A multi-level decision-making method and system for metal surface damage feature recognition

Abstract

The present invention belongs to the technical field of metal material surface damage identification, and provides a metal surface damage feature identification method and system with multi-level decision-making. The identification method includes extracting geometric features, frequency domain features, and fractal features from a three-dimensional image of a metal material damaged area sample; fusing the extracted geometric features, frequency domain features, and fractal number features to obtain a fused feature; processing the fractal feature using a first classifier, processing the geometric features and frequency domain features using a second classifier, and processing the fused feature using a third classifier to obtain the macroscopic category, microscopic category, and feature category of the metal material damaged area sample, respectively; determining the category and detailed features of the damaged area based on the correlation and classification confidence of the features corresponding to the macroscopic category, microscopic category, and feature category of the metal material damaged area sample, thereby achieving high-precision identification and classification of the macroscopic and microscopic features of the metal material surface damage.

Description

Technical Field

[0001] The present invention belongs to the technical field of metal material surface damage identification, and in particular relates to a metal surface damage feature identification method and system with multi-level decision-making. Background Art

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

[0003] The service performance of metal materials is crucial to product quality and safe operation. However, in practical applications, metal materials often fracture due to fatigue, corrosion, overload, and other factors, leading to equipment failure or accidents. Therefore, studying the characteristics of metal surface damage, revealing the material's failure behavior and mechanisms, and predicting its mechanical properties are of great significance for materials science research and engineering applications.

[0004] Traditional analysis of surface damage characteristics of metal components relies primarily on two-dimensional image processing technology, which uses microscopic image observation to perform qualitative macroscopic and microscopic analysis. However, this method has limitations in handling complex damage situations and revealing the details and irregularities of damage characteristics. It also struggles to fully reflect the distribution of internal defects and microcracks in the material, resulting in insufficient accuracy and reliability in the analysis results.

[0005] To address the inaccuracies and reliability issues associated with these two-dimensional image processing techniques, three-dimensional (3D) topography reconstruction and fractal analysis techniques are gaining increasing attention. Three-dimensional stereomicroscopy can characterize surface damage in metal materials from multiple dimensions. However, existing 3D fractal analysis methods still have room for improvement in computational accuracy, feature extraction, and classification algorithms, making them difficult to meet the needs of analyzing damaged areas on complex surfaces. Summary of the Invention

[0006] In order to solve the above technical problems, the present invention provides a metal surface damage feature recognition method and system with multi-level decision-making, which can achieve high-precision recognition and classification of macroscopic and microscopic features of metal material surface damage.

[0007] In order to achieve the above object, the present invention adopts the following technical solutions:

[0008] A first aspect of the present invention provides a multi-level decision-making method for identifying metal surface damage features.

[0009] In one or more embodiments, a multi-level decision-making method for identifying metal surface damage features is provided, including:

[0010] Obtain a three-dimensional image of a sample of a damaged area of ​​a metal material and extract corresponding geometric features, frequency domain features, and fractal features from it;

[0011] The extracted geometric features, frequency domain features and fractal number features are fused to obtain fused features;

[0012] Using a first classifier to process the fractal features, using a second classifier to process the geometric features and frequency domain features, and using a third classifier to process the fusion features, thereby obtaining macroscopic categories, microscopic categories, and feature categories of samples of damaged areas of metal materials, respectively;

[0013] Based on the correlation and classification confidence of the features corresponding to the macro-category, micro-category and feature category of the damaged area samples of metal materials, the category and detailed features of the damaged area are finally determined.

[0014] As an implementation method, the process of obtaining the fusion feature is as follows:

[0015] According to different fractal number features, the corresponding geometric features and frequency domain features are spliced ​​to obtain the corresponding first fusion feature vector;

[0016] Each first fusion feature vector is then concatenated with its corresponding fractal number feature to obtain the final fusion feature.

[0017] As an implementation method, corresponding weights are assigned according to the importance of geometric features, frequency domain features, and fractal number features, and fused features are obtained through weighted fusion.

[0018] As an implementation method, a pre-trained multimodal fusion network is used to fuse geometric features, frequency domain features, and fractal number features to obtain the final fusion features.

[0019] As an implementation method, the process of extracting geometric features from a three-dimensional image of a sample of a damaged area of ​​a metal material is as follows:

[0020] Perform longitudinal and annular / cross-sectional cutting on the three-dimensional image of the damaged area sample of the metal material;

[0021] Obtain the outermost edge segmentation curve on each cutting surface of the longitudinal cutting and the annular cross-section cutting;

[0022] The geometric characteristics of the outermost edge segmentation curve on each cutting surface are calculated, including curvature, degree of change, smoothness, arithmetic mean roughness, root roughness, maximum height and ten-point average roughness.

[0023] As an implementation method, the process of extracting frequency domain features from a three-dimensional image of a sample of a damaged area of ​​a metal material is as follows:

[0024] Perform longitudinal and annular cross-section cutting on the three-dimensional image of the damaged area sample of the metal material;

[0025] Obtain the outermost edge segmentation curve on each cutting surface of the longitudinal cutting and the annular cross-section cutting;

[0026] Each outermost edge segmentation curve is converted from the spatial domain to the frequency domain to obtain the corresponding frequency domain features.

