A surrounding rock loose circle drilling distributed optical fiber sound wave detection method and system

Abstract

This invention belongs to the field of surrounding rock loosening zone detection technology, specifically providing a distributed fiber optic acoustic wave detection method and system for surrounding rock loosening zones in boreholes. The method includes: acquiring raw vibration signals of the surrounding rock using a distributed fiber optic acoustic sensing system; preprocessing and extracting features from the raw vibration signals to obtain multi-parameter features, and combining the multi-parameter features to construct a fused feature vector; constructing an identification model based on the fused feature vector, and using an adaptive gating threshold mechanism to make decisions on the model output; obtaining the rock mass state category and discrimination confidence level at the target surrounding rock depth point, and drawing a comprehensive state profile of the surrounding rock. Addressing the limitations of traditional surrounding rock loosening zone detection methods, such as discontinuous monitoring, difficulty in picking up weak vibration signals, and inability to distinguish loosening frequency bands at different depths, this invention achieves non-destructive detection of the surrounding rock loosening zone range, determination of high-stress bearing areas, and detection of potential delamination and cavities.

Description

Technical Field

[0001] This invention belongs to the field of surrounding rock loosening zone detection technology, specifically relating to a distributed optical fiber acoustic wave detection method and system for surrounding rock loosening zone borehole drilling. Background Technology

[0002] Currently, monitoring the stability of surrounding rock in underground engineering projects such as coal mine roadways and tunnels mainly relies on methods such as acoustic wave testing, stress measurement, drill cuttings method, and microseismic monitoring. These methods have the following shortcomings: (1) Poor spatial continuity. Traditional methods are mostly point measurements, which make it difficult to obtain continuous depth distribution information. For example, the drill cuttings method can only provide local sampling data and cannot achieve continuous monitoring of the entire hole depth.

[0003] (2) Significant construction interference. The drill cuttings method requires repeated sampling and can only reflect the local stress state, which may lead to further disturbance of the surrounding rock; stress measurement requires embedded sensors, which are complicated to install.

[0004] (3) Limited accuracy and resolution. Acoustic wave testing is greatly affected by coupling conditions, making it difficult to effectively identify delamination and cavities, especially under complex geological conditions where signal attenuation is severe.

[0005] While distributed optical fiber acoustic sensing (DAS) technology has been used in seismic exploration and stress monitoring, an effective solution for loosening zone detection has yet to be developed. Traditional methods for detecting loosening zones in underground engineering surrounding rock lack multi-channel synchronous vibration frequency acquisition, weak vibration enhancement and pickup, and multi-frequency component separation. This makes it difficult to distinguish the loosening vibration characteristics of different locations and frequency bands in the surrounding rock. Not only can it not complete continuous monitoring along the entire borehole length and track the evolution of loosening zones over long periods, but it also lacks supporting detection methods and devices that can integrate multiple parameters such as wave velocity, frequency, mode, and coherence to improve the accuracy of loosening zone identification. Summary of the Invention

[0006] To address the problems existing in the prior art, this invention provides a distributed fiber optic acoustic wave detection method and system for loosened rock zones in boreholes. Utilizing an expandable long tube covered with optical fiber as the sensing element, and drawing on the photoelastic effect of fiber optics, the micro-vibrations and micro-deformations induced by the loosened rock zone are converted into optical phase signals. By inflating the fiber to ensure it adheres tightly to the borehole wall, and combining this with external excitation or environmental vibration sources to acquire response signals, continuous vibration and frequency signal acquisition across the entire area is achieved, overcoming the shortcomings of traditional single-point detection methods, such as discreteness and large blind zones. The loosened zone, bearing area, and delamination / cavity are identified through a multi-parameter joint criterion of wave velocity, frequency attenuation, vibration mode, and coherence. This method is not only suitable for one-time testing but can also be used for automated long-term monitoring during long-term operation of tunnels, utilizing mining machinery, blasting, or mine vibrations as vibration sources.

[0007] To achieve the above objectives, the present invention provides the following solution: A method for distributed fiber optic acoustic detection of loosened rock zones in boreholes includes: The original vibration signal of the surrounding rock was collected using a distributed fiber optic acoustic sensing system. The original vibration signal is preprocessed and features are extracted to obtain multi-parameter features; wherein, the multi-parameter features include wave velocity features, frequency decay features, vibration mode features, and coherence features; By fusing the multi-parameter features, a depth-aligned fused feature vector is constructed; A recognition model is constructed based on the fused feature vector and the improved lightweight gradient boosting decision tree. The improved lightweight gradient boosting decision tree uses a binary decision tree with depth constraints as the base learner, performs boosting by gradient unidirectional residual approximation, and adopts an adaptive gating threshold mechanism to make decisions on the posterior probability output by the recognition model. Based on the output of the identification model, the rock mass state category and discrimination confidence level of the target surrounding rock depth point are obtained; Based on the multi-parameter features, the rock mass state category and discrimination confidence level, and the borehole geological background information, a comprehensive state profile of the surrounding rock is drawn.

[0008] Preferably, the distributed fiber optic acoustic sensing system includes: An expandable long cylinder with distributed optical fibers arranged in a hybrid spiral and straight pattern on its outer wall, used for insertion into surrounding rock detection holes; An inflation system, connected to the expandable long cylinder, is used to inflate the expandable long cylinder with air to make it expand and fit against the hole wall; Excitation source, used to generate vibration signals in the surrounding rock; The DAS data acquisition unit, connected to the distributed optical fiber, is used to acquire the original vibration signal of the surrounding rock, with a sampling rate of not less than 5kHz and a gauge length adjustment range of 0.5~2m. The original vibration signal is stored in the form of a multi-channel strain time series, including timestamp, channel number and micro-strain value, and the sampling rate, gauge length, air pressure and trigger time are recorded.

