the accompanying drawings and specific embodiments. The warning threshold, historical modeling window length, and acoustic emission energy threshold mentioned below are parameters from an example verified in red sandstone experiments. For different lithologies, different sensor arrays, or field engineering conditions, they should be updated accordingly based on the forward quantile method and monitoring system calibration results provided in this invention.

[0036] Example 1: High-energy event early warning process based on three-dimensional fracture network and monitoring probability compensation This embodiment provides a continuous forward early warning method for high-energy events in rock masses. For example... Figure 1 As shown, the process sequentially includes acoustic emission data acquisition, time window setting, source parameter inversion, three-dimensional fracture network reconstruction, fracture energy density calculation, monitoring and detection probability modeling, probability compensation, and three-dimensional early warning area output.

[0037] S1. Acoustic Emission Data Acquisition and Strict Forward Time Window Setting: At least one acoustic emission sensor meeting the requirements for three-dimensional positioning is deployed around the rock mass to be monitored, continuously recording the acoustic emission waveforms during the loading process. For each acoustic emission event, at least the event occurrence time, the initial arrival information of the longitudinal waves from multiple sensors, the event spatial coordinates, the event energy, the waveform spectrum, and the status of whether each sensor successfully picked up the event are saved. An event is considered a reliable three-dimensional positioning event only if it is effectively picked up on at least four sensors.

[0038] Set the History Modeling Window (HMW) and Future Prediction Window (FPW) along the loading or engineering timeline, such as... Figure 2As shown. The historical modeling window only contains data before the prediction time t, used for source moment tensor inversion, source parameter calculation, three-dimensional fracture network reconstruction, fracture energy density calculation, detection probability field calculation, and threshold determination; the future prediction window is only used to verify or receive future high-energy events and does not participate in the current model parameter estimation. The baseline length of the historical modeling window is determined by the autocorrelation function of the acoustic emission event sequence. The autocorrelation function (ACF) of the event count or event energy sequence is calculated, and the time lag corresponding to when the ACF first enters the Bartlett confidence limit is used as the fluctuation scale (SOF). This time scale represents the duration during which the preceding micro-fracture event still has a significant correlation with subsequent fracture activity. In the 20 sets of red sandstone tests in this embodiment, the average SOF of the acoustic emission count is 10.2 s, so 10.2 s is used as the baseline HMW length of the fixed time window. For samples with insufficient number of events within the fixed time window, an event-type historical modeling window is constructed using 200 consecutive acoustic emission events. The length of the future prediction window (FPW) is set to be no greater than the HMW, and the sliding step size is one FPW length.

[0039] S2. Source Parameter Inversion and Rupture Surface Determination: For each acoustic emission event within the HMW, the multi-channel waveforms are sequentially processed, including waveform truncation, filtering within the sensor's effective frequency band, first arrival pickup, source location calculation, and low-frequency level calculation. Based on the far-field amplitudes recorded by each sensor, the source moment tensor is inverted to obtain its components. Subsequently, isotropic components, double-couple components, and other non-double-couple components are determined based on the moment tensor decomposition, and the rupture surface normal vector and slip vector are obtained.

[0040] To improve the stability of source mechanism inversion, events with similar waveforms and spatiotemporal characteristics are first clustered. Theoretical amplitudes are then calculated back within each cluster and compared with observed amplitudes. The input amplitudes and moment tensor results are iteratively updated until the normalized root mean square error (RMS) meets a preset condition. In this embodiment, an RMS value less than 0.1 is used as the iteration stopping condition.

[0041] The true fracture surface is determined based on the moment tensor solution and the improved Berthes-Coulomb assumption, combined with the Griffith-Coulomb criterion. The azimuth, dip, normal vector, and slip vector of the fracture surface are output for each event. The P-wave corner frequency f is used... c The source region wave velocity c and the Brune model coefficient k are used to calculate the source radius r, i.e. .

[0042] S3. Three-dimensional fracture network reconstruction: Each acoustic emission event is represented as a finite fracture surface centered on the source coordinates, with the source radius r defining the spatial scale, and its direction determined by the azimuth and dip angles of the actual fracture surface of the event. For any two events i and j within the HMW, spatial proximity and fracture surface intersection are determined sequentially, such as... Figure 3 As shown.

