Data acquisition method of rock fracture acoustic emission signal identification model

Abstract

The invention provides a data acquisition method of a rock fracture acoustic emission signal identification model. The data acquisition method comprises the following steps: acquisition of a pure rock fracture signal, acquisition of a noise signal, data preprocessing and generation of a synthetic monitoring signal. According to the data acquisition method of the rock fracture acoustic emission signal identification model provided by the invention, a multi-level signal superposition method is provided by systematically combining a physical simulation experiment, noise signal acquisition and a signal synthesis technology, and the method can generate representative and diversified training data, so that the data acquisition accuracy is improved. The method can significantly improve the recognition precision of a deep learning model on rock fracture signals in a complex noise environment, enhances the generalization performance of the model, and provides a solid technical support for rock burst prediction and prevention and control.

Description

Technical Field

[0001] The invention relates to the technical field of rock burst monitoring and prediction, and in particular to a data acquisition method for a rock fracture acoustic emission signal recognition model. Background Art

[0002] Rockburst is a major safety hazard in mines and underground projects, especially in my country, where rockbursts occur frequently in coal mines and deep mines, and the frequency of rockbursts continues to increase with the increase in mining depth. This poses a serious threat to personnel safety and project progress. Therefore, it is particularly urgent to study effective rockburst prevention and control measures.

[0003] Accurately predicting the occurrence time of rockburst is the key to preventing and controlling rockbursts. It can help take protective measures on site in advance and reduce disaster losses. The essence of rockburst is the interaction between the energy accumulated by mining disturbance and the rock medium. By quantitatively analyzing the process of rockburst initiation, explosion preparation and occurrence, the accuracy of prediction can be improved. However, due to the complexity of rockburst mechanism and the variability of main controlling factors, existing methods are difficult to fully meet actual needs. Therefore, exploring new quantitative analysis methods and obtaining accurate rockburst time prediction technology is the focus of current research.

[0004] With the advancement of data storage, transmission and processing technology, data-driven methods have become the "fourth paradigm" after experiments, theories and calculations. Machine learning, as a data-driven modeling method, has been widely used in various fields by extracting patterns from known data and predicting unknown situations. Deep learning models such as convolutional neural networks, recurrent neural networks and generative adversarial networks can directly process raw data, use their deep network structures to extract data features step by step, and have powerful data mining capabilities. However, the inventors of this application have found through research that the training data currently used for deep learning models lacks representativeness and diversity, and the recognition accuracy and generalization performance of rock fracture signals in complex noise environments are low. Summary of the invention

[0005] In view of the technical problems that the existing training data for deep learning models lacks representativeness and diversity, and the recognition accuracy and generalization performance of rock fracture signals in complex noise environments are low, the present invention provides a data acquisition method for a rock fracture acoustic emission signal recognition model.

[0006] In order to solve the above technical problems, the present invention adopts the following technical solutions:

[0007] A method for acquiring data of a rock fracture acoustic emission signal recognition model comprises the following steps:

[0008] S1. Acquisition of pure rock fracture signal:

[0009] S11. Experimental platform construction: Use TTG rock true triaxial test system and PCI-2 acoustic emission experimental device to build a laboratory rock burst test platform;

[0010] S12. Experimental material selection: Granite, sandstone and coal were selected as experimental materials to conduct physical simulation experiments of strain-type rock burst and disturbance-type rock burst;

[0011] S13, strain-type rockburst simulation experiment: First, three-dimensional synchronous loading is performed to the minimum principal stress value σ 1 =σ 2 =σ 3 , then keep σ 3 Constant, σ 1 and σ 2 Two-way simultaneous loading to the intermediate principal stress value σ 1 =σ 2 , and then keep σ 2 Constant, at σ 1 After loading to the maximum principal stress value in the direction and reaching the designed original rock stress value, suddenly unload σ 3 , until the rock burst occurs; during the experiment, the environmental noise is strictly controlled, and the loading equipment and cooling equipment are noise-reduced;

