A method and system for non-destructive testing of subgrade compaction degree based on acoustic-vibration coupling analysis

Abstract

A non-destructive testing method and system for roadbed compaction based on acoustic-vibration coupling analysis is disclosed. The method first collects raw acoustic-vibration data and extracts feature vectors to construct acoustic-vibration coupling features. A three-core composite kernel function, including time-domain, frequency-domain, and coupling sub-kernels, is designed to construct a multi-core fusion support vector regression model. Next, a parameter vector to be optimized is defined, and a minimization objective function is constructed to complete parameter optimization. The optimal parameters are substituted into the composite kernel function to obtain the optimal kernel function, which is then used to train the model. Finally, a new sample to be predicted is input, and the compaction detection result is output. This invention improves the model's generalization performance and prediction accuracy under complex conditions through acoustic-vibration multi-modal feature fusion and multi-core fusion regression model optimization design. Furthermore, by using parameter optimization and minimization of the objective function to determine the optimal kernel function for model training, the model's adaptability is enhanced, ensuring accurate and reliable detection results. Simultaneously, this invention does not require damage to the roadbed structure, significantly reducing the operational difficulty of roadbed detection.

Description

Technical Field

[0001] This invention relates to compaction testing methods, belonging to the field of road construction technology, and particularly to a non-destructive testing method and system for roadbed compaction based on acoustic-vibration coupling analysis. Background Technology

[0002] Subgrade compaction is one of the key indicators for measuring the quality of infrastructure construction such as highways and railways. It is directly related to the bearing capacity, stability and service life of the subgrade. Accurate testing of compaction is crucial to ensuring that the construction quality meets the design requirements.

[0003] Existing compaction testing methods mainly include the sand cone method, the ring cutter method, and the nuclear density and humidity meter method. The sand cone method is a widely used compaction testing method with high accuracy, but it is complex to operate, requiring multiple steps such as digging pits, filling sand, and weighing, which is time-consuming, labor-intensive, and causes significant damage to the roadbed. The ring cutter method is relatively simple to operate, but it requires high soil homogeneity and is difficult to adapt to complex field conditions. The nuclear density and humidity meter method uses the rays of radioactive isotopes to penetrate the roadbed material and calculates density and humidity by measuring the attenuation of the rays. Although the testing speed is fast, it poses radiation hazards, is sensitive to environmental conditions, and requires strict operation and management. Therefore, in order to improve the efficiency of compaction testing, reduce damage to the road surface, and reduce the maintenance difficulty of testing instruments, there is an urgent need for a fast and convenient non-destructive testing method for roadbed compaction to address the aforementioned shortcomings of existing technologies. Summary of the Invention

[0004] The purpose of this invention is to overcome the above-mentioned defects and problems in the prior art and to provide a non-destructive and convenient method and system for non-destructive testing of roadbed compaction based on acoustic-vibration coupling analysis.

[0005] To achieve the above objectives, the technical solution of this invention is: a non-destructive testing method for roadbed compaction based on acoustic-vibration coupling analysis, comprising:

[0006] Collect raw acoustic and vibration data and extract feature vectors to construct acoustic and vibration coupling features;

[0007] Based on the acoustic-vibration coupling characteristics, a three-core composite kernel function containing a time-domain sub-kernel, a frequency-domain sub-kernel, and a coupling sub-kernel is designed to construct a multi-core fusion support vector regression model;

[0008] Define the parameter vector to be optimized for the multi-core fusion support vector regression model, construct the objective function to minimize it and optimize the parameters to obtain the optimal parameters;

[0009] The optimal kernel function is obtained by substituting the optimal parameters into the composite kernel function, and a multi-kernel fusion support vector regression model is trained based on the optimal kernel function. The new sample to be predicted is input into the trained multi-kernel fusion support vector regression model to obtain the final compaction degree detection result.

[0010] Optionally, the step of acquiring raw acoustic and vibration data and extracting feature vectors to construct acoustic and vibration coupling features specifically includes:

[0011] Collect raw acoustic and vibration data of the roadbed and the corresponding actual compaction degree. and will As a supervisory label;

[0012] The raw acoustic vibration data are paired with the corresponding actual compaction degree to form data pairs. ;in For the sample The original data, For the sample The compaction degree monitoring label;

[0013] Feature vectors of the original acoustic-vibration data in the data pair are extracted, and the feature vectors are standardized and grouped to obtain acoustic-vibration coupling features. ;in: Features in the time domain; This refers to frequency domain / coupling characteristics.

[0014] Optionally, the design of a three-kernel composite kernel function based on acoustic-vibration coupling characteristics, comprising a time-domain sub-kernel, a frequency-domain sub-kernel, and a coupling sub-kernel, to construct a multi-kernel fusion support vector regression model specifically includes:

[0015] Based on the acoustic-vibration coupling characteristics, a time-domain sub-kernel, a frequency-domain sub-kernel, and a coupling sub-kernel are constructed and fused into a composite kernel function, the expression of which is as follows:

[0016] ;

[0017] in: It is a composite kernel function that characterizes the sample. and Overall similarity; For time-domain sub-kernel; For frequency domain sub-cores; For coupling subnuclei; , Samples Temporal eigenvectors ; , Samples Frequency domain / coupling eigenvectors ; , Samples and Eigenvectors in the coupled subspace; , , All are weights; Normalized time difference;

[0018] ;

[0019] in: It is an exponential function; For transpose; It is an anisotropic independent bandwidth matrix; This is the time decay coefficient;

[0020] ;

[0021] in: These are the polynomial kernel scaling factors; The inner product of frequency domain / coupled eigenvectors; It is of polynomial order;

[0022] ;

[0023] ;

[0024] in: , ; Cross-spectral amplitude; The coherence coefficient is denoted as .

