Rock failure stage identification method based on multi-channel acoustic emission feature fusion

Abstract

The invention relates to a rock failure stage identification method based on multi-channel acoustic emission feature fusion. The method comprises the following steps: collecting acoustic emission signals of a plurality of acoustic emission sensors arranged on a rock to be detected; counting the acoustic emission signals according to time windows with preset duration, and extracting acoustic emission characteristic parameters corresponding to each time window; and inputting the acoustic emission characteristic parameters corresponding to the acoustic emission sensors in the same time window into the trained rock failure stage identification model to obtain the failure stage of the rock to be detected in the current time window. Real-time and high-precision identification of a rock damage stage is realized through multi-channel acoustic emission signal fusion and a deep learning model; multiple sensors are used for collecting signals, the model automatically extracts high-dimensional features and enhances key information by means of an attention mechanism, and the recognition accuracy and robustness are remarkably improved; compared with a traditional method, the method can output judgment in real time, and provides effective support for engineering safety monitoring and early warning.

Description

Technical Field

[0001] This invention relates to the fields of rock mechanics and geotechnical engineering, and in particular to a method for identifying rock failure stages based on multi-channel acoustic emission feature fusion. Background Technology

[0002] Many mines in my country have entered the deep mining stage. However, engineering activities disrupt the original stress balance of the rock mass, leading to stress concentration and triggering ground pressure disasters such as rock fragmentation, tunnel spalling, and roof collapse. Rock mass is essentially a complex geological medium composed of rock and its internal structural planes, and its mechanical behavior depends on the initiation, propagation, and penetration of internal microcracks. During loading, rock typically undergoes multiple failure stages, including compaction, elasticity, stable fracturing, unstable fracturing, and ultimately macroscopic instability. Therefore, accurately identifying the failure stage of rock is crucial for ensuring safe production in deep mines.

[0003] Existing technologies for identifying the characteristic stress stages of rocks include the axial stiffness method, the circumferential strain response method, and the crack volumetric strain method. However, these methods require post-processing to obtain rock mechanical parameters for calculation. Especially for the crack propagation stage, which ends with peak stress, these factors prevent traditional methods from accurately determining the characteristic stress stage immediately during rock loading.

[0004] In other words, these traditional methods rely on complete stress-strain curves and post-processing calculations of rock mechanics parameters, making it impossible to achieve real-time online identification of rock failure stages. In actual engineering monitoring, the stress state of rocks is dynamically changing, and traditional methods struggle to capture the critical state from stable deformation to unstable failure in a timely manner, exhibiting significant lag and hindering early warning and prevention of geological disasters. Summary of the Invention

[0005] To address the technical problem of existing technologies failing to accurately identify rock failure stages in real time, this invention provides a method for identifying rock failure stages based on multi-channel acoustic emission feature fusion. The technical solution is as follows:

[0006] On the one hand, a method for identifying rock failure stages based on multi-channel acoustic emission feature fusion is provided. This method includes: acquiring acoustic emission signals from multiple acoustic emission sensors arranged on the rock to be detected; statistically analyzing the acoustic emission signals within a preset time window and extracting acoustic emission feature parameters corresponding to each time window; inputting the acoustic emission feature parameters corresponding to each acoustic emission sensor within the same time window into a trained rock failure stage identification model to obtain the failure stage of the rock to be detected within the current time window. The rock failure stage identification model includes: multiple deep feature extraction networks, each corresponding to a different acoustic emission sensor, used to extract high-level semantic features from the acoustic emission feature parameters corresponding to that sensor; each deep feature extraction network includes a dynamic fusion attention mechanism used to adaptively weight the channel dimension and feature dimension within the single signal; a fusion module that fuses the high-level semantic features output by each deep feature extraction network using multi-sensor features to obtain fused features; and a classification network that identifies the failure stage of the rock to be detected within the current time window based on the fused features.

