A method and system for in-situ detection of geological structure properties inside rock
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- KUNMING UNIV OF SCI & TECH
- Filing Date
- 2024-01-29
- Publication Date
- 2026-06-12
Smart Images
Figure CN117930325B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of rock mass engineering monitoring technology, specifically to an in-situ detection method and system for the geological structural properties inside rocks. Background Technology
[0002] In recent years, as major engineering projects have ventured deeper into the Earth, dynamic hazards in deep rock masses, such as rock bursts and landslides, have become increasingly severe, posing significant risks to these projects. Based on the specific conditions of the rock mass to be excavated, a dynamic hazard risk assessment is conducted, and corresponding excavation and support plans are formulated and implemented.
[0003] Detailed design and construction processes ensure the stability of the tunnel in each construction cycle. The geological structure characteristics of the rock mass to be excavated are crucial data for assessing dynamic hazard risks and designing excavation and support schemes accordingly.
[0004] Therefore, there is an urgent need to propose a safe and convenient method for rapid and accurate assessment of the geological structure properties of the rock mass to be excavated in front of the working face of deep roadways, to obtain the spatial distribution characteristics of the discontinuity of the rock mass to be excavated, and to provide key data for realizing dynamic rock mass dynamic disaster risk assessment and refined dynamic design and construction of excavation and support schemes.
[0005] Existing methods for detecting the internal geological structure properties of soil and rock generally suffer from drawbacks such as high detection costs, small detection range, low representativeness of results, and long inversion iteration time. In addition, the acquisition of detection results is limited by the environment, resulting in technical problems such as low accuracy and low acquisition efficiency. Summary of the Invention
[0006] This application provides an in-situ detection method and system for the geological structure properties inside rocks, which solves the shortcomings of existing technologies for detecting the geological structure properties inside rocks and soil, such as high detection cost, small detection range, low representativeness of results and long inversion iteration time. At the same time, the acquisition of detection results is limited by the environment, resulting in low accuracy and low acquisition efficiency.
[0007] In view of the above problems, this application provides an in-situ detection method for the geological structural properties inside rocks.
[0008] Firstly, this application provides an in-situ detection method for the geological structural properties inside rocks. The method includes: obtaining a target impact echo detection surface, wherein the target impact echo detection surface is generated by leveling and grinding the target rock mass monitoring area, and the target impact echo detection surface is provided with K impact echo measurement points; pre-deploying a target impact echo probe on each target impact echo detection surface, wherein the target impact echo probe is a single-excitation multi-channel synchronous receiving impact echo device; using the target impact echo probe to perform impact echo detection on the target impact echo detection surface to obtain target impact echo data; and performing time-frequency domain signal conversion on the target impact echo data to obtain the target impact echo signal time. The process involves: Spectrum analysis; Pre-constructing a deep learning model for predicting rock mass discontinuity; Synchronizing the target impact echo signal's temporal spectrum to the deep learning model for predicting rock mass discontinuity; Obtaining the target rock mass discontinuity prediction result, where the deep learning model is constructed based on an inversion deep learning network; Generating target rock mass discontinuity label data by collecting data from K impact echo measurement points; Pre-constructing a target loss function and synchronizing the target rock mass discontinuity prediction result and target rock mass discontinuity label data to the target loss function to obtain target gap information; Optimizing the deep learning model for predicting rock mass discontinuity based on the target gap information.
[0009] Secondly, this application provides an in-situ detection system for the geological structure properties inside rocks. The system includes: a detection surface module for obtaining a target impact echo detection surface, wherein the target impact echo detection surface is generated by smoothing and grinding the target rock mass monitoring area, and the target impact echo detection surface is provided with K impact echo measurement points, where K is a positive integer; a probe deployment module for pre-deploying target impact echo probes on the target impact echo detection surface, wherein the target impact echo probes are single-excitation multi-channel synchronous receiving impact echo devices; an impact echo detection module for using the target impact echo probes to perform impact echo detection on the target impact echo detection surface to obtain target impact echo data; and a signal conversion module for performing time-frequency domain signal conversion on the target impact echo data to obtain the target impact echo signal time-frequency domain. The system comprises the following modules: a spectrum prediction module and a discontinuity prediction module. The latter pre-constructs a deep learning model for predicting rock mass discontinuity, synchronizes the time spectrum of the target impact echo signal to this model, and obtains the prediction result. The deep learning model is constructed based on an inversion deep learning network. A data acquisition module collects data from the K impact echo measurement points to generate target rock mass discontinuity label data. A function construction module pre-constructs a target loss function and synchronizes the prediction result and label data to it to obtain target gap information. A model optimization module optimizes the deep learning model based on the target gap information.
