Method, system and equipment for identifying tight sandstone tensile and shear cracks

Abstract

The invention relates to a tight sandstone tensile and shear crack identification method, system and equipment, and the method comprises the steps: collecting the acoustic emission time-frequency diagram data of target tight sandstone, and inputting the acoustic emission time-frequency diagram data into a trained Transform model to identify the crack type. The Transform model is composed of a self-coding neural network of a self-attention mechanism, a full connection layer and a softmax layer. A representation alignment constraint is added in pre-training, and a decoder is limited to learn encoding information, so that an encoding task is constrained in a self-attention encoder, and the feature representation capability of the self-attention encoder is improved. Meanwhile, the self-attention encoder adopts a self-attention module, and time-frequency diagram information is comprehensively analyzed through self-correlation operation, so that the neural network more fully represents time-frequency diagram characteristics. The trained model can be used for effectively distinguishing tensile cracks and shear cracks, and the precision and efficiency of tight sandstone crack recognition are improved.

Description

Technical Field

[0001] The present application relates to the fields of rock mechanics, detection signal processing, artificial intelligence and machine learning technology, and in particular to a method, system and device for identifying tensile and shear cracks in dense sandstone. Background Art

[0002] The crack propagation direction mainly depends on the stress state of the crack surface. Clarifying the crack propagation type during rock failure is of great significance for revealing the rock failure mechanism and clarifying the rock failure type.

[0003] Traditional methods for identifying crack extension types include microscopic analysis and acoustic emission monitoring analysis. Microscopic analysis is to observe the morphology, direction and characteristics of cracks with a microscope (such as an optical microscope or an electron microscope) to distinguish between tensile cracks and shear cracks; acoustic emission monitoring analysis refers to the use of acoustic emission sensors to monitor the acoustic emission signals of crack formation and extension inside rocks in real time during rock testing, and to identify crack types by analyzing characteristic parameters such as the amplitude and frequency of the acoustic emission signals.

[0004] However, the traditional methods for identifying tensile and shear cracks in dense sandstone have problems such as low efficiency and low accuracy. Summary of the invention

[0005] Based on this, it is necessary to provide a method for identifying tensile and shear cracks in dense sandstone, a system for identifying tensile and shear cracks in dense sandstone, and a computer device to address the above technical problems.

[0006] In order to achieve the above objectives, the embodiments of the present invention adopt the following technical solutions:

[0007] On the one hand, a method for identifying tensile and shear cracks in dense sandstone is provided, comprising:

[0008] Collecting target acoustic emission data of target dense sandstone; the target acoustic emission data is a time-frequency diagram of acoustic emission of the target dense sandstone;

[0009] Input the target acoustic emission data into the trained Transformer model; the Transformer model includes an autoencoder neural network based on a self-attention mechanism, a fully connected layer network, and a softmax layer network; during the training process, representation alignment constraints are added to the autoencoder neural network;

[0010] The target crack type of the target acoustic emission data output by the Transformer model is obtained; the target crack type includes tensile crack and shear crack.

[0011] On the other hand, a system for identifying tensile and shear cracks in dense sandstone is also provided, comprising:

[0012] A collection module is used to collect target acoustic emission data of target dense sandstone; the target acoustic emission data is a time-frequency diagram of acoustic emission of the target dense sandstone;

[0013] The input module is used to input the target acoustic emission data into the trained Transformer model; the Transformer model includes an autoencoder neural network based on a self-attention mechanism, a fully connected layer network and a softmax layer network; during the training process, representation alignment constraints are added to the autoencoder neural network;

[0014] The output module is used to obtain the target crack type of the target acoustic emission data output by the Transformer model; the target crack type includes tensile crack and shear crack.

[0015] On the other hand, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the above-mentioned method for identifying tensile and shear cracks in dense sandstone when executing the computer program.

[0016] One or more technical solutions provided in the embodiments of the present application have at least the following technical effects or advantages:

[0017] The above-mentioned method, system and device for identifying tensile and shear cracks in dense sandstone collect the acoustic emission time-frequency graph data of the target dense sandstone and input it into the trained Transformer model to identify the crack type. The Transformer model includes an autoencoder neural network based on the self-attention mechanism, a fully connected layer network and a softmax layer network. In the pre-training process, a representation alignment constraint is added, that is, when training the autoencoder neural network, the decoder is restricted to learn the encoding information, and then all the encoding tasks are constrained in the self-attention encoder, which effectively improves the feature representation ability of the self-attention encoder; at the same time, a self-attention module is used in the self-attention encoder to comprehensively analyze the time-frequency graph information through autocorrelation operations, so that the neural network can more fully characterize the time-frequency graph features. After that, a training labeled data set is obtained through experiments, and then neural network training is carried out to obtain accurate identification of tensile and shear crack extension. Using the trained model for identification can effectively distinguish tensile cracks from shear cracks, improve the accuracy and efficiency of dense sandstone crack identification, and achieve accurate classification of crack types, with strong robustness and real-time recognition capabilities. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the conventional technology, the drawings required for use in the embodiments or the conventional technology descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0019] Figure 1 A schematic diagram of a flow chart of a method for identifying tensile and shear cracks in dense sandstone in one embodiment;

[0020] Figure 2 A schematic diagram of an autoencoding neural network structure based on a self-attention mechanism in one embodiment;

[0021] Figure 3 : is a time-frequency diagram generated based on a pre-trained model in one embodiment, wherein (a) represents an original image, (b) represents an input image, and (c) represents a reconstructed image;

[0022] Figure 4 Schematic diagram of a rock mechanics testing machine for performing a three-point bending test and a shear penetration test in one embodiment, wherein (a) is a schematic diagram of a testing machine for performing a three-point bending test, and (b) is a schematic diagram of a testing machine for performing a shear penetration test;

