Concrete stress state identification method based on acoustic emission waveform time-frequency characteristics

By combining acoustic emission technology with machine learning, and utilizing adaptive threshold wavelet denoising, DTW-K-medoids clustering, and dual-branch DCNN neural networks, the problems of flexibility and accuracy in monitoring the performance degradation of concrete structures were solved, and efficient identification of concrete stress state was achieved.

CN119004179APending Publication Date: 2024-11-22TONGJI UNIV
View PDF 0 Cites 3 Cited by

Patent Information

Application Number
CN202411087133.7
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-08
Publication Date
2024-11-22

AI Technical Summary

Technical Problem

Existing concrete structures frequently experience performance degradation and damage during use. Traditional stress monitoring methods suffer from limited monitoring range, short lifespan, high cost, and lack of flexibility, all of which affect structural safety.

Method used

By combining acoustic emission technology with machine learning, the stress state of concrete is identified through the time-frequency characteristics of acoustic emission waveforms. Adaptive threshold wavelet denoising, DTW-K-medoids clustering, fast Fourier transform and variational mode decomposition are used to construct a dual-branch DCNN neural network for feature fusion and recognition.

Benefits of technology

It enables flexible and efficient identification of concrete stress state, providing support for structural safety assessment and improving the flexibility and accuracy of monitoring.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119004179A_ABST
    Figure CN119004179A_ABST
Patent Text Reader

Abstract

The invention provides a concrete stress state identification method based on acoustic emission waveform time-frequency characteristics. The concrete stress state identification method comprises the following steps: carrying out a uniaxial graded loading test on a concrete sample; exciting the concrete test blocks during each stage of loading; collecting a waveform generated by the exciter; de-noising is carried out on the waveform; the de-noised waveform signals obtained through different graded loading are clustered, concrete stress state grades are obtained, and manual marking is carried out; fFT and VMD are carried out on the de-noised waveform signals, frequency domain and time domain signals of the signals are obtained respectively, and a data set for training is obtained; a double-branch DCNN is constructed, frequency domain and time domain features of the waveform signals are learned, and feature fusion is carried out through a global attention mechanism; inputting the waveform data set into the constructed neural network, and carrying out model training; and judging the safety state of the concrete structure by using the trained model. According to the method, the stress state grade corresponding to the concrete structure can be quickly and accurately identified, and support is provided for structural safety evaluation.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention belongs to the field of structural health monitoring, and in particular relates to a method for identifying concrete stress state based on time-frequency characteristics of acoustic emission waveforms. Background Art

[0002] With the acceleration of urbanization and sustained economic growth in my country, the scale of infrastructure construction has continued to expand, and various large and complex industrial and civil buildings have sprung up, among which concrete structures have occupied a dominant position. However, a large number of concrete structures have experienced performance degradation or even damage during use, which has aroused the industry's deep concern about the safety and durability of the structures. There are many reasons for this phenomenon, including the aging of the building itself, steel corrosion, material fatigue, etc. In addition, the loads borne by modern buildings are becoming more and more complex, and the interaction of various direct loads, indirect loads and accidental loads has further aggravated the structural resistance decline and damage accumulation, which in turn has a serious impact on the applicability, safety and durability of the structure. In this context, the importance of concrete stress monitoring technology has become increasingly prominent. The stress state of concrete is one of the key indicators reflecting the health of the structure. By real-time monitoring of the stress distribution and changes of concrete, the actual working status of the structure can be grasped in time, and early warning of potential risks can be provided. However, the current concrete stress state monitoring is achieved by pre-embedded stress gauges in the concrete structure during the construction process. This method has the disadvantages of limited monitoring range, short monitoring life, high cost, and lack of flexibility, and may affect the safety of the structure. Based on this, the present invention combines acoustic emission and machine learning technology to propose a concrete stress state identification method based on the time-frequency characteristics of acoustic emission waveforms, which can be extended to engineering practice. Summary of the invention

[0003] In order to solve the above problems, the present invention provides a method for identifying the stress state of concrete, which can combine acoustic emission technology with machine learning technology, identify the stress state of concrete according to the time-frequency characteristics of acoustic emission waveforms, and can be extended to engineering practice.