[0025] Optionally, the step of defining the parameter vector to be optimized for the multi-core fusion support vector regression model, constructing a minimum objective function and optimizing the parameters to obtain the optimal parameters specifically includes:

[0026] Define the parameter vector to be optimized in the multi-core fusion support vector regression model, and its expression is as follows:

[0027] ;

[0028] in: The parameter vector to be optimized; This is the penalty coefficient; Width of the insensitive area; Independent bandwidth for each feature in the time domain;

[0029] The bandwidth is initialized and the search boundary is determined by the following expression:

[0030] ;

[0031] ;

[0032] in: For the first Initial values ​​for the bandwidth of each time-domain feature; This is the median of the squared differences; For the first 5-dimensional standard deviation; For logarithmic field variables; This serves as the boundary for the interval search;

[0033] Based on the parameter vector to be optimized, a minimum objective function is constructed and the parameters are optimized to obtain the optimal parameters, as shown in the following expression:

[0034] ;

[0035] in: To minimize the objective function; Weighted root mean square error for cross-validation; For the weighting factor; This refers to the error variance;

[0036] ;

[0037] ;

[0038] in: For the first The true compaction degree of each sample; This is the predicted value for compaction degree; For sample weights; It is a coherence function; The coherence threshold;

[0039] ;

[0040] in: For a subset of samples; For the first Group error; The mean of the group error.

[0041] Optionally, the step of substituting the optimal parameters into the composite kernel function to obtain the optimal kernel function, and training a multi-kernel fusion support vector regression model based on the optimal kernel function; inputting the new sample to be predicted into the trained multi-kernel fusion support vector regression model to obtain the final compaction degree detection result, specifically includes:

[0042] Substituting the optimal parameters into the composite kernel function yields the optimal kernel function, whose expression is as follows:

[0043] ;

[0044] in: The optimal kernel function; , , All are optimal parameters;

[0045] Based on the optimal kernel function and data pairs, a multi-kernel fusion support vector regression model is trained, and new samples to be predicted are then used. The input is fed into the trained multi-core fusion support vector regression model to calculate the single-window predicted compaction degree, as shown in the following expression:

[0046] ;

[0047] in: Predict compaction degree using a single window; It is a set of support vectors; , All are Lagrange multipliers; is the feature vector of the sample to be predicted; For the first Support vectors; For bias terms;

[0048] The final compaction degree is obtained by weighted averaging the compaction degrees predicted from multiple overlapping windows under the same excitation, as expressed below:

[0049] ;

[0050] ;

[0051] in: This represents the final compaction degree; For the first Predicted compaction values ​​for each window; Number of windows; Window weight; The optimal time decay coefficient; For the first Normalized time difference of each window; For indicator functions; Coherent at the window's main frequency; The signal-to-noise ratio weights for the window.

[0052] Optionally, the time-domain features specifically include:

[0053] The speed of sound propagation is calculated based on the time difference between the peak of the transmitted and received first sound wave signal and the distance between the transmitting and receiving points. Its expression is as follows:

[0054] ;

[0055] in: The distance between the transmitting point and the receiving point. The time difference between the peaks;

[0056] Based on the initial signal amplitude of the transmitted sound wave and the amplitude of the received signal, the signal amplitude attenuation rate is calculated. Its expression is as follows:

[0057] ;

[0058] in: The initial signal amplitude; The amplitude of the received signal; It is a constant; The common logarithm;

[0059] Damping ratio calculated based on the logarithmic attenuation method of time-domain signal Its expression is as follows:

[0060] ;

[0061] in: The amplitude of the first sound wave peak. For the first Each acoustic wave peak amplitude; This is a correction factor.

[0062] Optionally, the frequency domain / coupling features specifically include:

[0063] Calculate the power spectral density of the acoustic signal With the power spectral density of the vibration signal And determine the specific dominant frequency corresponding to the acoustic spectral peak. Its expression is as follows:

[0064] ;

[0065] ;

[0066] ;

[0067] ;

[0068] in: The main frequency is a variable. For the sound wave signal Segmented windowed samples; For the vibration signal Segmented windowed samples; For window functions; The sampling length of each signal segment; The number of segments; The normalization constant for window energy; These are the basis functions of the discrete Fourier transform; As the independent variable;

[0069] Calculate energy concentration based on the power spectral density of the acoustic signal. This refers to the percentage of energy in the bandwidth near the main frequency relative to the total energy, expressed as follows:

[0070] ;

[0071] in: The bandwidth half-width of the variable main frequency;

[0072] Calculate the cross-spectral power density of the acoustic signal and the vibration signal. Cross-spectral amplitude at the dominant frequency Its expression is as follows:

[0073] ;

[0074] ;

[0075] in: For the first The windowed Fourier transform result of the acoustic wave signal; For the first The windowed Fourier transform result of the segment vibration signal; For complex conjugate; The normalization constant for window energy;

[0076] Calculate the coherence coefficient between the acoustic signal and the vibration signal near the dominant frequency. Average coherence with the neighborhood of the main frequency Its expression is as follows:

[0077] ;

[0078] ;

[0079] in: Let be the power spectral density of the vibration signal.