[0027] As an embodiment, the fractal features include the number of fractals corresponding to the surface of the sample of the damaged area of ​​the metal material covered by boxes of different shapes and the fitting coefficient of the number of boxes at different scales as the scale changes; the process is:

[0028] Use boxes of different shapes to cover the sample surface of the current metal material damage area and calculate the corresponding fractal number;

[0029] Calculate the fitting coefficient of the number of boxes at different scales as the scale changes.

[0030] The second aspect of the present invention provides a metal surface damage feature recognition system with multi-level decision-making.

[0031] In one or more embodiments, a multi-level decision-making metal surface damage feature recognition system includes:

[0032] A feature extraction module is used to obtain a three-dimensional image of a sample of a damaged area of ​​a metal material and extract corresponding geometric features, frequency domain features, and fractal features therefrom;

[0033] Feature fusion module, which is used to fuse the extracted geometric features, frequency domain features and fractal number features to obtain fused features;

[0034] a multi-level classification module, which is used to process the fractal features using a first classifier, process the geometric features and frequency domain features using a second classifier, and process the fusion features using a third classifier, thereby obtaining macroscopic categories, microscopic categories, and feature categories of the damaged area samples of the metal material, respectively;

[0035] The comprehensive decision-making module is used to determine the category and detailed features of the damaged area based on the correlation and classification confidence of the features corresponding to the macro category, micro category and feature category of the damaged area samples of the metal material.

[0036] As an implementation method, in the feature fusion module, the process of obtaining the fusion feature is as follows:

[0037] According to different fractal number features, the corresponding geometric features and frequency domain features are spliced ​​to obtain the corresponding first fusion feature vector;

[0038] Each first fusion feature vector is then concatenated with its corresponding fractal number feature to obtain the final fusion feature.

[0039] As an implementation mode, in the feature fusion module, corresponding weights are assigned according to the importance of geometric features, frequency domain features, and fractal number features, and fused features are obtained through weighted fusion.

[0040] As an implementation method, in the feature fusion module, a pre-trained multimodal fusion network is used to fuse geometric features, frequency domain features, and fractal number features to obtain the final fusion features.

[0041] A third aspect of the present invention provides a computer-readable storage medium.

[0042] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps in the multi-level decision-making metal surface damage feature identification method as described above.

[0043] A fourth aspect of the present invention provides an electronic device.

[0044] An electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the steps in the multi-level decision-making method for identifying metal surface damage characteristics as described above are implemented.

[0045] Compared with the prior art, the present invention has the following beneficial effects:

[0046] (1) The present invention combines multi-morphological fractal analysis, geometric feature extraction and frequency domain analysis to perform intelligent identification and classification of surface damage of metal materials. It comprehensively considers the macroscopic and microscopic features of the damaged area on the surface of metal materials, and then combines machine learning and deep learning algorithms to preliminarily identify macroscopic categories, microscopic categories and feature categories. Finally, a comprehensive decision is made on the macroscopic categories, microscopic categories and feature categories, achieving high-precision identification and classification of macroscopic and microscopic features of surface damage of metal materials, thereby providing strong technical support for the research and engineering application of failure behavior and mechanism of metal materials.

[0047] (2) The present invention uses a variety of morphological boxes and a multi-morphological fine covering method to achieve high-precision fractal number calculation of the damaged area on the surface of metal materials, and reveals the complexity and irregularity of the damaged area in the macroscopic aspect; in the microscopic aspects such as feature details and crack identification, the present invention further extracts the frequency domain and geometric characteristics of the material surface damage through Fourier transform and multi-scale geometric analysis technology, and conducts in-depth analysis of the microscopic details, thereby providing an accurate data basis for the comprehensive decision-making results, improving the accuracy of the metal surface damage feature identification results, and being beneficial for improving the accuracy of the judgment of the failure behavior of metal materials and the accuracy of predicting their mechanical properties. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] The accompanying drawings, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0049] Figure 1 1 is a flow chart of a multi-level decision-making method for identifying metal surface damage characteristics according to an embodiment of the present invention;

[0050] Figure 2 is a schematic diagram of the process of fusing features according to an embodiment of the present invention;

[0051] Figure 3 2 is a schematic structural diagram of a metal surface damage feature recognition system with multi-level decision-making according to an embodiment of the present invention;

[0052] Figure 4 is a schematic diagram of an electronic device according to an embodiment of the present invention;

[0053] Figure 5 is a schematic diagram of annular sections with different radii;

[0054] Figure 6 It is a schematic diagram of multiple cross-section cutting and cross-sectional cutting at the center of the circle. DETAILED DESCRIPTION

[0055] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0056] It should be noted that the following detailed descriptions are illustrative and intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which the present invention belongs.

[0057] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, they indicate the presence of features, steps, operations, devices, components and / or combinations thereof.

