Metal material nondestructive testing quantitative analysis method
By combining time-domain analysis, support vector machine classification, simulated acoustic wave propagation model and convolutional neural network, the problem of quantifying ultrasonic signals in complex materials was solved, achieving high-precision defect detection and evaluation, and improving the reliability of non-destructive testing.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-30
- Publication Date
- 2026-04-03
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
Existing nondestructive testing methods struggle to achieve real-time and accurate quantification of ultrasonic signals in complex materials, severely limiting the reliability of defect location and size assessment. This is especially true when material differences and defect types are complex, leading to unstable signal intensity-distance relationships and large errors in defect location and geometric size inversion.
By acquiring echo signals from an ultrasonic probe, and combining time-domain analysis, support vector machine classification, simulated sound wave propagation model, and convolutional neural network, a quantitative attenuation relationship model is constructed, iteratively optimized, and the defect location coordinates are refined. Finally, a size inversion algorithm is integrated to achieve three-dimensional spatial coordinates and high-precision size assessment.
It significantly improves the accuracy and reliability of defect detection in complex materials, enabling efficient defect type identification, location positioning, and size assessment, thereby enhancing the accuracy and reliability of detection.
Smart Images
Figure CN121784142A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of nondestructive testing technology, and in particular to a quantitative analysis method for nondestructive testing of metallic materials. Background Technology
[0002] Non-destructive testing (NDT) of materials is a core technology area for ensuring the safe operation of critical structures in aerospace, rail transportation, and energy equipment. It directly relates to the service life of major equipment and the prevention of sudden accidents, holding an irreplaceable and crucial position in the modern industrial system. Currently, most mainstream NDT methods rely on empirical threshold judgments or offline data analysis, making it difficult to achieve real-time and accurate quantification of the detection signal during its actual propagation. This severely restricts the reliability of defect location and size assessment.
[0003] Different materials, such as metals, composites, and ceramics, have vastly different internal structures. When sound waves propagate through these materials, they undergo complex attenuation, reflection, and refraction phenomena, which vary significantly depending on the material type and its internal state. When sound waves encounter defects such as cracks, pores, or inclusions, they trigger strong scattering, causing unpredictable changes in signal strength and waveform. This high material dependence of propagation characteristics directly leads to an extremely unstable relationship between signal strength and distance along the propagation path, further amplifying errors in the inversion of defect location and geometry. In practical testing, such as ultrasonic testing of a large, operating turbine disk, the echo signal received by the probe is often severely distorted due to the combined effects of local material anisotropy and defect scattering, making it difficult for operators to accurately separate the characteristic changes caused by defects from the mixed signal.
[0004] Because key parameters such as attenuation coefficient, reflectivity, and scattering parameters vary drastically with material and defect type during propagation, and these parameters are mutually coupled and influential, existing methods struggle to establish a reliable quantitative relationship between signal intensity attenuation and propagation distance. Therefore, accurately quantifying the intensity attenuation and scattering changes of ultrasonic signals along the propagation path in complex materials, thereby achieving high-precision reverse estimation of the three-dimensional spatial coordinates and geometric dimensions of defects, has become a critical problem that urgently needs to be addressed in the field of data quantification for nondestructive testing of materials. Summary of the Invention
[0005] This invention provides a method for quantitative analysis of nondestructive testing of metallic materials, mainly comprising: By collecting echo signal data of the ultrasonic probe on the target material, the initial signal intensity and waveform characteristics are extracted using time-domain analysis to obtain preliminary propagation path information; Based on the preliminary propagation path information, support vector machines are used to classify material types and internal defect types, determine the degree of influence of material differences on sound wave propagation, and obtain the classified material parameter set. If the classified material parameter set shows that the attenuation is complex and higher than the preset threshold, the scattering parameters are integrated by simulating the sound wave propagation model to determine the signal distortion compensation coefficient. After obtaining the signal distortion compensation coefficient, a convolutional neural network is used to process the compensated echo signal, extract multi-scale scattering features, and obtain a quantized attenuation relationship model. For the quantized attenuation relationship model, iterative optimization is performed by combining propagation path information to determine the preliminary estimated value of the defect location coordinates; By matching the preliminary estimated value with the signal strength data, if the matching degree is lower than the preset threshold, the complex scattering parameters are adjusted to obtain the coordinates of the refined defect location. Based on the location coordinates of the refined defects, the geometric dimensions of the defects are calculated using a dimensional inversion algorithm to determine the final three-dimensional spatial coordinates and high-precision dimensional evaluation results.