[0080] A non-destructive testing system for roadbed compaction based on acoustic-vibration coupling analysis, wherein the system is applied to the above-mentioned method, the system comprising:

[0081] The feature coupling module is used to acquire raw acoustic and vibration data and extract feature vectors to construct acoustic and vibration coupling features.

[0082] The model building module is used to design a three-core composite kernel function containing a time-domain sub-kernel, a frequency-domain sub-kernel, and a coupling sub-kernel based on acoustic-vibration coupling characteristics, and to build a multi-core fusion support vector regression model.

[0083] The parameter optimization module is used to define the parameter vector to be optimized in the multi-core fusion support vector regression model, construct the minimum objective function and perform parameter optimization to obtain the optimal parameters;

[0084] The compaction degree detection module is used to substitute the optimal parameters into the composite kernel function to obtain the optimal kernel function, and to train a multi-kernel fusion support vector regression model based on the optimal kernel function; the new sample to be predicted is input into the trained multi-kernel fusion support vector regression model to obtain the final compaction degree detection result.

[0085] A non-destructive testing device for roadbed compaction based on acoustic-vibration coupling analysis, the device comprising a processor and a memory;

[0086] The memory is used to store computer program code and to transmit the computer program code to the processor;

[0087] The processor is used to execute the above-described non-destructive testing method for roadbed compaction based on acoustic-vibration coupling analysis according to the instructions in the computer program code.

[0088] A computer-readable storage medium storing computer-executable instructions, which, when executed on a computer, implement the aforementioned non-destructive testing method for roadbed compaction based on acoustic-vibration coupling analysis.

[0089] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0090] This invention discloses a non-destructive testing method and system for roadbed compaction based on acoustic-vibration coupling analysis. The method first collects raw acoustic-vibration data and extracts feature vectors to construct acoustic-vibration coupling features. It then designs a three-core composite kernel function containing time-domain, frequency-domain, and coupling sub-kernels to construct a multi-core fusion support vector regression model. Next, it defines the parameter vector to be optimized and constructs a minimization objective function to complete parameter optimization. The optimal parameters are substituted into the composite kernel function to obtain the optimal kernel function, which is then used to train the model. Finally, new samples to be predicted are input, and the compaction detection results are output. In application, this design improves the generalization performance and prediction accuracy of the model under complex conditions through acoustic-vibration multi-modal feature fusion and multi-core fusion regression model optimization. Furthermore, by using parameter optimization and minimization of the objective function to determine the optimal kernel function for model training, the model's adaptability is enhanced, ensuring the accuracy and reliability of the detection results. Simultaneously, this invention does not require damage to the roadbed structure, significantly reducing the operational difficulty of roadbed detection. Attached Figure Description

[0091] Figure 1 This is a flowchart of the method of the present invention.

[0092] Figure 2This is a system structure diagram of the present invention.

[0093] Figure 3 This is a structural diagram of the device of the present invention.

[0094] In the figure: Feature coupling module 1, Model building module 2, Parameter optimization module 3, Compaction degree detection module 4, Processor 5, Memory 6, Computer program code 61. Detailed Implementation

[0095] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0096] Example 1:

[0097] See Figure 1 A non-destructive testing method for roadbed compaction based on acoustic-vibration coupling analysis, comprising:

[0098] Collect raw acoustic and vibration data and extract feature vectors to construct acoustic and vibration coupling features;

[0099] Furthermore, raw acoustic and vibration data were collected from multiple roadbed test sections (compaction degree 85%-98%), and the actual compaction degree at the corresponding test points was measured simultaneously. and will As a supervisory label;

[0100] The raw acoustic vibration data are paired with the corresponding actual compaction degree to form data pairs. ;in For the sample The original data, For the sample The compaction degree monitoring label;

[0101] Feature vectors of the original acoustic-vibration data in the data pair are extracted, and the feature vectors are standardized and grouped to obtain acoustic-vibration coupling features. ;in: For time domain features, For roadbed thickness (measurement input); This refers to frequency domain / coupling characteristics.

[0102] Furthermore, the specific methods for extracting the feature vectors are as follows:

[0103] The speed of sound propagation is calculated based on the time difference between the peak of the transmitted and received first sound wave signal and the distance between the transmitting and receiving points. Its expression is as follows:

[0104] ;

[0105] in: The distance between the transmitting point and the receiving point. The time difference between the peaks;

[0106] Based on the initial signal amplitude of the transmitted sound wave and the amplitude of the received signal, the signal amplitude attenuation rate is calculated. , representing the energy loss rate of sound waves propagating in the roadbed, is expressed as follows:

[0107] ;

[0108] in: The initial signal amplitude; The amplitude of the received signal; is a constant used to convert the natural logarithm of the amplitude ratio to decibels; The common logarithm;

[0109] The damping ratio is calculated based on the logarithmic attenuation method of the time-domain signal, and its expression is as follows:

[0110] ;

[0111] in: The amplitude of the first sound wave peak (obtained from the accelerometer). For the first Each acoustic wave peak amplitude; This is a correction factor;

[0112] The power spectral density of the acoustic signal is calculated using the Welch power spectral estimation method. With the power spectral density of the vibration signal And determine the specific dominant frequency corresponding to the acoustic spectral peak. Its expression is as follows:

[0113] ;

[0114] ;

[0115] ;

[0116] ;