[0058] Figure 1 FIG. 1 is a flow chart of a multi-level decision-making method for identifying metal surface damage characteristics in an embodiment of the present invention. Figure 1 The metal surface damage feature recognition method with multi-level decision-making in this embodiment may include:

[0059] S101, obtaining a three-dimensional image of a sample of a damaged area of ​​a metal material, and extracting corresponding geometric features, frequency domain features, and fractal features therefrom;

[0060] S102, fusing the extracted geometric features, frequency domain features, and fractal number features to obtain fused features;

[0061] S103, using a first classifier to process the fractal features, using a second classifier to process the geometric features and frequency domain features, and using a third classifier to process the fusion features, to obtain macroscopic categories, microscopic categories, and feature categories of the damaged area samples of the metal material, respectively;

[0062] S104, based on the correlation and classification confidence of the features corresponding to the macro-category, micro-category and feature category of the metal material damage area sample, the category and detailed features of the damage area are finally determined.

[0063] This embodiment combines multi-morphological fractal analysis, geometric feature extraction, and frequency domain analysis to perform intelligent identification and classification of surface damage of metal materials. It comprehensively considers the macroscopic and microscopic features of the damaged area on the surface of the metal material, and then combines machine learning and deep learning algorithms to preliminarily identify macroscopic categories, microscopic categories, and feature categories. Ultimately, a comprehensive decision is made on the macroscopic categories, microscopic categories, and feature categories, achieving high-precision identification and classification of macroscopic and microscopic features of surface damage of metal materials, thereby providing strong technical support for the research and engineering application of metal material failure behavior and mechanism.

[0064] In step S101, the morphology of the damaged area can be observed through a three-dimensional stereo microscope. The three-dimensional stereo microscope can provide stereoscopic vision and can capture cross-sectional images in any direction, so as to more intuitively observe and analyze the damaged area. It has the advantages of high resolution, large field of view and large depth of field, and can capture subtle features and structures of the damaged surface under conditions of a large field of view.

[0065] After acquiring 3D image data of the damaged area on the metal surface, the image data is preprocessed, including image cropping, grayscale conversion, and binarization. These preprocessing steps help improve the accuracy and effectiveness of subsequent calculations.

[0066] In step S101, the fractal features include the fractal numbers corresponding to the surface of the sample of the damaged area of ​​the metal material covered by boxes of different shapes and the fitting coefficients of the number of boxes at different scales as the scales change.

[0067] It should be noted here that the fractal dimension calculation has the following specific advantages:

[0068] A material's mechanical properties (such as tensile strength, yield strength, and impact toughness) are closely related to its fractal dimension. By analyzing the fractal dimension of a material's damaged surface, it's possible to predict the material's mechanical properties in practical applications. For example, materials with higher fractal dimensions typically have lower tensile and yield strengths due to the potential for more defects and microcracks within them.

[0069] Materials with higher fractal dimensions generally have better impact toughness because they can more effectively absorb and disperse impact energy. Fractal dimension can be used to study the fracture mechanisms of materials. For example, by analyzing the fractal dimension of a material's fracture surface under different conditions (such as temperature, stress, and environment), we can understand the material's fracture process and fracture pattern. Fractal dimension can help distinguish between brittle and ductile fractures and reveal microstructural changes during the fracture process.

[0070] In industrial production, fractal dimension can be used for quality control and failure analysis. By examining the fractal dimension of damaged areas on a product's surface, potential defects can be identified and appropriate measures can be taken to repair or improve the product. Furthermore, when material failure occurs, fractal dimension can help analyze the cause of the failure, providing a reliable scientific basis.

[0071] In step S101, the fractal feature extraction process is as follows:

[0072] Use boxes of different shapes to cover the sample surface of the current metal material damage area and calculate the corresponding fractal number;

[0073] Calculate the fitting coefficient of the number of boxes at different scales as the scale changes.

[0074] Specifically, in this embodiment, cubes of different sizes and shapes are used to cover the damaged surface. First, for example, a square box is selected. A larger box size δ = L is selected first, and then the box size is gradually reduced to cover the damaged surface at different scales. Commonly used box sizes can be reduced proportionally, for example, δ = L / 2, δ = L / 4, δ = L / 8, etc., where L is the maximum length of the damaged surface. The specific implementation steps are as follows:

[0075] (1) Initial partitioning: Divide the entire space into large cubic boxes with side length δmax.

[0076] (2) Recursive segmentation: Recursively segment each cubic box, reducing the box side length by half each time until the minimum box size δmin is reached.

[0077] (3) Calculate the number of boxes covered: At each recursive level, calculate the number of boxes required to completely cover the damaged surface and record the number of boxes N (δ i ).

[0078] The recursive segmentation algorithm formula is expressed as:

[0079]

[0080] Where H(·) is the Heaviside function, I(x, y, z) is the value of the fracture surface at point (x, y, z), and B(j, k, l) represents the (j, k, l)th box.

[0081] The Heaviside function, commonly denoted H(x), is a step function used in various applications in mathematics and engineering. It is defined as follows:

[0082]

[0083] In the present invention, N cube (δ i ) represents the square fractal number fitting. i Represents a box of the i-th scale.