[0007] The beneficial effects of the technical solution provided by the embodiments of the present invention include at least the following: The present invention achieves real-time and high-precision identification of rock failure stages by fusing multi-channel acoustic emission signals and combining them with a deep learning model; the method utilizes multiple acoustic emission sensors to collect multi-source acoustic emission signals during rock deformation, automatically extracts high-dimensional features related to failure stages through a deep learning model, and adaptively enhances key features with the help of an attention mechanism, significantly improving the accuracy and robustness of stage identification; compared with traditional methods that rely on post-processing parameters, the present invention can output failure stage identification results in real time during the loading process, providing timely and effective technical means for engineering safety monitoring and disaster early warning. Attached Figure Description

[0008] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying 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.

[0009] Figure 1 This is a flowchart of a rock failure stage identification method based on multi-channel acoustic emission feature fusion provided by an embodiment of the present invention;

[0010] Figure 2 This is an architecture diagram of a rock failure stage identification model provided in an embodiment of the present invention;

[0011] Figure 3This is an architecture diagram of a deep feature extraction network provided in an embodiment of the present invention;

[0012] Figure 4 This is an architecture diagram of a feature enhancement module provided in an embodiment of the present invention;

[0013] Figure 5 This is a flowchart illustrating the training process of a rock failure stage identification model provided in an embodiment of the present invention.

[0014] Figure 6 This is a flowchart of a rock failure stage category label determination method provided by an embodiment of the present invention;

[0015] Figure 7 This is a schematic diagram of a rock characteristic stress stage division based on the crack volumetric strain method provided in an embodiment of the present invention;

[0016] Figure 8 This is a comparison chart of the original labels of the test set and the model recognition results provided in an embodiment of the present invention. Detailed Implementation

[0017] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0018] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.

[0019] In the embodiments of this invention, the terms "image" and "picture" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, their intended meanings are consistent. Similarly, the terms "of," "corresponding (relevant)," and "corresponding" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, their intended meanings are consistent.

[0020] In this embodiment of the invention, sometimes a subscript such as W1 may be written in a non-subscript form such as W1. When the difference is not emphasized, the meaning they express is the same.

[0021] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0022] This invention provides a method for identifying rock failure stages based on multi-channel acoustic emission feature fusion. This method can be implemented using an electronic device, such as a terminal or a server. Figure 1 The flowchart shown is a method for identifying rock failure stages based on multi-channel acoustic emission feature fusion. The processing flow of this method may include steps S101 to S103.

[0023] Step S101: Collect acoustic emission signals from multiple acoustic emission sensors arranged on the rock to be tested. An acoustic emission sensor is a device used to capture elastic waves released during damage processes such as the generation and propagation of microcracks within a material. Due to the spatial inhomogeneity of damage within the rock, the signal received by a single sensor may be affected by factors such as propagation path and attenuation, leading to incomplete or distorted information. By arranging multiple sensors at different locations on the rock, acoustic emission events can be captured from multiple angles and in all directions, reducing missed detections and false alarms, and improving the reliability and coverage of the monitoring system.

[0024] Step S102: Statistically analyze the acoustic emission signals using a preset time window and extract the acoustic emission characteristic parameters corresponding to each time window.

[0025] Optionally, the acoustic emission characteristic parameters include at least two of the following: event counting parameters, energy parameters, time-domain waveform parameters, and frequency-domain parameters. By segmenting and statistically analyzing the acoustic emission signal within a set time window, characteristic parameters reflecting the dynamics of rock damage evolution can be extracted. These features describe the activity level, energy release, waveform characteristics, and spectral information of acoustic emission events from different dimensions, providing comprehensive and structured input for subsequent model identification and enhancing the model's ability to characterize the damage process.

