Fault slip early warning method and system based on acoustic emission waveform recognition

Abstract

The present invention relates to a fault slip early warning method and system based on acoustic emission waveform recognition. The method obtains the original acoustic emission waveform generated by the rock, and uses a trained acoustic emission waveform recognition model to predict the probability that the original acoustic emission waveform belongs to each type. A certain type of waveform is extracted, and when at least one of the total value, maximum value, minimum value or differential value of a set parameter X exceeds a threshold, an early warning of the risk of fault slip is issued. When training the acoustic emission waveform recognition model, based on the features in the original acoustic emission waveform, spectral data corresponding to the original acoustic emission waveform is spliced ​​in the layer dimension as input to subsequent layers, and high-order features are output after feature extraction and downsampling. The probability of the original acoustic emission waveform belonging to each type is obtained through classification in a fully connected layer.

Description

Technical Field

[0001] The present invention relates to the field of geotechnical engineering technology, and in particular to a fault slip early warning method and system based on acoustic emission waveform recognition. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] Acoustic emission (AE) refers to the phenomenon in which rock deforms internally under stress, fractures, and rapidly releases strain energy, generating stress waves. Stress waves generated at the fracture site propagate through the rock material and ultimately reach the rock surface, causing surface displacements that can be detected by AE sensors, allowing the AE signal to be collected and recorded by the instrument. AE has long been an important tool for indoor rock testing and on-site engineering monitoring. Existing technologies address the problem of extracting AE precursor information for intact rock failure, proposing early warning indicators, primarily of two types.

[0004] One is to perform statistical analysis on the time domain data of acoustic emission signals, and obtain precursor information before failure through single time domain parameters such as ring count, energy, amplitude, or secondary statistics based on time domain parameters such as cumulative ring count, cumulative energy, S value, b value, etc., which is used as an early warning indicator of rock failure;

[0005] Another type of method involves performing time-frequency transformations (Fourier transform, wavelet transform, etc.) on waveforms (short-term waveforms or waveform streams exceeding a threshold) to obtain frequency domain information of the acoustic emission signal. This method then provides early warning of rock damage based on changes in frequency domain parameters such as the dominant frequency, frequency domain amplitude, and centroid frequency. After the waveform undergoes time-frequency transformation, the frequency domain information is more discriminative, and therefore many early warning indicators are developed based on frequency domain data.

[0006] At present, most early warning indicators still focus on the characteristics of acoustic emission signals during the fracture process of intact rocks, and extract precursor information based on the time-frequency parameters of intact rock material damage. However, for rock masses with discontinuities such as faults, the slip failure process not only involves the initiation and propagation of tensile shear cracks caused by the fracture of rock materials, but also includes damage and fracture on discontinuities such as elastic contact of micro-asperities on the fault surface, fault creep or stick-slip. The time-frequency information of acoustic emission signals generated by different failure mechanisms will also be different. It is impossible to fully represent the rich spectral information through only a single fixed feature. Therefore, the early warning indicators of intact rocks cannot be fully applicable to the extraction and early warning of precursor information of rock fault slip failure. Summary of the Invention

[0007] In order to solve the technical problems existing in the above-mentioned background technology, the present invention provides a fault slip early warning method and system based on acoustic emission waveform recognition, which can consider the complex time-frequency information of acoustic emission, discriminately select waveform signals, and realize the acquisition of precursor information of rock damage containing discontinuities, thereby providing assistance for rock damage characteristic analysis and real-time fault slip early warning.

[0008] In order to achieve the above object, the present invention adopts the following technical solutions:

[0009] A first aspect of the present invention provides a fault slip early warning method based on acoustic emission waveform recognition, comprising the following steps:

[0010] Obtain the original acoustic emission waveform generated by the rock, and use the trained acoustic emission waveform recognition model to predict the probability that the original acoustic emission waveform belongs to each type;

[0011] Extract a certain type of waveform and issue a warning of fault slip risk when at least one of the total value, maximum value, minimum value, or differential value of the set parameter X exceeds a threshold;

[0012] Among them, when training the acoustic emission waveform recognition model, based on the features in the original acoustic emission waveform, the spectral data corresponding to the original acoustic emission waveform is spliced ​​in the layer dimension as the input of the subsequent layer, and high-order features are output after feature extraction and downsampling. The probability of the original acoustic emission waveform belonging to each type is obtained through classification in the fully connected layer.