[0006] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects: This invention discloses a quantitative analysis method for nondestructive testing of metallic materials, addressing complex business scenarios involving the identification, location, and dimensional assessment of internal material defects by integrating multi-level signal processing and intelligent algorithms. In scenarios where material differences lead to complex acoustic wave attenuation, this invention first extracts signal features through time-domain analysis, then uses support vector machines to classify materials and defect types to determine the degree of attenuation. When attenuation exceeds a threshold, an acoustic wave propagation model is constructed to calculate compensation coefficients, and a convolutional neural network is used to extract multi-scale scattering features, establishing a quantitative attenuation relationship model. Subsequently, through iterative optimization and scattering parameter adjustment, the defect location coordinates are refined, and finally, a dimensional inversion algorithm is integrated to achieve three-dimensional spatial coordinates and high-precision dimensional assessment. This invention significantly improves the accuracy and reliability of defect detection in complex materials through signal compensation and multi-model collaboration, providing an efficient solution for industrial nondestructive testing. Attached Figure Description
[0007] Figure 1 This is a flowchart of a non-destructive testing and quantitative analysis method for metallic materials according to the present invention.
[0008] Figure 2 This is a schematic diagram of a non-destructive testing and quantitative analysis method for metallic materials according to the present invention.
[0009] Figure 3 This is another schematic diagram of a non-destructive testing and quantitative analysis method for metallic materials according to the present invention. Detailed Implementation
[0010] To further understand the content of this invention, a detailed description of the invention is provided in conjunction with the accompanying drawings and embodiments. The specific embodiments described herein are for illustrative purposes only and are not intended to limit the invention. It should also be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings.
[0011] like Figures 1-3 This embodiment of a nondestructive testing and quantitative analysis method for metallic materials may specifically include: S101. By collecting echo signal data of the ultrasonic probe on the target material, the initial signal intensity and waveform characteristics are extracted using time-domain analysis to obtain preliminary propagation path information.
[0012] Echo signal data collected by an ultrasonic probe at multiple locations on the surface of the target material is acquired. Initial signal intensity peak and waveform characteristic parameters are extracted from the echo signal data. The attenuation amplitude of the echo signal is determined based on the initial signal intensity peak. If the attenuation amplitude exceeds a preset threshold, an internal defect in the material is identified. The time-of-flight difference of the waveform characteristics is calculated using time-of-flight analysis. The propagation path length of the ultrasonic wave in the target material is determined using the time-of-flight difference. The preliminary propagation path information is corrected based on the propagation path length and the initial signal intensity.
[0013] In the process of processing the echo signal data of the target material acquired by the ultrasonic probe using information technology, the echo signal is first digitally recorded using a high-precision data acquisition system at a sampling rate of 100MHz to ensure that subtle changes in the signal are captured. Assuming that the acquired signal duration is 10 microseconds and a total of 1000 sampling points are collected, with the signal amplitude ranging from -5V to 5V, the original signal is preprocessed using software algorithms. Wavelet denoising is used to remove high-frequency noise. The wavelet basis is set to Daubechies 4, and the decomposition level is 5. The signal-to-noise ratio of the denoised signal is calculated to be improved to over 20dB, laying the foundation for subsequent analysis. Next, the system extracts the initial signal strength and waveform features using time-domain analysis. It automatically identifies the initial peak value of the signal, assuming it occurs at the 200th sampling point with an amplitude of 3.2V. By calculating the rise and fall times, the rise time is found to be 0.5 microseconds and the fall time 0.7 microseconds. Simultaneously, the system extracts the signal envelope features and generates an envelope curve using the Hilbert transform algorithm. The calculated envelope peak attenuation rate is approximately 10% per microsecond, which is stored as waveform feature parameters in the database. Subsequently, based on these feature parameters, the system preliminarily calculates the propagation path information using time-domain reflection. Assuming the sound wave propagation speed in the target material is 5000 meters per second, and considering the 1.2 microsecond difference between the arrival time of the initial wave and the reflected wave, the propagation path length is calculated to be 6 millimeters. Combining this with a signal strength attenuation model, an attenuation coefficient of 0.1 per millimeter is set, suggesting a possible 0.6-millimeter defect in the path. The relevant data is automatically used to generate a path diagram and stored in the system's cloud for subsequent in-depth analysis and verification. Through the above series of automated processes, a complete logical chain is formed from signal acquisition to path information extraction, ensuring the accuracy and efficiency of data processing.