[0009] Preferably, the method for preprocessing the original vibration signal includes: The original vibration signal was bandpass filtered to obtain the main frequency bands reflecting the rock mass structure; Wavelet threshold denoising method is used to denoise the main frequency bands in order to suppress random noise; For data acquired by active excitation, the response signals of repeated excitations are time-domain aligned and superimposed by trigger alignment and signal averaging to suppress incoherent noise.

[0010] Preferably, methods for constructing fused feature vectors include: Based on the cross-correlation time delay estimation method, the travel time difference of vibration signals in adjacent sensing channels is calculated, a depth-velocity profile is constructed, and wave velocity characteristics are obtained. The amplitude spectrum is obtained by performing a fast Fourier transform on the preprocessed vibration signal. The logarithmic attenuation relationship of amplitude with propagation distance within a specified frequency band is fitted by linear regression method, and the attenuation coefficient is solved to obtain the frequency attenuation characteristics. Variational mode decomposition is performed on the preprocessed vibration signal to obtain the intrinsic mode components, and the energy proportion of each dominant mode is calculated to obtain the vibration mode characteristics. Based on the preprocessed vibration signal, the standardized cross-correlation coefficient is calculated in the time domain, the amplitude squared coherence function is calculated in the frequency domain, the depth-coherence curve is constructed, and the coherence characteristics are obtained. The fused feature vector is constructed based on the wave velocity feature, the frequency attenuation feature, the vibration mode feature, and the coherence feature.

[0011] Preferably, the comprehensive profile of the surrounding rock includes an animated view, a high-stress bearing zone, and delamination and cavities.

[0012] Preferably, it also includes extracting key information based on the comprehensive state profile of the surrounding rock to construct a structured text report, the structured text report including a comprehensive evaluation of the loosened zone, a delamination list, and a description of the depth range and parameter characteristics of the stable bearing zone; The comprehensive evaluation of the loosening zone includes the starting depth, ending depth, thickness, and average wave velocity reduction of the loosening zone. The delamination list includes the location, thickness, footwall and hanging wall lithology, and reliability rating of all identified discontinuities.

[0013] Preferably, the distributed fiber optic acoustic sensing system is further optimized based on the comprehensive rock condition profile. Based on the abnormal areas identified in the comprehensive rock condition profile, adjust the spatial gauge length of the distributed fiber optic acoustic sensing system. Adjust the sweep frequency range or excitation method of the excitation source according to the attenuation characteristics of the loosened ring or broken zone for a specific frequency band signal; Based on the trend of decreasing coherence or signal energy, coupling agent is replenished to a specified section through the built-in injection channel; Adjust the pressure setting of the air filling system according to the signal quality of irregular hole shape or alternating soft and hard rock sections; Based on the changes in the loosened zone boundary shown in the comprehensive state profile of the surrounding rock, adjust the monitoring frequency or deploy new detection holes.

[0014] The present invention also provides a distributed fiber optic acoustic detection system for boreholes in the loosened zone of surrounding rock, for implementing the method, comprising: The signal acquisition module is used to acquire the original vibration signal of the surrounding rock using a distributed fiber optic acoustic sensing system. At the same time, it is used with a photoelectric detection array to complete photoelectric conversion, ensuring the accuracy of weak vibration signal acquisition. The signal analysis module is used to preprocess and extract features from the original vibration signal to obtain multi-parameter features, including wave velocity features, frequency decay features, vibration mode features, and coherence features. A cross-correlation time delay algorithm is used to simultaneously extract vibration frequency and spatial location information to achieve joint detection of loosening zone position and frequency features. The feature fusion module is used to fuse the multi-parameter features and construct a depth-aligned fused feature vector; The recognition model construction module is used to construct a recognition model based on the fused feature vector and the improved lightweight gradient boosting decision tree. The improved lightweight gradient boosting decision tree uses a binary decision tree with depth constraints as the base learner, performs boosting by gradient unidirectional residual approximation, and adopts an adaptive gating threshold mechanism to make decisions on the posterior probability output by the recognition model. The state recognition module is used to obtain the rock mass state category and discrimination confidence level of the target surrounding rock depth point based on the output of the recognition model; The drawing module is used to draw a comprehensive state profile of the surrounding rock based on the multi-parameter features, the rock mass state category and the discrimination confidence level, and the borehole geological background information.

[0015] Compared with existing technologies, the advantages of this invention are as follows: It employs an expandable long tube covered with optical fiber as the sensing unit, and through inflation, tightly couples the optical fiber to the borehole wall. Relying on artificial excitation or natural seismic sources, it collects rock vibration response signals, effectively improving the sensitivity and accuracy of acquiring weak loosening signals in complex working conditions. This invention can analyze vibration signals in real time, distinguishing between shallow and deep loosening characteristics of the surrounding rock based on frequency differences, achieving continuous monitoring throughout the borehole, and accurately depicting the spatial distribution of the loosening zone. By constructing a multi-parameter joint criterion of wave velocity, frequency attenuation, vibration mode, and signal coherence, it accurately identifies the loosening zone and high-stress bearing area of ​​the surrounding rock, effectively detecting hidden defects such as rock delamination and cavities. This invention can accommodate both one-time non-destructive on-site testing and long-term automated monitoring based on engineering disturbances, and is not constrained by the borehole layout angle. It can achieve full-position detection of the roadway sides horizontally, the roof and floor vertically, and obliquely, avoiding the problems of poor angle adaptability and blind spots in traditional monitoring.