[0010] One or more technical solutions provided in this application have at least the following technical effects or advantages:
[0011] This application provides an in-situ detection method and system for the geological structural properties inside rocks. The method involves setting up an impact echo detection surface in the rock mass to be measured, deploying an array of echo monitoring probes on the detection surface, and performing impact echo measurements to obtain target impact echo data. The target impact echo data is then converted to a time-frequency domain signal to obtain the time spectrum of the target impact echo signal. A deep learning model maps the time spectrum data of the target impact echo signal to depth-related discontinuity data of the rock mass, thereby predicting the discontinuity properties of the rock mass. The model is trained by collecting data from K impact echo measurement points to generate target rock mass discontinuity label data. The target rock mass discontinuity prediction results and the target rock mass discontinuity label data are synchronized to a target loss function to obtain target gap information. Finally, the deep learning model for predicting rock mass discontinuity is optimized based on the target gap information. This invention addresses the shortcomings of existing technologies for detecting the internal geological structure properties of soil and rock, such as high detection costs, small detection range, low representativeness of results, and long inversion iteration time. It also solves the technical problems of low accuracy and low efficiency in obtaining detection results due to environmental limitations. This invention provides key data for dynamic rock mass disaster risk assessment and refined dynamic design and construction of excavation and support schemes, reducing subjectivity and errors to a certain extent. Attached Figure Description
[0012] Figure 1 This application provides a schematic flowchart of an in-situ detection method for the internal geological structure properties of rocks;
[0013] Figure 2 This application provides a schematic diagram of an in-situ detection system for the geological structure properties inside rocks;
[0014] Figure 3 This application provides a schematic flowchart of an impact echo detection method;
[0015] Figure 4 This application provides an example of a target impact echo detection surface;
[0016] Figure 5 This application provides an example of a target impact echo probe;
[0017] Figure 6 This application provides a schematic diagram of the layout for impact echo measurement points in a roadway.
[0018] Explanation of reference numerals in the attached diagram: Detection surface module 11, Probe deployment module 12, Impact echo detection module 13, Signal conversion module 14, Discontinuity prediction module 15, Data acquisition module 16, Function construction module 17, Model optimization module 18. Detailed Implementation
[0019] In recent years, the rapidly developing deep learning models have directly established a mapping function between observed data and target data through their powerful data feature extraction capabilities. This avoids the computationally expensive traditional waveform inversion calculation and enables the direct characterization of the geological structural properties of the rock mass through which ultrasonic waves propagate, achieving a good balance between detection efficiency and accuracy. It has significant application value in the rapid assessment of rock mass structural properties in underground rock engineering. This application provides an in-situ detection method and system for the geological structure properties inside rocks. The method involves obtaining a target impact echo detection surface, pre-deploying target impact echo probes on the surface, using these probes to detect impact echoes and obtain target impact echo data. The target impact echo data is then converted to a time-frequency domain signal to obtain the time spectrum of the target impact echo signal. This time spectrum is synchronized to a deep learning model for rock mass discontinuity prediction to predict rock mass discontinuity, resulting in a prediction result. Data is collected from K impact echo measurement points to generate target rock mass discontinuity label data. The prediction result and the target rock mass discontinuity label data are synchronized to a target loss function to obtain target gap information. Finally, the deep learning model for rock mass discontinuity prediction is optimized based on this gap information. This invention addresses the shortcomings of existing technologies for detecting the internal geological structure properties of soil and rock, such as high detection costs, small detection range, low representativeness of results, and long inversion iteration time. It also solves the technical problems of low accuracy and low efficiency in obtaining detection results due to environmental limitations. This invention provides key data for dynamic rock mass disaster risk assessment and refined dynamic design and construction of excavation and support schemes, reducing subjectivity and errors to a certain extent.
[0020] Example 1
[0021] like Figure 1 As shown, this application provides a method and system for in-situ detection of the geological structural properties inside rocks. The method includes:
[0022] A target impact echo detection surface is obtained, wherein the target impact echo detection surface is generated by flattening and grinding the target rock mass monitoring area, and the target impact echo detection surface is provided with K impact echo measurement points, where K is a positive integer;
[0023] like Figure 4The target impact echo detection surface refers to the face of the rock face. The monitoring area for the target rock mass is selected, ideally covering the entire area of the rock mass to be excavated. Based on the on-site rock mass characteristics, a suitable location is chosen. The selected surface should be smoothed using equipment such as angle grinders through polishing and grinding to ensure flatness and the validity of the impact echo data, facilitating impact echo detection. Using the waistline of the target impact echo detection surface generated by grinding as a baseline, two additional measuring lines are set at offsets of 0.5 meters above and below. Impact echo detection points are set at approximately 0.5-meter intervals on the rock wall along these measuring lines, resulting in multiple impact echo measurement points. Obtaining the target impact echo detection surface provides the basis for deploying target impact echo probes based on it.