[0014] S102. Based on the preliminary propagation path information, support vector machines are applied to classify material types and internal defect types, determine the degree of influence of material differences on sound wave propagation, and obtain the classified material parameter set. The specific calculations are as follows: ; Let m represent the material type obtained from the classification, and S represent the total number of support vectors. This represents the weight coefficient of the m-th class support vector i. Let K represent the label of the m-th class sample, K represent the kernel function, and P represent the initial propagation path information. This represents the support vector of the m-th class. This represents the m-th type of bias term; ; Let d represent the internal defect type obtained from classification, d represent the defect type category, and T represent the total number of support vectors. This represents the weight coefficient of the support vector j of class d. Let K represent the label of the d-th class sample, K represent the kernel function, and Q represent the initial propagation path information. This represents the support vector of class d. This represents the d-th type of bias term.
[0015] The initial propagation path information is input into a Support Vector Machine (SVM) model for training and labeling. The SVM model performs multi-class classification of material types. Simultaneously, the SVM model classifies and identifies internal defect types. The sound wave propagation attenuation coefficient corresponding to each material type is obtained from the classification results. If the material type classification results show significant differences, the propagation impedance value of the corresponding internal defect type is extracted. The impact of material differences on sound wave propagation is calculated based on the propagation attenuation coefficient and propagation impedance value. The path length parameter in the initial propagation path information is adjusted based on the impact magnitude to obtain a corrected set of material parameters.
[0016] In the process of classifying material type and internal defects based on preliminary propagation path information, the system first extracts the calculated propagation path data from the database. Assuming a path length of 8 mm, a sound wave propagation speed of 4500 m / s, and a reflection time difference of 1.8 microseconds, the algorithm automatically calculates the sound wave propagation time in the material to be 1.78 microseconds. Combined with a path attenuation coefficient of 0.08 per mm, the signal strength attenuation ratio is deduced to be 64%. Subsequently, the system inputs these parameters into a support vector machine model, using a radial basis function as the kernel function, setting the penalty parameter C to 10 and the kernel parameter gamma to 0.1. By learning from 500 samples in the training dataset, the system automatically classifies the material type as aluminum alloy and the internal defect type as porosity, achieving a classification confidence level of 85%. Next, the system analyzes the impact of material differences on sound wave propagation. Assuming the density of aluminum alloy is 2.7 g / cm³ and its acoustic impedance is 17.1 MPa·s / m, it compares this with the acoustic impedance of other materials in the database, such as steel (27.2 MPa·s / m). The calculated difference in sound wave reflectivity is 22%. Combining this with the characteristics of pores (a defect type), the system estimates an additional scattering loss of 15% due to pores. Therefore, the combined impact factor of material differences and defects on sound wave propagation is 0.37. Finally, the system automatically integrates the categorized material parameter set, including density, acoustic impedance, reflectivity, and scattering loss, into structured data, stores it on a cloud analysis platform, and links it to a material mechanical property database to generate a material parameter report. The report includes the quantified value of the impact factor (0.37) and a defect distribution probability map, providing data support for subsequent material performance evaluation. Through this automated process, the system achieves a complete logical chain from path information to material classification and impact analysis.
[0017] S103. If the classified material parameter set shows complex attenuation exceeding a preset threshold, then the scattering parameters are integrated using a simulated acoustic wave propagation model to determine the signal distortion compensation coefficient; including... ; c represents the signal distortion compensation coefficient. Let represent the i-th scattering parameter, and N represent the total number of scattering parameters. This formula calculates the compensation coefficient using the scattering parameters extracted from the simulated sound wave propagation model, which is used to offset signal distortion.
[0018] Scattering parameters are extracted using a simulated acoustic wave propagation model. Signal distortion compensation coefficients are calculated from these scattering parameters. The attenuation term in the attenuation parameters is then corrected based on the compensation coefficients. Multipath scattering components are separated from the compensated attenuation parameters. If the multipath scattering component exceeds a preset threshold, the remaining scattering intensity is determined. The impedance matching term is adjusted based on the remaining scattering intensity. The optimized set of material parameters is then obtained.