[0026] Optionally, event count parameters include at least one of event rate and cumulative event count; energy parameters include at least one of energy and cumulative energy; time-domain waveform parameters include at least one of ring count, cumulative ring count, rise time, average rise time, duration, and average duration; and frequency-domain parameters include at least one of average frequency and average center frequency. Event count parameters reflect the frequency of acoustic emission events, energy parameters characterize the energy released by damage, time-domain waveform parameters describe the waveform characteristics and duration of events, and frequency-domain parameters reveal the frequency distribution characteristics of events. The comprehensive use of multiple features can fully characterize the entire process of rock fracture from microcrack initiation to macroscopic fracturing, avoiding the information limitations of single features and improving the model's ability to distinguish different stages of damage.

[0027] Optionally, in subsequent embodiments, the acoustic emission characteristic parameters include all 12 parameters mentioned above, so the scale of the acoustic emission characteristic parameters is 12×1.

[0028] Step S103: Input the acoustic emission feature parameters corresponding to each acoustic emission sensor within the same time window into the trained rock failure stage identification model to obtain the failure stage of the rock to be detected within the current time window. The feature parameters extracted by each sensor at the same time are input into the trained identification model. The model processes the data from each sensor separately through its internal multi-path deep feature extraction network, then integrates the multi-source information through a fusion module, and finally outputs the failure stage of the rock at the current time by a classification network. This method achieves end-to-end identification from multi-channel acoustic emission signals to failure stage categories. The process is highly automated, requires no manual intervention, and is suitable for real-time on-site monitoring.

[0029] Please see Figure 2 Specifically, the aforementioned rock failure stage identification model includes: multiple deep feature extraction networks, a fusion module, and a classification network. Each deep feature extraction network corresponds to a specific acoustic emission sensor, forming a one-to-one relationship. Each deep feature extraction network processes the acoustic emission feature parameters corresponding to one sensor, performing high-level semantic feature extraction on those parameters. Each deep feature extraction network incorporates a dynamic fusion attention mechanism to adaptively weight the channel and feature dimensions within the single signal. The fusion module fuses the high-level semantic features output from each deep feature extraction network using multi-sensor features to obtain fused features. The classification network then identifies the failure stage of the rock under test within the current time window based on these fused features.

[0030] Each deep feature extraction network learns deep features from a single sensor signal. Its dynamic fusion attention mechanism adaptively enhances channels and feature dimensions important to the recognition task while suppressing irrelevant or noisy information. The fusion module integrates features from multiple sources, comprehensively utilizing monitoring information from multiple spatial points to improve the model's ability to perceive the overall damage state of the rock. The classification network performs stage discrimination based on the fused features, providing intuitive and reliable output results.

[0031] Please see Figure 3 Optionally, the deep feature extraction network includes an initial convolutional module and at least two feature enhancement modules connected in sequence, with adjacent feature enhancement modules connected by a downsampling convolutional module; wherein: the initial convolutional module is used to map the input single-channel acoustic emission feature parameters into a multi-channel initial feature map; the feature enhancement module is based on a residual network structure and introduces an attention mechanism to perform deep semantic feature extraction and adaptive weighting on the initial feature map; the downsampling convolutional module is used to spatially downsample the feature map and increase the channel dimension to expand the feature receptive field and extract higher-level abstract features.

[0032] Optionally, the initial convolutional module may include, for example, a convolutional layer, a batch normalization layer, and a ReLU activation layer. The convolutional layer consists of 64 3×1 kernels, responsible for mapping the single-channel input to a 64-channel feature map. After batch normalization and ReLU activation, a high-dimensional feature representation of 12×64 is formed. Multiple feature enhancement modules based on ResNet and a dynamic fusion attention mechanism are stacked sequentially after the initial convolutional module to perform deep feature learning. The feature enhancement modules are connected through downsampling convolutional modules. Figure 3 The single-channel deep feature extraction network in the model contains three stacked feature enhancement modules.

[0033] The deep feature extraction network employs a stacked structure of "initial convolution + multiple feature enhancement modules". The initial convolution module maps the original features to a high-dimensional space, the feature enhancement modules gradually extract and strengthen semantic features through residual connections and attention mechanisms, and the downsampling convolution module gradually expands the receptive field and abstracts higher-level feature representations. This structure effectively avoids the gradient vanishing problem and enhances the model's ability to capture deep patterns in acoustic emission signals.