[0117] in: The main frequency is a variable. For the sound wave signal Segmented windowed samples; For the vibration signal Segmented windowed samples; For window functions; The sampling length of each signal segment; The number of segments; The normalization constant for window energy; These are the basis functions of the discrete Fourier transform; As the independent variable;

[0118] In this embodiment, the Welch power spectrum estimation method is used to calculate the dominant frequency, energy concentration, cross-spectral power density, and coherence coefficient to improve the stability and reliability of frequency domain features. Compared with the traditional single-pass Fast Fourier Transform (FFT) direct calculation, the Welch method effectively reduces the variance of power spectrum estimation by segmenting the signal, windowing, and overlapping averaging, thereby reducing the impact of random noise and transient disturbances on the results. It also suppresses spectral leakage, resulting in clearer dominant frequency peaks and more realistic energy distribution. Furthermore, the cross-spectrum and coherence function are smoother and have higher confidence after multi-segment averaging, more accurately reflecting the coupling relationship between acoustic excitation and vibration response. The Welch method significantly improves the robustness and noise resistance of feature extraction while maintaining frequency resolution, providing more reliable input data for subsequent compaction inversion models.

[0119] Calculate energy concentration based on the power spectral density of the acoustic signal. That is, a specific clock frequency A certain bandwidth in the vicinity ( The percentage of energy from _____ to the total energy is expressed as follows:

[0120] ;

[0121] in: The bandwidth half-width of the variable main frequency; when A higher value indicates concentrated signal energy and stable frequency characteristics; conversely, a lower value indicates dispersed energy and uneven compaction layer characteristics.

[0122] Calculate the cross-spectral power density of the acoustic signal and the vibration signal. Cross-spectral amplitude at the dominant frequency This reflects the coupling strength between acoustic excitation and vibration response, and its expression is as follows:

[0123] ;

[0124] ;

[0125] in: For the first The windowed Fourier transform result of the acoustic wave signal; For the first The windowed Fourier transform result of the segment vibration signal; For complex conjugate; The normalization constant for window energy;

[0126] Calculate the coherence coefficient between the acoustic signal and the vibration signal near the dominant frequency. Average coherence with the neighborhood of the main frequency This is used to measure the degree of stable coupling of acoustic and vibration signals within the main frequency band, and its expression is as follows:

[0127] ;

[0128] ;

[0129] in: The power spectral density of the vibration signal; The range is 0-1, indicating that the two signals are in frequency ranges of 0-1. Consistency at the point.

[0130] Based on the acoustic-vibration coupling characteristics, a three-core composite kernel function containing a time-domain sub-kernel, a frequency-domain sub-kernel, and a coupling sub-kernel is designed to construct a multi-core fusion support vector regression model;

[0131] In this embodiment, to establish a nonlinear mapping relationship between acoustic-vibration coupling characteristics and compaction degree, a multi-kernel fusion support vector regression model is adopted. This model constructs a comprehensive kernel function to characterize the similarity between samples by simultaneously considering the differences in time domain, frequency domain, and coupling characteristics. This scheme expands the original structure, which consisted only of time-domain and frequency-domain kernel functions, into a three-kernel composite model containing a time-domain kernel, a frequency-domain kernel, and a coupling kernel. Specifically, the time-domain kernel uses a dynamic RBF kernel with a time decay factor, enabling the model to distinguish the importance of signals at different time intervals; anisotropic bandwidth is introduced to independently adjust the sensitivity of each time-domain feature; the frequency-domain kernel maintains a polynomial form but achieves adaptive parameter optimization to better fit the nonlinear energy distribution characteristics; the newly added coupling kernel is used to characterize the cross-spectral power and coherence relationship between acoustic waves and vibration signals, thereby strengthening the physical consistency of acoustic-vibration energy transfer. This improvement expands the model from a single mapping to a multi-modal feature fusion system, enabling stable identification of compaction degree variation patterns even in complex soil and noisy environments.

[0132] Furthermore, the design of a three-core composite kernel function based on acoustic-vibration coupling characteristics, comprising a time-domain sub-kernel, a frequency-domain sub-kernel, and a coupling sub-kernel, is used to construct a multi-core fusion support vector regression model, as detailed below:

[0133] Based on the acoustic-vibration coupling characteristics, a time-domain sub-kernel, a frequency-domain sub-kernel, and a coupling sub-kernel are constructed and fused into a composite kernel function, the expression of which is as follows:

[0134] ;

[0135] in: It is a composite kernel function that characterizes the sample. and Overall similarity; For time-domain sub-kernel; For frequency domain sub-cores; For coupling subnuclei; , Samples Temporal eigenvectors (including) ); , Samples Frequency domain / coupling eigenvectors (including) ); , Samples and Eigenvectors in the coupled subspace; , , All are weights, all are ≥0, and satisfy the following conditions: ; Normalized time difference;

[0136] ;

[0137] in: It is an exponential function; For transpose; It is an anisotropic independent bandwidth matrix; This is the time decay coefficient; , The center time of two samples (or windows) It is a time scale constant;

[0138] ;

[0139] in: These are the polynomial kernel scaling factors; The inner product of frequency domain / coupled eigenvectors; The order is a polynomial, typically 2 or 3; constant. This is a bias term used to prevent the kernel value from being too small when the inner product is small.

[0140] ;

[0141] ;

[0142] in: , ; Cross-spectral amplitude; Coherence coefficient; exponent To fix the quadratic form in order to enhance acoustic-vibration synchronization.