[0024] A target impact echo probe is pre-deployed on the target impact echo detection surface, wherein the target impact echo probe is a single-excitation multi-channel synchronous receiving type impact echo device;
[0025] This application employs an improved single-excitation, multi-channel synchronous receiving impulse echo device equipped with a test host. The stress wave exciter has a maximum transmission voltage of 1000V and a maximum impact force of 48N. The exciter surface is a circular metal sheet with a diameter of 8mm, and it has a built-in 8Ah battery, allowing for continuous operation for over 8 hours. The receiving end features a planar transducer with a main frequency of 50kHz and a sensitivity of 325pC / ms. -2 .like Figure 5 As shown in the diagram, the target impact echo probe is deployed on the target impact echo detection surface, arranged according to the impact echo points. The target impact echo probe primarily works by exciting an active seismic source and using sensors to receive stress wave data reflected after the stress wave contacts the geological structure within the rock mass. The reflected stress wave data is then processed using the propagation law of stress waves within the rock mass to achieve rock mass property imaging. The target impact echo probe transmits a signal in a single excitation and simultaneously receives echo signals through multiple channels. The transmitting device is located at the center of the entire target impact echo probe, and the receiving devices are arranged on the probe, filling the entire target impact echo probe. With the shock wave exciter as the center point of the rectangle, the receiving devices are arranged in a grid pattern with a spacing of 3.5-5 cm. The receiving devices are fixed by a retractable metal bracket to ensure that they are in close contact with the surface of the rock mass being measured, enabling multi-dimensional detection and analysis of target features. The deployment of the target impact echo probe provides the data foundation for subsequent acquisition of target impact echo data.
[0026] The target impact echo probe is used to detect the impact echo of the target impact echo detection surface to obtain target impact echo data.
[0027] Shock echo detection refers to the process of exciting stress wave signals through an exciter, which propagate perpendicularly from the surface of a rock mass into its interior. Upon encountering structural surfaces, these propagate within the rock mass, generating reflected waves, which are then output. A shock echo detector detects and receives these reflected waves, and a shock echo receiver array measures the impact echo. The measured results are then output as internal rock mass reflection data, which is used as the target shock echo data. To ensure the stability of the prediction results, multiple measurements can be taken in the same area. After measuring in one area, the instrument can be moved to the next measurement point to continue acquiring shock echo data. This provides the data foundation for subsequent time-frequency domain signal conversion of the shock echo to obtain the time spectrum of the target shock echo signal.
[0028] The target impact echo data is converted into a time-frequency domain signal to obtain the time spectrum of the target impact echo signal;
[0029] The target impact echo data undergoes preliminary processing. This involves converting the time-domain signal into a time-frequency domain signal using the Short-Time Fast Fourier Transform (SFT) method, followed by bandpass filtering. Here, the time domain refers to the signal's variation over time, represented in this application as the waveform of the target impact echo data over time, which can be continuous or discrete. The target impact echo data is then subjected to Fourier Transform to transform the time-domain signal into a time-frequency domain signal. The SFT is a mathematical tool for converting signals from the time domain to the time-frequency domain. The resulting time-frequency spectrum of the target impact echo signal provides a data foundation for subsequently constructing a deep learning model for predicting rock mass discontinuities.
[0030] A pre-constructed deep learning model for predicting rock mass discontinuity is used to synchronize the time spectrum of the target impact echo signal to the deep learning model for predicting rock mass discontinuity, thereby obtaining the target rock mass discontinuity prediction result. The deep learning model for predicting rock mass discontinuity is constructed based on an inversion deep learning network.
[0031] like Figure 3The flowchart of the shock echo detection method is shown. Inversion of the deep neural network model refers to the process of deriving input data from a given neural network model. In deep learning, neural network models are typically used to establish a mapping relationship from input data to output results. Inversion, on the other hand, is the reverse process, that is, reversing the input data from a pre-trained neural network model. Using the target shock echo signal's time-spectrum data as input and rock mass depth-related discontinuity data as label values, the deep neural network model is inverted to obtain the target rock mass discontinuity prediction results. By comparing the model-predicted rock mass discontinuity results with the actual measured discontinuity label data, the deep learning model parameters are updated using gradient backpropagation theory to reduce the discrepancy between the two. Finally, a deep neural network model capable of accurately predicting discontinuities is established and applied in practice. The target rock mass discontinuity prediction results provide a data foundation for subsequent construction and optimization of the inversion deep learning network.
[0032] Data is collected from the K impact echo measurement points to generate target rock mass discontinuity label data;
[0033] Signal data acquisition from impact echo measurement points involves establishing rock mass discontinuity labeling data using core samples and borehole camera data from corresponding measurement locations. Drilling is performed in the impact echo monitoring area, and cameras are taken within the boreholes to obtain borehole camera data. Geological structure analysis is then conducted based on this data to obtain depth-related rock mass geological structure image distributions. Core data analysis is then performed on these distributions to obtain core data. Deep learning algorithms are used to process ultrasonic waves reflected from different geological structures within the rock mass. The input is multi-channel impact echo time-spectrum data with dimensions l×k×n (l: time dimension of a single impact echo, k: frequency dimension of a single impact echo, n: number of receivers, i.e., the number of impact echoes acquired in a single excitation). An encoder-decoder model is then used to obtain depth-related rock mass discontinuity data. The obtained core data and borehole camera data are used to establish depth-correlated rock mass discontinuity data, which are then used as label values for the deep learning model. The error between the model prediction and the label values is calculated, and the deep learning model parameters are updated using gradient backpropagation technology. The model training is completed by minimizing the error through multiple iterations. This enables a simple and efficient prediction of the geological structure properties of rock mass using shock echo data. The generation of target rock mass discontinuity label data provides a data foundation for obtaining target gap information in the future.