[0013] The acoustic emission waveform recognition model is trained using the existing acoustic emission waveform data after preprocessing. The preprocessing process is as follows:

[0014] Select waveforms for training and testing from existing acoustic emission waveforms;

[0015] Obtain waveform spectrum based on discrete Fourier transform and fast Fourier transform;

[0016] The selected waveforms are classified into set categories according to the spectral characteristics, and an acoustic emission waveform recognition database is established.

[0017] The acoustic emission waveform recognition model takes the original acoustic emission waveform as the first input, performs convolution and pooling processing, and obtains the waveform time domain feature map;

[0018] The acoustic emission waveform spectrum is used as the second input and is spliced ​​with the obtained waveform time domain feature map to obtain the waveform time-frequency domain feature map;

[0019] Multiple convolution and pooling operations are performed on the waveform time-frequency domain feature map, and the waveform type probability prediction value is obtained through the fully connected layer and softmax function, that is, the probability that the original acoustic emission waveform belongs to each type.

[0020] During real-time recognition, the type corresponding to the maximum probability is the automatic recognition type of the original acoustic emission waveform.

[0021] The training process of the acoustic emission waveform recognition model is as follows:

[0022] Based on the data in the acoustic emission waveform recognition database, the data is divided into training set and test set according to the set ratio, and input into the acoustic emission waveform recognition model to obtain the probability prediction value of each waveform corresponding to each type;

[0023] Obtain the error between the probability prediction value and the true value, and update the size of each weight in the acoustic emission waveform recognition model according to the error value;

[0024] Repeatedly input the training waveform and update the weights in the acoustic emission waveform recognition model until the error meets the set conditions;

[0025] Input the test set waveform into the trained acoustic emission waveform recognition model to obtain the predicted type of each test set waveform;

[0026] The predicted type of the test set waveform is compared with the true type until the accuracy meets the requirements, and the trained acoustic emission waveform recognition model is obtained.

[0027] A second aspect of the present invention provides a system for implementing the above method, comprising:

[0028] The waveform type recognition unit is configured to: obtain the original acoustic emission waveform generated by the rock, and use the trained acoustic emission waveform recognition model to predict the probability that the original acoustic emission waveform belongs to each type;

[0029] The fault slip warning unit is configured to: extract a certain type of waveform and issue a warning indicating the presence of a fault slip risk when at least one of the total value, maximum value, minimum value, or differential value of a set parameter X exceeds a threshold;

[0030] Among them, when training the acoustic emission waveform recognition model, based on the features in the original acoustic emission waveform, the spectral data corresponding to the original acoustic emission waveform is spliced ​​in the layer dimension as the input of the subsequent layer, and high-order features are output after feature extraction and downsampling. The probability of the original acoustic emission waveform belonging to each type is obtained through classification in the fully connected layer.

[0031] A third aspect of the present invention provides a computer-readable storage medium.

[0032] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the fault slip early warning method based on acoustic emission waveform recognition as described above.

[0033] A fourth aspect of the present invention provides a computer device.

[0034] A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the steps of the fault slip early warning method based on acoustic emission waveform recognition as described above are implemented.