[0023] Figure 5 The actual processing diagram of a semi-circular disk three-point bending specimen and a shear penetration specimen in one embodiment, wherein (a) represents the actual processing diagram of a semi-circular disk three-point bending specimen, and (b) represents the actual processing diagram of a shear penetration specimen;

[0024] Figure 6 The cross-sectional view and three-dimensional view of the processing of a semi-circular disk three-point bending specimen in one embodiment, wherein (a) represents the cross-sectional view of the processing of a semi-circular disk three-point bending specimen, and (b) represents the three-dimensional view of the processing of a semi-circular disk three-point bending specimen;

[0025] Figure 7 The sectional view and three-dimensional view of the shear penetration specimen processing diagram in one embodiment, wherein (a) represents the sectional view of the shear penetration specimen processing diagram, and (b) represents the shear penetration specimen processing diagram. Figure 3 Vittu;

[0026] Figure 8 1 is a time-frequency diagram of the tensile crack and shear crack transformation in one embodiment, wherein (a) represents the time-frequency diagram of the tensile crack transformation, and (b) represents the time-frequency diagram of the shear crack transformation;

[0027] Fig. 9 The accuracy and loss distribution of the model training process in one embodiment;

[0028] Fig.10 1 is a diagram for identifying a tensile crack without acid treatment in an embodiment, wherein (a) represents a time-frequency diagram of a tensile crack without acid treatment, and (b) represents a weight diagram of a tensile crack without acid treatment;

[0029] Fig.11 1 is a diagram for identifying non-acidified shear cracks in an embodiment, wherein (a) represents a time-frequency diagram of non-acidified shear cracks, and (b) represents a weight diagram of non-acidified shear cracks;

[0030] Fig.12 1 is a diagram for identifying acid-treated tensile cracks in an embodiment, wherein (a) represents a time-frequency diagram of acid-treated tensile cracks, and (b) represents a weight diagram of acid-treated tensile cracks;

[0031] Fig.13 1 is a diagram for identifying acid shear in an embodiment, wherein (a) represents a time-frequency diagram of acid shear cracks, and (b) represents a weight diagram of acid shear cracks. DETAILED DESCRIPTION

[0032] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0033] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which this application belongs. The terms used in the specification of this application are only for the purpose of describing specific embodiments and are not intended to limit this application.

[0034] It should be noted that the reference to "embodiment" herein means that the specific features, structures or characteristics described in conjunction with the embodiment may be included in at least one embodiment of the present invention. The presentation of the phrase at various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments.

[0035] The following will describe the implementation of the present invention in detail with reference to the accompanying drawings in the embodiment diagram of the present invention.

[0036] In one embodiment, Figure 1 As shown, the embodiment of the present application provides a method for identifying tensile and shear cracks in dense sandstone, including the following processing steps S12-16:

[0037] S12, collecting target acoustic emission data of the target tight sandstone; the target acoustic emission data is a time-frequency diagram of acoustic emission of the target tight sandstone.

[0038] It can be understood that the target dense sandstone represents the specific dense sand to be studied and analyzed, and by monitoring the acoustic emission signals of the dense sandstone samples in the tensile or shear test, these signals are converted into acoustic emission time-frequency graphs, that is, time-frequency graphs, for analyzing and identifying the types of cracks. Specifically, these acoustic emission time-frequency graphs are input as input data to the trained Transformer model.

[0039] S14, inputting the target acoustic emission data into the trained Transformer model; the Transformer model includes an autoencoder neural network based on a self-attention mechanism, a fully connected layer network and a softmax layer network; during the training process, adding representation alignment constraints to the autoencoder neural network.

[0040] It can be understood that the collected acoustic emission time-frequency graph of the target tight sandstone is used as input data and input into a trained Transformer model, wherein the Transformer model includes an autoencoder neural network based on the self-attention mechanism, a fully connected layer network, and a softmax layer network. The self-attention mechanism is used to capture important features in the input data. The autoencoder neural network can encode (extract key features) and decode (reconstruct data) the input data, and the self-attention mechanism helps the model focus on relevant important features when processing data; the fully connected layer network is responsible for further processing the features extracted by the autoencoder neural network. The fully connected layer converts the features into more abstract representations through weights and biases, thereby providing support for the final classification task; the softmax layer network is the output layer of the Transformer model, which is used to convert the output of the fully connected layer into a probability distribution to achieve classification. The softmax layer will determine which type of crack (tensile crack or shear crack) the acoustic emission data belongs to based on the probability value of the input feature.

[0041] In addition, if Figure 2 As shown in the figure: Schematic diagram of the structure of the autoencoder neural network based on the self-attention mechanism. During the training process, the representation alignment constraint is added to the autoencoder neural network to restrict the decoder from learning the encoding information, and then all the encoding tasks are constrained in the self-attention encoder, which effectively improves the feature representation ability of the self-attention encoder. Specifically, the alignment loss function can use mean square error, cross entropy loss or contrast loss.

[0042] At the same time, the self-attention module is used in the self-attention encoder to comprehensively analyze the time-frequency graph information through autocorrelation operations, allowing the neural network to more fully characterize the time-frequency graph features. Figure 2 In the figure, Embedded Patches represents the embedding layer, Layer Norm represents the normalization layer, Multi-HeadAttention represents the self-attention module, and MLP represents the multi-layer perception module.

[0043] S16, obtaining a target crack type of the target acoustic emission data output by the Transformer model; the target crack type includes a tensile crack and a shear crack.

[0044] As you can understand, the trained Transformer model is used to process the target acoustic emission data. The model analyzes the input acoustic emission time-frequency graph and predicts the crack type corresponding to these data based on its learned features. The output of the model is the prediction result of the crack type of these acoustic emission data. The target crack type refers to the crack category that the model wants to identify. In this scenario, cracks are divided into two main types: tensile cracks and shear cracks.