[0012] S14, disturbance type rockburst simulation experiment: after loading to the initial stress in three directions, suddenly unloading on one side σ 3 , while in σ 1 The disturbance is applied in the three directions until a rock burst occurs; if no rock burst is observed, the initial triaxial stress is restored and σ is increased. 1 The test was repeated with the value of ; in this experiment, the servo control system in the TTG rock true triaxial test system pushed the variable frequency power wave to the cylinder of the true triaxial test machine to simulate the disturbance source;

[0013] S15. Data collection and recording: The rock fracture signals generated during the entire experiment are recorded using the PCI-2 acoustic emission experimental device, and the image data is used to assist in marking the exact time of rockburst, providing accurate data samples for model training;

[0014] S2. Acquisition of noise signal:

[0015] S21. On-site collection: Use an acoustic emission meter and a decibel meter to collect noise signals at the actual project site. During the collection process, a single noise environment close to the noise source should be selected to avoid interference from mixed noise;

[0016] S22. Laboratory reproduction: simulate and reproduce specific noise signals including drilling rig sounds and people talking in the laboratory;

[0017] S3, data preprocessing: performing signal denoising and feature extraction on the pure rock fracture signal obtained in step S1 and the noise signal obtained in step S2;

[0018] S4. Generation of synthetic monitoring signals: Using the Simulink library of MATLAB software, the pure rock fracture signal and the composite environmental noise signal are superimposed to generate a synthetic monitoring signal; by adjusting the noise intensity and frequency, a variety of monitoring signal samples are generated to improve the generalization performance of the deep learning model, including:

[0019] S41, superposition of strain-type rock burst signal and noise signal collected on site: superposition of rock fracture signal obtained in strain-type rock burst simulation experiment and noise signal collected on site;

[0020] S42, superposition of disturbance type rock burst signal and laboratory reproduced noise signal: superposition of rock fracture signal obtained in disturbance type rock burst simulation experiment and laboratory reproduced noise signal;

[0021] S43, synthesizing the superimposed signals obtained in steps S41 and S42 respectively to generate a synthetic monitoring signal, and when generating the synthetic monitoring signal, by adjusting the intensity coefficient and frequency parameters of the noise signal, different noise environments can be simulated, thereby generating diversified monitoring signal samples.

[0022] Furthermore, the power wave in step S14 is a sine wave, a square wave or a composite wave.

[0023] Further, the signal denoising in step S3 specifically includes:

[0024] Bandpass filtering: Apply bandpass filters to the collected pure rock fracture signals and noise signals. According to the frequency characteristics of the pure rock fracture signals, set an appropriate frequency range to remove noise beyond the target frequency band and retain the key components in the signal;

[0025] Wavelet transform denoising: The signal after bandpass filtering is subjected to wavelet transform, decomposed into sub-signals of different frequencies, and threshold processing is applied to remove high-frequency noise and retain the main signal components.

[0026] Further, the feature extraction in step S3 specifically includes:

[0027] Time domain feature extraction: Extract time domain features including signal peak, duration and energy from the denoised signal. These time domain features are used to describe the basic time domain properties of the signal.

[0028] Frequency domain feature extraction: Perform Fourier transform on the denoised signal to extract frequency domain features including main frequency and bandwidth. These frequency domain features are used to reveal the frequency characteristics of the signal.

[0029] Furthermore, the mathematical expression of the superposition of the strain-type rockburst signal and the noise signal collected on site in step S41 is:

[0030] S syn1 (t) = S strain (t)+α 1 ·S site_noise (t)

[0031] Among them, S syn1 (t) is the superposition signal of strain-type rockburst signal and field-collected noise signal, S strain (t) is the strain type rockburst signal, S site_noise (t) is the noise signal collected on site, α 1 Collect noise signal strength coefficient on site.