[0084] In order to further improve the calculation accuracy, a fractal density method is proposed. Fractal density is defined as the density of fracture surface points contained in each box. The specific steps are as follows:

[0085] 1): Calculate the density of each box: For each box, calculate the number of fracture surface points contained in it and normalize it to the density value.

[0086] 2): Density threshold screening: Set a density threshold and only count the number of boxes with a density greater than the threshold.

[0087] 3): Multi-scale density calculation: At different scales, calculate the number of boxes corresponding to different density thresholds to obtain multi-scale fractal density. Also recorded as: cube (δ i ) represents the square fractal number fitting. i Represents a box of the i-th scale.

[0088] The fractal density formula is expressed as:

[0089]

[0090] Among them, ρ th is the density threshold, and V(B(i, j, k)) is the box volume.

[0091] ρ(δ): fractal density at scale δ; N(δ): total number of boxes at scale δ; B(i, j, k): (i, j, k)th box; I(x, y, z): value of the surface of the damaged area at point (x, y, z) (if there is a point on the surface of the damaged area, then I(x, y, z) = 1, otherwise I(x, y, z) = 0); V(B(i, j, k)): volume of the (i, j, k)th box, usually δ 3 ρ th : density threshold; H(·): Heaviside function, used to determine whether a certain condition is met.

[0092] Find the type density of each box and splice it into a vector with a length of N cube (δ i ), forming the fractal density feature vector, denoted as T cube (δ i ).

[0093] Similarly, the present invention designs the use of spheres to cover the surface of the damaged area and calculates the fractal number. The specific steps are as follows:

[0094] To determine the size of the covering sphere, select multiple sphere radii δ of different sizes, e.g. Where L is the maximum length of the damaged surface. Ensure that the sphere size covers multiple scales from large to small to capture as many different features of the fracture surface as possible.

[0095] Calculate the number of spheres N(δ) at each scale, and for each scale δ, calculate the minimum number of spheres N(δ) required to completely cover the fracture surface. The specific method is as follows:

[0096] (1) Traversing the entire fracture surface: At each scale δ, a sphere is placed on the surface of the damaged area.

[0097] (2) Determine whether the sphere covers the fracture surface: For each sphere, determine whether it contains the point on the fracture surface. If it does, the count is increased by one.

[0098] The calculation formula is expressed as:

[0099]

[0100] Where H(·) is the Heaviside function, I(x, y, z) is the value of the fracture surface at point (x, y, z), and S(i, j, k) represents the (i, j, k)th sphere.

[0101] In the present invention, it is denoted as N sphere (δ i ), where sphere represents spherical, δ iRepresents a box of the i-th scale.

[0102] Next, calculate the fractal density of each sphere and calculate the density ρ of each sphere ijk (δ), which is determined by the ratio of the number of fracture points within the sphere to the volume of the sphere:

[0103]

[0104] where V(S(i, j, k)) is the volume of the (i, j, k)th sphere, usually

[0105] Perform Heaviside function screening and use the Heaviside function H(.) to determine whether the density of each sphere exceeds the set threshold ρ th :

[0106]

[0107] Accumulate the number of spheres that meet the conditions. For all spheres, calculate the number of spheres with density exceeding the threshold at a given scale δ:

[0108]

[0109] Finally, calculate the fractal density and normalize the number of spheres that meet the conditions to the density:

[0110]

[0111] The present invention calculates the type density of each box and splices it into a vector with a length of N. sphere (δ i ), forming the fractal density feature vector, denoted as T sphere (δ i ).

[0112] Next, the present invention calculates the fitting coefficient of the number of boxes (spheres, squares, etc.) at different scales as a function of scale as another feature. This coefficient reveals the complexity and irregularity of the fracture surface. The regularity of the number N(δ) as a function of scale δ can reveal the complexity and irregularity of the fracture surface.

[0113] The present invention designs logarithmic transformation to convert the above fractal data into linear eigenvalues, making data analysis more intuitive and simple. Specifically, for each type of cubic shape, solve ln(N(δ)) and The logarithmic graph can transform the originally complex nonlinear relationship into a linear relationship:

[0114] ln(N(δ))=-Dln(δ)+ln(B)

[0115] Calculate the fractal dimension D. The slope of the logarithmic graph is the negative value of the fractal dimension D. By plotting ln(N(δ)) vs. The slope a can be obtained by performing a linear regression analysis on the logarithmic plot of

[0116] ln(N(δ))=aln(δ)+b

[0117] Wherein, a=-D, and the absolute value of the slope α obtained by linear regression is the fractal dimension D. The fractal dimension D is an important parameter for measuring the complexity and self-similarity of the fracture surface, and is also called the simplex linear eigenvalue of the present invention.

[0118] Under the two shapes and different scales of the boxes proposed in the present invention, multiple sets of different simplex linear eigenvalues ​​are calculated, which are recorded as

[0119] in, represents the dimensions of the fitted box of the cube, represents the simplex linear eigenvalue at the fitting scale. Similarly, Fit the simplex linear eigenvalues ​​of the box to the sphere.