[0004] In order to achieve the above object, the present invention provides the following technical solutions:

[0005] A method for identifying concrete stress state based on time-frequency characteristics of acoustic emission waveforms comprises the following steps:

[0006] S1. Conduct uniaxial graded loading test on concrete specimens;

[0007] S2. Use the exciter to excite the concrete test blocks at various levels of loading in S1;

[0008] S3. Use the acoustic emission system to collect the waveform generated by the exciter in S2;

[0009] S4. De-noise the waveform collected by S3 using an adaptive threshold wavelet denoising method;

[0010] S5. Use the DTW-K-medoids clustering method to cluster the denoised waveform signals obtained by different graded loadings obtained in S4, obtain the concrete stress state level, and manually annotate it;

[0011] S6. Perform fast Fourier transform (FFT) and variational mode decomposition (VMD) on the denoised waveform signal after manual annotation in S5 to obtain frequency domain signals and time domain signals, respectively, to obtain a training data set;

[0012] S7. Construct a dual-branch DCNN neural network to learn the frequency domain and time domain features of the waveform signal respectively, and perform feature fusion through the global attention mechanism;

[0013] S8. Input the data set in S6 into the dual-branch DCNN neural network constructed in S7 to perform model training;

[0014] S9. Use the dual-branch DCNN neural network trained in S8 to identify the concrete stress state. When applied, it can identify and determine the safety state of the concrete structure in the application scenario.

[0015] Beneficial effects of the present invention:

[0016] The present invention combines acoustic emission technology with machine learning technology to flexibly and efficiently identify the stress state of concrete based on the time-frequency characteristics of the acoustic emission waveform, and provide support for structural safety evaluation. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 The present invention is a flow chart of a method for identifying concrete stress state based on time-frequency characteristics of acoustic emission waveforms according to an embodiment.

[0018] Figure 2 Schematic diagram of a concrete test block of the embodiment.

[0019] Figure 3 2 is a diagram of the concrete sample graded loading process of the embodiment.

[0020] Figure 4 It is a schematic diagram of the sensor installation and excitation surface of the embodiment.

[0021] Figure 5 is a schematic diagram of an actuator of an embodiment.

[0022] Figure 6 Schematic diagram of the adaptive threshold wavelet denoising effect of the embodiment.

[0023] Figure 7Schematic diagram of the change of the optimal parameter fitness function value of the adaptive threshold wavelet denoising in the embodiment with the number of iterations.

[0024] Figure 8 Schematic diagram of DTW-K-medoids clustering effect of concrete stress state in the embodiment.

[0025] Fig. 9 It is the FFT transform spectrum diagram of the excitation waveform of the embodiment.

[0026] Fig.10 3 is a VMD decomposition time-frequency diagram of the excitation waveform of the embodiment.

[0027] Fig.11 It is a schematic diagram of the accuracy of the concrete stress state recognition test set and training set according to the embodiment as the training cycle changes.

[0028] Fig.12 2 is a dual-branch DCNN neural network structure diagram of an embodiment. DETAILED DESCRIPTION

[0029] The technical solution provided by the present application will be further described below in conjunction with specific embodiments and accompanying drawings. The advantages and features of the present application will become more apparent with the following description.

[0030] Figure 1 A flow chart of a method for identifying the stress state of concrete is shown.

[0031] The method for identifying the stress state of concrete comprises:

[0032] S1. Conduct uniaxial graded loading test on concrete specimens;

[0033] S2. Use the exciter to excite the concrete test blocks at various levels of loading in S1;

[0034] S3. Use the acoustic emission system to collect the waveform generated by the exciter in S2;

[0035] S4. De-noise the waveform collected by S3 using an adaptive threshold wavelet denoising method;

[0036] S5. Use the DTW-K-medoids clustering method to cluster the denoised waveform signals obtained by different graded loadings obtained in S4, obtain the concrete stress state level, and manually annotate it;

[0037] S6. Perform FFT and VMD on the denoised waveform signal after manual annotation in S5 to obtain frequency domain signals and time domain signals respectively, and obtain a training data set;

[0038] S7. Construct a dual-branch DCNN neural network to learn the frequency domain and time domain features of the waveform signal respectively, and perform feature fusion through the global attention mechanism;

[0039] S8. Input the data set in S6 into the dual-branch DCNN neural network constructed in S7 to perform model training;

[0040] S9. Use the dual-branch DCNN neural network trained in S8 to identify the concrete stress state. When applied, it can identify and determine the safety state of the concrete structure in the application scenario.