[0035] Compared with the existing technology, one or more of the above technical solutions have the following beneficial effects:

[0036] At the same time, the original waveform and the spectrum corresponding to the waveform are used as input, and the characteristics of the original waveform and the spectrum data are spliced ​​in the layer dimension as the input of the subsequent layer, so as to consider the frequency domain information of the complex signal of the rock mass, and finally obtain the probability corresponding to the waveform type, so as to realize the classification and identification of rock damage changes, thereby obtaining the precursor information of rock fracture in a targeted manner and realizing real-time early warning of fault slip. Compared with traditional methods, it can consider the complex time-frequency information of acoustic emission, discriminately select waveform signals, and realize the acquisition of precursor information of rock damage containing discontinuities, which provides assistance for rock damage feature analysis and real-time early warning of fault slip. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] The accompanying drawings, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0038] Figure 1 is a schematic diagram of a fault slip early warning process based on acoustic emission waveform recognition provided by one or more embodiments of the present invention;

[0039] Figure 2 is a flow chart of a waveform recognition model training method provided by one or more embodiments of the present invention;

[0040] Figure 3 is a schematic diagram of the waveform recognition model structure provided by one or more embodiments of the present invention;

[0041] Figure 4 is a schematic diagram showing how the average error changes with the number of cycles during the waveform recognition model training process provided by one or more embodiments of the present invention;

[0042] Figure 5 Schematic diagram of loading a waveform recognition model indoor test specimen (including discontinuities) provided by one or more embodiments of the present invention;

[0043] Figure 6(a)-Figure 6(b) Schematic diagrams of the time domain and (fast Fourier transform) frequency domain of a Type I waveform provided by one or more embodiments of the present invention;

[0044] Figure 6(c)-Figure 6(d)Schematic diagrams of the time domain and (fast Fourier transform) frequency domain of a Type II waveform provided by one or more embodiments of the present invention;

[0045] Figure 6(e)-Figure 6(f) Schematic diagrams of the time domain and (fast Fourier transform) frequency domain of a Class III waveform provided by one or more embodiments of the present invention;

[0046] Figure 7(a)-Figure 7(c) Schematic diagrams of FFT amplitude distribution of three types of waveforms provided by one or more embodiments of the present invention;

[0047] Figure 8(a)-Figure 8(b) Schematic diagrams comparing the warning effects of the b value before and after processing provided by one or more embodiments of the present invention. DETAILED DESCRIPTION

[0048] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0049] It should be noted that the following detailed descriptions are exemplary and intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which the present invention belongs.

[0050] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, they indicate the presence of features, steps, operations, devices, components and / or combinations thereof.

[0051] As described in the background technology, for rock masses with discontinuities such as faults, the slip failure process not only involves the initiation and expansion of tensile shear cracks caused by the rupture of rock materials, but also includes damage and fracture on discontinuities such as elastic contact of micro-asperities on the fault surface, fault creep or stick-slip. The time-frequency information of acoustic emission signals generated by different failure mechanisms will also be different. It is impossible to fully represent the rich spectral information through only a single fixed feature. Therefore, the early warning indicators of intact rocks cannot be fully applicable to the extraction and early warning of precursor information of rock fault slip failure.

[0052] Therefore, the following embodiments provide a fault slip warning method and system based on acoustic emission waveform recognition, which can consider the complex time-frequency information of acoustic emissions, discriminately select waveform signals, and obtain precursor information of rock mass damage containing discontinuities, thereby providing assistance for rock damage feature analysis and real-time fault slip warning.

[0053] Example 1:

[0054] The fault slip early warning method based on acoustic emission waveform recognition includes the following steps:

[0055] Obtain the original acoustic emission waveform generated by the rock, and use the trained acoustic emission waveform recognition model to predict the probability that the original acoustic emission waveform belongs to each type;

[0056] Extract a certain type of waveform and issue a warning of fault slip risk when at least one of the total value, maximum value, minimum value, or differential value of the set parameter X exceeds a threshold;

[0057] Among them, when training the acoustic emission waveform recognition model, based on the features in the original acoustic emission waveform, the spectral data corresponding to the original acoustic emission waveform is spliced ​​in the layer dimension as the input of the subsequent layer, and high-order features are output after feature extraction and downsampling. The probability of the original acoustic emission waveform belonging to each type is obtained through classification in the fully connected layer.