[0019] When the system detects that the attenuation complexity index of the classified material parameter set exceeds the preset threshold of 0.45, it automatically triggers the acoustic wave propagation simulation module. It extracts the scattering coefficient (0.12), absorption coefficient (0.065), and equivalent defect diameter (1.2 mm) from the parameter set. First, it constructs a three-dimensional mesh model using the finite difference time-domain method, with a mesh resolution of 0.1 mm, simulating the propagation process of a 5 MHz longitudinal wave in the material. Through parallel computation, it iteratively solves the wave equation over 2000 time steps to obtain the scattered field distribution data. Subsequently, the system calculates the contribution of multiple scattering, using the Born approximation combined with the Monte Carlo method to simulate 5000 random acoustic ray paths. The average scattering loss is statistically calculated to be 18.4%, and compared with the single-scattering model, the additional signal distortion caused by multiple scattering is found to be 12.7%. Next, based on the scattered field energy distribution, the main lobe width amplification factor of 1.38 and the side lobe level increase of 9.6 dB are extracted. Combined with the preset signal-to-noise ratio requirements, the compensation filter parameters are fitted using a least-squares optimization algorithm, resulting in a signal distortion compensation factor of 0.72. This factor includes an amplitude recovery factor of 1.25 and a phase correction offset of -0.18 radians. Finally, the system automatically applies the compensation factor to the real-time signal processing link, restoring waveform integrity through convolution operations and reducing the updated attenuation complexity index to 0.39. The comparison data before and after compensation is stored in the database, providing a correction basis for subsequent quantitative defect assessment.
[0020] S104. After obtaining the signal distortion compensation coefficient, a convolutional neural network is used to process the compensated echo signal, extract multi-scale scattering features, and obtain a quantized attenuation relationship model. ; This represents the multi-scale scattering characteristics of the l-th layer. This represents the convolutional kernel of the l-th layer. This represents the compensated echo signal. ReLU represents the bias term, and ReLU represents the activation function; this formula represents the process of obtaining multi-scale scattering features by performing multi-layer convolution operations using a convolutional neural network.
[0021] Multi-scale scattering features are obtained by performing multi-layer convolution operations using a convolutional neural network on the compensated echo signal. Pooling downsampling is then performed on these multi-scale scattering features to determine the feature scale division. High-frequency scattering components and low-frequency attenuation components are separated from the feature scale division, resulting in a set of separated scattering features. The attenuation contribution value at each scale is calculated for the separated scattering feature set to determine the quantized attenuation contribution distribution. If the proportion of high-frequency components in the quantized attenuation contribution distribution is higher than a preset threshold, the high-frequency scattering components are suppressed, resulting in a set of suppressed scattering features. An attenuation relationship mapping function is constructed using the suppressed scattering feature set to obtain a quantized attenuation relationship model. This quantized attenuation relationship model is then used to inversely correct the compensated echo signal, resulting in a set of corrected attenuation parameters.
[0022] After acquiring the signal distortion compensation coefficients, the system automatically initiates the signal processing flow. First, the compensated echo signal is preprocessed with a sampling rate of 10 MHz and a signal length of 1024 sampling points. The time-domain signal is converted to the frequency domain using a Fast Fourier Transform (FFT), extracting the main components within the 2-8 MHz frequency range and filtering out low-frequency noise below 0.5 MHz to obtain the spectral energy distribution. The main frequency peak is located at 4.3 MHz, accounting for 62.7% of the energy. Subsequently, the system inputs the processed signal into a pre-trained convolutional neural network model. This model contains 5 convolutional layers and 3 pooling layers, with each convolutional kernel size of 3x3, a stride of 1, a pooling window of 2x2, and ReLU activation. Feature extraction is performed on the signal to capture scattering features at different scales, outputting a 128x64 dimension feature map containing local scattering intensity distribution and global energy attenuation patterns. The system further normalizes the feature map to a mean of 0.32 and a standard deviation of 0.15 to ensure feature stability. Next, the system uses a fully connected layer to map multi-scale features to an attenuation relationship model. A gradient descent optimization algorithm is employed with a learning rate of 0.001 and 500 iterations to calculate the parameters of the quantized attenuation relationship model. The correlation coefficient between the attenuation exponent and scattering intensity is 0.85, and the model prediction error is controlled within 5.2%. To ensure logical integrity, the system compares the model output with a database of internal material defects, setting the defect density parameter to 0.08 per cubic centimeter. Combined with historical data analysis, the model's adaptability to different material thicknesses (ranging from 5 to 20 millimeters) is verified. Finally, an attenuation relationship report is generated and automatically stored in the system's cloud, providing data support for subsequent material performance evaluation.
[0023] S105. For the quantized attenuation relationship model, iterative optimization is performed by combining propagation path information to determine the preliminary estimated value of the defect location coordinates.