[0041] The above steps are described in detail below.

[0042] S1. The uniaxial graded loading test on concrete specimens includes:

[0043] Use a uniaxial compressor to compress the concrete specimens at a fixed loading rate of 1 kN / s. The compression level starts from 0 kN and the interval is 30 kN until the concrete block is destroyed. After each loading to the corresponding level, the constant pressure must be maintained for a period of time. The compression levels are: 0, 30, 60, 90, 120, 150, 180, 210, 240, 270, 300, 330, 360, 390, 420, 450, 480 kN, a total of 17 levels. Figure 2 The concrete test block sample of the embodiment is shown in FIG. The selected concrete test block is labeled C40 and has a size of 100×100×100 mm. Figure 3 The concrete loading process of the embodiment is shown: the X-axis is time s, the section is 0 to 3000s. The Y-axis is force kN, the section is 0 to 600kN. It can be clearly seen from the figure that the process of graded loading is divided into 17 levels, and each level will be maintained for a period of time. The maximum level is 480kN.

[0044] S2. When the concrete specimen is subjected to uniaxial graded loading, an exciter is used to excite the concrete specimen at each level of loading. The exciter used is a spring pin. Figure 5 The figure is a physical picture of the exciter of the embodiment, and the exciter is a spring pin. The excitation magnitude of the spring pin is 24N each time. The position to be excited is the center position of any side of the concrete sample where the acoustic emission sensor is not arranged. The excitation method is to use the exciter to excite each compression level of the concrete sample 30 times when the concrete compression state reaches the corresponding level.

[0045] S3. Use an acoustic emission system to collect the waveform generated by the exciter, including: arranging an acoustic emission probe on the surface of the concrete test block, collecting the waveform signal generated by the exciter in the concrete sample in real time through the acoustic emission host, and storing it, and recording the compression grade corresponding to each waveform; the acoustic emission probe is one, arranged on any side except the compression surface of the concrete test block. Figure 4 The figure is a schematic diagram of the sensor installation and excitation surface of the concrete test block in the embodiment. An acoustic emission sensor is used and installed on the right side of the sample. The excitation surface is the front side of the concrete sample.

[0046] S4. Use the adaptive threshold wavelet denoising method to denoise the waveform collected in S3, specifically including: In the adaptive wavelet denoising method used, the denoising threshold formula is:

[0047]

[0048] where λ j is the wavelet denoising threshold formula for each level, ω k,j is the wavelet coefficient of each level (including approximate wavelet coefficient), δ j is the standard deviation of the noise at each level, j is the decomposition level, k is the location parameter, N j is the length of the wavelet coefficients at each level.

[0049] The wavelet function is:

[0050]

[0051] in is the wavelet function at each level, λ j is the wavelet denoising threshold formula for each level, ω k,j is the wavelet coefficient of each level (including approximate wavelet coefficient). α and β are shape adjustment factors, which can be obtained through optimization algorithm. Each waveform has its own most suitable α and β, but the value range of both is [0.01,3].

[0052] The optimization algorithm can use various optimization algorithms currently proposed, such as zebra optimization, sparrow optimization, particle swarm optimization algorithm, etc. In a preferred embodiment of the present invention, the optimization algorithm used is the zebra optimization algorithm, and the fitness function used is:

[0053]

[0054] Where F is the fitness function, σ 2 is the signal variance, H s is the spectral entropy of the signal, and H is the entropy of the signal. After denoising the original waveform, the denoised waveform data is recorded. Figure 6The following is an adaptive threshold wavelet denoising effect diagram of the embodiment: the X-axis is the number of sampling points, the range is 0 to 20000; the Y-axis is the amplitude, the range is 0 to 0.100 V. Compared with the original waveform signal, the denoised waveform signal obviously removes the Gaussian white noise, and the denoised signal has the same amplitude as the original signal, indicating that the denoising effect is good.

[0055] Figure 7 The following is a schematic diagram of the change of the fitness function value of the optimal parameter of adaptive threshold wavelet denoising with the number of iterations: the X-axis is the number of iterations, ranging from 0 to 50; the Y-axis is the fitness function value, ranging from 1.406×10 -7 to 1.414×10 -7 The fitness curve remains unchanged at first, then starts to rise exponentially in the 7th iteration, reaches the maximum value in the 8th iteration, and then remains unchanged, indicating that after the 8th iteration, the wavelet denoising parameters have reached the optimal solution.