[0058] Specifically:

[0059] like Figure 1-2 As shown in the figure, it includes four parts: acoustic emission waveform preprocessing, waveform recognition model training, real-time recognition of acoustic emission waveforms for fault monitoring, and fault slip warning.

[0060] Acoustic emission waveform preprocessing:

[0061] S1.1 Manually select waveforms for training and testing from existing acoustic emission waveforms;

[0062] S1.2 Based on the discrete Fourier transform (DFT) formula, the selected waveform is subjected to FFT (Fast Fourier Transform) to obtain the waveform spectrum, as shown in the following formula:

[0063]

[0064] in, N ≥ M, where N is the length of the DFT transform interval, x(n) is a finite sequence of length M, and X(k) is the discrete Fourier transform of x(n). In this example, M = 2046.

[0065] S1.3 Divide the selected waveform into a set number of types based on the spectrum characteristics. In this embodiment, the waveforms are divided into three categories: high-frequency waveforms, low-frequency waveforms, and other waveforms. For the time domain and frequency domain information of the three types of waveforms, see Figure 6(a)-Figure 6(f) , Figure 6(a)-Figure 6(b) is a low-frequency waveform, Figure 6(c)-Figure 6(d) High frequency waveform, Figure 6(e)-Figure 6(f) For other waveforms.

[0066] The low-frequency waveform has the following characteristics: the frequencies with a relatively high proportion (more than 80% of the maximum amplitude) are mainly concentrated in the range of 0 to 200 kHz, and the ratio of the main frequency (largest amplitude) to the secondary main frequency (amplitude second only to the main frequency) is greater than 0.5;

[0067] The high-frequency waveform has the following characteristics: the frequencies with a high proportion (more than 80% of the maximum amplitude) are mainly concentrated in the frequency band above 200kHz, and the ratio of the main frequency (largest amplitude) to the secondary main frequency (amplitude second only to the main frequency) is greater than 0.5;

[0068] Other waveforms have the following characteristics: the ratio of the main frequency (largest amplitude) to the secondary frequency (amplitude second only to the main frequency) is less than or equal to 0.5. Through manual selection, a total of 990 typical waveform samples were selected, of which the number of samples of each type was 330.

[0069] S1.4 uses one-hot encoding to encode the labels of the three types of waveforms, that is, the label of the first type of waveform is [1,0,0]; the label of the second type of waveform is [0,1,0]; the label of the third type of waveform is [0,0,1], and an acoustic emission waveform recognition database is established based on the labels.

[0070] Waveform recognition model training: This embodiment adopts an acoustic emission waveform recognition model based on a convolutional neural network, and its structure is as follows:

[0071] S2.1 The original acoustic emission waveform is used as the first input, and convolution and pooling operations are performed to obtain the waveform time domain feature map;

[0072] S2.2 The acoustic emission waveform spectrum is used as the second input and is spliced ​​with the waveform feature map to obtain the waveform time-frequency domain feature map;

[0073] S2.3 performs multiple convolution and pooling operations on the waveform time-frequency domain feature map, and obtains the waveform type probability prediction value through the fully connected layer and softmax function.

[0074] This embodiment is built using a convolutional neural network, and uses the original waveform and the spectrum corresponding to the waveform as input. Considering that the spectrum length is half of the original waveform, the original waveform is first pre-convolved and pooled to extract the features of the original waveform. The obtained features are then concatenated with the spectrum data in the layer dimension as input for subsequent layers.

[0075] In this embodiment, the Figure 3 The structure shown includes: original waveform input layer, pre-convolution layer, pre-pooling layer, spectrum input layer, convolution layer, pooling layer, flattening layer, fully connected layer, and output layer.

[0076] The specific data connection method is: original waveform input layer → pre-convolution layer → pre-pooling layer + spectrum input layer → 1#convolution layer → 1#pooling layer → 2#convolution layer → 2#pooling layer → 3#convolution layer → 3#pooling layer → 4#convolution layer → 4#pooling layer → 5#convolution layer → 5#pooling layer → flattening layer → 1#fully connected layer → 2#fully connected layer → output layer.