[0016] This invention solves the technical defects of traditional surrounding rock monitoring, such as low accuracy, weak adaptability to working conditions, and incomplete monitoring coverage. It features high sensing sensitivity, multi-dimensional discrimination parameters, flexible monitoring modes, and full-area detection without blind spots. It is applicable to various underground projects such as coal mine roadways, traffic tunnels, and water conservancy dams. It can provide effective technical support for rock mass stability assessment and long-term structural health monitoring, and has high engineering application and promotion value. Attached Figure Description

[0017] To more clearly illustrate the technical solution of the present invention, the drawings used in the embodiments are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart of a method according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the expandable elongated cylindrical structure according to an embodiment of the present invention; Figure 2 (a) in the diagram is a schematic of the detection scenario. Figure 2 (b) in the diagram is a schematic diagram of an expandable long cylinder; Figure 3 This is a depth-parameter curve (wave velocity, mode, coherence) of an embodiment of the present invention. Figure 4 This is a schematic diagram of multi-parameter fusion discrimination in an embodiment of the present invention.

[0019] Reference numerals: 11-hole wall, 12-coupling agent layer, 13-cylindrical body, 14-spiral fiber, 15-straight fiber, 1-detection hole, 2-air filling system, 3-excitation source. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0022] Example 1: like Figure 1 As shown, a distributed fiber optic acoustic wave detection method for the loosened zone of surrounding rock in boreholes includes: S1: Use a distributed fiber optic acoustic sensing system to collect the original vibration signal of the surrounding rock (hole wall 11).

[0023] A further implementation method is, such as Figure 2 As shown, Figure 2 (a) in the diagram is a schematic of the detection scenario. Figure 2 (b) is a schematic diagram of an expandable long cylinder; the distributed fiber optic acoustic sensing system includes: An expandable cylindrical tube with distributed optical fibers arranged in a hybrid spiral and linear pattern on its outer wall (e.g., Figure 2 The spiral optical fiber 14 and the straight optical fiber 15 are used for insertion into the surrounding rock detection hole 1; the cylindrical body 13 is made of flexible material (rubber / polymer composite) and can be inflated internally. Before insertion, a coupling agent (silicone grease / petroleum jelly) is applied to the outer wall, and a coupling agent layer 12 is designed; at the same time, an "internal injection channel" can be designed to inject low-viscosity coupling agent after inflation.

[0024] The inflation system 2 is connected to the expandable tube and is used to inflate the expandable tube to make it expand and fit against the hole wall 11; the inflation pressure is 0.2–0.3 MPa, and the optical fiber is protected by a double-layer airbag + flexible pad.

[0025] Excitation source 3 is used to generate vibration signals in the surrounding rock; the excitation methods include: a) instrumented hammering; b) electromagnetic exciter or vibration motor; c) using mining machinery vibration, blasting or mine tremor in long-term monitoring mode.

[0026] The DAS data acquisition unit, connected to distributed optical fiber, is used to acquire the original vibration signal of the surrounding rock, with a sampling rate of not less than 5kHz and a gauge length adjustment range of 0.5~2m.

[0027] The original vibration signal is stored in the form of a multi-channel strain time series, including timestamp, channel number and micro-strain value, and the sampling rate, gauge length, air pressure and trigger time are recorded.

[0028] Drill a test hole 1 radially in the surrounding rock of the tunnel. The hole depth is 3–5 times the radius of the tunnel, and the hole diameter is 42–50 mm.

[0029] During the acquisition process, the signal-to-noise ratio and channel consistency are monitored in real time to ensure signal quality. The obtained data can be directly connected to subsequent processing modules for feature extraction and analysis such as wave velocity, frequency attenuation, mode and coherence.

[0030] S2: The original vibration signal is preprocessed and features are extracted to obtain multi-parameter features, including wave velocity features, frequency decay features, vibration mode features, and coherence features. The multi-parameter fusion discrimination mechanism can improve accuracy and anti-interference ability.

[0031] A further implementation method for preprocessing the original vibration signal includes: performing bandpass filtering on the original vibration signal to obtain the main frequency bands reflecting the rock mass structure; using wavelet threshold denoising to denoise the main frequency bands to suppress random noise; and for data acquired by active excitation, performing time-domain alignment and superposition of the response signals from multiple repeated excitations by trigger alignment and signal averaging to suppress incoherent noise.

[0032] In this embodiment, the processing and feature extraction of the original vibration signal is a systematic process aimed at extracting key indicators for rock mass structure identification from massive strain data. The overall technical approach can be summarized as: preprocessing → feature extraction → parameter standardization, ultimately outputting a set of sensitive parameters for different rock mass states (intact, loose, delamination), which can be divided into the following three steps.

[0033] 1. Preprocessing.

[0034] The core of preprocessing is to improve the signal-to-noise ratio and extract effective vibration events. First, the original strain time series is bandpass filtered (e.g., 10-1000 Hz) to retain the main frequency bands reflecting the rock mass structure while suppressing low-frequency drift and high-frequency noise. Then, wavelet thresholding denoising (e.g., using the 'sym8' wavelet basis and soft thresholding rules) is employed to further separate the signal from random noise. For actively excited data, trigger alignment and signal averaging are used to time-domain align and superimpose the response signals from multiple repeated excitations, effectively suppressing incoherent noise and improving data stability.

[0035] 2. Multi-parameter feature extraction.

[0036] The preprocessed signal enters the parallel feature extraction channel to generate four categories of discrimination parameters: (1) Wave velocity: The travel time difference of the first arrival of the P-wave between adjacent sensing channels is accurately calculated by the cross-correlation time delay estimation method. Combined with the known gauge length ( ),according to The longitudinal wave velocity is calculated segment by segment. The spatial variation of the wave velocity directly maps to the distribution of the rock mass's elastic modulus.

[0037] (2) Frequency attenuation characteristics: Fast Fourier Transform (FFT) is performed on the signal segment of each channel to obtain its amplitude spectrum. By analyzing the attenuation law of the dominant frequency of the amplitude spectrum with the propagation distance, the attenuation coefficient is quantitatively calculated using the spectrum ratio method or the centroid frequency shift method. High attenuation zones typically correspond to fractured or fissured rock masses.