[0056] S5. Use the DTW-K-medoids clustering method to cluster the denoised waveform signals obtained by different graded loadings obtained in S4, obtain the concrete stress state level, and perform manual labeling, including: calculating the DTW (dynamic time bending) distance between all waveform signals, using the K-medoids clustering method to divide the waveform signals of different graded loadings into 4 categories, and according to the size of the graded pressure contained in the 4 categories, the 4 categories are recorded in order from small to large as follows: Level I: safe; Level II: relatively safe; Level III: relatively dangerous; Level IV: dangerous. The waveforms corresponding to each category are also labeled with the level number. Figure 8 The schematic diagram of DTW-K-medoids clustering effect of concrete stress state: among the four clustered levels, level I includes graded pressures of 0 and 30 kN; level II includes graded pressures of 60, 90, 120, 150, 180, 210, and 240 kN. Level III includes graded pressures of 270, 300, 330, 360, and 390 kN. Level IV includes graded pressures of 420, 450, and 480 kN.

[0057] S6. Perform fast Fourier transform (FFT) and variational mode decomposition (VMD) on the denoised waveform signals manually annotated in S5 to obtain frequency domain signals and time domain signals respectively, and obtain a training data set, specifically including: using FFT transformation to obtain the corresponding frequency spectrograms of all the denoised waveform signals obtained, and annotating them with the same level as the denoised signals; using VMD decomposition to obtain the corresponding time spectrograms of all the denoised waveform signals obtained, and annotating them with the same level as the denoised signals. Fig. 9 The FFT spectrum of the excitation waveform is as follows: the X-axis is the frequency Hz, ranging from 0 to 200,000; the Y-axis is the amplitude V, ranging from 0 to 0.025. Fig.10This is the time-frequency diagram of VMD decomposition of the excitation waveform: the X-axis is the number of sampling points, ranging from 0 to 20,000; the Y-axis is the amplitude, ranging from -1 to 1. The waveform signal reconstructed by VMD decomposition has a slightly lower amplitude than the original denoised signal.

[0058] S7. Construct a dual-branch DCNN neural network to learn the frequency domain and time domain features of the waveform signal respectively, and perform feature fusion through the global attention mechanism.

[0059] The constructed dual-branch DCNN neural network includes the following parts: Fig.12 shown):

[0060] DCNN branches: The network has two identical DCNN branches, processing FFT and VMD features respectively;

[0061] Each DCNN branch contains 4 convolutional layers, each followed by a ReLU activation function and a maximum pooling layer; the number of channels of the convolutional layer gradually increases (32→64→128→256) to extract more complex features; the final output retains the channel dimension and flattens the features.

[0062] Global attention mechanism module: Apply softmax to the features output by the DCNN branch to generate weights. Use these weights to perform weighted summation on the features to obtain a global feature representation.

[0063] Feature fusion module: concatenates the features of the two DCNN branches (FFT and VMD) processed by the attention mechanism module.

[0064] Classification layer: Use two fully connected layers for final classification;

[0065] The first fully connected layer keeps the feature dimension unchanged and uses ReLU activation;

[0066] The second fully connected layer maps the features to the dimension of the number of categories and uses Softmax for multi-category classification. The size of the graded pressure is output in 4 categories in ascending order, namely Level I: safe; Level II: relatively safe; Level III: relatively dangerous; Level IV: dangerous.

[0067] S8. Input the data set in S6 into the dual-branch DCNN neural network constructed in S7 for model training; specifically, using the previously labeled FFT and VMD data as training sets and inputting them into the constructed dual-branch DCNN neural network for training, obtaining the trained neural network model, and saving it. The loss function during training uses the cross entropy loss, and the specific formula is:

[0068]

[0069] Where L represents the overall loss, N represents the total number of samples in the dataset or batch, C represents the number of different categories, and y ij is a binary indicator (0 or 1) indicating whether category j is the correct classification for sample i, p ij is the probability that the model predicts that sample i belongs to category j.