[0045] In the above-mentioned identification method of tensile and shear cracks in dense sandstone, the acoustic emission time-frequency graph data of the target dense sandstone is collected and input into the trained Transformer model to identify the crack type. The Transformer model includes an autoencoder neural network based on the self-attention mechanism, a fully connected layer network and a softmax layer network. In the pre-training process, the representation alignment constraint is added, that is, when the autoencoder neural network is trained, the decoder is restricted to learn the encoding information, and then all the encoding tasks are constrained in the self-attention encoder, which effectively improves the feature representation ability of the self-attention encoder; at the same time, the self-attention module is used in the self-attention encoder, and the time-frequency graph information is comprehensively analyzed through the autocorrelation operation, so that the neural network can more fully characterize the time-frequency graph characteristics. After that, the training label data set is obtained through experiments, and then the neural network training is carried out to obtain accurate identification of tensile and shear crack extension. Using the trained model for identification, tensile cracks and shear cracks can be effectively distinguished, the accuracy and efficiency of dense sandstone crack identification are improved, and the accurate classification of crack types is achieved, which has strong robustness and real-time recognition capabilities.

[0046] In one embodiment, in the above-mentioned method for identifying tensile and shear cracks in dense sandstone, the autoencoder neural network includes a self-attention encoder and a decoder, the self-attention encoder includes an embedding layer, a normalization layer, a self-attention module and a multi-perception module, and the alignment loss function of the self-attention encoder is:

[0047]

[0048] Among them, Z m represents the mask block feature encoding obtained by the self-attention encoder, Represents the predicted feature encoding of the output of the self-attention encoder.

[0049] It can be understood that the autoencoder neural network includes a self-attention encoder and a decoder, wherein the main function of the self-attention encoder is to extract and encode the input data through the self-attention mechanism. The self-attention mechanism allows the encoder to capture global information and long-range dependencies when processing the sequence, rather than just the local context. This mechanism allows each input element to pay attention to other elements of the entire input sequence, thereby more comprehensively understanding the patterns and relationships of the data. Through layer-by-layer processing of the embedding layer, the self-attention module, the normalization layer, and the multi-layer perception module, the self-attention encoder can convert the input data into high-dimensional feature representations. These representations contain rich information of the input sequence, laying the foundation for subsequent decoding or other tasks. The role of the decoder is to decode the feature representation generated by the encoder into a target output sequence or target form. In this application, the decoder is used to reconstruct the acoustic emission time-frequency diagram.

[0050] The alignment loss function ensures that the features learned by the encoder are closer to the real target by directly comparing the predicted feature encoding output by the encoder with the actual mask block feature encoding. This direct supervisory signal helps the encoder to be more accurate in the feature extraction stage. The alignment loss function promotes the consistency of feature representation, enabling the encoder to better learn and capture the potential patterns and structures of the input data. This optimization process can improve the model's ability to understand the data, thereby achieving higher accuracy and robustness in the task.

[0051] In one embodiment, in the above-mentioned method for identifying tensile and shear cracks in dense sandstone, the training process of the Transformer model includes a pre-training process, and the pre-training process includes the steps of:

[0052] The existing pre-training acoustic emission data are respectively segmented to obtain small image blocks of the pre-training acoustic emission data, and the small image blocks of the pre-training acoustic emission data are fused encoded by image encoding and position encoding to obtain a pre-training data set.

[0053] It can be understood that after the pre-training acoustic emission data is divided into small image blocks, each small block is subjected to a fusion process of image coding and position coding, thereby generating a fusion feature coding containing image and position information, and finally constructing a pre-training data set.

[0054] After randomly masking the pre-training dataset, the pre-training data of each unmasked and corresponding masked block are respectively input into the self-attention encoder to obtain each predicted feature encoding and the corresponding feature encoding of each masked block.

[0055] It can be understood that after the pre-training data set is randomly masked, the unmasked part and the corresponding mask block are input into the self-attention encoder respectively. The unmasked part generates the predicted feature code through the encoder, while the mask block generates the real mask block feature code through the encoder.

[0056] The self-attention encoder is optimized according to the alignment loss function, each predicted feature encoding and the corresponding feature encoding of each mask block.

[0057] It can be understood that the alignment loss function is used to compare each predicted feature encoding with the corresponding mask block feature encoding, measure the difference between them, and optimize the parameters of the self-attention encoder by minimizing this difference, so that the encoder can more accurately predict the features of the mask part. This process strengthens the encoder's understanding and learning ability of the input features, thereby improving the model's ability to represent the time-frequency graph features.

[0058] According to each prediction feature code, each pre-training acoustic emission data is reconstructed to obtain each pre-training reconstructed acoustic emission data.

[0059] It can be understood that according to each prediction feature code, it is converted into a corresponding acoustic emission time-frequency diagram. That is, the pre-trained acoustic emission data is reconstructed through the decoding process, and finally the pre-trained reconstructed acoustic emission data with a structure similar to the original data is obtained.

[0060] The autoencoder neural network is optimized according to the reconstruction loss function, the pre-trained reconstructed acoustic emission data and the pre-trained acoustic emission data.

[0061] It can be understood that the reconstruction loss function is used to compare the pre-trained reconstructed AE data with the original pre-trained AE data and calculate the difference between them. By minimizing this reconstruction error, the parameters of the autoencoder neural network are optimized. This process aims to improve the reconstruction ability of the network so that it can learn and reproduce the characteristics of the original AE data more accurately. The decoder only focuses on how to accurately reconstruct the input data without making too many adjustments to the feature representation generated by the encoder.

[0062] In order to verify the learning effect of extracting time-frequency graph features, the missing information restoration method is used to test the time-frequency graph feature extraction ability of the Transformer model. That is, only part of the time-frequency graph information is given and the model is allowed to restore all the time-frequency graph information based on the existing prior information. The closer the restoration degree is to the real image, the better the model learning effect is. The test results are as follows: Figure 3 : The time-frequency diagram generated based on the pre-trained model, where (a) represents the original image, (b) represents the input image, and (c) represents the reconstructed image.