[0120] In order to analyze the fractal characteristics more accurately, the present invention obtains polymorphic linear eigenvalues ​​and uses the number and density of different types of covering units (spheres, cubes) as input variables to construct a regression model.

[0121] Assume the regression model is:

[0122] ln(N total (δ))=a1 ln(δ sphere )+a2 ln(δ box )+a3 ln(ρ sphere )+a4 ln(ρ box )+b

[0123] Where: N total (δ) is the total number of different coverage units at a given scale δ; δ sphere and δ box is the scale of the spherical and cubic covering units. sphere and ρ box is the density of spherical and cubic cover cells.

[0124] Through multivariate regression analysis, the coefficient a of each parameter can be determined i and the intercept b. The fractal dimension D can be expressed by a combination of these coefficients.

[0125] By extracting parameter features from the macroscopic fractal dimension, valuable parameters can be extracted from the complexity and irregularity of the damaged surface. These macroscopic fractal features not only reflect the geometric characteristics of the material at the macroscopic scale, but also reveal the potential defects and microcrack distribution in the material's microstructure.

[0126] In step S101, the process of extracting geometric features from the three-dimensional image of the damaged area sample of the metal material is as follows:

[0127] Perform longitudinal and annular cross-section cutting on the three-dimensional image of the damaged area sample of the metal material;

[0128] Obtain the outermost edge segmentation curve on each cutting surface of the longitudinal cutting and the annular cross-section cutting;

[0129] The geometric characteristics of the outermost edge segmentation curve on each cutting surface are calculated, including curvature, degree of change, smoothness, arithmetic mean roughness, root roughness, maximum height and ten-point average roughness.

[0130] Cutting samples across different sections, especially annular sections, can reveal microcracks and potential defects within the material. Analyzing the fractal features on these cut surfaces can reveal potential weaknesses in the material under loading or environmental influences, allowing for the evaluation of its performance and reliability. The acquisition and analysis of edge segmentation curves not only characterizes the macroscopic fracture characteristics of the material but also reveals information such as its microstructure and grain orientation. This is of great significance to materials science research and engineering applications, helping to further understand the fracture mechanisms of materials and predict their failure modes.

[0131] The damaged area is segmented into rings according to different segmentation radii, and the radius is set manually. Figure 5 As shown. Multiple cross-sections of the center of the circle are cut, and the cross-section cutting diagram is shown as follows Figure 6 shown.

[0132] Analyze the geometric characteristics of the cross-section at different scales, including curvature, angles, and smoothness. These characteristics can be captured by varying the analysis scale (e.g., filtering or downsampling) to capture variations in the cross-section at different scales. Comprehensively measure the vertical cross-sectional characteristics of the fracture surface, providing a deep understanding of its morphological features and their role in the material fracture process. These measurements can be used not only for basic geometric analysis but also in conjunction with material mechanical properties to reveal the relationship between fracture mechanisms and material properties, providing a comprehensive understanding of the geometric characteristics.

[0133] Preprocess the fracture surface or curve data to ensure data continuity and processability. Processing steps include denoising: using filters (such as Gaussian filters) to remove noise so that geometric features can be more clearly observed. Smoothing: using smoothing techniques such as moving averages or spline interpolation to reduce the effects of high-frequency noise.

[0134] Downsampling: Changing the resolution of the data by reducing the number of sampling points, thereby generating a rougher curve. Filtering is then performed, using a low-pass filter to smooth the curve and remove high-frequency components, thereby generating curves of different scales.

[0135] For each scale of the curve or surface, the following geometric features are calculated:

[0136] Curvature: For a two-dimensional curve, curvature can be calculated by derivatives. For example, for a parameterized curve (x(t), y(t)), the curvature K can be expressed as:

[0137]

[0138] In multi-scale analysis, the change of curvature can reveal the geometric complexity of the curve at different scales.

[0139] Angle change: Calculate the angle of direction change between adjacent points. At different scales, the angle change can reflect the degree of folding of the curve. For example, by calculating the angle between the direction vectors of each adjacent point, the local bending characteristics of the curve can be obtained. It is denoted as K j , where j represents the curvature value at the jth scale.

[0140] Smoothness: The smoothness of a curve can be evaluated by calculating local derivatives (such as first-order derivatives or second-order derivatives). The higher the smoothness, the smoother the curve changes.

[0141] Arithmetic mean roughness (Ra): Ra is the most commonly used roughness parameter, which represents the arithmetic mean of the deviation of a surface or profile from the mean line. It reflects the average height of surface fluctuations.

[0142] Calculation formula:

[0143] Among them, y i is the height deviation of the i-th sampling point relative to the mean line, and N is the total number of sampling points.

[0144] The calculation steps of arithmetic mean roughness (Ra) are as follows:

[0145] Get the height data y of the surface or section i ;

[0146] Calculate the absolute value of all data points' deviation from the mean;

[0147] All absolute deviations are averaged to obtain the Ra value.