[0038] (3) Vibration modes: Variational mode decomposition is applied to adaptively decompose the non-stationary vibration signal into a series of intrinsic mode functions (IMFs). The energy proportion and instantaneous frequency of each IMF component are calculated, and the dominant mode energy and mode complexity characterizing the overall vibration characteristics of the rock mass are extracted. Loss of structural integrity will lead to a decrease in dominant mode energy and mode disorder.

[0039] (4) Coherence: Within a set time window, calculate the cross-correlation coefficient or spectral coherence coefficient between two adjacent sensing channels. The coherence (Coh) value is between 0 and 1 and is used to quantify the consistency of the vibration waveform between two measuring points. Low coherence indicates the presence of discontinuities in the rock mass (such as delamination or cavities), leading to interruption or disorder in the vibration transmission path.

[0040] 3. Parameter standardization.

[0041] To ensure comparability between different parameters and subsequent fusion, the extracted raw parameter values ​​are subjected to depth alignment and normalization. Typically, the measured values ​​from the intact rock mass segment near the borehole opening (i.e., the P-wave velocity measurement value of the intact rock mass segment near the borehole opening) are used. Measurement value of attenuation coefficient of intact rock mass section at borehole opening Using this as a reference benchmark, calculate the relative values ​​of parameters at each depth point (such as wave velocity ratio). attenuation coefficient ratio This forms a multi-parameter feature profile that is continuously distributed along the borehole depth.

[0042] Specifically, the method for extracting multi-parameter features in this embodiment includes: Based on the cross-correlation time delay estimation method, the travel time difference of vibration signals from adjacent sensing channels is calculated to construct a depth-velocity profile and obtain wave velocity characteristics. Specifically, wave velocity is a direct indicator reflecting the dynamic elastic modulus and integrity of the rock mass. During the analysis, the first arrival travel time determined by the cross-correlation function is first corrected for residuals to eliminate the influence of excitation point error and instrument delay. Subsequently, the P-wave velocity values ​​at each depth point are calculated. The focus of the analysis is to establish a depth-velocity profile of wave velocity and identify its spatial variation characteristics: in the intact bearing zone, the wave velocity is high and the curve is gentle; in the loosened zone, due to fracture development and stress release, the wave velocity exhibits a significant negative gradient anomaly, typically decreasing to below 70%-85% of the reference value. The sliding window averaging method is used to smooth local fluctuations, and the first derivative is used to help locate the wave velocity abrupt change interface, which often corresponds to the boundary of the loosened zone or the delamination location. Wave velocity criteria: , For depth Longitudinal wave velocity of the rock mass; The spatial gauge length of the DAS system; For depth The vibration travel time difference between adjacent channels. It was determined to be a loose ring.

[0043] The preprocessed vibration signal was subjected to a Fast Fourier Transform (FFT) to obtain the amplitude spectrum. A linear regression method was used to fit the logarithmic attenuation relationship of amplitude with propagation distance within a specified frequency band, and the attenuation coefficient was solved to obtain the frequency attenuation characteristics. Specifically, this parameter is used to assess the rock mass's ability to absorb and scatter vibration energy, and is particularly sensitive for identifying fractured zones and fissure zones. The analysis is based on the amplitude spectrum, using a linear regression method to fit the logarithmic attenuation relationship of amplitude with propagation distance within a specified frequency band, and solving for the attenuation coefficient (…). To improve stability, multiple center frequencies are often used for parallel calculations, and the statistical median is taken. High attenuation region ( A value exceeding 20% ​​above the reference value indicates a fractured rock mass, dense fissures, or the presence of weak interlayers. Analysis must be combined with wave velocity results for comprehensive judgment to avoid misjudgment due to variations in water content or lithology caused by a single parameter. Frequency attenuation criterion: , like This area was identified as a fractured zone. For frequency ,depth Signal amplitude; The initial amplitude at the excitation end; For frequency The corresponding attenuation coefficient; The coordinates are along the borehole depth.

[0044] The preprocessed vibration signal is subjected to variational mode decomposition (VMD) to obtain intrinsic mode components (IMFs). The energy proportion of each dominant mode is calculated to obtain vibration mode characteristics. Specifically, vibration modes reveal the overall and local vibration characteristics of the rock mass structure under excitation. In the analysis, the signal is first subjected to VMD to obtain a series of intrinsic mode functions (IMFs). Subsequently, the energy entropy and instantaneous frequency stability of each IMF component are calculated. Intact rock masses typically exhibit energy concentration in the first 1-2 dominant modes, with stable instantaneous frequencies. When the rock mass becomes loose or delamination occurs, modal energy disperses (the energy proportion of the dominant mode decreases by more than 15%), higher-order modes are excited, and the instantaneous frequency exhibits a sharp jump. This parameter has a unique advantage in identifying hidden structural surfaces (such as unopened cracks). Vibration mode criteria: , For depth The dominant mode energy ratio; The dominant mode component energy; For the first Energy of each modal component.

[0045] Based on the preprocessed vibration signal, the standardized cross-correlation coefficient is calculated in the time domain, and the amplitude-squared coherence function is calculated in the frequency domain to construct a depth-coherence curve and obtain coherence characteristics. Specifically, coherence measures the similarity and phase continuity of vibration waveforms between adjacent measurement points, and is a direct tool for detecting discontinuities (such as delamination and cavities). The analysis is performed in parallel in the time and frequency domains: the standardized cross-correlation coefficient is calculated in the time domain, and the amplitude-squared coherence function is calculated in the frequency domain. The two results are combined to generate a depth-coherence curve. A coherence value below 0.4 is considered a significant anomaly, indicating the presence of a structural surface at that depth that causes wavefield distortion or interruption. To eliminate the influence of random noise, the analysis uses statistical averaging of multiple event results, and significance tests (such as t-tests) are performed on coherence abrupt change points to ensure the reliability of anomaly detection. Coherence criteria: = , like The presence of delamination / void was determined. For depth Signal coherence; For channel With channel The cross-power spectrum; For channel Self-power spectrum; For channel Self-power spectrum.