[0032] Furthermore, the mathematical expression of the superposition of the disturbance type rock burst signal and the laboratory reproduction noise signal in step S42 is:

[0033] S syn2 (t) = S disturbance (t)+α 2 ·S lib_noise (t)

[0034] Among them, S syn2 (t) is the superposition signal of the disturbance type rock burst signal and the laboratory reproduced noise signal, S disturbance (t) is the disturbance type rock burst signal, S lib_noise (t) is the laboratory reproduction noise signal, α 2 The signal strength coefficient for laboratory reproduction noise.

[0035] Further, the mathematical expression of the synthetic monitoring signal finally generated in step S43 is:

[0036] S final (t) = β 1 ·S syn1 (t)+β 2 ·S syn2 (t)

[0037] Among them, S final (t) is the final generated synthetic monitoring signal; β 1 and β 2 is the adjusted weight coefficient used to balance the superimposed signal S syn1 (t) and S syn2 (t) contribution.

[0038] Compared with the prior art, the data acquisition method of the rock fracture acoustic emission signal recognition model provided by the present invention systematically combines physical simulation experiments, noise signal acquisition and signal synthesis technology, and proposes a multi-level signal superposition method. This method can generate representative and diverse training data, which can significantly improve the recognition accuracy of the deep learning model for rock fracture signals in complex noise environments, enhance the generalization performance of the model, and provide solid technical support for rock burst prediction and prevention. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 It is a flow chart of a data acquisition method for a rock fracture acoustic emission signal recognition model provided by the present invention.

[0040] Figure 2 It is a schematic diagram of a strain-type rockburst physical simulation experimental device and a loading and unloading path provided by the present invention.

[0041] Figure 3 It is a schematic diagram of a disturbance type rock burst physical simulation experimental device and a loading and unloading path provided by the present invention. DETAILED DESCRIPTION

[0042] In order to make the technical means, creative features, objectives and effects achieved by the present invention easy to understand, the present invention is further explained below with reference to specific diagrams.

[0043] Please refer to Figure 1 As shown, the present invention provides a data acquisition method for a rock fracture acoustic emission signal recognition model, comprising the following steps:

[0044] S1. Acquisition of pure rock fracture signal:

[0045] S11. Experimental platform construction: A laboratory rock burst test platform is constructed using the TTG rock true triaxial test system and the PCI-2 acoustic emission experimental device; wherein the true triaxial test is to place rock or soil samples on three mutually perpendicular principal axes and apply corresponding loads to simulate the real stress state of underground rock and soil, wherein the three principal axes are the horizontal x-axis, the y-axis perpendicular to the x-axis, and the z-axis perpendicular to the xy plane. The specific structural composition and working principle of the TTG rock true triaxial test system and the PCI-2 acoustic emission experimental device are well known to those skilled in the art, and thus will not be described in detail here;

[0046] S12. Experimental material selection: Granite, sandstone and coal were selected as experimental materials to conduct physical simulation experiments of strain-type rock burst and disturbance-type rock burst;

[0047] S13, strain-type rockburst simulation experiment: First, three-dimensional synchronous loading is performed to the minimum principal stress value σ 1 =σ 2 =σ3 , then keep σ 3 Constant, σ 1 and σ 2 Two-way simultaneous loading to the intermediate principal stress value σ 1 =σ 2 , and then keep σ 2 Constant, at σ 1 After loading to the maximum principal stress value in the direction and reaching the designed original rock stress value, suddenly unload σ 3 Until the rock burst occurs, please refer to the specific physical simulation experimental device and loading and unloading path. Figure 2 As shown; During the experiment, the environmental noise was strictly controlled, and the loading equipment and cooling equipment were subjected to noise reduction treatment;

[0048] S14, disturbance type rockburst simulation experiment: after loading to the initial stress in three directions, suddenly unloading on one side σ 3 , while in σ 1 The disturbance is applied in the three directions until a rock burst occurs; if no rock burst is observed, the initial triaxial stress is restored and σ is increased. 1 For the specific physical simulation experimental device and loading and unloading path, please refer to Figure 3 As shown; During the experiment, the servo control system in the TTG rock true triaxial test system pushes variable frequency power waves such as sine waves, square waves or composite waves to the true triaxial test machine cylinder to simulate the disturbance source;

