A non-contact surface roughness detection method and system for free-form surface milling

Abstract

The application discloses a kind of non-contact surface roughness detection method and system for free-form surface milling, it is related to roughness detection technical field, ideal workpiece model is generated by CAD simulation, and the optimal laser detector motion path LMP is planned out with this, and along the path synchronous acquisition and standardization processing, obtain the theoretical scattering spot image data T and actual scattering spot image data I accurately matched with actual workpiece, eliminate the measurement error introduced due to probe angle change, ensure the consistency of data and the reliability of subsequent analysis. By calculating to obtain difference covariance scalar D and anisotropy index A, realize the accurate quantification of surface scattering deviation total amount and identify tool marks and other key directional defects. Finally, the surface quality index Q is calculated, and the preset surface quality index threshold Q th Comparison is generated to generate detection evaluation result AIR, realizes the timely feedback and closed-loop control of workpiece detection result.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of roughness detection, in particular to a non-contact surface roughness detection method and system for free-form surface milling. BACKGROUND

[0002] In modern industrial manufacturing and quality control systems, precision detection technology is the cornerstone of ensuring product performance and reliability. This technology covers a wide range from macroscopic size measurement to microscopic physical property analysis. The accurate evaluation of the surface topography of parts is directly related to core performance indicators such as friction and wear, fatigue resistance, corrosion resistance, and fitting accuracy. Among the many parameters of surface topography, surface roughness detection is particularly critical. It is widely used in various precision manufacturing fields, especially in the aerospace, precision mold, automotive industry, and high-end medical equipment industries. The quality of the complex component surface formed by computer numerical control milling often directly determines the final performance and service life of the entire high-end equipment. For example, the overall blisk of an aircraft engine, the key crankshaft of a car engine, and artificial joints implanted in the human body all have irregular, asymmetric complex geometric shapes, i.e., so-called "free-form surfaces." Therefore, how to efficiently, accurately, and non-destructively detect the surface roughness of such free-form surfaces formed after milling has given rise to an urgent need for corresponding detection methods and systems.

[0003] Currently, in the above-mentioned specific field, the detection of the surface roughness of free-form surfaces after milling still mainly relies on traditional contact measurement methods, with a probe-type profilometer as a typical representative. Although this method is mature and the results are traceable, its inherent defects have become increasingly prominent in meeting the needs of modern high-end manufacturing. First, its detection efficiency is low, and the probe needs to be physically drawn slowly on the workpiece surface to collect data point by point. Full-size detection of large-area or complex surfaces is almost impossible, and sampling detection is usually used, which brings risks and blind spots to quality control. Second, the measurement process is invasive, and the high-hardness probe tip may cause permanent scratches on the finished surface, especially on surfaces made of soft metal or requiring high smoothness, directly leading to the scrap of costly workpieces. Third, due to the physical size and mechanical structure of the probe, it often cannot track the free-form surface with sharp changes in curvature, steep slopes, or small grooves, resulting in inaccurate and unreliable measurement results. This has become a key technical bottleneck restricting the development of high-end complex curved component manufacturing towards higher quality, higher efficiency, and higher intelligence. SUMMARY

[0004] In view of the deficiencies of the prior art, the application provides a non-contact surface roughness detection method and system for free-form surface milling, which solves the problems mentioned in the background art.

[0005] To achieve the above object, the application is implemented by the following technical scheme: a non-contact surface roughness detection method for free-form surface milling, comprising the following steps:

[0006] S1, based on the milling operation program, milling simulation processing is performed in the CAD simulation software to obtain a processed free-form surface workpiece model, and a surface normal vector set NVC of the free-form surface workpiece model in a generated three-dimensional coordinate system is obtained;

[0007] S2, based on the surface normal vector set NVC, a laser detector motion path LMP is generated, and the laser scattering method is used according to the laser detector motion path LMP to collect an original image data set Raw and perform standardization processing to obtain theoretical scattering spot image data T and actual scattering spot image data I for the free-form surface workpiece model and the actual free-form surface workpiece;

[0008] S3, based on the theoretical scattering spot image data T and the original scattering spot image data I, difference covariance feature extraction is performed to obtain a difference covariance scalar D;

[0009] S4, based on the theoretical scattering spot image data T and the original scattering spot image data I, residual image data G is generated and residual energy centroid coordinates (x', y') are calculated, and second-order central moments (μ 20 , μ 02 , μ 11 ) are calculated according to the residual image data G and the residual energy centroid coordinates (x', y');

[0010] S5, based on the second-order central moments (μ 20 , μ 02 , μ 11 ), normalization calculation is performed to obtain an anisotropy index A to evaluate the difference of the stretching degree of each direction of the residual image data G;

[0011] S6, based on the difference covariance scalar D and the anisotropy index A, a surface quality index Q is calculated, and a comparison is made with a preset surface quality index threshold Q th to generate a detection evaluation result AIR.