[0063] from Figure 3As can be seen in the figure, when a picture with 75% random occlusion is input to the model, the pre-trained model uses the input prior information (unoccluded part) and reconstructs the time-frequency diagram based on the learned time-frequency diagram features. By comparing the original image and the reconstructed image, it can be seen that the model can basically restore the entire time-frequency diagram, indicating that the pre-trained model has learned the changing law of the acoustic emission waveform and can infer the information of other parts through partial information. Although, there are still some details that are not completely restored, such as the area under the high energy part in the figure. There are multiple grid-like distributions in the original image, which are not reflected in the reconstructed image. This may be caused by two reasons. First, the input prior information is not strongly correlated with the waveform information of this part, and the information of this part cannot be obtained based on the known time-frequency diagram; in addition, it may be that the model has not learned the relationship between this part of the information and the prior information. This requires further increasing the sample of training data, while increasing the number of model parameters and increasing the dimension of the embedding layer, and then training a more powerful pre-trained model for extracting time-frequency diagram information.

[0064] In one embodiment, in the above-mentioned method for identifying tensile and shear cracks in dense sandstone, the reconstruction loss function is:

[0065]

[0066] Where n represents the number of images in the training batch (i.e., batch size), P represents the image set of the training batch, and μ x Represents the mean of the processed image, μ y represents the mean of the real image, σ x represents the variance of the processed image, σ y represents the variance of the real image, σ xy represents the covariance of the processed image and the true image, C1 represents the first constant, and C2 represents the second constant (C1 and C2 are set to maintain stability and prevent the denominator from being divided by zero).

[0067] During the training process, we only calculate the loss of the encoding alignment constraint without gradient transfer, and finally get the total loss function as follows:

[0068]

[0069] Among them, L recon is the reconstruction loss function, λ is the adjustment coefficient, sg[·] means no gradient transfer, Z m represents the mask block feature encoding obtained by the self-attention encoder, Represents the predicted feature encoding of the output of the self-attention encoder.

[0070] In one embodiment, the above-mentioned method for identifying tensile and shear cracks in dense sandstone, after the Transformer model is pre-trained, further comprises the following steps:

[0071] Tensile and shear fracture tests are carried out on the training dense sandstone, and each tensile fracture acoustic emission data and each shear fracture acoustic emission data of the training dense sandstone are collected as the training labeled data set; the tensile fracture acoustic emission data is the acoustic emission time-frequency diagram of the tensile fracture test of the training dense sandstone, and the shear fracture acoustic emission data is the acoustic emission time-frequency diagram of the shear fracture test of the training dense sandstone.

[0072] It can be understood that in different application scenarios, the steps for obtaining training labeled data sets are slightly different. For example, in the scenario of identifying tensile and shear cracks in dense sandstone, it is only necessary to collect the tensile fracture acoustic emission data and the shear fracture acoustic emission data of the training dense sandstone as the training labeled data set, train the model, and finally apply the trained model to identify the tensile and shear cracks in dense sandstone. In the scenario of spatiotemporal expansion analysis of CO2 fracturing cracks, it is necessary to collect the tensile fracture acoustic emission data and the shear fracture acoustic emission data of the training dense sandstone before and after acidification as the training labeled data set (the acidification process is to use soil acid (12% HCl + 3% HF) for 24 hours), and finally apply the trained model to identify the types of tensile and shear cracks in the CO2 fracturing process.

[0073] In this embodiment, taking the application scenario of identifying tensile and shear cracks in dense sandstone as an example, in order to obtain a training labeled data set, the acoustic emission data generated by the three-point bending test and the shear penetration test can be obtained experimentally. The specific experimental steps can be: (1) respectively processing and manufacturing the semi-circular three-point bending specimen and the shear penetration specimen of the training dense sandstone and drying them for later use; (2) carrying out the three-point bending test and the shear penetration test, and monitoring and saving the acoustic emission data in the experimental process in real time; (3) using the wavelet transformation method to convert the acoustic emission time-frequency diagram into a two-dimensional image.

[0074] Specifically, a rock mechanics testing machine can be used to carry out a semi-circular disk three-point bending test and a shear penetration test by means of displacement loading control, and the displacement loading rate can be set at 0.02 mm / min. Figure 4 A rock mechanics testing machine for performing a three-point bending test and a shear penetration test is shown, wherein (a) is a schematic diagram of a testing machine for performing a three-point bending test, and (b) is a schematic diagram of a testing machine for performing a shear penetration test; Figure 5 The figures show the actual processing diagrams of the semi-circular disk three-point bending specimen and the shear penetration specimen, wherein (a) shows the actual processing diagram of the semi-circular disk three-point bending specimen, and (b) shows the actual processing diagram of the shear penetration specimen; Figure 6The figure shows the cross-sectional view and three-dimensional view of the processing of the semi-circular disk three-point bending specimen, wherein (a) shows the cross-sectional view of the processing of the semi-circular disk three-point bending specimen, and (b) shows the three-dimensional view of the processing of the semi-circular disk three-point bending specimen; Figure 7 The cross-sectional view and three-dimensional view of the shear penetration specimen processing diagram are shown, where (a) shows the cross-sectional view of the shear penetration specimen processing diagram, and (b) shows the shear penetration specimen processing diagram. Figure 3 Dimensional diagram; the shear penetration specimen is a cube with a height of 40 mm (it can also be processed into a cylinder). A diamond drill with a diameter of 25 mm is used to preform circular cracks with depths of a = 4 mm and b = 24 mm on two parallel sections of the specimen. The crack width t is 1.5 mm, and the central area cylinder diameter ID is 22 mm. After the tensile and shear fracture tests on the training dense sandstone, Figure 8 The time-frequency diagrams after the transformation of tensile and shear cracks are shown in Figure 1, where (a) is the time-frequency diagram after the transformation of tensile cracks, and (b) is the time-frequency diagram after the transformation of shear cracks. The horizontal axis in the figure represents the duration, which is 1024 μs in total, and the vertical axis represents the frequency, which ranges from 0 to 500 kHz. The acoustic emission data obtained from the three-point bending test are marked as tensile cracks, and the acoustic emission data obtained from the shear penetration test are marked as shear cracks.