[0049] S15. Data collection and recording: The rock fracture signals generated during the entire experiment are recorded using the PCI-2 acoustic emission experimental device, and the image data is used to assist in marking the exact time of rockburst, providing accurate data samples for model training;

[0050] S2. Acquisition of noise signal:

[0051] S21. On-site collection: Use an acoustic emission meter and a decibel meter to collect noise signals at the actual project site. During the collection process, a single noise environment close to the noise source should be selected to avoid interference from mixed noise;

[0052] S22, laboratory reproduction: simulate and reproduce specific noise signals including drilling rig sound and personnel conversation sound in the laboratory; the specific noise signal can be achieved by making noise through laboratory equipment and collecting its acoustic emission signal;

[0053] S3, data preprocessing: performing signal denoising and feature extraction on the pure rock fracture signal obtained in step S1 and the noise signal obtained in step S2;

[0054] S4. Generation of synthetic monitoring signals: Using the Simulink library of MATLAB software, the pure rock fracture signal and the composite environmental noise signal are superimposed to generate a synthetic monitoring signal; by adjusting the noise intensity and frequency, a variety of monitoring signal samples are generated to improve the generalization performance of the deep learning model, including:

[0055] S41, superposition of strain-type rock burst signal and noise signal collected on site: superposition of rock fracture signal obtained in strain-type rock burst simulation experiment and noise signal collected on site;

[0056] S42, superposition of disturbance type rock burst signal and laboratory reproduced noise signal: superposition of rock fracture signal obtained in disturbance type rock burst simulation experiment and laboratory reproduced noise signal;

[0057] S43, synthesizing the superimposed signals obtained in steps S41 and S42 respectively to generate a synthetic monitoring signal, and when generating the synthetic monitoring signal, by adjusting the intensity coefficient and frequency parameters of the noise signal, different noise environments can be simulated, thereby generating diversified monitoring signal samples.

[0058] In order to simulate the complex environmental noise conditions in actual engineering, the present invention proposes two signal superposition schemes as described in steps S41 and S42.

[0059] As a specific embodiment, the signal denoising in step S3 specifically includes:

[0060] Bandpass filtering: Apply bandpass filters to the collected pure rock fracture signals and noise signals. According to the frequency characteristics of the pure rock fracture signals, set an appropriate frequency range to remove noise beyond the target frequency band and retain the key components in the signal;

[0061] Wavelet transform denoising: The signal after bandpass filtering is subjected to wavelet transform, decomposed into sub-signals of different frequencies, and threshold processing is applied to remove high-frequency noise and retain the main signal components.

[0062] As a specific embodiment, the feature extraction in step S3 specifically includes:

[0063] Time domain feature extraction: Extract time domain features including signal peak, duration and energy from the denoised signal. These time domain features are used to describe the basic time domain properties of the signal.

[0064] Frequency domain feature extraction: Perform Fourier transform on the denoised signal to extract frequency domain features including main frequency and bandwidth. These frequency domain features are used to reveal the frequency characteristics of the signal.

[0065] As a specific embodiment, the mathematical expression of the superposition of the strain-type rockburst signal and the field-collected noise signal in step S41 is:

[0066] S syn1 (t) = S strain (t)+α 1 ·S site_noise (t)

[0067] Among them, S syn1 (t) is the superposition signal of strain-type rockburst signal and field-collected noise signal, S strain (t) is the strain type rockburst signal, S site_noise (t) is the noise signal collected on site, α 1 Collect noise signal strength coefficient on site.

[0068] As a specific embodiment, the mathematical expression of the superposition of the disturbance type rock burst signal and the laboratory reproduction noise signal in step S42 is:

[0069] S syn2 (t) = S disturbance (t)+α 2 ·S lib_noise (t)

[0070] Among them, S syn2 (t) is the superposition signal of the disturbance type rock burst signal and the laboratory reproduced noise signal, S disturbance (t) is the disturbance type rock burst signal, S lib_noise (t) is the laboratory reproduction noise signal, α 2 The signal strength coefficient for laboratory reproduction noise.