[0043] First, calculate the Euclidean distance between the centers of the two seismic sources. : ;when > When the spatial scale of the two fissures is insufficient for them to contact, they are determined to be disconnected; when ≤ Then, the intersection line is obtained based on the plane equations of the two finite fracture surfaces, and it is determined whether the intersection line is simultaneously within the effective radius of the two fracture surfaces. Only when both the center distance condition and the geometric intersection condition are satisfied are the two fractures considered to be connected.

[0044] Construct a graph G=(V,E) from the source centers of all events, where V is the set of nodes and E is the set of connected edges. If fracture i and fracture j satisfy the aforementioned double connectivity condition, then an edge is established between node i and node j. After traversing all event pairs, a complete three-dimensional fracture network graph structure within the HMW is obtained. To display the fracture propagation path, the source centers can be connected along the shortest path between nodes within the connected components, forming a three-dimensional fracture network skeleton. Locations with higher node degrees or where multiple connected branches converge correspond to regions with stronger fracture interactions.

[0045] S4. Fracture Energy Density Calculation and 3D Imaging: For any node j in the 3D fracture network, extract n events that are directly connected to this node or connected through a defined topological computing unit, and calculate the total acoustic emission energy of these events. The fracture aperture parameters are calculated based on the source moment tensor, P-wave corner frequency, S-wave velocity, Lamé constant, and the normal and sliding directions of the rupture surface. The formula is used... Calculate the fracture energy density at node j, where This represents the equivalent aperture parameter of the corresponding connected fracture. The numerator of the fracture energy density represents the radiation energy of events connected to the fracture, and the denominator represents the geometric expansion scale of the connecting fracture channels. As local microcracks continue to accumulate, interact, and form through fractures, the energy accumulation level per unit fracture expansion scale increases, and the fracture energy density increases accordingly. Therefore, fracture energy density can be used to link the evolution of microscopic fracture networks with the occurrence of macroscopic high-energy events.

[0046] Among them, the equivalent aperture parameters of the two-phase intersecting fractures A and B, and the aperture of a single fracture. Calculate according to the following formula:

[0047]

[0048] in, This is the trace of the focal moment tensor. This is the P-wave corner frequency. These are constants related to the source model. The S-wave velocity in the source region. and Let be the Lamé constant, n and v be the unit vectors normal to the rupture surface and the unit vectors in the sliding direction, respectively, and θ be the corresponding angle parameter. In this embodiment... Take 1.32.

[0049] The fracture energy density values ​​at each node, along with the node's three-dimensional coordinates, are input into a three-dimensional kernel density estimate. Spatial smoothing is then applied to the discrete node indices to obtain a continuous three-dimensional fracture energy density spatial field, such as... Figure 4 As shown, regions with high fracture energy density represent localized energy concentrations near potential fracture surfaces and can serve as candidate locations for high-energy events.

[0050] S5. Calculation of Single Sensor Pickup Probability and System 3D Detection Probability Field: Due to the different radiation intensities of the same source mechanism in different propagation directions, the pickup capability of a fixed sensor array for events at different locations and with different rupture mechanisms is not uniform. This embodiment uses a probability-based full magnitude method to probabilistically describe the detection capability of the monitoring system, such as... Figure 5 As shown.

[0051] For the g-th sensor, all known events within the HMW are mapped to the energy amplitude-source distance (DM) space according to the energy amplitude lgE and the source distance R. If the sensor successfully picks up an event, the sample is marked as "picked"; otherwise, it is marked as "not picked up". A neighborhood with a normalized search radius is taken around each grid node, and the number of picked events N+ and the number of unpicked events N- are counted. To ensure statistical stability, a minimum sample size Nmin is set; in this embodiment, Nmin is 9. When the total number of events in the neighborhood is not less than Nmin, the single-sensor picking probability PD is the ratio of N+ to N++N-, i.e.

[0052] When there are insufficient samples in the neighborhood, samples are supplemented from adjacent regions in the DM space, and a monotonic constraint consistent with wave propagation attenuation is applied, meaning that the picking probability must not decrease when the energy increases or the distance decreases.

[0053] Finally, the PD distribution of each sensor under different lgE and R conditions was obtained.