[0034] Please see Figure 4 Optionally, the feature enhancement module includes at least one residual unit and a dynamically fused attention unit connected in sequence. The residual unit, through shortcut paths containing skip connections and a main path composed of convolutional layers, is used to achieve residual learning and deep feature extraction. The dynamically fused attention unit adaptively weights the features output by the residual unit in both the channel and feature dimensions to enhance key features related to rock failure stages. In the feature enhancement module, the residual unit retains original feature information through skip connections, mitigating network degradation; the dynamically fused attention unit calculates adaptive weights in both the channel and feature dimensions, highlighting the feature components that contribute most to stage identification. This mechanism enables the model to focus on acoustic emission features most relevant to rock failure evolution, improving recognition accuracy and generalization ability.

[0035] Optionally, the dynamic fusion attention unit includes cascaded channel attention subunits and feature attention subunits, wherein the channel attention subunit is used to generate adaptive weights for each channel of the input feature map and to weight all channels; the feature attention subunit is used to generate adaptive weights for each feature dimension of the feature map after channel attention weighting and to weight all feature dimensions.

[0036] Specifically, the input features of the channel attention subunit are denoted as... Where T is the number of AE features (AE stands for acoustic emission features, and its number is, for example, 12), and C is the number of convolutional channels (for example, 64). In the channel attention subunit, the input features are first subjected to global average pooling along the feature dimension:

[0037] ,

[0038] In the formula, z c This represents the mean global response of the c-th channel.

[0039] Subsequently, channel weights are generated using a two-layer fully connected network with a compression ratio of 8:

[0040] ,

[0041] In the formula, W1 and W2 are learnable weight matrices, ReLU is the activation function, and σ is the Sigmoid function. The output is constrained to the interval [0,1] as the adaptive importance weight for each channel. The importance weights for each channel.

[0042] The weighted features are:

[0043] .

[0044] Feature attention subunit with the above X ca As input, the feature activity sequence is first obtained by taking the mean along the convolution channel dimension:

[0045] .

[0046] Next, the AE feature weights are generated through another two fully connected layers:

[0047] ,

[0048] In the formula, W3 and W4 are learnable weight matrices. The attention weights for each AE feature.

[0049] The final weighted output is:

[0050] .

[0051] This mechanism enhances the discriminative convolutional feature map through channel attention and dynamically weights key AE parameters through feature attention, thereby significantly improving the model's accuracy in identifying rock characteristic stress stages.

[0052] Optionally, the high-level semantic features output by each deep feature extraction network are fused using multi-sensor features to obtain fused features, including: (1) adding the high-level semantic features output by each deep feature extraction network element by element to obtain the summed feature map. Assuming there are M deep feature extraction networks and their input features are... The output features are Feature map after addition (2) Perform global average pooling along the feature dimension of the summed feature map to obtain the fused feature, the calculation formula of which is as follows:

[0053] ,

[0054] In the formula, This represents all channel features of the t-th AE feature.

[0055] Multi-sensor feature fusion employs an element-wise addition method, which simply and efficiently integrates multi-source feature information while preserving the original contribution of each sensor feature. Subsequently, global average pooling is used to compress along the feature dimension, resulting in a compact fused representation. This reduces computational cost while retaining the overall statistical characteristics of multi-source information, providing a highly discriminative input for classification.

[0056] Optionally, the classification network includes a dropout layer and a classification head connected in sequence. The dropout layer performs random dropout regularization on the fused features at a preset dropout rate. The classification head performs a fully connected transformation and probability normalization on the regularized features, outputting the probability of the detected rock belonging to each damage stage, and taking the damage stage corresponding to the highest probability as the damage stage of the detected rock in the current time window. The dropout layer in the classification network randomly drops some neurons with a certain probability, effectively preventing model overfitting and enhancing generalization ability. The classification head maps the fused features to the category space through a fully connected layer, and outputs the confidence of each stage after probability normalization. Finally, the stage corresponding to the highest probability is taken as the recognition result, and the output is clear and easy to understand.