[0143] Define the parameter vector to be optimized for the multi-core fusion support vector regression model, construct the objective function to minimize it and optimize the parameters to obtain the optimal parameters;

[0144] Furthermore, the parameter vector to be optimized for the multi-core fusion support vector regression model is defined, a minimum objective function is constructed, and parameter optimization is performed to obtain the optimal parameters, specifically including:

[0145] Define the parameter vector to be optimized in the multi-core fusion support vector regression model, and its expression is as follows:

[0146] ;

[0147] in: The parameter vector to be optimized; This is the penalty coefficient; Width of the insensitive area; These are the independent bandwidths for each feature in the time domain; This is the time decay coefficient; These are the polynomial kernel scaling factors; It is of polynomial order; , , All are weights;

[0148] The bandwidth is initialized and the search boundary is determined by the following expression:

[0149] ;

[0150] ;

[0151] in: For the first Initial values ​​for the bandwidth of each time-domain feature; This is the median of the squared differences; For the first 5-dimensional standard deviation; For logarithmic field variables; This serves as the boundary for the interval search;

[0152] In this embodiment, the traditional single-objective Bayesian optimization is extended to a multi-objective robust optimization mechanism. A generalization variance constraint term is introduced to simultaneously consider prediction accuracy and consistency across operating conditions. Anisotropic bandwidth, time decay coefficient, and multi-kernel weights are added to the parameter set to make the optimization space more physically meaningful and adaptive. In order to avoid optimization instability caused by differences in parameter scales, bandwidth initialization and search boundary constraints are performed so that the bandwidth and kernel parameters can be automatically adjusted according to the sample distribution. Through hierarchical cross-validation and sample weighting strategies, the model can maintain stable prediction accuracy and generalization performance under different roadbed materials, moisture content and excitation conditions.

[0153] Based on the parameter vector to be optimized, a minimum objective function is constructed and the parameters are optimized to obtain the optimal parameters, as shown in the following expression:

[0154] ;

[0155] in: To minimize the objective function; Weighted root mean square error for cross-validation; For the weighting factor; Error variance grouped by material / operating condition (generalization consistency);

[0156] ;

[0157] ;

[0158] in: For the first The true compaction degree of each sample; This is the predicted value for compaction degree; Sample weights (take the in-band coherence mean or threshold indicator); It is a coherence function; The coherence threshold (e.g., 0.6-0.7);

[0159] ;

[0160] in: For a subset of samples; For the first Group error (e.g., RMSE); The mean of the group error.

[0161] The optimal kernel function is obtained by substituting the optimal parameters into the composite kernel function, and a multi-kernel fusion support vector regression model is trained based on the optimal kernel function. The new sample to be predicted is input into the trained multi-kernel fusion support vector regression model to obtain the final compaction degree detection result.

[0162] Furthermore, obtaining the final compaction test result specifically includes:

[0163] Substituting the optimal parameters into the composite kernel function yields the optimal kernel function, whose expression is as follows:

[0164] ;

[0165] in: The optimal kernel function; , , All are optimal parameters;

[0166] Based on the optimal kernel function and data pairs, a multi-kernel fusion support vector regression model is trained, and new samples to be predicted are then used. The input is fed into the trained multi-core fusion support vector regression model to calculate the single-window predicted compaction degree, as shown in the following expression:

[0167] ;

[0168] in: Predict compaction degree using a single window; It is a set of support vectors; , All are Lagrange multipliers; is the feature vector of the sample to be predicted; For the first Support vectors; For bias terms;

[0169] The final compaction degree is obtained by weighted averaging the compaction degrees predicted from multiple overlapping windows under the same excitation, as expressed below:

[0170] ;

[0171] ;

[0172] in: This represents the final compaction degree; For the first Predicted compaction values ​​for each window; Number of windows; Window weight; The optimal time decay coefficient; For the first Normalized time difference of each window; For indicator functions; Coherent at the window's main frequency; The signal-to-noise ratio weights for the window.

[0173] In this embodiment, the optimal kernel function uses coupled sub-kernels and multi-window weighting, and introduces a multi-window weighted fusion strategy based on the original single-window output. Specifically, the detection signal is divided into several overlapping time windows, each window independently calculates the compaction degree prediction value, and then the weight coefficients are determined based on time proximity, signal coherence, and signal-to-noise ratio. The time decay term is used to assign higher weights to samples in the recent time period, and a coherence threshold is set to filter out low-confidence signals, making the overall result insensitive to abnormal fluctuations. Finally, the output adopts a weighted average form to obtain a continuous and smooth compaction degree estimation curve, which greatly improves the robustness and stability of the results and ensures the reliability of the model in the real-time detection environment.

[0174] In this embodiment, based on the detection method of this scheme, a corresponding system can be designed to implement it. This system includes an acoustic excitation device, a sensor array, a signal amplifier and conditioner, a multi-channel data acquisition unit, an embedded signal processor, and a human-machine interface terminal. The system's operation flow is as follows:

[0175] Equipment initialization: The system is started, and the embedded signal processor completes sensor calibration and AD initialization; the acoustic excitation device is preheated and self-tested; the operator inputs the detection task number and roadbed type (roadbed type is also one of the input features) through the human-machine interface terminal.

[0176] Acoustic excitation: The embedded signal processor sends a trigger signal, and the acoustic excitation device sends a short pulse; the excitation energy is injected into the foundation under test through the piezoelectric device to excite the acoustic vibration response.