[0054] For a monitoring network containing m effective sensors, calculate and sum the probabilities that an event is picked up by exactly 0, 1, 2, and 3 sensors, then subtract the sum from 1 to obtain the system detection probability P that the event is picked up by at least four sensors. EThe above calculations are performed on each grid node in the three-dimensional monitoring space and the energy amplitude of a given event to obtain the detection probability field of the three-dimensional system. ,Right now:

[0055] S6. Probability Compensation and High-Risk Area Identification: For each acoustic emission node in the three-dimensional fracture network, the system detection probability P is read at its coordinates and energy amplitude. E If P E Below the lower limit of detection probability P min Then replace it with P min In this embodiment, P is used. min A value of 0.1 is used as the lower limit of the detection probability. Then, according to the formula... The probability compensation index is obtained by dividing the original fracture energy density by the effective detection probability, such as... Figure 6 As shown, this processing can improve the underestimated fracture energy density in directions with low detection probability, compensating for the incomplete 3D fracture network caused by directional missed detections. A 3D spatial imaging process with the same fracture energy density is performed on the probability-compensated fracture energy density to obtain the probability-compensated 3D risk field.

[0056] S7. At any given prediction time, only the probability-compensated fracture energy density distribution within the current high-power microwave (HMW) is statistically analyzed, and the 75th percentile is taken as the warning threshold for the next power-shifted wave (FPW). Grid regions exceeding the threshold are marked as high-risk areas. The high-risk areas can output their three-dimensional coordinate range, center location, and risk index value to indicate the location of potential high-energy events within the future prediction window. After each FPW is completed, the time window is moved forward by one FPW length, and the above steps are repeated to form a continuous rolling warning system.

[0057] Example 2: Acoustic emission test verification of red sandstone under compression and tension shear To verify the applicability of the method of this invention to high-energy events in rock masses under different shear loading conditions, compressive-shear and tensile-shear acoustic emission tests were conducted on red sandstone. Representative stress states such as weak compressive-shear, strong compressive-shear, strong tensile-shear, and weak tensile-shear were set up in the tests to cover mixed failure modes in mine overburden that are dominated by shear and superimposed with tensile or compressive components.

[0058] The lateral compressive stresses for the compression-shear test were set to 0.4 MPa, 0.8 MPa, 1.2 MPa, 1.6 MPa, and 2.0 MPa, and the lateral tensile stresses for the tension-shear test were set to 0.2 MPa, 0.4 MPa, 0.6 MPa, 0.8 MPa, and 1.0 MPa. Both types of tests used two vertical shear loading rates of 0.05 mm / s and 0.1 mm / s, resulting in a total of 20 sets of test conditions.

[0059] Acoustic emission data were acquired using a PCI-2 system with eight RS-54A sensors. The acoustic emission threshold was 40 dB, the preamplifier gain was 40 dB, the sampling frequency was 2 MSPS, and the PDT, HDT, and HLT were set to 50 μs, 200 μs, and 300 μs, respectively. Before the formal experiment, energy calibration was performed on the same red sandstone using the standard lead core breaking method. The system's effective detection limit was approximately 50 aJ, and the average background noise energy was below 100 aJ. Therefore, events with energy greater than 100 aJ were considered valid crack propagation acoustic emission events, as shown in Table 1.

[0060] Table 1. Main parameters for acoustic emission acquisition and time window in the embodiments.

[0061] Autocorrelation analysis was performed on the acoustic emission events of 20 sets of experiments. The results showed that the shot number of acoustic emission (SOF) was greater than 10 s in 14 of the 20 sets of experiments, and the average SOF of all samples was 10.2 s. The temporal clustering of the shot number was stronger than that of the absolute energy sequence. Therefore, 10.2 s was used as the reference length of HMW in this embodiment, and the forward time window was set as described in Example 1.

[0062] Moment tensor inversion, rupture surface attitude calculation, and 3D fracture network reconstruction were performed on events within the high-magnitude microwave background (HMW). Two fractures were graphically connected only when the distance between their centers was no greater than the sum of their source radii and a 3D intersection line existed on the rupture surfaces. Fracture energy density was calculated for the connected network, and 3D kernel density imaging was performed.

[0063] To quantitatively define high-energy events, the Gutenberg-Richter relation was established based on 12,867 valid acoustic emission events from all 20 samples. As the event energy enters the unstable fracture penetration stage, the b-value decreases significantly. In the entire sample statistics, when the event energy exceeds 8 × 10⁻⁶... 4 After aJ, the b value continued to decrease, and at 10 5 The critical inflection point is reached at aJ, therefore the energy should be no less than 10. 5 The event aJ is defined as a high-energy event; for three events that do not have 10... 5 For samples with events greater than aJ, the supplementary threshold corresponding to the inflection point of their respective b-values ​​is used, and a value of 10 is uniformly taken. 5 aJ. A total of 92 high-energy events were obtained according to the above criteria for blind forward validation.