[0070] The accuracy rate is used in the evaluation of the model effect of the dual-branch DCNN neural network. The accuracy rate is obtained by dividing the number of correct recognitions by the total number of recognitions. It is used to evaluate the effect of the model. A model accuracy rate of 0.95 or above indicates a good effect. Fig.11 The following is a schematic diagram of the accuracy of the concrete stress state recognition test set and training set as the training cycle changes: the X-axis is the number of iterations, the segment is 0 to 500: the Y-axis is the accuracy, the segment is 0 to 1. The accuracy of the training set and the test set increases with the increase in the number of iterations. When the number of iterations reaches about 250, the accuracy remains basically unchanged, which means that the model has been trained. The accuracy of the training set reaches 0.99, and the accuracy of the test set reaches 0.96. This shows that the neural network model works very well and can accurately identify the stress state of concrete samples.

[0071] S9 uses the dual-branch DCNN neural network trained in S8 to identify the concrete stress state and ultimately determine the safety state of the concrete structure.

[0072] In a preferred embodiment of the present invention, the excitation waveforms of concrete test blocks in stress states of four categories are collected, and after denoising, FFT transformation and VMD decomposition, the obtained data are input into the trained neural network model in S8 for recognition, and the safety state recognition result of the concrete is obtained:

[0073] Level I: safe; Level II: relatively safe; Level III: relatively dangerous; Level IV: dangerous. The corresponding level among these four levels.

[0074] The above description is only a description of the preferred embodiments of the present application, and is not intended to limit the scope of the present application. Any changes or modifications made by any person skilled in the art based on the above disclosed technical contents shall be deemed as equivalent effective embodiments and shall fall within the scope of protection of the technical solution of the present application.

Claims

1. A method for identifying concrete stress state based on the time-frequency characteristics of acoustic emission waveforms. It is characterized in that The steps are as follows: S1. Conduct uniaxial graded loading test on concrete specimens; S2. Use the exciter to excite the concrete test blocks at various levels of loading in S1; S3. Use the acoustic emission system to collect the waveform generated by the exciter in S2; S4. De-noise the waveform collected by S3 using an adaptive threshold wavelet denoising method; S5. Use the DTW-K-medoids clustering method to cluster the denoised waveform signals obtained by different graded loadings obtained in S4, obtain the concrete stress state level, and manually annotate it; S6. Perform fast Fourier transform FFT and variational mode decomposition VMD on the denoised waveform signal after manual annotation in S5, and obtain frequency domain signal and time domain signal respectively, so as to obtain a data set for training; S7. Construct a dual-branch DCNN neural network to learn the frequency domain and time domain features of the waveform signal respectively, and perform feature fusion through the global attention mechanism; S8. Input the data set in S6 into the dual-branch DCNN neural network constructed in S7 to perform model training; S9. Use the dual-branch DCNN neural network trained in S8 to identify the concrete stress state. When applied, it can identify and determine the safety state of the concrete structure in the application scenario.

2. A method for identifying concrete stress state based on time-frequency characteristics of acoustic emission waveform according to claim 1, It is characterized in that In S1, a uniaxial graded loading test was conducted on the concrete specimens: a uniaxial compressor was used to compress the sandstone at a fixed loading rate of 1 kN / s, with the compression grade starting from 0 kN and the interval being 30 kN until the concrete block was destroyed; after each loading to the corresponding grade, a constant pressure was maintained for a period of time; The concrete test blocks are C20, C30, C40, C50 and other commonly used engineering grade concrete, all of which are cubic structures with a test block size of 100×100×100mm.

3. A method for identifying concrete stress state based on time-frequency characteristics of acoustic emission waveform according to claim 1, It is characterized in that In S4, the acquired waveform is denoised using an adaptive threshold wavelet denoising method, which includes: In the adaptive wavelet denoising method used, the denoising threshold formula is: where λ j is the wavelet denoising threshold formula for each level, ω k,j is the wavelet coefficient at each level, δ j is the standard deviation of the noise at each level, j is the decomposition level, k is the location parameter, N j is the length of the wavelet coefficients at each level; The wavelet function is: in is the wavelet function at each level, λ j is the wavelet denoising threshold formula for each level, ω k,j is the wavelet coefficient of each level, α and β are shape adjustment factors, obtained through optimization algorithm, each waveform has its own most suitable α and β, but both values ​​are in the range of [0.01,3]; optimization algorithms include but are not limited to zebra optimization, sparrow optimization, particle swarm optimization algorithm; in various optimization algorithms, the fitness functions used are: Where F is the fitness function, σ 2 is the signal variance, H s is the spectral entropy of the signal, H is the entropy of the signal; After denoising the original waveform, the denoised waveform data is recorded.