[0012] Preferably, S1 comprises S11;

[0013] S11, based on the milling operation program, input the program code for driving the numerical control machine tool and the original CAD model of the workpiece into the CAD simulation software, simulate the milling process of the numerical control machine tool, obtain the free-form surface workpiece model, and calculate and generate the surface normal vector set NVC of the free-form surface workpiece model in the three-dimensional coordinate system by surface scanning of the free-form surface workpiece model by the CAD simulation software.

[0014] Preferably, S2 comprises S21;

[0015] S21, based on the surface normal vector set NVC, generate the laser probe motion path LMP perpendicular to the surface normal vector set NVC by the CAD simulation software, simulate the moving laser scattering probe in the CAD simulation software, and collect the original theoretical scattering spot gray image T of M*N pixels of the free-form surface workpiece model. R ;

[0016] In the numerical control machine tool, input the laser probe motion path LMP, collect the M*N pixel original actual scattering spot gray image I of the actual free-form surface workpiece R , combine the original theoretical scattering spot image T R to generate the original image data set Raw;

[0017] According to the original image data set Raw, the gray value of each pixel point of the original theoretical scattering spot gray image T R and the original actual scattering spot gray image I R is normalized and standardized to obtain the processed theoretical scattering spot image data T and the actual scattering spot image data I.

[0018] Preferably, S3 comprises S31;

[0019] S31, based on the theoretical scattering spot image T and the original scattering spot image I, by calculating the logarithmic intensity covariance of the actual scattering spot image data I and the theoretical scattering spot image data T, obtaining the difference covariance scalar D, and by evaluating the difference degree of the actual scattering spot image data I and the theoretical scattering spot image data T, quantifying the scattering deviation caused by the real roughness;

[0020] Wherein, the difference covariance scalar D is calculated as follows:

[0021] ;

[0022] ;

[0023] ;

[0024] In the formula, I(x, y) represents the normalized gray value of the point at coordinate (x, y) in the actual scattered light spot image data I, T(x, y) represents the normalized gray value of the point at coordinate (x, y) in the theoretical scattered light spot image data T, LI(x, y) represents the logarithmic operation on the normalized gray value of the actual scattered light spot image data I, LT(x, y) represents the logarithmic operation on the normalized gray value of the theoretical scattered light spot image data T, I' represents the average value of the normalized gray values ​​of all pixels in the actual scattered light spot image data I, T' represents the average value of the normalized gray values ​​of all pixels in the theoretical scattered light spot image data T, ln represents the logarithmic function with the natural constant e as the base, and ∑ represents the iterative summation operation.

[0025] Preferably, S4 includes S41 and S42;

[0026] S41. Based on the theoretical scattered spot image data T and the original scattered spot image data I, calculate the absolute value of the gray-level normalized value of each pixel in the theoretical scattered spot image data T and the original scattered spot image data I, and generate residual image data G.

[0027] Calculate the zero-order moment μ of the residual image data G based on the residual image data G. 00 and the first moment (μ) 10 μ 01 ), and based on the zeroth moment μ of the residual image data G 00 and the first moment (μ) 10 μ 01 Calculate the centroid coordinates (x', y') of the residual image data G;

[0028] Where, μ 10 μ represents the horizontal component of the centroid of the residual image data G. 01 This represents the vertical component of the centroid of the residual image data G.

[0029] Preferably, in step S42, the second-order central moment (μ) of the residual image data G is calculated based on the residual image data G and the residual energy centroid coordinates (x', y'). 20 μ 02 μ 11 );

[0030] Where, μ 20 μ represents the horizontal stretching of the residual image data G. 02 μ represents the vertical stretching of the residual image data G. 11 This represents the diagonal correlation degree of the residual image data G.