[0077] Table 1 is Figure 3 Parameters of the waveform recognition model;

[0078] Table 1: Waveform recognition model parameters;

[0079]

[0080]

[0081] like Figure 3 As shown in Table 1, the waveform input length varies depending on the sampling settings of the acoustic emission system. For one waveform sample in this embodiment, the waveform recognition model first performs pre-convolution and pre-pooling on the original acoustic emission waveform with a length of 2048 sampling points (the number of layers is 1 at this time). At this time, a waveform feature sequence with a length of 1024 sampling points and a layer number of 1 is obtained. Subsequently, it is spliced ​​with the waveform spectrum data with a length of 1024 sampling points and a layer number of 1 to obtain a waveform time-frequency domain feature sequence with a length of 1024 sampling points but a layer number of 2.

[0082] Subsequently, after feature extraction in each convolutional layer and downsampling in each pooling layer, the output feature length becomes shorter and shorter, and the number of layers increases. This indicates that the model gradually extracts the high-order features contained in the waveform and spectrum, and classifies them through the fully connected layer based on these high-order features, and finally obtains a feature value of length 3. After normalization by the softmax function, the three numbers of the output layer Y obtained correspond to the probability predicted by the model that the waveform belongs to each type.

[0083] The Softmax function is shown below:

[0084]

[0085] Where y k ′ is the probability value of the k-th waveform output, k = 1, 2, 3; a k It is the kth eigenvalue output by the 2# fully connected layer.

[0086] This function can convert the eigenvalue a in the range of negative infinity to positive infinity k (In the formula, k is a subscript) is scaled to three numbers ranging from 0 to 1, and the sum is 1, which is used to represent the probability of each type.

[0087] For example, if a waveform is input into a waveform recognition model and the output is [0.1, 0.3, 0.6], it can be understood as "the waveform recognition model believes that the waveform has a 10% chance of being type I, a 30% chance of being type II, and a 60% chance of being type III."

[0088] During real-time recognition, the type corresponding to the maximum probability is taken as the automatic recognition type of the waveform.

[0089] In this embodiment, the ReLU function is used as the activation function to introduce nonlinear terms into the waveform recognition model, as shown in the following formula:

[0090]

[0091] The steps for training the acoustic emission waveform recognition model are as follows:

[0092] S2.3.1 Model Initialization. Randomly assign values ​​to the convolution kernel parameters of the convolutional layer and the parameters of the hidden units of the fully connected layer in the waveform recognition model; randomly divide the acoustic emission waveform recognition database into a training set and a test set according to a certain ratio; in this embodiment, the ratio is 7:3, that is, 700 waveforms are used for training and 290 waveforms are used for testing;

[0093] S2.3.2 Input the acoustic emission waveforms in the training set into the acoustic emission waveform recognition model to obtain the probability prediction value of each waveform corresponding to each type;

[0094] S2.3.3 Calculate the error between the probability prediction value and the true value, and update the weights in the acoustic emission waveform recognition model based on the error value;

[0095] Specifically, the cross entropy function is used to measure the error between the model output value and the true value, and according to the size of the error value, the weights in each convolutional layer and fully connected layer in the acoustic emission waveform recognition model are updated based on the gradient.

[0096] The cross entropy function is shown below:

[0097]

[0098] Where y k is the label value of the k-th category of the waveform, k = 1, 2, 3.

[0099] S2.3.4 Repeatedly input the training waveform and update the weights in the acoustic emission waveform recognition model until the error meets the set conditions;

[0100] S2.3.5 Input the test set waveform into the trained acoustic emission waveform recognition model and obtain the predicted type of each test set waveform;

[0101] In this embodiment, the training completion condition is set to 200 training iterations. Figure 4 The figure below shows the average error of training samples as a function of the number of iterations during waveform recognition model training. It can be seen that the error is high when the model is initially randomly set with parameters, but it decreases rapidly as training progresses, ultimately dropping to around 0.007.