[0046] S3: Integrate the aforementioned multi-parameter features to construct a depth-aligned fused feature vector; specifically, construct the fused feature vector based on wave velocity features, frequency attenuation features, vibration mode features, and coherence features. Align and normalize the four parameter sequences (wave velocity ratio, attenuation coefficient ratio, dominant mode energy ratio, and coherence) aligned along the borehole depth direction, and combine them into a four-dimensional fused feature vector: , in, For normalized wave velocity ratio, This is the normalized attenuation coefficient ratio. To normalize the dominant mode energy ratio, For depth Coherence degree.

[0047] S4: Construct a recognition model based on the fused feature vector and the improved lightweight gradient boosting decision tree. The improved lightweight gradient boosting decision tree uses a binary decision tree with depth constraints as the base learner, performs boosting by gradient unidirectional residual approximation, and adopts an adaptive gating threshold mechanism to make decisions on the posterior probability output by the recognition model. Specifically, this embodiment provides an improved lightweight gradient boosting decision tree, which upgrades the original random forest or gradient boosting decision tree into an improved lightweight gradient boosting decision tree ensemble model. It uses a depth-constrained binary decision tree as the base learner and performs boosting by unidirectional gradient residual approximation. The input is a four-dimensional fused feature vector. The output is the posterior probability of three types of rock mass states: intact rock mass, loosened zone, and delamination / cavity. The model can automatically calculate feature importance and is more sensitive to the identification of loose zone boundaries, weak delamination layers, and small-scale voids. It supports incremental learning and lightweight deployment.

[0048] Model training employs a multi-objective fusion loss function. Balancing classification accuracy, sample imbalance, and generalization ability: , In the formula: This is a weighted cross-entropy loss used to address class imbalance.

[0049] To focus on loss, it is used to enhance learning of hard-to-distinguish samples.

[0050] Label smoothing loss is used to suppress overfitting.

[0051] To replace the fixed threshold and enhance the model's generalization ability under complex geological conditions, an adaptive gating threshold mechanism is introduced: After the model outputs the rock mass state probability, a dynamic adaptive threshold is used to make the decision: setting an adaptive discrimination threshold. ,default =0.5; if If the probability of the highest probability is found, the rock mass state category will be output. If the highest probability is lower than the threshold, or the difference between the probabilities of the two categories is less than 0.1, it will be marked as an undetermined area, and manual judgment will be made in conjunction with the single parameter curve. In long-term monitoring, the high-confidence identification results will be used as pseudo-labels to dynamically update the gating threshold and improve the robustness of the judgment under complex geological conditions.

[0052] S5: Based on the output of the recognition model, the rock mass state category and discrimination confidence of the target surrounding rock depth point are obtained; specifically, the model input and output are as follows: (1) Model input: the above fused feature vector (2) Model output: Rock mass state category label at this depth point, including: Category I (intact bearing zone), Category II (loose zone), Category III (delamination / cavity).

[0053] The model optimization process is based on a closed-loop system: new data acquisition → performance evaluation → model update → redeployment. Its core objective is to improve the model's generalization ability, recognition accuracy, and stability.

[0054] The model iterative optimization process mainly consists of the following steps: 1. Data pool update and quality screening: During long-term monitoring, the system continuously stores newly collected and preprocessed data, along with their preliminary identification results (after manual sampling or high-confidence automatic labeling), into the historical database, forming a constantly expanding training data pool. Before being stored in the database, data undergoes quality screening through consistency checks (e.g., low conflict levels in multi-parameter discrimination results).

[0055] 2. Incremental learning and model retraining: An incremental learning strategy is employed, triggered periodically (e.g., monthly) or when a significant decline in model performance is detected. The system extracts the latest representative samples from the data pool and combines them with the original training set to retrain the model. During training, model hyperparameters (such as the number and depth of decision trees) can be fine-tuned and optimized simultaneously, and the fused feature vector is dynamically adjusted based on feature importance analysis results. Weighting coefficients to .

[0056] 3. Performance Evaluation and Verification: The performance of the updated model was evaluated using an independent validation dataset (never used in training). Key evaluation metrics included overall classification accuracy, precision and recall for each category, and confusion matrix analysis. Particular attention was paid to whether the recognition ability for a few categories, such as "out-of-class / hole" classifications, was improved.

[0057] The convergence condition of the model is determined by the following three sets of combined conditions. The model is considered to have converged if any one of them is met: (1) After three consecutive iterations of optimization, the overall classification accuracy of the model on the validation set changed by less than 1%, and there was no performance degradation.

[0058] (2) The loss function value during model training has become stable during the iteration process, and the loss value of the last five iterations has decreased by less than the preset threshold (e.g., 0.001).

[0059] This embodiment uses an independent validation dataset to evaluate the performance of the updated model. The core evaluation metrics include: overall classification accuracy, precision and recall for each category, and confusion matrix analysis, with particular attention to whether the recognition ability for a few categories such as delamination / holes has been improved.

[0060] The model training uses a multi-objective fusion loss function for optimization, which takes into account classification accuracy, sample imbalance and generalization ability.

[0061] 1) Weighted cross-entropy loss: To address the sample imbalance issue in the three-class classification of surrounding rock conditions—intact rock mass, loosened zone, and delamination / cavity—higher weights are applied to outlier categories with fewer samples. The expression is: , In the formula: This represents the total number of depth point samples.

[0062] This represents the number of surrounding rock condition categories, with a value of 3.

[0063] For the i-th sample The actual label of the class, with a value of 0 or 1.

[0064] To predict whether the i-th sample belongs to the i-th sample, the model is needed. The probability of a class.