[0034] A target loss function is pre-constructed, and the target rock mass discontinuity prediction results and the target rock mass discontinuity label data are synchronized to the target loss function to obtain target gap information;
[0035] Synchronizing the target rock mass discontinuity prediction results and target rock mass discontinuity label data to the target loss function refers to the process of calculating the loss function between the predicted rock mass depth-related discontinuity evaluation output by the inversion deep neural network model and the depth-related discontinuity results calculated from borehole camera results. The loss function is used to evaluate the difference between the discontinuity evaluation value established from actual borehole measurement results and the model prediction value. The loss function is pre-constructed, and a function value comparison calculation formula is used. The target rock mass discontinuity prediction results and target rock mass discontinuity label data are input into the target loss function to calculate the numerical difference between them. This is the deviation between the data calculated by the model and the actual situation, i.e., the target gap information, providing a data foundation for subsequent optimization of the deep learning model for rock mass discontinuity prediction.
[0036] The deep learning model for predicting rock mass discontinuity is optimized based on the target gap information.
[0037] The network parameters are updated and optimized based on the target gap information to minimize the difference between the data calculated based on the loss function and the label values, thereby determining the model parameters. The updated model is then tested by conducting impact echo tests on the surfaces of other rock masses to be excavated in the project, and the results are input into the trained model to obtain the corresponding rock mass depth-related discontinuity prediction values. By testing and predicting multiple points on the rock mass surface, and based on the spatial location of the predicted values, an interpolation algorithm is used to measure the discontinuity of the entire rock mass, characterizing its geological structural properties. The updated and optimized deep learning model for rock mass discontinuity prediction can more accurately analyze the rock mass, thereby improving the accuracy of real-time rock mass quality assessment in underground rock mass engineering construction, and providing important data for excavation risk level assessment and subsequent dynamic and refined excavation and support scheme design.
[0038] Furthermore, this application also includes:
[0039] Obtain the target waistline of the target impact echo detection surface;
[0040] A preset measurement line offset index is used, and the target waistline is used as the reference measurement line. The measurement line offset index is used as a constraint to obtain the first measurement line and the second measurement line.
[0041] A preset measurement point offset index is used as a constraint to locate the measurement points of the target waistline, the first measurement line, and the second measurement line, thereby obtaining the K impact echo measurement points.
[0042] The appropriate location should generally be chosen in an area where the sensor can couple well with the rock wall, and where there are few or no micro-cracks or structural planes on the rock surface. For example... Figure 6 The schematic diagram of the impact echo measurement point layout in the tunnel is shown. In the tunnels and chambers of underground engineering, the waistline of the working face is used as the reference measurement line. A measurement line offset index is set, and the offset distance is obtained through the measurement line offset index. After offsetting by 0.5m in both directions, two other measurement lines are determined, resulting in the first measurement line and the second measurement line. At every offset distance on the measurement lines, appropriate rock surfaces are selected to deploy impact echo measurements on the rock walls to obtain the measurement point locations. Based on the measurement point locations, multiple impact echo measurement points are obtained. The acquisition of impact echo measurement points provides a basis for the subsequent deployment of target impact echo probes.
[0043] Furthermore, this application also includes:
[0044] Drilling and core sampling were performed at the K impact echo measurement points to obtain K target cores;
[0045] Geological logging was performed based on the K target rock cores to determine the attitude of the rock mass structural planes, and to obtain the measurement parameters of the K rock core structural planes and the K target boreholes;
[0046] Drilling images were taken of the K target boreholes to obtain the K target borehole image results;
[0047] The target rock mass discontinuity label data is constructed based on the measurement parameters of the K core structural surfaces and the imaging results of the K target boreholes.
[0048] Core samples were obtained by drilling at multiple impact echo measurement points. The collected samples were then used to acquire multiple target core samples. Based on these target core samples, the orientation of the core structural planes was measured, yielding structural plane measurement results. These results included multiple core structural plane measurement parameters and multiple target boreholes corresponding to these parameters. Borehole photography was performed on these boreholes, and the results were obtained. Combining these borehole photography results with the core structural plane measurement parameters, depth-related rock mass discontinuity data was constructed. This data provides a foundation for subsequently constructing a target loss function and obtaining target gap information.
[0049] Furthermore, this application also includes:
[0050] The target impact echo probe is composed of a target stress wave exciter and a target stress wave receiver.