[0031] Preferably, S5 includes S51;

[0032] S51, Based on the second-order central moment (μ)20 μ 02 μ 11 ), by calculating the second central moment (μ 20 μ 02 μ 11 The largest eigenvalue λ of the covariance matrix C formed by ) MAX and the minimum eigenvalue λ MIN And combined with the largest eigenvalue λ MAX and the minimum eigenvalue λ MIN The difference is used as the normalization numerator, combined with the largest eigenvalue λ. MAX and the minimum eigenvalue λ MIN The sum of the values ​​is used as the normalized denominator to calculate the anisotropy index A, which is used to evaluate the differences in the degree of stretching of the residual image data G in each direction.

[0033] The covariance matrix C is expressed as follows:

[0034] ;

[0035] The anisotropy index A is calculated using the following formula:

[0036] ;

[0037] ;

[0038] .

[0039] Preferably, S6 includes S61 and S62;

[0040] S61. Based on the differential covariance scalar D and the anisotropy index A, a surface quality index algorithm is constructed. For the anisotropy index A, a nonlinear penalty function f(A) is designed and combined with the differential covariance scalar D to calculate the surface quality index Q.

[0041] The surface quality index algorithm expression is as follows:

[0042] ;

[0043] ;

[0044] In the formula, ln represents a logarithmic function with the natural constant e as the base.

[0045] Preferably, in step S62, the surface quality index Q is compared with a preset surface quality index threshold Q. th Compare;

[0046] If the surface quality index Q < the preset surface quality index threshold Q thIf the surface quality index Q is greater than or equal to a preset surface quality index threshold Q

[0047] If the surface quality index Q is greater than or equal to a preset surface quality index threshold Q th If the surface quality index Q is greater than or equal to a preset surface quality index threshold Q

[0048] A non-contact surface roughness detection system for free-form surface milling includes a simulation module, a data acquisition module, a differential covariance feature extraction module, a second moment calculation module of residual image, an anisotropy index analysis module and a comprehensive evaluation module.

[0049] The simulation module simulates milling in a CAD simulation software based on a milling program to obtain a machined free-form surface workpiece model, and generates a set of surface normal vectors NVC in a three-dimensional coordinate system according to the free-form surface workpiece model.

[0050] The data acquisition module generates a laser probe motion path LMP based on the set of surface normal vectors NVC, and uses a laser scattering method to collect a set of raw image data Raw and obtain theoretical scattering spot image data T and actual scattering spot image data I through standardization processing according to the laser probe motion path LMP for the free-form surface workpiece model and an actual free-form surface workpiece.

[0051] The differential covariance feature extraction module extracts differential covariance features based on the theoretical scattering spot image data T and the raw scattering spot image data I to obtain a differential covariance scalar D.

[0052] The second moment calculation module of residual image generates residual image data G and calculates residual energy centroid coordinates (x', y') based on the theoretical scattering spot image data T and the raw scattering spot image data I, and calculates second central moments (μ 20 , μ 02 , μ 11 ) according to the residual image data G and the residual energy centroid coordinates (x', y').

[0053] The anisotropy index analysis module performs normalization calculation based on the second central moments (μ 20 , μ 02 , μ 11 ) to obtain an anisotropy index A, and evaluates the difference in stretching degree of the residual image data G in each direction.

[0054] The comprehensive evaluation module calculates a surface quality index Q based on the differential covariance scalar D and the anisotropy index A, and compares the surface quality index Q with a surface quality index threshold Q th to generate a detection evaluation result AIR.

[0055] The application provides a non-contact surface roughness detection method and system for free-form surface milling, which has the following beneficial effects:

[0056] (1) First, a theoretically perfect free-form surface workpiece model is established as a digital reference through CAD simulation processing, and based on this, the optical scattering characteristics are accurately compared with the actual workpiece under the same geometric path. The application creatively analyzes the surface quality from two dimensions of amplitude and directionality, calculates the differential covariance scalar D representing the total amount of scattering deviation, and quantifies the anisotropy index A of surface texture stretching characteristics, and finally fuses the two key features into a comprehensive surface quality index Q. This evaluation method based on digital twin comparison and multi-dimensional feature fusion solves the problem that traditional methods cannot balance detection accuracy, efficiency and non-destructiveness, and realizes objective, comprehensive and automated accurate judgment of complex surface quality.