[0075] Split the training labeled dataset into training dataset and test dataset.

[0076] It can be understood that after the generated time-frequency graphs are mixed, most of them are randomly extracted as training sets, for example, about 75%-90% can be extracted as training data sets for training Transformer models, that is, to let the model learn the characteristics and patterns of the data, and the remaining part is used as a test data set to evaluate the performance of the model and check whether the model can be well generalized to unseen data. Specifically, in this embodiment, after the generated time-frequency graphs are mixed, 80% are randomly extracted as training data sets, and the remaining 20% ​​are used as test data sets.

[0077] The Transformer model is trained for multiple rounds based on the training data set to optimize the parameters of the Transformer model. The Transformer model is verified based on the test data set to evaluate the accuracy of the Transformer model.

[0078] It can be understood that when training the Transformer model, the model will continuously adjust its internal parameters to minimize the error between the predicted output and the true label by using the training data set in multiple epochs. After training, the model is verified using an unseen test data set, and the generalization ability and accuracy of the model are measured through evaluation indicators (such as accuracy, etc.).

[0079] Specifically, Fig. 9The accuracy and loss distribution of the model training process are shown in the figure, where Accuracy represents the correct rate, Loss represents the loss, Train accuracy represents the training accuracy, Valid accuracy represents the validation accuracy, Train loss represents the training loss, and Valid loss represents the validation loss. The training data set was trained for 60 epochs (training rounds). The model was unstable in the first 30 epochs, indicating that the model was constantly adjusting the feature weights to seek higher accuracy. After 35 epochs of training, the accuracy of the model on the test set stabilized at around 95% and no longer increased, while the accuracy of the training set continued to increase, indicating that the model has been overfitted, and the neural network model began to use the unique noise information in the acoustic emission time-frequency diagram as the basis for recognition to further improve the accuracy. This phenomenon can be alleviated by further increasing the amount of training data. In this training, the training with the highest accuracy (accuracy of 95.4%) was finally selected as the final model of this training.

[0080] In addition, by obtaining the attention of the Transformer to analyze the feature weight distribution of the neural network, the neural network discrimination basis is established, and then the difference between the tensile and shear failure of sandstone is obtained. The identification basis method is as follows:

[0081] ① Calculate the score of each attention layer in each layer of Transformer based on correlation;

[0082] ② By combining correlation and gradient information, the negative effects are eliminated through iteration;

[0083] ③ Integrate the information into the attention map to obtain a class-specific visualization of the self-attention model;

[0084] ④ Finally, the judgment basis of the self-attention model can be displayed in the generated thermal map, and the difference in the time-frequency diagram of rock fracture with or without acidification is obtained.

[0085] Fig.10 : represents the identification basis diagram of tensile cracks without acidification (in the method of this embodiment, the dense sandstone is not acidified, and in order to distinguish it from the tensile crack identification basis diagram after acidification in other embodiments below, it is named as the identification basis diagram of tensile cracks without acidification), wherein (a) represents the time-frequency diagram of tensile cracks without acidification, (b) represents the weight diagram of tensile cracks without acidification, Fig.11 The figure represents the identification basis of non-acidified shear cracks, where (a) represents the time-frequency diagram of non-acidified shear cracks, and (b) represents the weight diagram of non-acidified shear cracks. Figure 10-11 The bright area is an important basis for model recognition. Fig.10 and Fig.11It can be seen that the main identification basis of tensile and shear cracks is distributed in the low-frequency area, and compared with shear cracks, the acoustic emission frequency diagram generated by tensile cracks has a large amount of high energy distribution in the high-frequency part. At the same time, the frequency of the main identification area of ​​shear cracks by the model is often lower than that of tensile cracks. Fig.10 The tensile crack in the model is Fig.11 ) The main basis for distinguishing the time-frequency diagram is the low-frequency area around the high energy, which may be determined by the rupture mode of the shear fracture. The fracture surface of the shear fracture is generally large, and the waveform generated during rupture has low frequency and high energy.

[0086] In one embodiment, the above-mentioned method for identifying tensile and shear cracks in dense sandstone, the steps of performing tensile and shear fracture tests on training dense sandstone, and collecting each tensile fracture acoustic emission data and each shear fracture acoustic emission data of the training dense sandstone as a training labeled data set include: performing tensile and shear fracture tests on the training dense sandstone before and after acidification, and collecting each tensile fracture acoustic emission data and each shear fracture acoustic emission data of the training dense sandstone before and after acidification as a training labeled data set; the acidification process is to use soil acid (12% HCl+3% HF) for 24 hours.

[0087] It can be understood that in the scenario of spatiotemporal crack propagation analysis of CO2 fracturing, it is necessary to collect the tensile fracture acoustic emission data and shear fracture acoustic emission data of the training tight sandstone before and after acidification as the training labeled data set (the acidification process is to use soil acid (12% HCl + 3% HF) for 24 hours).

[0088] The specific test steps can be: (1) respectively processing and manufacturing a semi-circular three-point bending specimen and a shear penetration specimen of training dense sandstone and drying them for later use; (2) conducting tensile and shear fracture acoustic emission tests on acid-treated and non-acid-treated dense sandstone, wherein the acid-treated specimen is treated with soil acid (12HCl+3% HF) for 24 hours and the acoustic emission data during the experiment are monitored and saved in real time; (3) using the wavelet transformation method to convert the acoustic emission time-frequency diagram into a two-dimensional image.