[0065] For class-adaptive weights, the calculation is inversely proportional to the number of samples: , For the first Number of samples per class.

[0066] 2) Focus on losses: To reduce the loss contribution of easily separable samples and enhance the learning of difficult-to-separate samples such as loose zone boundaries and weak delamination layers, the expression is: , In the formula: This is the focus factor, used to reduce the weight of simple samples and amplify the loss of difficult samples; its value is 2.

[0067] 3) Label smoothing regularization loss: To suppress model overfitting and overconfidence, and improve the robustness of discrimination under complex geological conditions, the expression is: , , In the formula: This is the label smoothing coefficient, with a value of 0.1. This represents the i-th sample and the c-th softened label after label smoothing.

[0068] 4) Multi-objective fusion loss function:

[0069] Recommendation weight: =0.4, =0.4, =0.2.

[0070] 5) Model convergence criteria: ① After three consecutive iterations of optimization, the overall classification accuracy of the validation set changed by less than 1%, and there was no performance degradation.

[0071] ② Multi-objective fusion loss function The decrease was less than 0.001 in each of the five consecutive iterations, and the loss curve tended to stabilize.

[0072] ③ The model's predicted probability entropy on new data remains stable, with no significant statistical difference from historical data.

[0073] (3) The model’s prediction results on new data have stable uncertainty measures (such as the entropy of the prediction probability) and no statistically significant difference from historical data (which can be judged by hypothesis testing).

[0074] S6: Based on multi-parameter characteristics, rock mass condition categories and discrimination confidence levels, and borehole geological background information, a comprehensive state profile of the surrounding rock is drawn. A further implementation method includes an animated view, high-stress bearing zones, and delamination and cavities in the comprehensive state profile of the surrounding rock.

[0075] Specifically, this step first gathers three core data sets: (1) curves of four major parameters continuously distributed along the borehole depth (Z-axis), namely wave velocity ratio. attenuation coefficient ratio Modal energy ratio and coherence (2) The output of the model, namely the state category (intact bearing area, loose zone, delamination / void) and its discrimination confidence for each depth point. (3) Drilling and geological background information, including borehole location, depth, borehole diameter, lithological columnar section, etc.

[0076] After acquiring the above data, a multi-Y-axis coordinate system was used to plot the data under the same depth datum. The main plot area displays curves for the four main parameters, distinguished by different colors and line types, all sharing a common depth axis. Each curve is smoothed, and data is labeled at key inflection points. Along the depth axis, the status labeling area uses different colored fills or highlight strips to visually label the surrounding rock status areas identified by the intelligent model (e.g., green represents the bearing zone, yellow represents the loosened zone, and red represents delamination / cavity). Standard legend symbols clearly display "loosened zone thickness," "bearing zone range," and "location and scale of delamination / cavity," such as... Figure 3 , Figure 4 As shown.

[0077] The system automatically identifies and highlights abnormal parameter areas (such as sudden drops in wave velocity and low coherence troughs), and marks them with shading or arrows. For delamination / voids identified by the model, the system accurately marks the top and bottom plate depths and estimated thicknesses on the graph. In the parameter curve graph, the discrimination threshold lines for each parameter (such as the wave velocity ratio 0.75 line and the coherence 0.4 line) are drawn in dashed form, providing intuitive reference criteria. Simultaneously, profiles from previous or adjacent boreholes can be overlaid as a comparative background.

[0078] A further implementation method includes extracting key information from the comprehensive state profile of the surrounding rock to construct a structured text report. The structured text report includes a comprehensive evaluation of the loosened zone, a delamination list, and a description of the depth range and parameter characteristics of the stable bearing zone. The comprehensive evaluation of the loosened zone includes the starting depth, ending depth, thickness, and average wave velocity reduction of the loosened zone. The delamination list includes the location, thickness, footing and hanging wall lithology, and reliability rating of all identified discontinuities.

[0079] A further implementation method includes optimizing the distributed fiber optic acoustic sensing system based on the comprehensive rock condition profile. This step constitutes a key link in the "monitoring-analysis-optimization" closed loop. Its core lies in performing a series of targeted parameter adjustments and maintenance operations on the detection devices deployed in the borehole based on the rock mass structure information revealed by the generated comprehensive rock condition profile. The aim is to continuously optimize monitoring performance, adapt to the evolution of the surrounding rock condition, and guide subsequent engineering decisions. Following the principles of "zonal diagnosis, key reinforcement, and dynamic adaptation," specific optimization methods include: Based on the anomaly areas identified in the comprehensive rock condition profile, the spatial gauge length of the distributed fiber optic acoustic sensing system (DAS) is adjusted. Specifically, based on the distribution of anomaly areas revealed in the profile, the spatial gauge length of the DAS is dynamically adjusted to optimize spatial resolution and signal-to-noise ratio. This is done in several ways, depending on the different zones in the profile: (1) Intensive scanning of key areas: For identified delamination, voids or strongly fractured areas, in subsequent monitoring, the gauge length of the depth segment and its upper and lower affected areas (e.g., 1m before and after) will be manually or automatically switched to the minimum level (e.g. 0.5m) to obtain higher resolution vibration signals and accurately characterize the defect boundary and internal structure.

[0080] (2) Simplified scanning of stable areas: For sections with stable parameters and determined to be complete bearing areas, a larger gauge length (such as 1.5-2m) can be used for scanning to improve monitoring efficiency and ensure sufficient signal-to-noise ratio.

[0081] (3) Tracking the front edge of the loosening zone: At the transition front edge between the loosening zone and the bearing area, a moderate gauge length (such as 1m) is used for key monitoring to clearly capture the possible movement of the loosening zone boundary over time.