[0051] A preset detection surface meshing threshold is set, and the target impact echo detection surface is divided into a target detection surface mesh based on the detection surface meshing threshold;
[0052] A set of grid endpoints is determined based on the target detection surface grid, wherein the set of grid endpoints includes the grid center endpoint and M grid endpoints;
[0053] The target stress wave exciter is placed at the center endpoint of the grid, and the set of target stress wave receivers is placed at the grid endpoint set to construct a stress wave receiver array;
[0054] The target stress wave exciter emits stress waves and generates reflected waves at the target impact echo detection surface;
[0055] In the stress wave receiver array, the target stress wave receiver set receives reflected waves to generate the target impact echo data, wherein the target impact echo data includes M channels of impact echo data.
[0056] The target shock echo probe includes a target stress wave exciter and a target stress wave receiver. A detection surface gridding threshold is set, and the detection surface is divided into a target detection surface grid based on this threshold. The grid endpoint set is then determined based on the grid, including the grid center endpoint and multiple grid endpoints. This involves determining the rows and columns of the center position and its surrounding areas, specifying the positions of the transmitting and receiving devices, and placing the target stress wave exciter at the grid center endpoint. The target stress wave receiver set is then arranged according to the grid endpoint set, and the deployed target stress wave exciter and receivers together form a stress wave receiver array. Stress waves are emitted from the target stress wave exciter and reflected at the target impact echo detection surface. The target stress wave receiver set in the stress wave receiver array receives the reflected waves to generate the target impact echo data. The impact echo data of each channel is different. The impact echo data of multiple channels are combined to construct the target impact echo data. The target impact echo data provides the data basis for subsequent time-frequency domain signal conversion to obtain the time spectrum of the target impact echo signal.
[0057] Furthermore, this application also includes:
[0058] Collect and acquire core data, borehole camera data, and shock echo data from multiple sample monitoring areas;
[0059] Based on the multiple sample core data and the multiple sample borehole camera data, multiple sample rock mass discontinuity label data were obtained;
[0060] A pre-constructed deep learning model for predicting rock mass discontinuity is provided. The deep learning model for predicting rock mass discontinuity includes an encoder module, a decoder module, and a residual connection module. The encoder module and the decoder module are connected based on the residual connection module.
[0061] The multiple sample impact echo data and the multiple sample rock mass discontinuity label data are identified and divided to obtain the model training dataset, the model testing dataset, and the model validation dataset.
[0062] The encoder module and the decoder module are trained under supervision using the model training dataset. After training, the encoder module and the decoder module are verified and tested using the model test dataset and the model validation dataset.
[0063] Core data, borehole camera data, and shock echo data were collected from the monitoring area to obtain multiple samples of core data, borehole camera data, and shock echo data. Based on these samples, multiple samples of rock mass discontinuity label data were obtained. A pre-built deep learning model for predicting rock mass discontinuity was then constructed. This pre-built deep learning model includes an encoder module, a decoder module, and a residual connection module. The encoder and decoder modules are connected based on the residual connection module. The encoder module is used to extract high-order features of the time-frequency domain signal; the decoder module is used for feature mapping; and the residual connection module is used to preserve features. The multiple samples of shock echo data and rock mass discontinuity label data were then labeled and divided to obtain a model training dataset, a model testing dataset, and a model validation dataset. The encoder and decoder modules were then trained under supervised supervision using the model training dataset. Specifically, based on the training set data, firstly... The encoder module extracts high-order time-frequency domain features from the shock echo spectral data, which are then mapped to a depth-related discontinuity feature map via the decoder module. Finally, the residual connection module preserves the initial waveform signal features. The results are then validated against a model validation dataset. The validation results are used to further construct a deep learning model for predicting rock mass discontinuity. The dataset is obtained by processing core data from the monitoring area combined with borehole camera data to obtain corresponding depth-related discontinuity label data for the rock mass. The discontinuity data ranges from 0 to 1, where continuous rock mass is 0 and completely discontinuous is 1. The rock mass discontinuity label data and the corresponding shock echo data are integrated to establish a sample dataset, which is then randomly divided into a training dataset for model training and a validation dataset for model validation. The training dataset is input into the deep learning model for training. After training, the model is applied to other shock echo data to predict the discontinuity of the rock mass under test, providing a foundation for obtaining subsequent prediction results of the target rock mass discontinuity.
[0064] Furthermore, this application also includes:
[0065] The model output accuracy requirement is preset, and it is determined whether the output accuracy of the deep learning model for predicting rock mass discontinuity meets the model output accuracy requirement.
[0066] If the output accuracy of the deep learning model for predicting rock mass discontinuity meets the model output accuracy requirement, then the deep learning model for predicting rock mass discontinuity is obtained.
[0067] The output rate of the preset model is set, and models that meet the accuracy requirements are output. Unqualified preset models are reconstructed to improve the accuracy of the preset models. By analyzing the accuracy of the preset models, the obtained accuracy is compared with the preset accuracy, and the comparison results are output. The comparison results are judged. The setting of the accuracy requirements of the preset model improves the completeness of the preset rock mass discontinuity prediction deep learning model, and improves the accuracy of the target rock mass discontinuity prediction results calculated by the preset rock mass discontinuity prediction deep learning model, further increasing the accuracy of the construction of the rock mass discontinuity prediction deep learning model.