[0057] (2) Based on the actual milling operation program, CAD simulation is carried out to prospectively generate a free-form surface workpiece model consistent with the theoretical processing intention, and a high-density surface normal vector set NVC is extracted from it. A perfect digital twin reference is constructed to provide accurate geometric basis for all subsequent operations. Based on this geometric reference, an optimized laser detector motion path LMP can be automatically generated, ensuring that the detector always maintains the best posture perpendicular to the measured surface during scanning, and fundamentally eliminates measurement errors introduced by probe angle changes. More importantly, the system strictly follows this path to synchronously obtain theoretical scattering spot image data T from the virtual model and actual scattering spot image data I from the actual workpiece, and standardizes both. This ensures that the data pairs used for subsequent comparative analysis are collected under the same geometric and environmental reference, greatly ensuring the consistency of the data and the reliability of the subsequent analysis.

[0058] (3) By calculating the difference covariance scalar D, the accurate quantification of the surface roughness amplitude information is realized. More importantly, the present application does not stop at a single amplitude evaluation, but generates residual image data G and performs rigorous image matrix analysis thereon, and further calculates an anisotropy index A. The index successfully quantifies the directionality of the surface texture, and can intelligently distinguish random, less harmful rough morphology from directional, potentially stress-concentrated or lubrication-failed tool marks and other serious defects. Finally, through a comprehensive evaluation algorithm, the difference covariance scalar D representing the amplitude and the anisotropy index A representing the directionality are organically integrated, and a comprehensive surface quality index Q is calculated. This multi-dimensional evaluation is much more scientific and reliable than single-parameter judgment, and the final generated detection evaluation result AIR can provide clear and executable decision instructions for production, create conditions for realizing timely feedback and closed-loop control of workpieces, and has significant engineering application value and technical progress significance. BRIEF DESCRIPTION OF DRAWINGS

[0059] Figure 1 It is a free-form surface milling non-contact surface roughness detection method step schematic diagram of the present application;

[0060] Figure 2 It is a free-form surface milling non-contact surface roughness detection system block diagram schematic diagram of the present application;

[0061] Figure 3 It is a three-dimensional correlation diagram of the surface quality index Q, the difference covariance scalar D and the anisotropy index A. DETAILED DESCRIPTION

[0062] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor are within the scope of protection of the present application.

[0063] Embodiment 1

[0064] The present application provides a free-form surface milling non-contact surface roughness detection method, please refer to Figure 1 , comprising the following steps:

[0065] S1, based on the milling operation program, milling simulation machining is performed in the CAD simulation software to obtain a machined free-form surface workpiece model, and a surface normal vector set NVC of the free-form surface workpiece model in the generated three-dimensional coordinate system is obtained;

[0066] S2, based on the set of surface normal vectors NVC, generate the laser probe motion path LMP, use the laser scattering method according to the laser probe motion path LMP to collect the original image data set Raw and perform standardization processing to obtain the theoretical scattering spot image data T and the actual scattering spot image data I on the free-form surface workpiece model and the actual free-form surface workpiece;

[0067] S3, based on the theoretical scattering spot image data T and the original scattering spot image data I, perform differential covariance feature extraction to obtain the differential covariance scalar D;

[0068] S4, based on the theoretical scattering spot image data T and the original scattering spot image data I, generate the residual image data G and calculate the residual energy centroid coordinates (x', y'); according to the residual image data G and the residual energy centroid coordinates (x', y'), calculate the second-order central moments (μ 20 , μ 02 , μ 11 );

[0069] S5, based on the second-order central moments (μ 20 , μ 02 , μ 11 ), perform normalization calculation to obtain the anisotropy index A to evaluate the difference of the stretching degree of each direction of the residual image data G;

[0070] S6, based on the differential covariance scalar D and the anisotropy index A, calculate the surface quality index Q and compare it with the preset surface quality index threshold Q th to generate the detection evaluation result AIR.