[0057] Please see Figure 5 Optionally, the rock failure stage identification model is trained through the following steps S501~S504.

[0058] Step S501: Obtain a sample dataset labeled with rock failure stage tags. Rock failure stage tags will be mentioned later; for example, numbers 1, 2, 3, and 4 can be used as category tags for the four stages. The aforementioned sample dataset was obtained through uniaxial compression experiments on limestone samples (specifications: Φ50×100mm) under displacement control mode (loading rate 0.002mm / s). Two acoustic emission sensors were used to monitor the deformation and failure process in real time, recording acoustic emission signals at a sampling rate of 1MHz. The peak stress point was used as the rock failure point, and acoustic emission signals from the entire compression process before rock failure were selected. The tags corresponding to these acoustic emission signals will be described in detail in subsequent embodiments.

[0059] Optionally, the sample dataset can be divided into a training set, a validation set, and a test set in a certain ratio (6:2:2). The training set is used for learning the deep learning model, the validation set is used to adjust the model parameters, and the test set is used to evaluate the model performance.

[0060] Step S502: Determine the normalized statistics of each acoustic emission characteristic parameter based on the sample dataset. For example, the normalized statistics can be determined based on the maximum value X of the acoustic emission characteristic parameters contained in the sample dataset. max and minimum value X min As a normalized statistic.

[0061] Step S503: Normalize the sample data using normalization statistics. The specific formula is as follows:

[0062] ,

[0063] In the formula, x i These are the original values ​​of the acoustic emission characteristic parameters. This is the normalized value.

[0064] Step S504: Using normalized sample data and a loss function incorporating a class weight balancing strategy, train the initial rock failure stage identification model to obtain a trained rock failure stage identification model. The class weight calculation formula is as follows:

[0065] ,

[0066] In the formula, N is the total number of samples, n c Let w be the number of samples in class c, and K be the number of classes (K=4 in this embodiment). In a specific embodiment of the present invention, a total of 3480 samples were obtained, including 1118 samples in class 1, 644 samples in class 2, 1064 samples in class 3, and 654 samples in class 4. Based on the above formula, the weights of each class in this embodiment can be calculated as follows: w1=0.778, w2=1.351, w3=0.818, w4=1.330.

[0067] Because normalization is introduced during training, the rock failure stage identification model obtained after training also needs to use normalized statistics to perform the same normalization processing on the input acoustic emission feature parameters during inference. That is, after obtaining the acoustic emission feature parameters in step S102, the acoustic emission feature parameters are normalized using preset normalized statistics; in step S103, the normalized acoustic emission feature parameters corresponding to each acoustic emission sensor within the same time window are input into the trained rock failure stage identification model.

[0068] During model training, acoustic emission features are normalized to accelerate model convergence and improve numerical stability. A class weight balancing strategy is introduced to mitigate training bias caused by uneven sample sizes at different stages, thereby enhancing the model's ability to recognize a minority of classes. The trained model uses the same normalization parameters during inference to ensure consistency and reliability in the recognition process.

[0069] Please see Figure 6 Optionally, in the step of obtaining a sample dataset labeled with rock failure stages, the rock failure stage labels are determined through the following steps: S601, synchronously collect axial stress, axial strain, and transverse strain of the rock sample during loading; S602, calculate volumetric strain based on axial strain and transverse strain; S603, calculate crack volumetric strain based on axial stress, volumetric strain, and the rock elastic modulus and Poisson's ratio calculated from the stress-strain curve; S604, identify characteristic inflection points of crack volumetric strain based on the curve of crack volumetric strain versus axial stress, and divide multiple characteristic stress stages of the rock based on these characteristic inflection points; S605, assign each characteristic stress stage as a category label to the acoustic emission signal samples collected within the corresponding time period. By synchronously collecting axial and transverse strain, combining stress data to calculate crack volumetric strain, and using characteristic inflection points (such as inflection points and extreme points) on the curve, the characteristic stress stages of the rock are objectively divided. This method is based on rock mechanics principles, and the division results have clear physical meaning, providing accurate and reliable category labels for acoustic emission samples and laying the data foundation for model training. The four categories obtained are compaction stage, linear elastic stage, stable crack propagation stage, and unstable crack propagation stage, which correspond to category labels 1, 2, 3, and 4, respectively.