[0102] S2.3.6 Input the test set waveforms into the trained acoustic emission waveform recognition model to obtain the predicted type of each test set waveform;

[0103] S2.3.7 Compare the predicted type of the test set waveform with the actual type. If the accuracy rate does not meet the requirements, retrain until the accuracy rate meets the requirements. The training of the waveform recognition model is completed, and a mature acoustic emission waveform recognition model is obtained.

[0104] In this embodiment, the confusion matrix of the test sample recognition results is shown in Table 2;

[0105] Table 2: Confusion matrix of waveform recognition model test results;

[0106]

[0107] It can be found that the recognition accuracy and recall rate of the model for each type of signal are both over 90%; the overall accuracy of the model is over 95%, indicating that the model trained this time is very accurate and can meet the needs of practical applications. k and recall R k Calculate according to the following formula:

[0108]

[0109]

[0110]

[0111] Where N is the total number of samples; TP k is the number of samples of the kth class that are correctly identified; FP is the number of samples of non-kth class that are mistakenly identified as the kth class; FN k is the number of samples of the kth class that are mistakenly identified as non-kth classes.

[0112] Real-time identification of fault monitoring acoustic emission waveforms, including:

[0113] S3.1 Place acoustic emission sensors near the joints of the indoor test or the faults of the field monitoring to monitor the acoustic emission waveform of the rock mass in real time; the rock mass loading method and sensor placement method are as follows: Figure 5 As shown;

[0114] S3.2 inputs the real-time monitored acoustic emission waveform into a well-trained acoustic emission waveform recognition model to obtain the predicted type corresponding to the waveform in real time.

[0115] Fault slip early warning, including:

[0116] S4.1 calculates a warning parameter of the Xth type of waveform for the real-time identification type of acoustic emission waveform; in this embodiment, the FFT (Fast Fourier Transform) amplitude of the three types of waveforms is calculated, such as Figure 7(a)-Figure 7(c) As shown in the figure, the third type of waveform shows a large amount of aggregation and sudden increase before the peak, which has good early warning potential;

[0117] S4.2 When the total value / maximum value / minimum value / difference value of the X parameter exceeds the threshold, it is considered that there is a risk of fault slip and an early warning is issued.

[0118] In this embodiment, the b-value of the type III waveform is used as the warning indicator, and the b-value is calculated as follows:

[0119] lgN=ab×(A dB / 20)

[0120] Among them, b is the physical quantity of the acoustic emission activity level, A dB is the amplitude of the acoustic emission expressed in decibels, A dB =20lgA max , A max is the amplitude of the acoustic emission expressed in microvolts, N is the statistical cumulative frequency of acoustic emissions under the magnitude interval, and a is a constant.

[0121] like Figure 8(a)-Figure 8(b) As shown, when the calculated b value is greater than 10 (90% of the maximum b value), it is considered that there is a risk of fault slip and an early warning is issued.

[0122] It has been verified that the b value calculated by this embodiment reaches a maximum value before the stress peak, and has a better warning effect than that obtained before waveform processing, and can provide a timely fault warning.

[0123] Other indicators of acoustic emission can also be selected as early warning indicators, including S value, RA value-AF value, frequency centroid value, etc., but these early warning indicators are all early warning indicators for intact rock (excluding joints). In this embodiment, the application of b value is the most extensive.

[0124] Other early warning indicators

[0125] (1) S value

[0126]

[0127] Among them, m siis the magnitude of each acoustic emission event, m s is the magnitude of the maximum acoustic emission event amplitude in the statistical window.

[0128] (2) RA value-AF value

[0129]

[0130] Among them, AE count AE is the ring count. dutation is the duration, AE rise is the duration, AE amplitude When used for rock failure early warning, RA and AF are often used together.

[0131] (3) Frequency centroid

[0132]

[0133] Among them, f i and a i They are the frequency spectrum and amplitude spectrum in the frequency domain of the waveform after Fourier transform.