[0082] Based on the attenuation characteristics of the loose ring or broken zone for a specific frequency band signal, adjust the sweep frequency range or excitation method of the excitation source 3. Specifically, if the cross-sectional view shows that the loose ring or broken zone attenuates the signal of a specific frequency band (such as mid-to-high frequency), then in subsequent active testing, the sweep frequency range of the electromagnetic vibrator or the material of the hammer pad can be adjusted to strengthen the excitation of the frequency band sensitive to defects, or a low-frequency steady-state excitation that can penetrate the broken zone can be adopted.

[0083] If a large-scale delamination risk is found in a certain side wall or roof area, it is recommended that artificial vibration points be strategically placed in locations that can more effectively stimulate the response in that area during subsequent monitoring.

[0084] Based on a trend of decreasing coherence or signal energy, coupling agent is replenished to a designated section via the built-in injection channel. Specifically, the coupling agent replenishment warning is as follows: if a trend of decreasing coherence or signal energy is observed in a specific depth segment in the continuous profile generated by long-term monitoring, while the rock mass condition assessment remains unchanged, the system warns that coupling agent drying or fiber micro-debonding may occur in that segment. Based on this, a plan can be formulated to specifically replenish the coupling agent (such as low-viscosity silicone grease) in that segment via the device's built-in injection channel.

[0085] Adjust the pressure setting of the air filling system 2 according to the signal quality of irregular hole shape or rock mass with alternating soft and hard sections; fine-tuning of contact pressure: if the signal quality is poor in irregular hole shape or in sections with alternating soft and hard rock mass, the pressure setting of the air filling system 2 can be slightly adjusted within a safe range (such as 0.2-0.3 MPa) to improve the uniformity of the bonding between the optical fiber and the hole wall 11.

[0086] If the profile shows a large-scale signal anomaly without geological basis, it may trigger a diagnostic procedure for the fiber optic link itself (such as bending loss or breakpoints).

[0087] Based on the changes in the loosened zone boundary shown in the comprehensive rock condition profile, adjust the monitoring frequency or deploy new detection borehole 1. If the profile shows that the bottom of the existing borehole is still in the loosened zone or new anomalies are found, it verifies the rationality of the current borehole depth (3-5 times the tunnel radius), or suggests that deeper or new detection borehole 1 needs to be deployed in the adjacent area. Monitoring frequency adjustment: During the stable period of the surrounding rock condition, the active detection frequency can be appropriately reduced; while after the loosened zone expands rapidly or new delamination is found, the monitoring frequency is automatically increased, entering the enhanced monitoring mode.

[0088] According to the method of the present invention, two specific examples are provided: Example 1: Drilling in coal mine roadways: Aperture 42 mm, depth 8 m; Equipment: outer fiber optic tube, inflated to 0.25 MPa; DAS: sampling rate 10 kHz, gauge length 2 m.

[0089] Results: Wave velocity decreased by 25% in the 0–2 m range, with significant frequency attenuation and a 20% decrease in modes → loosening zone; wave velocity recovered in the 2–5 m range, with high coherence → bearing zone; coherence dropped to 0.3 in the 5.5 m range → delamination.

[0090] Example 2: Long-term monitoring of tunnels: Drilling: 50 mm diameter, 10 m depth; DAS: 8 kHz sampling rate, 1 m gauge length; Vibration: long-term mining machinery vibration.

[0091] Results: Wave velocity decreased to 0.8 times at 0–3 m → loosening zone; stable at 3–7 m → bearing zone; coherence decreased at 8 m → cavity. Long-term monitoring showed that the loosening zone expanded over time, with a clear evolutionary pattern.

[0092] Example 2: This invention also provides a distributed fiber optic acoustic detection system for boreholes in the loosened rock zone, used to implement the method of Embodiment 1, comprising: The signal acquisition module is used to acquire the original vibration signal of the surrounding rock using a distributed fiber optic acoustic sensing system. It is also used in conjunction with a photoelectric detection array to complete photoelectric conversion, ensuring the accuracy of weak vibration signal acquisition.

[0093] The signal analysis module is used to preprocess and extract features from the original vibration signal to obtain multi-parameter features, including wave velocity features, frequency decay features, vibration mode features, and coherence features. A cross-correlation time delay algorithm is used to simultaneously extract vibration frequency and spatial location information, realizing joint detection of loosening zone location and frequency features.

[0094] The feature fusion module is used to fuse the multi-parameter features to construct a depth-aligned fused feature vector.

[0095] The recognition model construction module is used to construct a recognition model based on the fused feature vector and the improved lightweight gradient boosting decision tree. The improved lightweight gradient boosting decision tree uses a binary decision tree with depth constraints as the base learner, performs boosting by gradient unidirectional residual approximation, and adopts an adaptive gating threshold mechanism to make decisions on the posterior probability output by the recognition model.

[0096] The state recognition module is used to obtain the rock mass state category and discrimination confidence level of the target surrounding rock depth point based on the output of the recognition model.

[0097] The drawing module is used to draw a comprehensive state profile of the surrounding rock based on multi-parameter features, rock mass state category and discrimination confidence level, and borehole geological background information.

[0098] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made by those skilled in the art to the technical solutions of the present invention without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. A method for distributed fiber optic acoustic wave detection in boreholes of loosened rock zones, characterized in that, include: The original vibration signal of the surrounding rock was collected using a distributed fiber optic acoustic sensing system. The original vibration signal is preprocessed and features are extracted to obtain multi-parameter features; wherein, the multi-parameter features include wave velocity features, frequency decay features, vibration mode features, and coherence features; By fusing the multi-parameter features, a depth-aligned fused feature vector is constructed; A recognition model is constructed based on the fused feature vector and the improved lightweight gradient boosting decision tree. The improved lightweight gradient boosting decision tree uses a binary decision tree with depth constraints as the base learner, performs boosting by gradient unidirectional residual approximation, and adopts an adaptive gating threshold mechanism to make decisions on the posterior probability output by the recognition model. Based on the output of the identification model, the rock mass state category and discrimination confidence level of the target surrounding rock depth point are obtained; Based on the multi-parameter features, the rock mass state category and discrimination confidence level, and the borehole geological background information, a comprehensive state profile of the surrounding rock is drawn.