[0068] Furthermore, this application also includes:
[0069] Based on the encoder module, high-order waveform features are extracted from the M channel impulse echo data to obtain M high-order waveform feature maps.
[0070] Based on the decoder module, the mapping processing of the M waveform high-order feature maps is performed to obtain the M channel rock mass discontinuity prediction results;
[0071] Based on the residual connection module, the initial waveform signal features of the M channel impulse echo data are extracted to obtain M initial waveform signal features;
[0072] The residual connection module connects the M initial waveform signal features to the decoder module;
[0073] The decoder module generates M channel feature images based on the predicted rock discontinuity results of the M channels and the features of the M initial waveform signals;
[0074] The M channel feature images are aggregated to generate the target rock mass discontinuity prediction result, wherein the target rock mass discontinuity prediction result is a depth correlation vector.
[0075] First, the encoder module extracts high-order waveform features from the shock echo data of each channel, with one channel corresponding to one stress wave receiver. Then, the decoder module maps the feature map containing these high-order waveform features to depth-related rock discontinuity data, obtaining prediction results for rock discontinuities across multiple channels. To effectively fuse the high-order features with the original features, a residual concatenation module extracts features from the initial waveform signal during the decoding process, obtaining multiple initial waveform signal features. These features are then connected to the corresponding deconvolution layers of the decoder. Finally, the decoder generates feature images for multiple channels, which are then processed... The output module aggregates the depth correlation vectors of the corresponding rock mass discontinuity attributes in the space to obtain the target rock mass discontinuity prediction result. The encoder module consists of four cascaded convolutional layers to downsample the input shock echo data. The decoder module consists of four cascaded deconvolutional layers, and the residual connection module connects them to the corresponding feature map in the decoder for upsampling. The deep learning model for rock mass discontinuity prediction constructed by the encoder module, decoder module and residual connection module is more accurate and complete, and the target rock mass discontinuity prediction result output by the model is also more accurate.
[0076] Example 2
[0077] Based on the same inventive concept as the in-situ detection method for the internal geological structure properties of rocks in the foregoing embodiments, such as Figure 2 As shown, this application provides an in-situ detection system for the geological structural properties inside rocks, the system comprising:
[0078] Detection surface module 11: Obtain target impact echo detection surface, wherein the target impact echo detection surface is generated by smoothing and grinding the target rock mass monitoring area, and the target impact echo detection surface is provided with K impact echo measurement points;
[0079] Probe deployment module 12: Pre-deploys target impact echo probes on the target impact echo detection surface, wherein the target impact echo probe is a single-excitation multi-channel synchronous receiving type impact echo device;
[0080] Shock echo detection module 13: Uses the target shock echo probe to detect the target shock echo detection surface and obtains target shock echo data;
[0081] Signal conversion module 14: Performs time-frequency domain signal conversion on the target impact echo data to obtain the time spectrum of the target impact echo signal;
[0082] Discontinuity prediction module 15: pre-constructs a deep learning model for predicting rock mass discontinuity, synchronizes the time spectrum of the target impact echo signal to the deep learning model for predicting rock mass discontinuity, and obtains the target rock mass discontinuity prediction result, wherein the deep learning model for predicting rock mass discontinuity is constructed based on an inversion deep learning network;
[0083] Data acquisition module 16: Generates target rock mass discontinuity label data by acquiring data from the K impact echo measurement points;
[0084] Function construction module 17: pre-constructs the target loss function and synchronizes the target rock mass discontinuity prediction result and the target rock mass discontinuity label data to the target loss function to obtain target gap information;
[0085] Model optimization module 18: Optimizes the deep learning model for predicting rock mass discontinuity based on the target gap information.
[0086] Furthermore, the detection surface module 11 includes the following execution steps:
[0087] Obtain the target waistline of the target impact echo detection surface;
[0088] A preset measurement line offset index is used, and the target waistline is used as the reference measurement line. The measurement line offset index is used as a constraint to obtain the first measurement line and the second measurement line.
[0089] A preset measurement point offset index is used as a constraint to locate the measurement points of the target waistline, the first measurement line, and the second measurement line, thereby obtaining the K impact echo measurement points.
[0090] Furthermore, the data acquisition module 16 includes the following execution steps:
[0091] Drilling and core sampling were performed at the K impact echo measurement points to obtain K target cores;
[0092] Based on the K target rock cores, rock core structure surface measurements are performed to obtain K rock core structure surface measurement parameters and K target boreholes;
[0093] Drilling images were taken of the K target boreholes to obtain the K target borehole image results;
[0094] The target rock mass discontinuity label data is constructed based on the measurement parameters of the K core structural surfaces and the imaging results of the K target boreholes.
[0095] Furthermore, the shock echo detection module 13 includes the following execution steps:
[0096] The target impact echo probe is composed of a target stress wave exciter and a target stress wave receiver.