[0089] Specifically, a semi-circular three-point bending specimen and a shear penetration specimen were processed by a rock mechanics testing machine, and tensile and shear crack acoustic emission experiments were carried out before and after acidification. The acid-treated specimens were treated with soil acid (12% HCl + 3% HF) for 24 hours. The tensile and shear crack extension tests adopted a displacement loading mode with a loading rate of 0.02 mm / min. The collected acoustic emission time-frequency graph was transformed into time-frequency using the wavelet transform method, and then the one-dimensional features of the acoustic emission waveform were transformed into two-dimensional features to facilitate subsequent extraction and utilization. The acoustic emission data obtained from the three-point bending test were marked as type I cracks, and the acoustic emission data obtained from the shear penetration test were marked as type II cracks. After mixing the time-frequency graphs generated by them, 80% were randomly selected as the training data set, and the remaining 20% ​​were used as the test data set. The specific data distribution is shown in Table 1: The number of AE waveforms obtained.

[0090] Table 1:

[0091]

[0092] By acquiring the attention of the Transformer, the feature weight distribution of the neural network can be analyzed, and the neural network discrimination basis can be established. The identification basis of the non-acidified tensile crack is shown in the figure. Fig.10 As shown; the identification basis of non-acidified shear cracks is shown in the figure Fig.11 As shown; Fig.12 The figure represents the identification basis of acid-treated tensile cracks, wherein (a) represents the time-frequency diagram of acid-treated tensile cracks, and (b) represents the weight diagram of acid-treated tensile cracks; Fig.13 The diagram represents the identification basis of acid shear, where (a) represents the time-frequency diagram of acid shear cracks, and (b) represents the weight diagram of acid shear cracks. Figure 10-13 The bright area is an important basis for model recognition. Figure 10-13 It can be seen that compared with shear cracks, the acoustic emission time-frequency diagram generated by tensile cracks has a large amount of high energy distribution in the high-frequency part. At the same time, the acoustic emission waveform generated by rock fracture after acidification has more energy distributed in the high-frequency part.

[0093] exist Figure 10-13 In the distribution diagram of identification basis in , it is found that the main identification basis of the model for unacidified tensile and shear cracks is distributed in the low-frequency area. The identification basis of the model for cracks after acidification is mainly distributed in the high-frequency area, especially the acoustic emission time-frequency diagram of tensile cracks after acidification. The frequency of the main identification area of ​​the model for shear cracks is often lower than that of tensile cracks. For example, compared with Fig.10 The model is for acid-free shear Fig.11 The main basis for the identification of the time-frequency diagram is the low-frequency area around the high energy. After acidification, the model shifts the key identification area of ​​the acoustic emission time-frequency diagram generated by shear cracks to the high frequency ( Fig.13 ).

[0094] In one embodiment, in the above-mentioned method for identifying tensile and shear cracks in tight sandstone, the step of collecting target acoustic emission data of the target tight sandstone includes: performing acid treatment and CO2 fracturing test on the target tight sandstone to obtain target acoustic emission data of the target tight sandstone. In the step of obtaining the target crack type of the target acoustic emission data output by the Transformer model, the target crack type includes tensile cracks after acidizing and shear cracks after acidizing; after the step of obtaining the target crack type of the target acoustic emission data, it also includes: performing spatiotemporal expansion analysis according to the target crack type.

[0095] It can be understood that in the scenario of CO2 fracturing crack spatiotemporal expansion analysis, after pre-training, in the application process, the specific steps are: acid treatment and CO2 fracturing test are performed on the target dense sandstone, the target acoustic emission data of the target dense sandstone is obtained, and the target crack type of the target acoustic emission data output by the Transformer model is obtained (identified as tensile cracks after acidization or shear cracks after acidization). Finally, the CO2 fracturing crack spatiotemporal expansion analysis can be performed according to the output crack type.

[0096] Specifically, in this embodiment, cracks mainly occur at about 1000s and 1500s. At about 1000s, cracks mainly occur around the borehole, and two cracks mainly occur at about 1500s. Among them, acidification stretching mainly occurs in the early stage of macroscopic rupture, such as the red event near the CO2 injection port occurs at about 100s, and the remaining acidification stretching events occur between 1400-1500s. This stage is the microcrack expansion stage inside the sandstone. At the same time, the location of the acidification stretching generated in this stage is mainly concentrated in the area away from the borehole. In addition, the acidification shear fracture is also mainly divided into two parts. The first part is the CO2 pressure increase stage of about 1000s, and the other part is between 1300-1500s. This stage is the stage where the fluid pressure produces a series of pressure drops (rapid crack expansion). At the same time, the acidification shear events in the entire fracturing process are mainly concentrated after 1300s according to the occurrence time, and two more obvious cracks are formed. Unacidified stretching also contributes a large number of cracks. Unacidified shear cracks are mainly divided into two parts. The first part is generated in the stress adjustment stage before 200s, and the other part is between 1300-1500s, which is the stage where the fluid pressure produces a series of pressure drops (rapid crack expansion). Based on acoustic emission positioning analysis, the temporal and spatial distribution of different types of cracks in the entire fracturing process can be obtained.

[0097] It should be understood that although Figure 1The steps in the flowchart are shown in sequence as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified in this document, there is no strict order restriction for the execution of these steps, and these steps can be executed in other orders. Moreover, Figure 1 At least part of the steps may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily executed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least part of the sub-steps or stages of other steps.

[0098] In one embodiment, a system for identifying tensile and shear cracks in dense sandstone is provided, comprising:

[0099] A collection module is used to collect target acoustic emission data of target dense sandstone; the target acoustic emission data is a time-frequency diagram of acoustic emission of the target dense sandstone;

[0100] The input module is used to input the target acoustic emission data into the trained Transformer model; the Transformer model includes an autoencoder neural network based on a self-attention mechanism, a fully connected layer network and a softmax layer network; during the training process, representation alignment constraints are added to the autoencoder neural network;

[0101] The output module is used to obtain the target crack type of the target acoustic emission data output by the Transformer model; the target crack type includes tensile crack and shear crack.