[0024] Based on the quantized attenuation relationship model and propagation path information, the path loss correction value is calculated to obtain the initial path adjustment parameters. The propagation distance parameters are then corrected using these initial path adjustment parameters to determine the attenuation coefficient after distance compensation. The attenuation coefficient after distance compensation is then incorporated into the medium inhomogeneity for weighted calculation to obtain the inhomogeneity-corrected model parameters. The path angle deviation is calculated for the inhomogeneity-corrected model parameters to determine the angle-corrected propagation path. The attenuation coefficient adjustment is extracted from the angle-corrected propagation path to obtain the adjusted quantized attenuation relationship model. The adjusted quantized attenuation relationship model is then used to calculate the coordinate error distribution and determine the preliminary estimate of the defect location coordinates.
[0025] ; Let N represent the coordinate error distribution, N represent the number of sampling points, and Q_a represent the adjusted quantization attenuation model. This represents the distance of the i-th path. This represents the corresponding observation attenuation value, used to determine the preliminary estimate of the defect location coordinates and optimize the convergence conditions.
[0026] If the coordinate error distribution exceeds the optimization convergence condition, the path loss correction value is returned to perform iterative optimization calculation to obtain the converged defect location coordinates.
[0027] Upon receiving the echo signal, the system automatically initiates the propagation path analysis process. First, the signal undergoes time-domain preprocessing with a sampling rate of 12 MHz and a signal length of 2048 sampling points. The signal is decomposed to four levels using wavelet transform, and the Daubechies wavelet basis function is selected to extract sub-band energy in the 3-9 MHz frequency band. Background noise below 0.3 MHz is filtered out, yielding the energy distribution. The peak frequency of the main sub-band is 5.1 MHz, accounting for 58.4% of the energy. Subsequently, the system combines the propagation path information with a sound wave propagation model to calculate the refraction and reflection coefficients of the signal in the material. The material sound velocity is set to 5900 m / s, and the path length ranges from 10 to 25 mm. A path feature matrix with dimensions of 256x128 is generated, containing propagation time delay and intensity attenuation information. Next, the system inputs the path feature matrix into a recurrent neural network. The model contains two LSTM layers with 64 units per layer and a time step of 32. The tanh activation function is used to capture the dynamic features of path dependence. The output feature vector has a dimension of 64, a mean of 0.28, and a standard deviation of 0.12. The system uses a Bayesian optimization algorithm to iteratively update the quantization decay relationship model with a learning rate of 0.002 and 400 iterations. The correlation coefficient between the decay coefficient and the path length is calculated to be 0.79, and the model error is controlled within 6.1%. To estimate the defect location coordinates, the system combines the propagation time difference and path geometric constraints, employing a triangulation algorithm. The defect depth is set to a range of 3 to 15 mm, and preliminary coordinate estimates are obtained with a lateral error of 0.45 mm and a longitudinal error of 0.62 mm. The system compares the estimated results with a material defect database, with a defect density of 0.06 g / cm³. It verifies the robustness of the model under different material densities, ranging from 2.5 to 7.8 g / cm³. Finally, it generates a defect location report, which is automatically uploaded to the cloud database to provide a reference for subsequent quality inspection.
[0028] S106. By matching the preliminary estimated value with the signal strength data, if the matching degree is lower than the preset threshold, adjust the scattering complexity parameters to obtain the coordinates of the refined defect location. ; This represents the correction amount for complex scattering parameters. These represent the support vector coefficients in a support vector machine regression model. Represents the kernel function. denoted as the input feature vector containing the initial settings of complex scattering parameters, b represents the model bias term, and m represents the number of support vectors.
[0029] A matching degree is obtained by comparing the preliminary estimated value with the signal strength data. Based on the matching degree, a preset threshold is compared to determine whether to initiate scattering complexity parameter adjustment. If the matching degree is lower than the preset threshold, the initial setting value of the scattering complexity parameter is extracted. A support vector machine is used to perform regression prediction on the initial setting value of the scattering complexity parameter to determine the correction amount. The scattering complexity parameter is updated using the correction amount to obtain the updated scattering complexity parameter. The defect location coordinates are recalculated using the updated scattering complexity parameter to obtain the refined defect location coordinates. Residual analysis is performed based on the refined defect location coordinates and the signal strength data to determine the residual distribution.