[0148] Where j represents the average roughness value at the jth scale.

[0149] Root roughness (Rq): Also known as root mean square height, it is the root mean square of all square deviations. It is more sensitive to large deviations than Ra, so it can reflect the actual situation of surface fluctuations.

[0150] Calculation formula:

[0151] Among them, y i is the height deviation of the ith sampling point, and N is the total number of sampling points.

[0152] The calculation steps are:

[0153] Get the height data y of the surface or section i ;

[0154] Square the deviation of each data point;

[0155] Calculate the average of these squared values;

[0156] Taking the square root of the average value gives the Rq value.

[0157] Where j represents the average roughness value at the jth scale.

[0158] Maximum height (Rz):

[0159] It is the average of the height differences between the five highest peaks and the five lowest valleys within the measurement area. It is a measure of the maximum fluctuations in a surface or profile.

[0160] Calculation formula:

[0161]

[0162] Among them, P i is the height of the five highest peaks, V i is the height of the five lowest valleys.

[0163] The calculation process of the maximum height (Rz) is:

[0164] Five highest and five lowest points were identified on the profile;

[0165] Calculate the height difference between the highest point and the lowest point;

[0166] The five height differences were averaged to obtain the Rz value.

[0167] Where j represents the average roughness value at the jth scale.

[0168] Ten-point average roughness (Rz (JIS)): This is a roughness parameter that represents the average of the differences between the five highest peaks and the five lowest valleys.

[0169] Calculation formula:

[0170] P i is the height of the five highest peaks, V i is the height of the five lowest valleys

[0171] The present invention is denoted as Rz j (JIS), where j represents the average roughness value at the jth scale.

[0172] Profile Peak (Rp) and Profile Valley (Rv): Rp represents the maximum peak height on a surface or profile relative to the mean line, while Rv represents the maximum valley depth relative to the mean line.

[0173] Calculation formula:

[0174] Rp = maximum peak = max(y i ); Rv = maximum valley value = min(y i );

[0175] Among them, y i is the height value of the sampling point.

[0176] By calculating these characteristic parameters, the fluctuation of the fracture surface or profile curve can be quantified, providing important data features for quality control, material research, fault classification, etc.

[0177] The present invention is denoted as Rv j and Rp j , where j represents the average roughness value at the jth scale.

[0178] Specifically, the process of extracting frequency domain features from the three-dimensional image of the damaged area sample of the metal material is as follows:

[0179] Perform longitudinal and annular cross-section cutting on the three-dimensional image of the damaged area sample of the metal material;

[0180] Obtain the outermost edge segmentation curve on each cutting surface of the longitudinal cutting and the annular cross-section cutting;

[0181] Each outermost edge segmentation curve is converted from the spatial domain to the frequency domain to obtain the corresponding frequency domain features.

[0182] The main frequency component in the frequency domain corresponds to the main fluctuation characteristics of the profile curve, which can be used to measure the periodicity and wavelength of the profile. The characteristic is recorded as F icross , where cross represents the cross-section curve, i represents the Fourier eigenvector of the i-th curve, and similarly, F i circle The circle represents the cross-cutting curve.

[0183] In some optional embodiments, according to Figure 2 , the process of obtaining fusion features is:

[0184] S201: According to different fractal number features, the corresponding geometric features and frequency domain features are spliced ​​to obtain the corresponding first fusion feature vector;

[0185] Features after splicing, for example:

[0186] [K j , Ra j , Rq j , Rz j , Rz j (JIS), Rv j , Rp j ]. Traverse different i to obtain different Features. Traverse different to obtain different [K j , Ra j , Rq j , Rz j , Rz j (JIS), Rv j , Rp j ]feature.

[0187] S202: Concatenate each first fusion feature vector and its corresponding fractal number feature to obtain a final fusion feature.

[0188] The features obtained by multiple traversals are spliced ​​together, such as:

[0189]

[0190] In other embodiments, corresponding weights are assigned according to the importance of geometric features, frequency domain features, and fractal number features, and fused features are obtained through weighted fusion.

[0191] In other optional embodiments, a pre-trained multimodal fusion network is used to fuse the geometric features, frequency domain features, and fractal number features to obtain the final fused features. For example, a dedicated neural network is constructed, which takes the fractal features, geometric features, and frequency domain features as inputs and fuses them through a multi-layer perceptron (MLP).

[0192] In other optional embodiments, the feature fusion operation is followed by a feature selection layer. To reduce feature dimensionality and improve classification efficiency, the feature selection layer uses algorithms such as principal component analysis (PCA), maximum relevance minimum redundancy (mRMR), or recursive feature elimination (RFE) to select the most representative feature subset from the fused features. The feature selection layer is followed by a feature classification layer, which inputs the selected features into a classifier model for classification.

[0193] In step S103, the first classifier, the second classifier and the third classifier may be implemented using a support vector machine (SVM), or other network models such as a random forest (RF), which will not be described in detail here.

[0194] In step S104 , the process of determining the category and detailed features of the damaged area can be implemented by using a voting method or other decision-making layer networks.