[0071] In this embodiment, first, by pre-generating the set of surface normal vectors NVC and planning the laser probe motion path LMP in this way, the detection process has high adaptability and pose accuracy for complex free-form surfaces, effectively solving the measurement distortion problem caused by large curvature change in the background technology. Secondly, the entire detection process uses the laser scattering method, completely avoiding physical contact and fundamentally eliminating the risk of causing permanent scratches on the high-surface-finish workpiece surface. More importantly, this method not only calculates the differential covariance scalar D representing the roughness amplitude, but also innovatively introduces the anisotropy index A to quantify the directionality of the surface texture. This two-dimensional evaluation can more deeply understand the surface quality to distinguish defects of different properties such as random pitted surface and directional knife marks, making up for the shortcomings of traditional methods that can only provide a single parameter and incomplete information. Finally, by fusing the calculation of the surface quality index Q and generating clear detection evaluation results AIR, clear and executable decision instructions can be provided for production, creating conditions for timely feedback and closed-loop control of the workpiece, and having significant engineering application value and technical progress significance.

[0072] Embodiment 2

[0073] This embodiment is an explanation and illustration in Embodiment 1, please refer to Figure 1 , specifically: S1 includes S11;

[0074] S11, based on the milling operation program, input the program code for driving the numerical control machine tool and the original CAD model of the workpiece into the CAD simulation software, simulate the milling process of the numerical control machine tool, obtain the free-form surface workpiece model, and calculate and generate the surface normal vector set NVC of the free-form surface workpiece model in the three-dimensional coordinate system by the CAD simulation software;

[0075] S2 includes S21;

[0076] S21, based on the surface normal vector set NVC, generate the laser probe motion path LMP perpendicular to the surface normal vector set NVC by the CAD simulation software, simulate the moving laser scattering probe in the CAD simulation software, and collect the M*N pixel original theoretical scattering spot gray image T R ;

[0077] In the numerical control machine tool, input the laser probe motion path LMP, collect the M*N pixel original actual scattering spot gray image I R of the actual free-form surface workpiece, and generate the original image data set Raw in combination with the original theoretical scattering spot image T R ;

[0078] According to the original image data set Raw, the gray value of each pixel point of the original theoretical scattering spot gray image T R and the original actual scattering spot gray image I R is normalized and standardized to obtain the processed theoretical scattering spot image data T and the actual scattering spot image data I;

[0079] Wherein, the normalization calculation formula is as follows:

[0080] ;

[0081] ;

[0082] In the formula, I(x, y) represents the gray normalization value of the coordinate (x, y) point in the actual scattering spot image data I, T(x, y) represents the gray normalization value of the coordinate (x, y) point in the theoretical scattering spot image data T, I R (x, y) represents the gray value of the coordinate (x, y) point in the original actual scattering spot image data I R , and T R(x, y) represents the gray value of the coordinate (x, y) point in the original theoretical scattering spot image data T R , MAX(I R , T R ) represents the maximum gray value among the original actual scattering spot image data I R and the original theoretical scattering spot image data T R , MIN(I R , T R ) represents the minimum gray value among the original actual scattering spot image data I R and the original theoretical scattering spot image data T R , and ∑ represents a traversal operation.

[0083] In this embodiment, by directly using the actual program code for driving the numerical control machine tool for simulation, the generated free-form surface workpiece model is ensured to be maximally faithful to the final machining intention, so that the surface normal vector set NVC has extremely high theoretical accuracy and traceability. On this basis, the embodiment constructs a strict "twin" data acquisition channel: it not only plans a laser detector motion path LMP that is completely consistent and perpendicular to the surface for the virtually acquired original theoretical scattering spot gray image TR and the physically acquired original actual scattering spot gray image IR, but also finally obtains the theoretical scattering spot image data T and the actual scattering spot image data I by uniformly normalizing both, which completely eliminates systematic deviations such as environmental light and sensor gain. This method ensures the fidelity of the reference model from the source and ensures the consistency of the path and the purity of the data during the acquisition process, which provides an unparalleled high-quality data basis for subsequent steps of high-credibility feature extraction and comprehensive evaluation.