[0102] The above-mentioned identification system of tensile and shear cracks in dense sandstone collects the acoustic emission time-frequency graph data of the target dense sandstone and inputs it into the trained Transformer model to identify the crack type. The Transformer model includes an autoencoder neural network based on the self-attention mechanism, a fully connected layer network and a softmax layer network. In the pre-training process, the representation alignment constraint is added, that is, when the autoencoder neural network is trained, the decoder is restricted to learn the encoding information, and then all the encoding tasks are constrained in the self-attention encoder, which effectively improves the feature representation ability of the self-attention encoder; at the same time, the self-attention module is used in the self-attention encoder to comprehensively analyze the time-frequency graph information through autocorrelation operations, so that the neural network can more fully characterize the time-frequency graph features. After that, the training labeled data set is obtained through experiments, and then the neural network training is carried out to obtain accurate identification of tensile and shear crack extension. Using the trained model for identification, it can effectively distinguish tensile cracks from shear cracks, improve the accuracy and efficiency of dense sandstone crack identification, and realize accurate classification of crack types, with strong robustness and real-time recognition capabilities.

[0103] In one embodiment, the above-mentioned system for identifying tensile and shear cracks in dense sandstone further includes a training module for training the Transformer model.

[0104] In one embodiment, in the above-mentioned identification system of tensile and shear cracks in dense sandstone, the autoencoder neural network includes a self-attention encoder and a decoder, the self-attention encoder includes an embedding layer, a normalization layer, a self-attention module and a multi-perception module, and the alignment loss function of the self-attention encoder is:

[0105]

[0106] Among them, Z m represents the mask block feature encoding obtained by the self-attention encoder, Represents the predicted feature encoding of the output of the self-attention encoder.

[0107] In one embodiment, the above-mentioned system for identifying tensile and shear cracks in dense sandstone includes a pre-training module in the training module, and the pre-training module includes:

[0108] The acquisition submodule is used to perform image segmentation on each existing pre-training acoustic emission data, obtain image blocks of each pre-training acoustic emission data, and perform fusion encoding of image encoding and position encoding on each image block of the pre-training acoustic emission data to obtain a pre-training data set;

[0109] An input submodule is used to randomly mask the pre-trained data set, and then input the pre-trained data of each unmasked block and the corresponding masked block into the self-attention encoder to obtain each predicted feature code and the corresponding feature code of each masked block;

[0110] A first optimization submodule is used to optimize the self-attention encoder according to the alignment loss function, each predicted feature encoding and each corresponding mask block feature encoding;

[0111] A reconstruction submodule, used to reconstruct each pre-trained acoustic emission data according to each prediction feature code, and obtain each pre-trained reconstructed acoustic emission data;

[0112] The second optimization submodule is used to optimize the autoencoder neural network according to the reconstruction loss function, each pre-trained reconstructed acoustic emission data and each pre-trained acoustic emission data.

[0113] In one embodiment, the reconstruction loss function of the above-mentioned identification system of tensile and shear cracks in dense sandstone is:

[0114]

[0115] Where n is the number of images in the training batch, P is the image set of the training batch, and μ xRepresents the mean of the processed image, μ y represents the mean of the real image, σ x represents the variance of the processed image, σ y represents the variance of the real image, σ xy represents the covariance between the processed image and the real image, C1 represents the first constant, and C2 represents the second constant.

[0116] In one embodiment, the training module of the above-mentioned system for identifying tensile and shear cracks in dense sandstone further includes:

[0117] The training label data set acquisition module is used to perform tensile and shear fracture tests on the training dense sandstone, and collect various tensile fracture acoustic emission data and various shear fracture acoustic emission data of the training dense sandstone as the training label data set; the tensile fracture acoustic emission data is the acoustic emission time-frequency diagram of the tensile fracture test of the training dense sandstone, and the shear fracture acoustic emission data is the acoustic emission time-frequency diagram of the shear fracture test of the training dense sandstone;

[0118] A splitting module is used to split the training labeled data set into a training data set and a test data set;

[0119] The round training module is used to train the Transformer model for multiple rounds according to the training data set to optimize the parameters of the Transformer model;

[0120] The verification module is used to verify the Transformer model based on the test data set and evaluate the accuracy of the Transformer model.

[0121] In one embodiment, the above-mentioned system for identifying tensile and shear cracks in dense sandstone, in a training label data set acquisition module, performs tensile and shear fracture tests on the training dense sandstone before and after acidification, and collects each tensile fracture acoustic emission data and each shear fracture acoustic emission data of the training dense sandstone before and after acidification as a training label data set; the acidification process is to use soil acid (12% HCl + 3% HF) for 24 hours.

[0122] In one embodiment, the above-mentioned identification system of tensile and shear cracks in dense sandstone, in a collection module, performs acid treatment and CO2 fracturing test on the target dense sandstone to obtain target acoustic emission data of the target dense sandstone. In the output module, the target crack types include tensile cracks after acidification and shear cracks after acidification. It also includes an analysis module for performing spatiotemporal expansion analysis according to the target crack type.

[0123] The specific definition of the identification system of tensile and shear cracks in dense sandstone can be found in the definition of the identification method of tensile and shear cracks in dense sandstone mentioned above, which will not be repeated here. Each module in the above-mentioned identification system of tensile and shear cracks in dense sandstone can be implemented in whole or in part by software, hardware and a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory in the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0124] In one embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the following steps are implemented:

[0125] Collecting target acoustic emission data of target dense sandstone; the target acoustic emission data is a time-frequency diagram of acoustic emission of the target dense sandstone;

[0126] Input the target acoustic emission data into the trained Transformer model; the Transformer model includes an autoencoder neural network based on a self-attention mechanism, a fully connected layer network, and a softmax layer network; during the training process, representation alignment constraints are added to the autoencoder neural network;

[0127] The target crack type of the target acoustic emission data output by the Transformer model is obtained; the target crack type includes tensile crack and shear crack.