[0134] The above method considers the frequency domain information of complex rock signals and classifies and identifies changes in rock damage, thereby obtaining targeted precursor information of rock fracture and realizing real-time early warning of fault slip. Compared with traditional methods, it can consider the complex time-frequency information of acoustic emission, discriminately select waveform signals, and obtain precursor information of rock damage containing discontinuities, providing assistance for rock damage characteristic analysis and real-time early warning of fault slip.

[0135] Example 2:

[0136] A system for implementing the above method includes:

[0137] The waveform type recognition unit is configured to: obtain the original acoustic emission waveform generated by the rock, and use the trained acoustic emission waveform recognition model to predict the probability that the original acoustic emission waveform belongs to each type;

[0138] The fault slip warning unit is configured to: extract a certain type of waveform and issue a warning indicating the presence of a fault slip risk when at least one of the total value, maximum value, minimum value, or differential value of a set parameter X exceeds a threshold;

[0139] Among them, when training the acoustic emission waveform recognition model, based on the features in the original acoustic emission waveform, the spectral data corresponding to the original acoustic emission waveform is spliced ​​in the layer dimension as the input of the subsequent layer, and high-order features are output after feature extraction and downsampling. The probability of the original acoustic emission waveform belonging to each type is obtained through classification in the fully connected layer.

[0140] Example 3:

[0141] This embodiment provides a computer-readable storage medium having a computer program stored thereon. When the program is executed by a processor, the steps of the fault slip early warning method based on acoustic emission waveform recognition as described in the first embodiment above are implemented.

[0142] Example 4:

[0143] This embodiment provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the steps of the fault slip early warning method based on acoustic emission waveform recognition as described in the first embodiment are implemented.

[0144] The steps or modules involved in Examples 2 to 4 above correspond to those in Example 1. For detailed implementations, please refer to the relevant description of Example 1. The term "computer-readable storage medium" should be understood to mean a single medium or multiple media containing one or more instruction sets; it should also be understood to include any medium capable of storing, encoding, or carrying an instruction set for execution by a processor and causing the processor to perform any of the methods of the present invention.

[0145] The foregoing description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Those skilled in the art will readily appreciate that various modifications and variations of the present invention are possible. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention are intended to be within the scope of protection of the present invention.

Claims

1. A fault slip early warning method based on acoustic emission waveform recognition, characterized in that: The following steps are involved: The original acoustic emission waveform generated by the rock is obtained, and the trained acoustic emission waveform recognition model is used to predict the probability that the original acoustic emission waveform belongs to each type. The waveform is divided into a set number of types according to the spectral characteristics, including high-frequency waveforms, low-frequency waveforms, and other waveforms. Low-frequency waveforms are defined as: frequencies exceeding 80% of the maximum amplitude are between 0 and 200 kHz, and the ratio of the main frequency to the secondary main frequency is greater than 0.5; high-frequency waveforms are defined as: frequencies exceeding 80% of the maximum amplitude are in the frequency range above 200 kHz, and the ratio of the main frequency to the secondary main frequency is greater than 0.5; other waveforms are defined as: the ratio of the main frequency to the secondary main frequency is less than or equal to 0.5; Extract a certain type of waveform and issue a warning of fault slip risk when at least one of the total value, maximum value, minimum value, or differential value of the set parameter X exceeds a threshold; Among them, when training the acoustic emission waveform recognition model, based on the features in the original acoustic emission waveform, the spectral data corresponding to the original acoustic emission waveform is spliced ​​in the layer dimension as the input of the subsequent layer, and high-order features are output after feature extraction and downsampling. The probability of the original acoustic emission waveform belonging to each type is obtained through classification in the fully connected layer; during real-time recognition, the type corresponding to the maximum probability is the automatic recognition type of the original acoustic emission waveform.