4. A method for identifying concrete stress state based on time-frequency characteristics of acoustic emission waveform according to claim 1, It is characterized in that In S5, the DTW-K-medoids clustering method is used to cluster the denoised waveform signals obtained by different graded loading to obtain the concrete stress state level and perform manual annotation, including: The DTW distances between all waveform signals were calculated, and the K-medoids clustering method was used to classify the waveform signals with different graded loading into four categories. According to the size of the graded pressure contained in the four categories, the four categories were recorded in ascending order as Grade I: safe; Grade II: relatively safe; Grade III: relatively dangerous; Grade IV: dangerous; The waveforms corresponding to each category are also labeled with their level numbers.

5. A method for identifying concrete stress state based on time-frequency characteristics of acoustic emission waveform according to claim 1, It is characterized in that In S6, the denoised waveform signal after manual annotation is subjected to FFT and VMD to obtain the frequency domain and time domain signals of the signal, respectively, including: For all the denoised waveform signals obtained, the corresponding frequency spectrograms are obtained by using FFT transformation, and are marked with the same level as the denoised signal; for all the denoised waveform signals obtained, the corresponding time spectrograms are obtained by using VMD decomposition, and are marked with the same level as the denoised signal.

6. A method for identifying concrete stress state based on time-frequency characteristics of acoustic emission waveform according to claim 1, It is characterized in that In S7, a dual-branch DCNN neural network is constructed to learn the frequency domain and time domain features of the waveform signal respectively, and perform feature fusion through the global attention mechanism. The constructed dual-branch DCNN neural network includes the following parts: DCNN branches: The network has two identical DCNN branches, processing FFT and VMD features respectively; Each DCNN branch contains 4 convolutional layers, each followed by a ReLU activation function and a maximum pooling layer; the number of channels in the convolutional layer gradually increases to extract more complex features; the final output retains the channel dimension and flattens the features; Global attention mechanism module: Apply softmax to the features output by the DCNN branch to generate weights; use these weights to perform weighted summation on the features to obtain a global feature representation; Feature fusion module: It combines the features of the two DCNN branches, namely FFT and VMD, after being processed by the attention mechanism module; Classification layer: Two fully connected layers are used for final classification, where: The first fully connected layer keeps the feature dimension unchanged and uses ReLU activation; The second fully connected layer maps the features to the dimension of the number of categories and uses Softmax for multi-category classification. The size of the graded pressure is output in 4 categories in ascending order, namely Level I: safe; Level II: relatively safe; Level III: relatively dangerous; Level IV: dangerous.

7. A method for identifying concrete stress state based on time-frequency characteristics of acoustic emission waveform according to claim 1, It is characterized in that In S8, the waveform data set is input into the constructed neural network to perform model training, which includes: Use the previously labeled FFT and VMD data as a training set and input it into the dual-branch DCNN neural network constructed in S7 for training, obtain the trained neural network model, and save it; The loss function during training uses cross entropy loss, and the specific formula is: Where L represents the overall loss; N represents the total number of samples in the dataset or batch; C represents the number of different categories; y ij is a binary indicator, 0 or 1, indicating whether category j is the correct classification of sample i; p ij is the probability that the model predicts that sample i belongs to category j.

8. A method for identifying concrete stress state based on time-frequency characteristics of acoustic emission waveform according to claim 1, It is characterized in that In S9, the trained model is used to identify the concrete stress state and determine the safety state of the concrete structure in the application scenario, including: The excitation waveform of the concrete specimen with unknown stress state is collected, and after denoising, FFT transformation and VMD decomposition, the obtained data is input into the trained neural network model for recognition; According to the identification results, the safety status of concrete is classified into the corresponding level among the four levels: Level I: safe; Level II: relatively safe; Level III: relatively dangerous; Level IV: dangerous.

Citation Information

Cited By

  • System and method for testing wave velocity in concrete pressure-bearing state based on machine learning

    CN119470650A

  • Hydraulic concrete structure state monitoring method and system based on acoustic emission and deep learning

    CN119691583A