[0071] As a specific embodiment, the noise signal strength coefficient α 1 and α 2 The value range of α is usually 0.1 to 0.5, and the specific value is adjusted according to the signal-to-noise ratio (SNR) in the actual application scenario. 1 and α 2 It can be appropriately increased to 0.5-0.8 to ensure that the superimposed signal can truly reflect the degree of interference of field noise on rock fracture signals.

[0072] As a specific embodiment, the mathematical expression of the synthetic monitoring signal finally generated in step S43 is:

[0073] S final (t) = β 1 ·S syn1 (t)+β 2 ·S syn2 (t)

[0074] Among them, Sfinal (t) is the final generated synthetic monitoring signal; β 1 and β 2 is the adjusted weight coefficient used to balance the superimposed signal S syn1 (t) and S syn2 (t) contribution.

[0075] During model training, the weight coefficient β 1 and β 2 The value range of β is usually between 0.2 and 0.8, and is adjusted according to the importance of different tasks or features. In some specific scenarios, such as when a task is particularly critical to the performance of the model, β 1 and β 2 The value of can be set greater than 1 to enhance its relative weight in the loss function. Through experimental verification, the β that makes the model perform best is finally determined. 1 and β 2 value.

[0076] Compared with the prior art, the data acquisition method of the rock fracture acoustic emission signal recognition model provided by the present invention systematically combines physical simulation experiments, noise signal acquisition and signal synthesis technology, and proposes a multi-level signal superposition method. This method can generate representative and diverse training data, which can significantly improve the recognition accuracy of the deep learning model for rock fracture signals in complex noise environments, enhance the generalization performance of the model, and provide solid technical support for rock burst prediction and prevention.

[0077] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the purpose and scope of the technical solution of the present invention, which should be included in the scope of the claims of the present invention.

Claims

1. A data acquisition method for a rock fracture acoustic emission signal recognition model, characterized in that: The following steps are involved: S1. Acquisition of pure rock fracture signal: S11. Experimental platform construction: Use TTG rock true triaxial test system and PCI-2 acoustic emission experimental device to build a laboratory rock burst test platform; S12. Experimental material selection: Granite, sandstone and coal were selected as experimental materials to conduct physical simulation experiments of strain-type rock burst and disturbance-type rock burst; S13, strain-type rockburst simulation experiment: first, load synchronously in three directions to the minimum principal stress value σ1=σ2=σ3, then keep σ3 constant, load synchronously in two directions σ1 and σ2 to the intermediate principal stress value σ1=σ2, then keep σ2 constant, load to the maximum principal stress value in the σ1 direction, and after reaching the designed original rock stress value, suddenly unload σ3 on one side until the rockburst occurs; during this experiment, strictly control the environmental noise, and perform noise reduction treatment on the loading equipment and cooling equipment; S14, disturbance type rock burst simulation experiment: after loading to the initial triaxial stress, suddenly unload σ3 on one side, and apply disturbance in the σ1 direction until rock burst occurs; if rock burst is not observed, restore the initial triaxial stress, increase the value of σ1, and repeat the test; in this experiment, the servo control system in the TTG rock true triaxial test system pushes a variable frequency power wave to the true triaxial test machine cylinder to simulate the disturbance source; S15. Data collection and recording: The rock fracture signals generated during the entire experiment are recorded using the PCI-2 acoustic emission experimental device, and the image data is used to assist in marking the exact time of rockburst, providing accurate data samples for model training; S2. Acquisition of noise signal: S21. On-site collection: Use an acoustic emission meter and a decibel meter to collect noise signals at the actual project site. During the collection process, a single noise environment close to the noise source should be selected to avoid interference from mixed noise; S22. Laboratory reproduction: simulate and reproduce specific noise signals including drilling rig sounds and people talking in the laboratory; S3, data preprocessing: performing signal denoising and feature extraction on the pure rock fracture signal obtained in step S1 and the noise signal obtained in step S2; S4. Generation of synthetic monitoring signals: Using the Simulink library of MATLAB software, pure rock fracture signals and composite environmental noise signals are superimposed to generate synthetic monitoring signals; By adjusting the noise intensity and frequency, we can generate diverse monitoring signal samples and improve the generalization performance of deep learning models. Specifically, we can: S41, superposition of strain-type rock burst signal and noise signal collected on site: superposition of rock fracture signal obtained in strain-type rock burst simulation experiment and noise signal collected on site; S42, superposition of disturbance type rock burst signal and laboratory reproduced noise signal: superposition of rock fracture signal obtained in disturbance type rock burst simulation experiment and laboratory reproduced noise signal; S43, synthesizing the superimposed signals obtained in steps S41 and S42 respectively to generate a synthetic monitoring signal, and when generating the synthetic monitoring signal, by adjusting the intensity coefficient and frequency parameters of the noise signal, different noise environments can be simulated, thereby generating diversified monitoring signal samples.