[0128] It can be understood that in addition to the memory and processor involved above, the above-mentioned computer device also includes other software and hardware components not listed in this specification. The specific components can be determined according to the model of the specific computer device in different application scenarios, and this specification will not list them one by one in detail.

[0129] In one embodiment, when the processor executes the computer program, the processor may also implement the steps or sub-steps added in each embodiment of the above-mentioned method for identifying tensile and shear cracks in dense sandstone.

[0130] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0131] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A method for identifying tensile and shear cracks in dense sandstone, characterized in that: Includes steps: Collecting target acoustic emission data of target dense sandstone; the target acoustic emission data is an acoustic emission time-frequency diagram of the target dense sandstone; The target acoustic emission data is input into a trained Transformer model; the Transformer model includes an autoencoder neural network based on a self-attention mechanism, a fully connected layer network and a softmax layer network; during the training process, a representation alignment constraint is added to the autoencoder neural network; The target crack type of the target acoustic emission data output by the Transformer model is obtained; the target crack type includes tensile crack and shear crack.

2. The method for identifying tensile and shear cracks in dense sandstone according to claim 1, characterized in that: The autoencoder neural network includes a self-attention encoder and a decoder. The self-attention encoder includes an embedding layer, a normalization layer, a self-attention module and a multi-perception module. The alignment loss function of the self-attention encoder is: Among them, Z m represents the mask block feature encoding obtained by the self-attention encoder, Represents the predicted feature encoding of the output of the self-attention encoder.

3. The method for identifying tensile and shear cracks in dense sandstone according to claim 2, characterized in that: The training process of the Transformer model includes a pre-training process, and the pre-training process includes the following steps: Performing image segmentation on each existing pre-training acoustic emission data to obtain image blocks of each pre-training acoustic emission data, and performing fusion encoding of image encoding and position encoding on each image block of the pre-training acoustic emission data to obtain a pre-training data set; After randomly masking the pre-training data set, the pre-training data of each unmasked block and the corresponding masked block are respectively input into the self-attention encoder to obtain each predicted feature code and each corresponding masked block feature code; Optimizing the self-attention encoder according to the alignment loss function, each of the predicted feature codes and each of the corresponding mask block feature codes; Reconstructing each of the pre-trained acoustic emission data according to each of the predicted feature codes to obtain each of the pre-trained reconstructed acoustic emission data; The autoencoder neural network is optimized according to the reconstruction loss function, each of the pre-trained reconstructed acoustic emission data and each of the pre-trained acoustic emission data.

4. The method for identifying tensile and shear cracks in dense sandstone according to claim 3, characterized in that: The reconstruction loss function is: Where n is the number of images in the training batch, P is the image set of the training batch, and μ x Represents the mean of the processed image, μ y represents the mean of the real image, σ x represents the variance of the processed image, σ y represents the variance of the real image, σ xy represents the covariance between the processed image and the real image, C1 represents the first constant, and C2 represents the second constant.

5. The method for identifying tensile and shear cracks in dense sandstone according to claim 3 or 4, characterized in that: After the Transformer model performs the pre-training process, the following steps are further included: Conducting tensile and shear fracture tests on the training dense sandstone, collecting each tensile fracture acoustic emission data and each shear fracture acoustic emission data of the training dense sandstone as a training labeled data set; the tensile fracture acoustic emission data is a tensile fracture test acoustic emission time-frequency diagram of the training dense sandstone, and the shear fracture acoustic emission data is a shear fracture test acoustic emission time-frequency diagram of the training dense sandstone; Splitting the training labeled data set into a training data set and a test data set; Performing multiple rounds of training on the Transformer model according to the training data set to optimize the parameters of the Transformer model; The Transformer model is verified according to the test data set to evaluate the accuracy of the Transformer model.

6. The method for identifying tensile and shear cracks in dense sandstone according to claim 5, characterized in that: The step of performing tensile and shear fracture tests on the training dense sandstone and collecting each tensile fracture acoustic emission data and each shear fracture acoustic emission data of the training dense sandstone as a training labeled data set comprises: The training dense sandstone is subjected to tensile and shear fracture tests before and after acidification, and each tensile fracture acoustic emission data and each shear fracture acoustic emission data of the training dense sandstone before and after acidification are collected as a training labeled data set; the acidification process is to use soil acid (12% HCl+3% HF) for 24 hours.

7. The method for identifying tensile and shear cracks in dense sandstone according to claim 6, characterized in that: The step of collecting target acoustic emission data of the target tight sandstone includes: performing acid treatment and CO2 fracturing test on the target tight sandstone to obtain the target acoustic emission data of the target tight sandstone; In the step of obtaining the target crack type of the target acoustic emission data output by the Transformer model, the target crack type includes tensile cracks after acidizing and shear cracks after acidizing; After the step of obtaining the target crack type of the target acoustic emission data, the method further includes: performing a time-space expansion analysis according to the target crack type.

8. A system for identifying tensile and shear cracks in dense sandstone, characterized in that: include: A collection module, used for collecting target acoustic emission data of target dense sandstone; The target acoustic emission data is an acoustic emission time-frequency diagram of the target tight sandstone; An input module, used to input the target acoustic emission data into a trained Transformer model; the Transformer model includes an autoencoder neural network based on a self-attention mechanism, a fully connected layer network and a softmax layer network; during the training process, a representation alignment constraint is added to the autoencoder neural network; An output module, used to obtain a target crack type of the target acoustic emission data output by the Transformer model; The target crack types include tensile cracks and shear cracks.

9. The identification system for tensile and shear cracks in dense sandstone according to claim 8, characterized in that: It also includes a training module for training the Transformer model.

10. A computer device comprising a memory and a processor, characterized in that: The memory stores a computer program, and when the processor executes the computer program, the steps of the method for identifying tensile and shear cracks in dense sandstone described in any one of claims 1 to 7 are implemented.

Images