[0177] Response signal acquisition: The sensor array synchronously acquires response signals. The data is amplified by the signal amplifier and conditioner and then transmitted to the embedded signal processor by the multi-channel data acquisition unit. All signals are sampled in the full time domain to accurately preserve the original characteristics. After acquisition, the signals enter the buffer storage area.

[0178] Signal alignment and preprocessing: The embedded signal processor reads multiple data streams and performs time delay compensation using the cross-correlation function method; wavelet denoising and bandpass filtering are performed to suppress environmental and instrument noise; signal amplitude is standardized to avoid deviations introduced by equipment differences.

[0179] Feature extraction: sound wave propagation speed (time difference of arrival of the first wave); signal amplitude attenuation rate (amplitude-to-mean ratio); damping ratio (logarithmic attenuation method); power spectral density of the acoustic signal and the specific dominant frequency corresponding to the spectral peak (Welch power spectrum estimation method); energy concentration (energy proportion within the dominant frequency band); cross-spectral amplitude at the dominant frequency (Welch cross-spectral estimation method); coherence coefficient and average coherence (Welch coherence function method). See the table below for explanations of the feature types and their correlation with compaction degree.

[0180]

[0181] Feature normalization and model prediction: All features are normalized and then input into the SVR model; the SVR model calculates the compaction degree K value (%) of the current detection point.

[0182] Results display and storage: The human-computer interaction terminal displays the K value, main frequency, waveform and its position on the "compaction degree-material type" two-dimensional graph in real time. Operators can make construction quality judgments and feedback based on the displayed results. At the same time, all data is packaged and stored in the internal database for subsequent use.

[0183] This scheme acquires acoustic and vibration response signals of roadbed materials under acoustic excitation. Combining characteristics such as acoustic propagation speed, energy attenuation, vibration acceleration, and damping, an improved multi-kernel support vector regression (SVR) algorithm is used to establish a compaction inversion model. The model introduces a time-domain dynamic RBF kernel, a frequency-domain polynomial kernel, and an acoustic-vibration coupling kernel at the kernel function level, realizing multi-modal fusion of acoustic propagation characteristics and vibration response characteristics. Furthermore, through the time attenuation weight and anisotropic bandwidth (ARD) mechanism, the model can adaptively distinguish the importance of different features, effectively suppress noise interference, and achieve non-destructive, rapid, and high-precision detection of compaction.

[0184] In terms of feature extraction, Welch average power spectrum estimation is used to replace the traditional FFT method, which robustly estimates key features such as dominant frequency, energy concentration, cross-spectral power density, and coherence coefficient. Smoothing and weighting are performed within the frequency band to improve the stability of frequency domain parameters in complex noise environments. The model input includes sound wave propagation parameters (velocity, attenuation), vibration response parameters (dominant frequency, damping, energy distribution), and acoustic-vibration coupling parameters (cross-spectral power density, coherence coefficient). Combined with material type and layer thickness information, multi-dimensional input features are formed. Through multi-kernel fusion and parameter adaptive optimization, the model can maintain high generalization performance and prediction accuracy under different roadbed structures, different moisture contents, and different densities.

[0185] In the data fusion and output stage, a multi-window weighted averaging strategy is adopted to smooth and robustly process the results. By dividing the detection signal into time windows and weighting the prediction results of each window according to time proximity, coherence and signal-to-noise ratio, the influence of instantaneous noise and random error in a single excitation can be significantly reduced. At the same time, multi-sensor collaborative measurement using a ring array and spatial mean filtering are used to achieve synchronous noise reduction and outlier suppression of multi-channel signals, thereby ensuring the stability and repeatability of the model under complex field conditions.

[0186] Example 2:

[0187] See Figure 2 A non-destructive testing system for roadbed compaction based on acoustic-vibration coupling analysis is provided. This system is applied to the method described in Example 1, and the system includes:

[0188] Feature coupling module 1 is used to collect raw acoustic and vibration data and extract feature vectors to construct acoustic and vibration coupling features;

[0189] Model building module 2 is used to design a three-core composite kernel function containing a time-domain sub-kernel, a frequency-domain sub-kernel, and a coupling sub-kernel based on acoustic-vibration coupling characteristics, and to build a multi-core fusion support vector regression model;

[0190] Parameter optimization module 3 is used to define the parameter vector to be optimized in the multi-core fusion support vector regression model, construct the minimization objective function and perform parameter optimization to obtain the optimal parameters;

[0191] The compaction degree detection module 4 is used to substitute the optimal parameters into the composite kernel function to obtain the optimal kernel function, and to train the multi-kernel fusion support vector regression model based on the optimal kernel function; the new sample to be predicted is input into the trained multi-kernel fusion support vector regression model to obtain the final compaction degree detection result.

[0192] Furthermore, for the specific implementation steps of the feature coupling module 1, model construction module 2, parameter optimization module 3, and compaction degree detection module 4, please refer to the corresponding description in Example 1, which will not be repeated here.

[0193] Example 3:

[0194] See Figure 3 A non-destructive testing device for roadbed compaction based on acoustic-vibration coupling analysis, the device comprising a processor 5 and a memory 6;

[0195] The memory 6 is used to store computer program code 61 and transmit the computer program code 61 to the processor 5;

[0196] The processor 5 is used to execute the non-destructive testing method for roadbed compaction based on acoustic-vibration coupling analysis as described in Embodiment 1 according to the instructions in the computer program code 61.