[0070] Figure 8 The diagram illustrates a comparison between the original labels on the test set and the model recognition results. The rock failure stage identification method in this invention achieves an accuracy rate of up to 91.38%.

[0071] In summary, this invention provides a method for identifying rock failure stages based on multi-channel acoustic emission feature fusion. Through multi-sensor collaborative monitoring, multi-dimensional feature extraction, and adaptive fusion and classification using a deep learning model, it achieves real-time and accurate identification of the rock failure process. This method overcomes the limitations of traditional methods that rely on post-processing and cannot be applied in real time, significantly improving the automation and accuracy of stage identification. It provides strong technical support for rock mass stability monitoring and disaster early warning in deep mines, tunnel engineering, and other fields, and has significant engineering application value.

[0072] The above embodiments can be implemented, in whole or in part, by software, hardware (such as circuits), firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of the present invention are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. A semiconductor medium can be a solid-state drive.

[0073] It should be understood that the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. A and B can be singular or plural. Additionally, the character " / " in this article generally indicates an "or" relationship between the preceding and following related objects, but it can also represent an "and / or" relationship. Please refer to the context for a more accurate understanding.

[0074] In this invention, "at least one" means one or more, and "more than one" means two or more. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of a single item or a plurality of items. For example, at least one of a, b, or c can represent: a, b, c, ab, ac, bc, or abc, where a, b, and c can be a single item or multiple items.

[0075] It should be understood that, in various embodiments of the present invention, the order of the above-mentioned process numbers does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0076] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0077] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the devices, apparatuses, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0078] In the several embodiments provided by this invention, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another device, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0079] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0080] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0081] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0082] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for identifying rock failure stages based on multi-channel acoustic emission feature fusion, characterized in that, The method includes: Acoustic emission signals from multiple acoustic emission sensors deployed on the rock to be tested were collected; The acoustic emission signals are statistically analyzed in a preset time window, and the acoustic emission feature parameters corresponding to each time window are extracted. The acoustic emission feature parameters corresponding to each acoustic emission sensor within the same time window are input into a trained rock failure stage identification model to obtain the failure stage of the rock to be detected within the current time window. The rock failure stage identification model includes: Multiple deep feature extraction networks, each corresponding to a single acoustic emission sensor, are used to perform high-level semantic feature extraction on the acoustic emission feature parameters corresponding to that sensor. Each deep feature extraction network includes a dynamic fusion attention mechanism, which is used to adaptively weight the channel dimension and feature dimension within the single signal. The fusion module fuses the high-level semantic features output from each deep feature extraction network with multi-sensor features to obtain fused features; and A classification network identifies the destruction stage of the rock to be detected within the current time window based on the fused features.

2. The rock failure stage identification method based on multi-channel acoustic emission feature fusion according to claim 1, characterized in that, The acoustic emission characteristic parameters include at least two of the following: event counting parameters, energy parameters, time-domain waveform parameters, and frequency-domain parameters.

3. The rock failure stage identification method based on multi-channel acoustic emission feature fusion according to claim 2, characterized in that, The event counting parameters include at least one of event rate and cumulative event count; The energy parameters include at least one of energy and cumulative energy; The time-domain waveform parameters include at least one of ring count, cumulative ring count, rise time, average rise time, duration, and average duration. The frequency domain parameters include at least one of the average frequency and the average center frequency.