[0030] After obtaining the initial defect coordinates, the system automatically extracts the peak amplitude sequence of the corresponding echo signal, calculating an average intensity of -12.3 dB and a standard deviation of 2.1 dB. The initial coordinates are then substituted into the scattering model, and the Kirchhoff approximation is used to calculate the theoretical scattering intensity. With the equivalent defect radius set at 0.8 mm and the material attenuation coefficient at 0.035 neper / mm, the theoretical intensity is calculated to be -10.7 dB. If the measured intensity deviates from the theoretical value by more than 1.6 dB, an adaptive adjustment process is initiated: a complex scattering parameter α (initially set to 1.2) is introduced, and gradient descent is used for iterative optimization with a step size of 0.0005. The objective function is the mean square error between the measured and theoretical intensities. After 12 iterations, α converges to 1.47. At this point, the recalculated scattering intensity is -11.9 dB, reducing the deviation from the measured value to 0.4 dB. The system then uses the updated α correction to adjust the equivalent path length on the propagation path, obtaining a refined longitudinal depth of 7.82 mm (previously estimated at 8.1 mm) and a lateral position of 14.36 mm (previously 14.2 mm). Finally, the refined coordinates are matched and verified with the phased array focusing data of the multi-probe array, with a correlation coefficient of 0.92. After confirming the reliability of the position, it is automatically saved to the local defect file and marked as "refined" for subsequent use by the 3D reconstruction module.
[0031] S107. Based on the location coordinates of the refining defects, the geometric dimensions of the defects are calculated using the dimensional inversion algorithm to determine the final three-dimensional spatial coordinates and high-precision dimensional evaluation results.
[0032] By refining coordinate and position data and employing a pre-established mapping model, the initial three-dimensional spatial distribution is derived, yielding preliminary spatial coordinate information. Based on this preliminary spatial coordinate information, and combined with geometric dimension calculation methods, multi-dimensional data decomposition processing is performed to determine the preliminary size range of the defect. For this preliminary size range, a support vector machine model is used to classify the size data, obtaining a more accurate geometric dimension description. Using this accurate geometric dimension description, combined with the three-dimensional spatial distribution, coordinate correction calculations are performed to obtain high-precision spatial coordinate data. Based on this high-precision spatial coordinate data, if the corrected coordinate offset exceeds a preset threshold, the position data undergoes secondary processing to determine if it meets the size evaluation criteria. If the position data meets the size evaluation criteria, multi-dimensional data integration is used to generate the final evaluation result, determining the complete spatial description of the defect. Based on the final evaluation result, data storage and formatting processing is performed on the complete spatial description of the defect to obtain structured information suitable for subsequent analysis.
[0033] After acquiring the coordinates of the refining defect location, the system automatically invokes the size inversion module. First, it extracts the scattering cross-section sequence of the defect at different incident angles from the multi-angle phased array echo data, calculating the average equivalent scattering cross-section to be 0.42 mm². Then, a size inversion algorithm based on the Born approximation is used to construct a three-dimensional mesh model centered on the refining coordinates. The defect is assumed to be an ellipsoid with initial major axis a = 1.2 mm, minor axis b = c = 0.7 mm, and the material acoustic impedance ratio set to 1.15. The objective function, i.e., the root mean square error between the measured and theoretical scattering cross-sections, is optimized using the Levenberg-Marquardt nonlinear least squares method. The initial iteration step size is 0.001, and after 18 iterations, the parameters converge to a = 1.45 mm, b = 0.82 mm, and c = 0.79 mm. At this point, the root mean square error between the theoretical and measured scattering cross-sections decreases to 0.031 mm². The system then calculates the defect volume as 0.395 mm³ based on the optimized ellipsoid parameters. Through coordinate transformation, it aligns the ellipsoid center with the refining coordinates, obtaining the final three-dimensional spatial coordinates (x=14.36 mm, y=7.82 mm, z=0.00 mm) and the major axis vector (0.92, 0.28, 0.26), thus determining the defect to be an approximately elliptical, flattened defect. Finally, the size inversion results and refining coordinates are saved together to the defect file, marked as "size assessment completed," and a high-precision three-dimensional defect model is generated for subsequent damage assessment and life prediction modules.
[0034] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.
Claims
1. A non-destructive testing and quantitative analysis method for metallic materials, characterized in that, The method includes: By collecting echo signal data of the ultrasonic probe on the target material, the initial signal intensity and waveform characteristics are extracted using time-domain analysis to obtain preliminary propagation path information; Based on the preliminary propagation path information, support vector machines are used to classify material types and internal defect types, determine the degree of influence of material differences on sound wave propagation, and obtain the classified material parameter set. If the classified material parameter set shows that the attenuation is complex and higher than the preset threshold, the scattering parameters are integrated by simulating the sound wave propagation model to determine the signal distortion compensation coefficient. After obtaining the signal distortion compensation coefficient, a convolutional neural network is used to process the compensated echo signal, extract multi-scale scattering features, and obtain a quantized attenuation relationship model. For the quantized attenuation relationship model, iterative optimization is performed by combining propagation path information to determine the preliminary estimated value of the defect location coordinates; By matching the preliminary estimated value with the signal strength data, if the matching degree is lower than the preset threshold, the complex scattering parameters are adjusted to obtain the coordinates of the refined defect location. Based on the location coordinates of the refined defects, the geometric dimensions of the defects are calculated using a dimensional inversion algorithm to determine the final three-dimensional spatial coordinates and high-precision dimensional evaluation results.