[0197] This embodiment also includes a computer-readable storage medium storing computer-executable instructions. When the computer-executable instructions are executed on a computer, the non-destructive testing method for roadbed compaction based on acoustic-vibration coupling analysis described in Embodiment 1 is implemented.

[0198] Generally, the computer instructions for implementing the method of the present invention can be carried on any combination of one or more computer-readable storage media. Non-transitory computer-readable storage media can include any computer-readable medium except for the signal itself, which is temporarily propagating.

[0199] Computer-readable storage media can be, for example, but not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatuses, or any combination thereof. More specific examples (a non-exhaustive list) of computer-readable storage media include: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EKROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this invention, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0200] Computer program code for performing the operations of this invention can be written in one or more programming languages ​​or a combination thereof. These programming languages ​​include object-oriented programming languages—such as Java, Smarttalk, and C++—as well as conventional procedural programming languages—such as the "C" language or similar programming languages. In particular, Python, suitable for neural network computation, and platform frameworks such as TensorFlow and PyTorch can be used. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer or to an external computer (e.g., via the Internet using an Internet service provider) through any type of network, including a local area network (LAN) or a wide area network (WAN).

[0201] The aforementioned equipment and non-transitory computer-readable storage media can be found in the detailed description of a non-destructive testing method for roadbed compaction based on acoustic-vibration coupling analysis and its beneficial effects, which will not be repeated here.

[0202] Although embodiments of the present invention have been shown and described above, it should be understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A non-destructive testing method for roadbed compaction degree based on acoustic-vibration coupling analysis, characterized in that, include: Collect raw acoustic and vibration data and extract feature vectors to construct acoustic and vibration coupling features; Based on the acoustic-vibration coupling characteristics, a three-core composite kernel function containing a time-domain sub-kernel, a frequency-domain sub-kernel, and a coupling sub-kernel is designed to construct a multi-core fusion support vector regression model; Define the parameter vector to be optimized for the multi-core fusion support vector regression model, construct the objective function to minimize it and optimize the parameters to obtain the optimal parameters; The optimal kernel function is obtained by substituting the optimal parameters into the composite kernel function, and a multi-kernel fusion support vector regression model is trained based on the optimal kernel function. The new sample to be predicted is input into the trained multi-kernel fusion support vector regression model to obtain the final compaction degree detection result.

2. The non-destructive testing method for roadbed compaction based on acoustic-vibration coupling analysis according to claim 1, characterized in that: The process of collecting raw acoustic and vibration data, extracting feature vectors, and constructing acoustic-vibration coupling features specifically includes: Collect raw acoustic and vibration data of the roadbed and the corresponding actual compaction degree. and will As a supervisory label; The raw acoustic vibration data are paired with the corresponding actual compaction degree to form data pairs. ;in For the sample The original data, For the sample The compaction degree monitoring label; Feature vectors of the original acoustic-vibration data in the data pair are extracted, and the feature vectors are standardized and grouped to obtain acoustic-vibration coupling features. ;in: Features in the time domain; This refers to frequency domain / coupling characteristics.

3. The non-destructive testing method for roadbed compaction based on acoustic-vibration coupling analysis according to claim 2, characterized in that: The design of a three-kernel composite kernel function based on acoustic-vibration coupling characteristics, comprising a time-domain sub-kernel, a frequency-domain sub-kernel, and a coupling sub-kernel, constructs a multi-kernel fusion support vector regression model, specifically including: Based on the acoustic-vibration coupling characteristics, a time-domain sub-kernel, a frequency-domain sub-kernel, and a coupling sub-kernel are constructed and fused into a composite kernel function, the expression of which is as follows: ； in: It is a composite kernel function that characterizes the sample. and Overall similarity; For time-domain sub-kernel; For frequency domain sub-cores; For coupling subnuclei; , Samples Temporal eigenvectors ; , Samples Frequency domain / coupling eigenvectors ; , Samples and Eigenvectors in the coupled subspace; , , All are weights; Normalized time difference; ； in: It is an exponential function; For transpose; It is an anisotropic independent bandwidth matrix; This is the time decay coefficient; ； in: These are the polynomial kernel scaling factors; The inner product of frequency domain / coupled eigenvectors; It is of polynomial order; ； ； in: , ; Cross-spectral amplitude; The coherence coefficient is denoted as .

4. The non-destructive testing method for roadbed compaction based on acoustic-vibration coupling analysis according to claim 3, characterized in that: The process of defining the parameter vector to be optimized in the multi-core fusion support vector regression model, constructing a minimum objective function, and optimizing the parameters to obtain the optimal parameters specifically includes: Define the parameter vector to be optimized in the multi-core fusion support vector regression model, and its expression is as follows: ； in: The parameter vector to be optimized; This is the penalty coefficient; Width of the insensitive area; Independent bandwidth for each feature in the time domain; The bandwidth is initialized and the search boundary is determined by the following expression: ； ； in: For the first Initial values ​​for the bandwidth of each time-domain feature; This is the median of the squared differences; For the first 5-dimensional standard deviation; For logarithmic field variables; This serves as the boundary for the interval search; Based on the parameter vector to be optimized, a minimum objective function is constructed and the parameters are optimized to obtain the optimal parameters, as shown in the following expression: ； in: To minimize the objective function; Weighted root mean square error for cross-validation; For the weighting factor; This refers to the error variance; ； ； in: For the first The true compaction degree of each sample; This is the predicted value for compaction degree; For sample weights; It is a coherence function; The coherence threshold; ； in: For a subset of samples; For the first Group error; The mean of the group error.