2. The distributed fiber optic acoustic wave detection method for the loosened zone of surrounding rock in boreholes according to claim 1, characterized in that, The distributed fiber optic acoustic sensing system includes: An expandable long cylinder with distributed optical fibers arranged in a hybrid spiral and straight pattern on its outer wall, used for insertion into surrounding rock detection holes; An inflation system, connected to the expandable cylinder, is used to inflate the expandable cylinder with air to cause it to expand and fit against the hole wall. Excitation source, used to generate vibration signals in the surrounding rock; The DAS data acquisition unit, connected to the distributed optical fiber, is used to acquire the original vibration signal of the surrounding rock, with a sampling rate of not less than 5kHz and a gauge length adjustment range of 0.5~2m. The original vibration signal is stored in the form of a multi-channel strain time series, including timestamp, channel number and micro-strain value, and the sampling rate, gauge length, air pressure and trigger time are recorded.

3. The distributed fiber optic acoustic wave detection method for the loosened zone of surrounding rock in boreholes according to claim 1, characterized in that, The method for preprocessing the original vibration signal includes: The original vibration signal was bandpass filtered to obtain the main frequency bands reflecting the rock mass structure; Wavelet threshold denoising method is used to denoise the main frequency bands in order to suppress random noise; For data acquired by active excitation, the response signals of repeated excitations are time-domain aligned and superimposed by trigger alignment and signal averaging to suppress incoherent noise.

4. The distributed fiber optic acoustic wave detection method for the loosened zone of surrounding rock in boreholes according to claim 1, characterized in that, Methods for constructing fused feature vectors include: Based on the cross-correlation time delay estimation method, the travel time difference of vibration signals in adjacent sensing channels is calculated, a depth-velocity profile is constructed, and wave velocity characteristics are obtained. The amplitude spectrum is obtained by performing a fast Fourier transform on the preprocessed vibration signal. The logarithmic attenuation relationship of amplitude with propagation distance within a specified frequency band is fitted by linear regression method, and the attenuation coefficient is solved to obtain the frequency attenuation characteristics. Variational mode decomposition is performed on the preprocessed vibration signal to obtain the intrinsic mode components, and the energy proportion of each dominant mode is calculated to obtain the vibration mode characteristics. Based on the preprocessed vibration signal, the standardized cross-correlation coefficient is calculated in the time domain, the amplitude squared coherence function is calculated in the frequency domain, the depth-coherence curve is constructed, and the coherence characteristics are obtained. The fused feature vector is constructed based on the wave velocity feature, the frequency attenuation feature, the vibration mode feature, and the coherence feature.

5. The distributed fiber optic acoustic wave detection method for the loosened zone of surrounding rock in boreholes according to claim 1, characterized in that, The comprehensive profile of the surrounding rock includes an animated diagram, high-stress bearing areas, and delamination and cavities.

6. The distributed fiber optic acoustic wave detection method for the loosened zone of surrounding rock in boreholes according to claim 1, characterized in that, It also includes extracting key information based on the comprehensive state profile of the surrounding rock to construct a structured text report, which includes a comprehensive evaluation of the loosened zone, a delamination list, and a description of the depth range and parameter characteristics of the stable bearing zone; The comprehensive evaluation of the loosening zone includes the starting depth, ending depth, thickness, and average wave velocity reduction of the loosening zone. The delamination list includes the location, thickness, footwall and hanging wall lithology, and reliability rating of all identified discontinuities.

7. The distributed fiber optic acoustic wave detection method for the loosened zone of surrounding rock according to claim 2, characterized in that, It also includes optimizing the distributed fiber optic acoustic sensing system based on the comprehensive rock condition profile: Based on the abnormal areas identified in the comprehensive rock condition profile, adjust the spatial gauge length of the distributed fiber optic acoustic sensing system. Adjust the sweep frequency range or excitation method of the excitation source according to the attenuation characteristics of the loosened ring or broken zone for a specific frequency band signal; Based on the trend of decreasing coherence or signal energy, coupling agent is replenished to a specified section through the built-in injection channel; Adjust the pressure setting of the air filling system according to the signal quality of irregular hole shape or alternating soft and hard rock sections; Based on the changes in the loosened zone boundary shown in the comprehensive state profile of the surrounding rock, adjust the monitoring frequency or deploy new detection holes.

8. A distributed fiber optic acoustic detection system for boreholes in the loosened zone of surrounding rock, used to implement the method described in any one of claims 1-7, characterized in that, include: The signal acquisition module is used to acquire the original vibration signal of the surrounding rock using a distributed fiber optic acoustic sensing system, and at the same time, it is used in conjunction with a photoelectric detection array to complete the photoelectric conversion. The signal analysis module is used to preprocess and extract features from the original vibration signal to obtain multi-parameter features; wherein, the multi-parameter features include wave velocity features, frequency decay features, vibration mode features, and coherence features; The feature fusion module is used to fuse the multi-parameter features and construct a depth-aligned fused feature vector; The recognition model construction module is used to construct a recognition model based on the fused feature vector and the improved lightweight gradient boosting decision tree. The improved lightweight gradient boosting decision tree uses a binary decision tree with depth constraints as the base learner, performs boosting by gradient unidirectional residual approximation, and adopts an adaptive gating threshold mechanism to make decisions on the posterior probability output by the recognition model. The state recognition module is used to obtain the rock mass state category and discrimination confidence level of the target surrounding rock depth point based on the output of the recognition model; The drawing module is used to draw a comprehensive state profile of the surrounding rock based on the multi-parameter features, the rock mass state category and the discrimination confidence level, and the borehole geological background information.

Images