2. The fault slip early warning method based on acoustic emission waveform recognition according to claim 1, characterized in that: The acoustic emission waveform recognition model is trained using the existing acoustic emission waveform data after preprocessing. The preprocessing process is as follows: Select waveforms for training and testing from existing acoustic emission waveforms; Obtain waveform spectrum based on discrete Fourier transform and fast Fourier transform; The selected waveforms are classified into set categories according to the spectral characteristics, and an acoustic emission waveform recognition database is established.

3. The fault slip early warning method based on acoustic emission waveform recognition according to claim 1, characterized in that: The acoustic emission waveform recognition model takes the original acoustic emission waveform as the first input, performs convolution and pooling processing, and obtains a waveform time domain feature map.

4. The fault slip early warning method based on acoustic emission waveform recognition according to claim 3, characterized in that: The acoustic emission waveform recognition model takes the acoustic emission waveform spectrum as a second input and splices it with the obtained waveform time domain feature map to obtain the waveform time-frequency domain feature map; Multiple convolution and pooling operations are performed on the waveform time-frequency domain feature map to obtain the waveform type probability prediction value, that is, the probability that the original acoustic emission waveform belongs to each type.

5. The fault slip early warning method based on acoustic emission waveform recognition according to claim 1, characterized in that: The training process of the acoustic emission waveform recognition model includes: Based on the data in the acoustic emission waveform recognition database, the data is divided into training set and test set according to the set ratio, and input into the acoustic emission waveform recognition model to obtain the probability prediction value of each waveform corresponding to each type; Obtain the error between the probability prediction value and the true value, and update the size of each weight in the acoustic emission waveform recognition model according to the error value; The training waveform is repeatedly input and the weights in the acoustic emission waveform recognition model are updated until the error meets the set conditions.

6. The fault slip early warning method based on acoustic emission waveform recognition according to claim 5, characterized in that: The training process of the acoustic emission waveform recognition model also includes: Input the test set waveform into the acoustic emission waveform recognition model to obtain the predicted type of each test set waveform; The predicted type of the test set waveform is compared with the true type until the accuracy meets the requirements, and the trained acoustic emission waveform recognition model is obtained.

7. The fault slip early warning system based on acoustic emission waveform recognition is characterized by: include: The waveform type recognition unit is configured to: obtain the original acoustic emission waveform generated by the rock, and use the trained acoustic emission waveform recognition model to predict the probability that the original acoustic emission waveform belongs to each type; wherein the waveform is divided into a set number of types according to the spectral characteristics, including high-frequency waveforms, low-frequency waveforms, and other waveforms. The low-frequency waveform is: the frequency of more than 80% of the maximum amplitude is between 0 and 200 kHz, and the ratio of the main frequency to the sub-main frequency is greater than 0.5; the high-frequency waveform is: the frequency of more than 80% of the maximum amplitude is in the frequency band above 200 kHz, and the ratio of the main frequency to the sub-main frequency is greater than 0.5; the other waveforms are: the ratio of the main frequency to the sub-main frequency is less than or equal to 0.5; The fault slip warning unit is configured to: extract a certain type of waveform and issue a warning indicating the presence of a fault slip risk when at least one of the total value, maximum value, minimum value, or differential value of a set parameter X exceeds a threshold; Among them, when training the acoustic emission waveform recognition model, based on the features in the original acoustic emission waveform, the spectral data corresponding to the original acoustic emission waveform is spliced ​​in the layer dimension as the input of the subsequent layer, and high-order features are output after feature extraction and downsampling. The probability of the original acoustic emission waveform belonging to each type is obtained through classification by the fully connected layer; during real-time recognition, the type corresponding to the maximum probability is taken as the automatic recognition type of the original acoustic emission waveform.

8. A computer-readable storage medium having a computer program stored thereon, wherein when the program is executed by a processor, the steps of the fault slip early warning method based on acoustic emission waveform recognition according to any one of claims 1 to 6 are implemented.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the steps of the fault slip early warning method based on acoustic emission waveform recognition as described in any one of claims 1 to 6 are implemented.

Images