[0097] A preset detection surface meshing threshold is set, and the target impact echo detection surface is divided into a target detection surface mesh based on the detection surface meshing threshold;
[0098] A set of grid endpoints is determined based on the target detection surface grid, wherein the set of grid endpoints includes the grid center endpoint and M grid endpoints;
[0099] The target stress wave exciter is placed at the center endpoint of the grid, and the set of target stress wave receivers is placed at the grid endpoint set to construct a stress wave receiver array;
[0100] The target stress wave exciter emits stress waves and generates reflected waves at the target impact echo detection surface;
[0101] In the stress wave receiver array, the target stress wave receiver set receives reflected waves to generate the target impact echo data, wherein the target impact echo data includes M channels of impact echo data.
[0102] Furthermore, the discontinuity prediction module 15 includes the following execution steps:
[0103] Collect and acquire core data, borehole camera data, and shock echo data from multiple sample monitoring areas;
[0104] Based on the multiple sample core data and the multiple sample borehole camera data, multiple sample rock mass discontinuity label data were obtained;
[0105] A pre-constructed deep learning model for predicting rock mass discontinuity is provided. The deep learning model for predicting rock mass discontinuity includes an encoder module, a decoder module, and a residual connection module. The encoder module and the decoder module are connected based on the residual connection module.
[0106] The multiple sample impact echo data and the multiple sample rock mass discontinuity label data are identified and divided to obtain the model training dataset, the model testing dataset, and the model validation dataset.
[0107] The encoder module and the decoder module are trained under supervision using the model training dataset. After training, the encoder module and the decoder module are verified and tested using the model test dataset and the model validation dataset.
[0108] Furthermore, the discontinuity prediction module 15 includes the following execution steps:
[0109] The model output accuracy requirement is preset, and it is determined whether the output accuracy of the deep learning model for predicting rock mass discontinuity meets the model output accuracy requirement.
[0110] If the output accuracy of the deep learning model for predicting rock mass discontinuity meets the model output accuracy requirement, then the deep learning model for predicting rock mass discontinuity is obtained.
[0111] Furthermore, the discontinuity prediction module 15 includes the following execution steps:
[0112] Based on the encoder module, high-order waveform features are extracted from the M channel impulse echo data to obtain M high-order waveform feature maps.
[0113] Based on the decoder module, the mapping processing of the M waveform high-order feature maps is performed to obtain the M channel rock mass discontinuity prediction results;
[0114] Based on the residual connection module, the initial waveform signal features of the M channel impulse echo data are extracted to obtain M initial waveform signal features;
[0115] The residual connection module connects the M initial waveform signal features to the decoder module;
[0116] The decoder module generates M channel feature images based on the predicted rock discontinuity results of the M channels and the features of the M initial waveform signals;
[0117] The M channel feature images are aggregated to generate the target rock mass discontinuity prediction result, wherein the target rock mass discontinuity prediction result is a depth correlation vector.
[0118] Through the foregoing detailed description of a method for in-situ detection of the internal geological structure properties of rocks, those skilled in the art can clearly understand that this embodiment provides a method for in-situ detection of the internal geological structure properties of rocks. As for the apparatus disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and relevant details can be found in the method section.
[0119] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
Claims
1. A method for in-situ detection of internal geological structure properties of a rock, characterized in that, The method includes: A target impact echo detection surface is obtained, wherein the target impact echo detection surface is generated by flattening and grinding the target rock mass monitoring area, and the target impact echo detection surface is provided with K impact echo measurement points, where K is a positive integer; A target impact echo probe is pre-deployed on the target impact echo detection surface, wherein the target impact echo probe is a single-excitation multi-channel synchronous receiving type impact echo device; The target impact echo probe is used to detect the impact echo of the target impact echo detection surface to obtain target impact echo data. The target impact echo data is converted into a time-frequency domain signal to obtain the time spectrum of the target impact echo signal; A pre-constructed deep learning model for predicting rock mass discontinuity is used to synchronize the time spectrum of the target impact echo signal to the deep learning model for predicting rock mass discontinuity, thereby obtaining the target rock mass discontinuity prediction result. The deep learning model for predicting rock mass discontinuity is constructed based on an inversion deep learning network. Data is collected from the K impact echo measurement points to generate target rock mass discontinuity label data; A target loss function is pre-constructed, and the target rock mass discontinuity prediction results and the target rock mass discontinuity label data are synchronized to the target loss function to obtain target gap information; The deep learning model for predicting rock mass discontinuity is optimized based on the target gap information.
2. The method of claim 1, wherein, The target impact echo detection surface is provided with K impact echo measurement points, and the method further includes: Obtain the target waistline of the target impact echo detection surface; A preset measurement line offset index is used, and the target waistline is used as the reference measurement line. The measurement line offset index is used as a constraint to obtain the first measurement line and the second measurement line. A preset measurement point offset index is used as a constraint to locate the measurement points of the target waistline, the first measurement line, and the second measurement line, thereby obtaining the K impact echo measurement points.