5. The non-destructive testing method for roadbed compaction based on acoustic-vibration coupling analysis according to claim 4, characterized in that: The optimal kernel function is obtained by substituting the optimal parameters into the composite kernel function, and a multi-kernel fusion support vector regression model is trained based on the optimal kernel function. The new samples to be predicted are input into the trained multi-core fusion support vector regression model to obtain the final compaction degree detection results, which include: Substituting the optimal parameters into the composite kernel function yields the optimal kernel function, whose expression is as follows: ； in: The optimal kernel function; , , All are optimal parameters; Based on the optimal kernel function and data pairs, a multi-kernel fusion support vector regression model is trained, and new samples to be predicted are then used. The input is fed into the trained multi-core fusion support vector regression model to calculate the single-window predicted compaction degree, as shown in the following expression: ； in: Predict compaction degree using a single window; It is a set of support vectors; , All are Lagrange multipliers; is the feature vector of the sample to be predicted; For the first Support vectors; For bias terms; The final compaction degree is obtained by weighted averaging the compaction degrees predicted from multiple overlapping windows under the same excitation, as expressed below: ； ； in: This represents the final compaction degree; For the first Predicted compaction values ​​for each window; Number of windows; Window weight; The optimal time decay coefficient; For the first Normalized time difference of each window; For indicator functions; Coherent at the window's main frequency; The signal-to-noise ratio weights for the window.

6. The non-destructive testing method for roadbed compaction based on acoustic-vibration coupling analysis according to claim 2, characterized in that: The time-domain features specifically include: Calculate the speed of sound propagation based on the time difference between the peak of the transmitted and received first sound wave signal and the distance between the transmitting and receiving points. Its expression is as follows: ； in: The distance between the transmitting point and the receiving point. The time difference between the peaks; Calculate the signal amplitude attenuation rate based on the initial signal amplitude of the transmitted sound wave and the amplitude of the received signal. Its expression is as follows: ； in: The initial signal amplitude; The amplitude of the received signal; It is a constant; The common logarithm; Damping ratio calculated based on the logarithmic attenuation method of time-domain signal Its expression is as follows: ； in: The amplitude of the first sound wave peak. For the first Each acoustic wave peak amplitude; This is a correction factor.

7. The non-destructive testing method for roadbed compaction based on acoustic-vibration coupling analysis according to claim 6, characterized in that: The frequency domain / coupling features specifically include: Calculate the power spectral density of the acoustic signal With the power spectral density of the vibration signal And determine the specific dominant frequency corresponding to the acoustic spectral peak. Its expression is as follows: ； ； ； ； in: The main frequency is a variable. For the sound wave signal Segmented windowed samples; For the vibration signal Segmented windowed samples; For window functions; The sampling length of each signal segment; The number of segments; The normalization constant for window energy; These are the basis functions of the discrete Fourier transform; As the independent variable; Calculate energy concentration based on the power spectral density of the acoustic signal. This refers to the percentage of energy in the bandwidth near the main frequency relative to the total energy, expressed as follows: ； in: The bandwidth half-width of the variable main frequency; Calculate the cross-spectral power density of the acoustic signal and the vibration signal. Cross-spectral amplitude at the dominant frequency Its expression is as follows: ； ； in: For the first The windowed Fourier transform result of the acoustic wave signal; For the first The windowed Fourier transform result of the segment vibration signal; For complex conjugate; The normalization constant for window energy; Calculate the coherence coefficient between the acoustic signal and the vibration signal near the dominant frequency. Average coherence with the neighborhood of the main frequency Its expression is as follows: ； ； in: Let be the power spectral density of the vibration signal.

8. A non-destructive testing system for roadbed compaction based on acoustic-vibration coupling analysis, characterized in that, The system is applied to the method according to any one of claims 1-7, the system comprising: Feature coupling module (1) is used to collect raw acoustic and vibration data and extract feature vectors to construct acoustic and vibration coupling features; Model building module (2) is used to design a three-core composite kernel function containing time-domain sub-kernel, frequency-domain sub-kernel and coupling sub-kernel based on acoustic-vibration coupling characteristics, and to build a multi-core fusion support vector regression model; The parameter optimization module (3) is used to define the parameter vector to be optimized in the multi-core fusion support vector regression model, construct the minimization objective function and perform parameter optimization to obtain the optimal parameters; The compaction detection module (4) is used to substitute the optimal parameters into the composite kernel function to obtain the optimal kernel function, and to train the multi-kernel fusion support vector regression model based on the optimal kernel function; the new sample to be predicted is input into the trained multi-kernel fusion support vector regression model to obtain the final compaction detection result.

9. A non-destructive testing device for roadbed compaction based on acoustic-vibration coupling analysis, characterized in that: The device includes a processor (5) and a memory (6); The memory (6) is used to store computer program code (61) and to transmit the computer program code (61) to the processor (5). The processor (5) is used to execute the non-destructive testing method for roadbed compaction based on acoustic-vibration coupling analysis according to any one of the instructions in the computer program code (61).

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions, which, when executed on a computer, implement the non-destructive testing method for roadbed compaction based on acoustic-vibration coupling analysis as described in any one of claims 1-7.

Images