2. The method for quantitative analysis of nondestructive testing of metallic materials according to claim 1, characterized in that, The process involves acquiring echo signal data from the ultrasonic probe on the target material, extracting the initial signal intensity and waveform characteristics using time-domain analysis, and obtaining preliminary propagation path information, including: Acquire echo signal data from multiple locations on the surface of the target material using an ultrasonic probe; Extract the initial signal strength peak value and waveform characteristic parameters from the echo signal data; The attenuation amplitude of the echo signal is determined based on the initial signal strength peak value; If the attenuation exceeds a preset threshold, it is determined that there is an internal defect in the material. The time-of-flight difference of waveform characteristics is calculated using time-domain analysis methods; The propagation path length of ultrasonic waves in the target material is determined by the time difference of flight. The initial propagation path information is corrected based on the propagation path length and initial signal strength.
3. The method for quantitative analysis of nondestructive testing of metallic materials according to claim 1, characterized in that, Based on the preliminary propagation path information, a support vector machine is applied to classify material type and internal defect type, determine the degree of influence of material differences on sound wave propagation, and obtain the classified material parameter set; the specific calculation is as follows: ； Let m represent the material type obtained from the classification, and S represent the total number of support vectors. This represents the weight coefficient of the m-th class support vector i. Let K represent the label of the m-th class sample, K represent the kernel function, and P represent the initial propagation path information. This represents the support vector of the m-th class. This represents the m-th type of bias term; ； Let d represent the internal defect type obtained from the classification, d represent the defect type category, and T represent the total number of support vectors. This represents the weight coefficient of the support vector j of class d. Let K represent the label of the d-th sample, K represent the kernel function, and Q represent the initial propagation path information. This represents the support vector of class d. This represents the type d bias term; This includes: training and labeling the support vector machine model based on the initial propagation path information; A support vector machine model is used to perform multi-class classification of material types. The internal defect types are simultaneously classified and identified using a support vector machine model. Obtain the sound wave propagation attenuation coefficient corresponding to the material type from the classification results; If the material type classification results show significant differences, then extract the propagation impedance value of the corresponding internal defect type; The influence of material differences on sound wave propagation is calculated based on the propagation attenuation coefficient and propagation impedance value. By adjusting the path length parameter in the initial propagation path information based on the magnitude of the impact, a corrected set of material parameters is obtained.
4. The method for quantitative analysis of nondestructive testing of metallic materials according to claim 1, characterized in that, If the classified material parameter set shows complex attenuation exceeding a preset threshold, then the scattering parameters are integrated using a simulated sound wave propagation model to determine the signal distortion compensation coefficient, including: ； c represents the signal distortion compensation coefficient. Let represent the i-th scattering parameter, and N represent the total number of scattering parameters; this formula calculates the compensation coefficient using the scattering parameters extracted from the simulated sound wave propagation model, which is used to offset signal distortion; Scattering parameters were extracted using a simulated sound wave propagation model; Calculate the signal distortion compensation coefficient using scattering parameters; The attenuation term in the attenuation parameter is corrected based on the compensation coefficient; Separate the multipath scattering components from the compensated attenuation parameters; If the multipath scattering component is higher than a preset threshold, then the remaining scattering intensity is determined; The impedance matching term is adjusted by the residual scattering intensity; Obtain the optimized set of material parameters.
5. The method for quantitative analysis of nondestructive testing of metallic materials according to claim 1, characterized in that, After obtaining the signal distortion compensation coefficient, a convolutional neural network is used to process the compensated echo signal, extract multi-scale scattering features, and obtain a quantized attenuation relationship model; the formula is as follows: ； This represents the multi-scale scattering characteristics of the l-th layer. This represents the convolutional kernel of the l-th layer. This represents the compensated echo signal. ReLU represents the bias term, and ReLU represents the activation function. This formula represents the process of obtaining multi-scale scattering features by performing multi-layer convolution operations using a convolutional neural network; This includes: using a convolutional neural network to perform multi-layer convolution operations based on the compensated echo signal to obtain multi-scale scattering features; Pooling downsampling is performed using multi-scale scattering features to determine the feature scale division; The high-frequency scattering component and the low-frequency attenuation component are separated from the characteristic scale division to obtain the separated scattering feature set; Calculate the attenuation contribution value at each scale for the separated scattering feature set, and determine the quantized attenuation contribution distribution; If the proportion of high-frequency components in the quantized attenuation contribution distribution is higher than a preset threshold, then the high-frequency scattering components are suppressed to obtain the suppressed scattering feature set. By constructing an attenuation relationship mapping function from the suppressed scattering feature set, a quantized attenuation relationship model is obtained; The echo signal after compensation is corrected by using a quantized attenuation relationship model to obtain the set of corrected attenuation parameters.