3. The method of claim 2, wherein, By acquiring data from the K impact echo measurement points, target rock mass discontinuity label data is generated. The method further includes: Drilling and core sampling were performed at the K impact echo measurement points to obtain K target cores; Based on the K target rock cores, rock core structure surface measurements are performed to obtain K rock core structure surface measurement parameters and K target boreholes; Drilling images were taken of the K target boreholes to obtain the K target borehole image results; The target rock mass discontinuity label data is constructed based on the measurement parameters of the K core structural surfaces and the imaging results of the K target boreholes.
4. The method as described in claim 1, characterized in that, The method further includes: using the target impact echo probe to detect the impact echo on the target impact echo detection surface to obtain target impact echo data; and using the target impact echo probe to detect the impact echo on the target impact echo detection surface to obtain target impact echo data. The target impact echo probe is composed of a target stress wave exciter and a target stress wave receiver. A preset detection surface meshing threshold is set, and the target impact echo detection surface is divided into a target detection surface mesh based on the detection surface meshing threshold; A set of grid endpoints is determined based on the target detection surface grid, wherein the set of grid endpoints includes the grid center endpoint and M grid endpoints; The target stress wave exciter is placed at the center endpoint of the grid, and the set of target stress wave receivers is placed at the grid endpoint set to construct a stress wave receiver array; The target stress wave exciter emits stress waves and generates reflected waves at the target impact echo detection surface; In the stress wave receiver array, the target stress wave receiver set receives reflected waves to generate the target impact echo data, wherein the target impact echo data includes M channels of impact echo data.
5. The method as described in claim 4, characterized in that, The method further includes: (1) Pre-constructing a deep learning model for predicting rock mass discontinuities. Collect and acquire core data, borehole camera data, and shock echo data from multiple sample monitoring areas; Based on the multiple sample core data and the multiple sample borehole camera data, multiple sample rock mass discontinuity label data were obtained; A pre-constructed deep learning model for predicting rock mass discontinuity is provided. The deep learning model for predicting rock mass discontinuity includes an encoder module, a decoder module, and a residual connection module. The encoder module and the decoder module are connected based on the residual connection module. The multiple sample impact echo data and the multiple sample rock mass discontinuity label data are identified and divided to obtain the model training dataset, the model testing dataset, and the model validation dataset. The encoder module and the decoder module are trained under supervision using the model training dataset. After training, the encoder module and the decoder module are verified and tested using the model test dataset and the model validation dataset.
6. The method as described in claim 5, characterized in that, The method further includes: The model output accuracy requirement is preset, and it is determined whether the output accuracy of the deep learning model for predicting rock mass discontinuity meets the model output accuracy requirement. If the output accuracy of the deep learning model for predicting rock mass discontinuity meets the model output accuracy requirement, then the deep learning model for predicting rock mass discontinuity is obtained.
7. The method as described in claim 5, characterized in that, The method further includes synchronizing the time spectrum of the target impact echo signal to the deep learning model for rock mass discontinuity prediction to predict rock mass discontinuity, thereby obtaining the target rock mass discontinuity prediction result. Based on the encoder module, high-order waveform features are extracted from the M channel impulse echo data to obtain M high-order waveform feature maps. Based on the decoder module, the mapping processing of the M waveform high-order feature maps is performed to obtain the M channel rock mass discontinuity prediction results; Based on the residual connection module, the initial waveform signal features of the M channel impulse echo data are extracted to obtain M initial waveform signal features; The residual connection module connects the M initial waveform signal features to the decoder module; The decoder module generates M channel feature images based on the predicted rock discontinuity results of the M channels and the features of the M initial waveform signals; The M channel feature images are aggregated to generate the target rock mass discontinuity prediction result, wherein the target rock mass discontinuity prediction result is a depth correlation vector.
8. An in-situ detection system for the geological structural properties inside rocks, characterized in that, The system includes: Detection surface module: Obtains target impact echo detection surface, wherein the target impact echo detection surface is generated by smoothing and grinding the target rock mass monitoring area, and the target impact echo detection surface is provided with K impact echo measurement points, where K is a positive integer; Probe deployment module: Target impact echo probes are pre-deployed on the target impact echo detection surface, wherein the target impact echo probe is a single-excitation multi-channel synchronous receiving type impact echo device; Shock echo detection module: The target shock echo probe is used to detect the shock echo on the target shock echo detection surface to obtain target shock echo data; Signal conversion module: performs time-frequency domain signal conversion on the target impact echo data to obtain the time spectrum of the target impact echo signal; Discontinuity prediction module: A deep learning model for predicting rock mass discontinuity is pre-constructed. The time spectrum of the target impact echo signal is synchronized to the deep learning model for predicting rock mass discontinuity to obtain the target rock mass discontinuity prediction result. The deep learning model for predicting rock mass discontinuity is constructed based on an inversion deep learning network. Data acquisition module: Generates target rock mass discontinuity label data by acquiring data from the K impact echo measurement points; Function construction module: pre-constructs the target loss function and synchronizes the target rock mass discontinuity prediction results and the target rock mass discontinuity label data to the target loss function to obtain target gap information; Model optimization module: Optimizes the deep learning model for predicting rock mass discontinuity based on the target gap information.