[0195] Among them, the relevance and classification confidence of the features corresponding to the macroscopic category, microscopic category and feature category of the damaged area samples of the metal material can be obtained in advance based on training.

[0196] Figure 3 This is a schematic diagram of the structure of a metal surface damage feature recognition system with multi-level decision-making in an embodiment of the present invention. Figure 1 The metal surface damage feature recognition method based on multi-level decision-making is corresponding, such as Figure 3 As shown, the metal surface damage feature recognition system with multi-level decision-making in this embodiment may include:

[0197] A feature extraction module 301 is used to obtain a three-dimensional image of a sample of a damaged area of ​​a metal material and extract corresponding geometric features, frequency domain features and fractal features therefrom;

[0198] A feature fusion module 302 is used to fuse the extracted geometric features, frequency domain features, and fractal number features to obtain fused features;

[0199] A multi-level classification module 303 is configured to process the fractal features using a first classifier, process the geometric features and frequency domain features using a second classifier, and process the fusion features using a third classifier, thereby obtaining macroscopic categories, microscopic categories, and feature categories of the damaged area samples of the metal material, respectively.

[0200] The comprehensive decision module 304 is used to finally determine the category and detailed features of the damage area based on the correlation and classification confidence of the features corresponding to the macro category, micro category and feature category of the metal material damage area sample.

[0201] Specifically, in the feature fusion module 302, the process of obtaining the fusion feature is as follows:

[0202] According to different fractal number features, the corresponding geometric features and frequency domain features are spliced ​​to obtain the corresponding first fusion feature vector;

[0203] Each first fusion feature vector is then concatenated with its corresponding fractal number feature to obtain the final fusion feature.

[0204] In other embodiments, in the feature fusion module 302, corresponding weights are assigned according to the importance of geometric features, frequency domain features, and fractal number features, and fused features are obtained through weighted fusion.

[0205] In other optional embodiments, in the feature fusion module 302, a pre-trained multimodal fusion network is used to fuse the geometric features, frequency domain features, and fractal number features to obtain the final fusion features.

[0206] It should be noted here that, Figure 3 The various modules in the multi-level decision-making metal surface damage feature recognition system are Figure 1 Each step in the multi-level decision-making metal surface damage feature recognition method corresponds one to one, and the specific implementation process is the same, which will not be repeated here.

[0207] Reference Figure 4 , a schematic diagram of an electronic device is given. It should be noted that, Figure 4 The electronic device 400 shown is only an example and should not limit the functions and scope of use of the embodiments of the present invention.

[0208] like Figure 4 As shown, electronic device 400 includes a central processing unit (CPU) 401, which can perform various appropriate actions and processes according to the program stored in read-only memory (ROM) 402 or the program loaded from storage portion 408 into random access memory (RAM) 403. Various programs and data required for system operation are also stored in RAM 403. Central processing unit 401, ROM 402 and RAM 403 are connected to each other via bus 404. Input / output (I / O) interface 405 is also connected to bus 404.

[0209] The following components are connected to the I / O interface 405: an input section 406 including a keyboard, a mouse, and the like; an output section 407 including devices such as a cathode ray tube (CRT), a liquid crystal display (LCD), and a speaker; a storage section 408 including devices such as a hard disk; and a communication section 409 including a network interface card such as a local area network (LAN) card or a modem. The communication section 409 performs communication processing via a network such as the Internet. A drive 410 is also connected to the I / O interface 405 as needed. Removable media 411, such as a magnetic disk, an optical disk, a magneto-optical disk, or a semiconductor memory, is installed in the drive 410 as needed, so that computer programs read therefrom can be installed into the storage section 408 as needed.

[0210] When the central processing unit 401 in the electronic device of this embodiment executes the program, the following is achieved: Figure 1 The steps in the multi-level decision-making method for metal surface damage feature recognition are shown.

[0211] In particular, according to an embodiment of the present application, the process described above with reference to the flowchart can be implemented as a computer software program. For example, an embodiment of the present application includes a computer program product, which includes a computer program carried on a computer readable medium, the computer program including a computer program for executing Figure 1 In such an embodiment, the computer program can be downloaded and installed from a network via the communication section 409 and / or installed from the removable medium 411. When the computer program is executed by the central processing unit 401, the various functions defined in the apparatus of the present application are performed.

[0212] in, Figure 1 The computer program instructions corresponding to the method shown can also be stored in a computer readable memory that can guide a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture including an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0213] Those skilled in the art will appreciate that all or part of the processes in the above-described method embodiments can be implemented by instructing related hardware through a computer program. The program can be stored in a computer-readable storage medium, and when executed, the program can include the processes in the above-described method embodiments. The storage medium can be a magnetic disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM).

[0214] The foregoing description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Those skilled in the art will readily appreciate that various modifications and variations of the present invention are possible. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention are intended to be within the scope of protection of the present invention.