[0064] For each high-energy event to be verified, only acoustic emission events within the HMW before its occurrence are used to construct a three-dimensional fracture network, calculate the fracture energy density, establish a detection probability field, and determine the threshold. The target high-energy event itself and its subsequent data are not included in the model.

[0065] Before compensation, with a fracture energy density threshold of 0.6, the blind forward prediction accuracy for 92 high-energy events was 77%. After calculating the three-dimensional detection probability field for each HMW using the detection probability method, compensation was performed according to the probability compensation formula. Under the same threshold of 0.6, the prediction accuracy increased to 84%, the false negative rate decreased from 23% to 16%, and the relative false positive rate decreased from 28% to 25%. A comparison of the prediction results before and after compensation is shown below. Figure 7 As shown in Table 2.

[0066] Table 2 Comparison of high-energy event early warning performance before and after probability compensation.

[0067] After 1000 Bootstrap resampling operations on 92 high-energy events, the 95% confidence interval for the prediction accuracy before compensation was 67.4% to 84.8%, and the 95% confidence interval for the prediction accuracy after compensation was 75.8% to 90.9%. The 95% confidence interval for the difference in accuracy between the two was 1.1% to 13.0%, indicating that the improvement brought by probability compensation is stable.

[0068] Further Monte Carlo analysis was performed 1000 times on the azimuth error of the fracture surface. After randomly applying azimuth perturbations ranging from 0 to 20 degrees to the acoustic emission events participating in the fracture energy density network reconstruction, the coefficient of variation of the predicted high-risk volume increased from 3.2% at 0 degrees to 12.3% at 20 degrees. When the azimuth error reached 20 degrees, the spatial overlap rate between the core high-risk area and the unperturbed results was still 72.4%, and the average overlap rate for each error level was 85.8%. According to the azimuth error grouping, the average blind forward prediction accuracy of high-energy events decreased from 82.3% for the group with less than 5 degrees to 71.0% for the group with 15 to 20 degrees, indicating that improving the inversion accuracy of the fracture surface is beneficial to further improving the spatial early warning performance of this method.

[0069] The above experimental results show that reconstructing a three-dimensional fracture network using acoustic emission source parameters can extend the judgment of potential high-energy event locations from two-dimensional regions to high-risk regions near three-dimensional fracture surfaces, while detection probability compensation can reduce the event missed detection bias caused by source radiation directionality and fixed sensor layout.

[0070] It should be noted that the 0.6 threshold in Example 2 was only verified under laboratory conditions for red sandstone; for coal, granite, limestone, and other rock masses, their fracture evolution and energy release characteristics are different, and they should be recalibrated using corresponding experiments or historical field data. The above-mentioned field rolling threshold method is used to avoid directly extrapolating specific laboratory thresholds to different engineering conditions.

[0071] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for early warning of high-energy events in rock masses based on acoustic emission probability compensation and fracture reconstruction, characterized in that, Includes the following steps: S1. Establish an acoustic emission monitoring system to collect acoustic emission waveforms and event data generated during rock loading, mining or excavation, and establish a forward sliding time window according to the time axis. The forward sliding time window includes a non-overlapping historical modeling window and a future prediction window. S2. Perform waveform preprocessing, time-based picking, and 3D positioning on acoustic emission events within the historical modeling window to obtain event spatial coordinates, event energy, and spectral parameters. Perform source moment tensor inversion on the events and determine the true rupture surface to obtain the azimuth, dip, normal vector, slip vector, and source radius of the rupture surface. S3. Using the center of the event source as a node, determine the fracture connectivity of any two events in turn. When the Euclidean distance between the two source centers is not greater than the sum of the radii of the two event sources, and the corresponding rupture surfaces of the two events have a geometric intersection in three-dimensional space, determine that the fractures of the two events are connected and establish an edge between the two nodes. Traverse all event pairs in the historical modeling window and reconstruct the three-dimensional fracture network using graph theory. S4. Calculate the fracture energy density of each node based on the three-dimensional fracture network; S5. Based on the waveform picking results of each acoustic emission sensor, project the events within the historical modeling window onto the energy amplitude-source distance space with the event energy amplitude as the horizontal axis and the source distance as the vertical axis. Statistically calculate the picking probability of a single sensor under different energy amplitudes and source distances. Based on the condition that an event must be picked up by at least four sensors to perform three-dimensional positioning, calculate the three-dimensional detection probability field of the acoustic emission monitoring system under different spatial locations and different event energy amplitudes. S6. Based on the three-dimensional detection probability field, perform probability compensation on the fracture energy density to obtain the probability-compensated fracture energy density. S7. Determine the warning threshold using only the probability-compensated fracture energy density distribution within the current historical modeling window. Identify three-dimensional regions where the probability-compensated fracture energy density exceeds the warning threshold as high-risk regions. Output the spatial warning results of high-energy rock events within the future prediction window. Then, update the historical modeling window and the future prediction window along the time axis.