6. The method for quantitative analysis of nondestructive testing of metallic materials according to claim 1, characterized in that, The preliminary estimate of the defect location coordinates, obtained by iteratively optimizing the quantized attenuation relationship model in conjunction with propagation path information, includes: The path loss correction value is calculated based on the quantized attenuation relationship model and the propagation path information to obtain the initial path adjustment parameters; The propagation distance parameter is corrected by adjusting the initial path parameter, and the attenuation coefficient after distance compensation is determined. By incorporating the attenuation coefficient after distance compensation into the medium inhomogeneity and performing a weighted calculation, the model parameters after inhomogeneity correction are obtained. Calculate the path angle deviation for the model parameters after non-uniform correction, and determine the propagation path after angle correction; The attenuation coefficient adjustment is extracted from the propagation path after angle correction to obtain the adjusted quantized attenuation relationship model; The coordinate error distribution is calculated using the adjusted quantization attenuation relationship model to determine the preliminary estimated value of the defect location coordinates; ； Let N represent the coordinate error distribution, N represent the number of sampling points, and Q_a represent the adjusted quantization attenuation model. This represents the distance of the i-th path. This represents the corresponding observation attenuation value, used to determine the preliminary estimate of the defect location coordinates and optimize the convergence conditions; If the coordinate error distribution exceeds the optimization convergence condition, the path loss correction value is returned to perform iterative optimization calculation to obtain the converged defect location coordinates.
7. The method for quantitative analysis of nondestructive testing of metallic materials according to claim 1, characterized in that, The process involves matching the preliminary estimated value with the signal strength data. If the matching degree is lower than a preset threshold, the scattering complexity parameters are adjusted to obtain the coordinates of the refined defect location. ； This represents the correction amount for complex scattering parameters. Represents the support vector coefficients in a support vector machine regression model. Represents the kernel function. denoted as the input feature vector containing the initial settings of complex scattering parameters, b represents the model bias term, and m represents the number of support vectors; include: The matching degree value is obtained by comparing the preliminary estimate with the signal strength data; Based on the comparison of the matching degree value with a preset threshold, determine whether to initiate the adjustment of complex scattering parameters. If the matching degree value is lower than the preset threshold, the initial setting value of the complex scattering parameters is extracted; A support vector machine is used to perform regression prediction on the initial set values of the scattering complex parameters to determine the correction amount of the scattering complex parameters; The scattering complexity parameters are updated by the correction amount of the scattering complexity parameters to obtain the updated scattering complexity parameters; The defect location coordinates are recalculated using the updated scattering complexity parameters to obtain the refined defect location coordinates; Residual analysis is performed based on the location coordinates of refining defects and signal strength data to determine the residual distribution.
8. The method for quantitative analysis of nondestructive testing of metallic materials according to claim 1, characterized in that, The process of calculating the defect's geometric dimensions based on the refined defect's location coordinates, using a dimensional inversion algorithm, and determining the final three-dimensional spatial coordinates and high-precision dimensional evaluation results includes: By refining the coordinate and position data and using a pre-established mapping model, the initial three-dimensional spatial distribution is derived, and preliminary spatial coordinate information is obtained. Based on the preliminary spatial coordinate information and combined with the geometric dimension calculation method, multidimensional data decomposition processing is performed to determine the preliminary size range of the defect; For the initial size range, a support vector machine model is used to classify the size data to obtain a more accurate description of the geometric dimensions; By accurately describing the geometric dimensions and combining them with the three-dimensional spatial distribution, coordinate correction calculations are performed to obtain high-precision spatial coordinate data. Based on high-precision spatial coordinate data, if the corrected coordinate offset exceeds the preset threshold, the position data will be processed a second time to determine whether it meets the size evaluation standard. If the location data meets the size evaluation criteria, the final evaluation result is generated through multi-dimensional data integration to determine the complete spatial description of the defect; Based on the final evaluation results, a complete spatial description of the defects is obtained, and data storage and formatting processes are performed to acquire structured information that can be used for subsequent analysis.