Claims

1. A multi-level decision-making method for identifying metal surface damage features, characterized in that: include: Obtain a three-dimensional image of a sample of a damaged area of ​​a metal material and extract corresponding geometric features, frequency domain features, and fractal features from it; The process of extracting geometric features from the three-dimensional image of the damaged area sample of the metal material is as follows: Perform longitudinal and annular cross-section cutting on the three-dimensional image of the damaged area sample of the metal material; Obtain the outermost edge segmentation curve on each cutting surface of the longitudinal cutting and the annular cross-section cutting; Calculate the geometric characteristics of the outermost edge segmentation curve on each cutting surface, such as curvature, degree change, smoothness, arithmetic mean roughness, root roughness, maximum height and ten-point average roughness; The process of extracting frequency domain features from the three-dimensional image of the damaged area sample of the metal material is as follows: Perform longitudinal and annular cross-section cutting on the three-dimensional image of the damaged area sample of the metal material; Obtain the outermost edge segmentation curve on each cutting surface of the longitudinal cutting and the annular cross-section cutting; Convert each outermost edge segmentation curve from the spatial domain to the frequency domain to obtain the corresponding frequency domain features; The fractal features include the number of fractals corresponding to the surface of the sample of the damaged area of ​​the metal material covered by boxes of different shapes and the fitting coefficient of the number of boxes at different scales as the scale changes; The process is: Use boxes of different shapes to cover the sample surface of the current metal material damage area and calculate the corresponding fractal number; When choosing a square box, we first select a box size δ = L, then gradually reduce the box size to cover the damaged surface at different scales and calculate the fractal density; When a spherical box is selected to cover the surface of the damaged area and the fractal number is calculated, the specific steps are as follows; Determine the size of the coverage sphere and select multiple sphere radii of different sizes to ensure that the sphere sizes cover multiple scales from large to small; Calculate the number of spheres at each scale, and for each scale, calculate the minimum number of spheres required to completely cover the fracture surface and the fractal density; Calculate the fitting coefficients of the number of square and spherical boxes at different scales as the scale changes; The extracted geometric features, frequency domain features and fractal number features are fused to obtain fused features; Using a first classifier to process the fractal features, using a second classifier to process the geometric features and frequency domain features, and using a third classifier to process the fusion features, thereby obtaining macroscopic categories, microscopic categories, and feature categories of samples of damaged areas of metal materials, respectively; Based on the correlation and classification confidence of the features corresponding to the macro-category, micro-category and feature category of the damaged area samples of metal materials, the category and detailed features of the damaged area are finally determined.

2. The metal surface damage feature recognition method with multi-level decision-making as claimed in claim 1, characterized in that: The process of obtaining fusion features is: According to different fractal number features, the corresponding geometric features and frequency domain features are spliced ​​to obtain the corresponding first fusion feature vector; Each first fusion feature vector is then concatenated with its corresponding fractal number feature to obtain the final fusion feature.

3. The metal surface damage feature recognition method with multi-level decision-making as claimed in claim 1, characterized in that: According to the importance of geometric features, frequency domain features and fractal number features, corresponding weights are assigned and fusion features are obtained through weighted fusion; Or use a pre-trained multimodal fusion network to fuse geometric features, frequency domain features, and fractal number features to obtain the final fusion features.

4. A multi-level decision-making metal surface damage feature recognition system, characterized in that: A metal surface damage feature recognition method using the multi-level decision-making method according to any one of claims 1 to 3, comprising: A feature extraction module is used to obtain a three-dimensional image of a sample of a damaged area of ​​a metal material and extract corresponding geometric features, frequency domain features, and fractal features therefrom; Feature fusion module, which is used to fuse the extracted geometric features, frequency domain features and fractal number features to obtain fused features; a multi-level classification module, which is used to process the fractal features using a first classifier, process the geometric features and frequency domain features using a second classifier, and process the fusion features using a third classifier, thereby obtaining macroscopic categories, microscopic categories, and feature categories of the damaged area samples of the metal material, respectively; The comprehensive decision-making module is used to determine the category and detailed features of the damaged area based on the correlation and classification confidence of the features corresponding to the macro category, micro category and feature category of the damaged area samples of the metal material.

5. The metal surface damage feature recognition system with multi-level decision-making as claimed in claim 4, characterized in that: In the feature fusion module, the process of obtaining the fusion feature is as follows: According to different fractal number features, the corresponding geometric features and frequency domain features are spliced ​​to obtain the corresponding first fusion feature vector; Each first fusion feature vector is then concatenated with its corresponding fractal number feature to obtain the final fusion feature; Alternatively, in the feature fusion module, corresponding weights are assigned according to the importance of geometric features, frequency domain features, and fractal number features, and fusion features are obtained through weighted fusion; Alternatively, in the feature fusion module, a pre-trained multimodal fusion network is used to fuse the geometric features, frequency domain features, and fractal number features to obtain the final fusion features.

6. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the multi-level decision-making metal surface damage feature recognition method according to any one of claims 1 to 3 are implemented.

7. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the steps of the metal surface damage feature identification method with multi-level decision-making are implemented as described in any one of claims 1-3.

Images