2. The method according to claim 1, characterized in that, In step S1, the length of the historical modeling window is determined based on the autocorrelation function of the acoustic emission event sequence. The time lag corresponding to the first entry of the autocorrelation function into the Bartlett confidence limit is used as the fluctuation scale, and this fluctuation scale is used as the reference length of the fixed-time historical modeling window. When the number of events within the fixed-time window is insufficient to stably reconstruct the fracture network, an event window containing a preset number of consecutive events is used to replace the fixed-time historical modeling window. The length of the future prediction window is not greater than the length of the historical modeling window, and the sliding step size is taken as the length of the future prediction window, so that the current prediction only uses the acoustic emission events that have occurred previously.

3. The method according to claim 1, characterized in that, The source moment tensor inversion in step S2 includes waveform preprocessing, frequency filtering, P-wave first arrival picking, source location, low-frequency level calculation, initial moment tensor inversion, moment tensor iterative correction based on event clustering, and determination of the actual rupture surface. When the inversion error meets the preset conditions, the source moment tensor, source type, and rupture surface attitude are output. The source radius is calculated using the Brune source model, and the source radius r satisfies... Where c is the wave velocity in the source region, k is the Brune source model coefficient, and f c This is the longitudinal wave corner frequency.

4. The method according to claim 1, characterized in that, In step S3, the existence of a geometric intersection line between the rupture surfaces corresponding to the two events in three-dimensional space means that the finite rupture surfaces determined by the azimuth, dip angle and spatial position of the two rupture surfaces have an intersection line within the effective radius of the two seismic sources.

5. The method according to claim 1, characterized in that, In step S4, the fracture energy density is the ratio of the total acoustic emission energy of the event corresponding to the fracture connected to the node to the sum of the equivalent aperture parameters of the connected fractures. The fracture energy density spatial field is formed by three-dimensional kernel density estimation. The equivalent aperture parameters of the fracture are calculated based on the trace of the source moment tensor, the P-wave corner frequency, the shear wave velocity in the source region, the Lamé constant, and the dot product of the unit vector of the fracture surface normal and the unit vector of the sliding direction.

6. The method according to claim 1, characterized in that, In step S5, the single sensor picking probability is determined as follows: at each grid node in the energy amplitude-source distance space, the number of picked events and the number of unpicked events within the search radius are counted. When the total number of events is not less than the minimum sample size, the picking probability is the ratio of the number of picked events to the total number of events. When the total number of events is less than the minimum sample size, corresponding events are added from the energy amplitude-source distance space to form a conservative estimate, and a monotonic physical constraint is applied that the picking probability must not decrease when the energy increases or the distance decreases.

7. The method according to claim 1, characterized in that, In step S5, the three-dimensional detection probability field is the probability that an event is picked up by four or more sensors. It is obtained by summing the probabilities that the event is picked up by exactly zero, one, two, and three sensors and then taking the complement. The joint picking probability of each sensor combination is the product of the picking probabilities of the selected sensors and the product of the non-selected sensors' non-picking probabilities.

8. The method according to claim 1, characterized in that, In step S6, probability compensation involves dividing the original fracture energy density by the larger of the detection probability at the corresponding spatial location and energy level and the preset lower limit of the detection probability. Specifically, this means: probability-compensating the fracture energy density. satisfy ,in The original fracture energy density, Let be the system detection probability at the spatial coordinates and energy level of the i-th acoustic emission event. This represents the lower limit of the detection probability.

9. The method according to claim 1, characterized in that, In step S7, the warning threshold is the 75th percentile of the probability-compensated fracture energy density within the current historical modeling window, which corresponds to the warning threshold of the future prediction window, thus avoiding the use of future event information for ex-post threshold fitting.

10. The method according to claim 5, characterized in that, In step S4, the three-dimensional kernel density estimation uses the fracture energy density value of each node and the three-dimensional coordinates of the node as inputs, and performs spatial smoothing on the discrete node indexes to obtain a continuous three-dimensional fracture energy density spatial field.

Images