4. The rock failure stage identification method based on multi-channel acoustic emission feature fusion according to claim 1, characterized in that, The deep feature extraction network includes an initial convolutional module and at least two feature enhancement modules connected in sequence, with adjacent feature enhancement modules connected by downsampling convolutional modules; wherein: The initial convolution module is used to map the input single-channel acoustic emission feature parameters into a multi-channel initial feature map; The feature enhancement module, based on a residual network structure and incorporating an attention mechanism, is used to perform deep semantic feature extraction and adaptive weighting on the initial feature map. The downsampling convolution module is used to spatially downsample the feature map and increase the channel dimension to expand the feature receptive field and extract higher-level abstract features.

5. The rock failure stage identification method based on multi-channel acoustic emission feature fusion according to claim 4, characterized in that, The feature enhancement module includes at least one residual unit and a dynamic fusion attention unit connected in sequence; wherein: The residual unit, through a shortcut path containing skip connections and a main path composed of convolutional layers, is used to achieve residual learning and deep feature extraction. The dynamic fusion attention unit is used to adaptively weight the features output by the residual unit in the channel dimension and the feature dimension to enhance key features related to the rock failure stage.

6. The rock failure stage identification method based on multi-channel acoustic emission feature fusion according to claim 5, characterized in that, The dynamic fusion attention unit includes cascaded components: The channel attention subunit is used to generate adaptive weights for each channel of the input feature map and to weight all channels. The feature attention subunit is used to generate adaptive weights for each feature dimension of the feature map after channel attention weighting, and to weight all feature dimensions.

7. The rock failure stage identification method based on multi-channel acoustic emission feature fusion according to claim 1, characterized in that, The high-level semantic features output by each deep feature extraction network are fused with multi-sensor features to obtain fused features, including: The high-level semantic features output by each deep feature extraction network are added element by element to obtain the summed feature map; The fused features are obtained by performing a global average pooling operation along the feature dimension of the summed feature map.

8. The rock failure stage identification method based on multi-channel acoustic emission feature fusion according to claim 1, characterized in that, The classification network includes a discard layer and a classification head connected in sequence; The discard layer is used to perform random discard regularization processing on the fused features at a preset discard rate; The classification head is used to perform fully connected transformation and probability normalization on the regularized features, output the category probability of the rock to be detected belonging to each destruction stage, and take the destruction stage corresponding to the highest probability as the destruction stage of the rock to be detected in the current time window.

9. The rock failure stage identification method based on multi-channel acoustic emission feature fusion according to claim 1, characterized in that, The rock failure stage identification model was trained in the following manner: Obtain a sample dataset labeled with rock failure stages; Based on the sample dataset, normalized statistics for each acoustic emission characteristic parameter are determined; The sample data were normalized using the normalized statistics. Using normalized sample data and employing a loss function that incorporates a class weight balancing strategy, the initial rock failure stage identification model is trained to obtain a well-trained rock failure stage identification model. The rock failure stage identification model obtained after training needs to use the normalized statistics to perform the same normalization processing on the input acoustic emission feature parameters during inference.

10. The rock failure stage identification method based on multi-channel acoustic emission feature fusion according to claim 9, characterized in that, In the step of obtaining a sample dataset labeled with rock failure stage tags, the rock failure stage tags are determined through the following steps: Simultaneously collect axial stress, axial strain, and transverse strain of rock samples during the loading process; Calculate the volumetric strain based on the axial strain and transverse strain; Based on the axial stress, the volumetric strain, and the rock elastic modulus and Poisson's ratio calculated from the stress-strain curve, the crack volumetric strain is calculated. Based on the curve of the change of crack volumetric strain with axial stress, the characteristic inflection point of crack volumetric strain is identified, and multiple rock characteristic stress stages are divided based on the characteristic inflection point. Each characteristic stress stage is used as a category label and assigned to the acoustic emission signal samples collected within the corresponding time period.

Images