2. The data acquisition method of the rock fracture acoustic emission signal recognition model according to claim 1 is characterized in that: The power wave in step S14 is a sine wave, a square wave or a composite wave.

3. The data acquisition method for rock fracture acoustic emission signal recognition model according to claim 1 is characterized in that: The signal denoising in step S3 specifically includes: Bandpass filtering: Apply bandpass filters to the collected pure rock fracture signals and noise signals. According to the frequency characteristics of the pure rock fracture signals, set an appropriate frequency range to remove noise beyond the target frequency band and retain the key components in the signal; Wavelet transform denoising: The signal after bandpass filtering is subjected to wavelet transform, decomposed into sub-signals of different frequencies, and threshold processing is applied to remove high-frequency noise and retain the main signal components.

4. The data acquisition method for rock fracture acoustic emission signal recognition model according to claim 1 is characterized in that: The feature extraction in step S3 specifically includes: Time domain feature extraction: Extract time domain features including signal peak, duration and energy from the denoised signal. These time domain features are used to describe the basic time domain properties of the signal. Frequency domain feature extraction: Perform Fourier transform on the denoised signal to extract frequency domain features including main frequency and bandwidth. These frequency domain features are used to reveal the frequency characteristics of the signal.

5. The data acquisition method for rock fracture acoustic emission signal recognition model according to claim 1 is characterized in that: The mathematical expression of the superposition of the strain-type rockburst signal and the noise signal collected on site in step S41 is: S syn1 (t)=S strain (t)+α1·S site_noise (t) Among them, S syn1 (t) is the superposition signal of strain-type rockburst signal and field-collected noise signal, S strain (t) is the strain type rockburst signal, S site_noise (t) is the noise signal collected on site, and α1 is the intensity coefficient of the noise signal collected on site.

6. The data acquisition method for rock fracture acoustic emission signal recognition model according to claim 5 is characterized in that: The mathematical expression of the superposition of the disturbance type rock burst signal and the laboratory reproduction noise signal in step S42 is: S syn2 (t)=S disturbance (t)+α2·S lib_noise (t) Among them, S syn2 (t) is the superposition signal of the disturbance type rock burst signal and the laboratory reproduced noise signal, S disturbance (t) is the disturbance type rock burst signal, S lib_noise (t) is the laboratory reproduction noise signal, and α2 is the laboratory reproduction noise signal intensity coefficient.

7. The data acquisition method for rock fracture acoustic emission signal recognition model according to claim 6 is characterized in that: The mathematical expression of the synthetic monitoring signal finally generated in step S43 is: S final (t)=β1·S syn1 (t)+β2·S syn2 (t) Among them, S final (t) is the final generated synthetic monitoring signal; β1 and β2 are the adjusted weight coefficients used to balance the superposition signal S syn1 (t) and S syn2 (t) contribution.

Images