[0084] Embodiment 3

[0085] This embodiment is an explanation and description in embodiment 2, please refer to Figure 1 , specifically: S3 includes S31;

[0086] S31, based on the theoretical scattering spot image T and the original scattering spot image I, the difference covariance scalar D is obtained by calculating the logarithmic intensity covariance of the actual scattering spot image data I and the theoretical scattering spot image data T, the difference between the actual scattering spot image data I and the theoretical scattering spot image data T is evaluated, and the scattering deviation caused by the real roughness is quantified;

[0087] Wherein, the difference covariance scalar D is calculated according to the following formula:

[0088] ;

[0089] ;

[0090] ;

[0091] In the formula, I(x, y) represents the gray-level normalized value of the point at coordinate (x, y) in the actual scattered light spot image data I, T(x, y) represents the gray-level normalized value of the point at coordinate (x, y) in the theoretical scattered light spot image data T, LI(x, y) represents the logarithmic operation on the gray-level normalized value of the actual scattered light spot image data I, LT(x, y) represents the logarithmic operation on the gray-level normalized value of the theoretical scattered light spot image data T, I' represents the average gray-level normalized value of all pixels in the actual scattered light spot image data I, T' represents the average gray-level normalized value of all pixels in the theoretical scattered light spot image data T, ln represents the logarithmic function with the natural constant e as the base, and ∑ represents the iterative summation operation;

[0092] S4 includes S41 and S42;

[0093] S41. Based on the theoretical scattered spot image data T and the original scattered spot image data I, calculate the absolute value of the gray-level normalized value of each pixel in the theoretical scattered spot image data T and the original scattered spot image data I, and generate residual image data G.

[0094] The formula for calculating the normalized gray value G(x,y) of the coordinate (x,y) point in the residual image data G is as follows:

[0095] ;

[0096] In the formula, ∑ represents the traversal operation, and || represents the absolute value operation;

[0097] Calculate the zero-order moment μ of the residual image data G based on the residual image data G. 00 and the first moment (μ) 10 μ 01 ), and based on the zeroth moment μ of the residual image data G 00 and the first moment (μ) 10 μ 01 Calculate the centroid coordinates (x', y') of the residual image data G;

[0098] Where, μ 10 μ represents the horizontal component of the centroid of the residual image data G. 01 The vertical component of the centroid of the residual image data G, and its zeroth moment μ 00 and the first moment (μ) 10 μ 01 The formulas for calculating the centroid coordinates (x', y') and the centroid coordinates (x', y') are as follows:

[0099] ;

[0100] ;

[0101] ;

[0102] ;

[0103] S42, calculate the second order central moments (μ 20 , μ 02 , μ 11 ) of the residual image data G according to the residual image data G and the residual energy centroid coordinates (x', y');

[0104] wherein μ 20 represents the horizontal extension of the residual image data G, μ 02 represents the vertical extension of the residual image data G, and μ 11 represents the diagonal correlation of the residual image data G, and the second order central moments (μ 20 , μ 02 , μ 11 ) are calculated according to the following expressions:

[0105] ;

[0106] ;

[0107] ;

[0108] wherein ∑ represents the traversal summation operation.

[0109] S5 comprises S51;

[0110] S51, based on the second order central moments (μ 20 , μ 02 , μ 11 ), calculate the maximum eigenvalue λ 20 and the minimum eigenvalue λ 02 of the covariance matrix C composed of the second order central moments (μ 11 , μ MAX , μ MIN ), and combine the difference between the maximum eigenvalue λ MAX and the minimum eigenvalue λ MIN as the normalization numerator, and combine the sum of the maximum eigenvalue λ MAX and the minimum eigenvalue λ MIN as the normalization denominator, to obtain the anisotropy index A, to evaluate the difference of the stretching degree of each direction of the residual image data G;

[0111] wherein the covariance matrix C is expressed as follows:

[0112] ;

[0113] wherein the anisotropy index A is calculated as follows:

[0114] ;

[0115] ;

[0116] .

[0117] In this embodiment, the difference covariance scalar D is calculated by using the logarithmic intensity covariance, and by logarithmically compressing the light intensity signal, the excessive influence of the occasional extremely bright or extremely dark pixel points on the overall calculation is effectively suppressed, and the robustness of the algorithm to noise and outliers is enhanced. At the same time, this processing also improves the sensitivity to weak texture changes in the dark area of the image. Therefore, the difference covariance scalar D not only quantifies the difference, but also is a more stable and more real surface micro-undulation amplitude index in statistics. When processing the anisotropy index A representing the directionality, a classical mathematical tool of eigenvalue analysis of the covariance matrix is introduced instead of a simple geometric ratio. The construction idea is to regard the second-order central moments (μ 20 , μ 02 , μ 11 ) of the residual image data G as a complete system describing its energy distribution, and by solving the maximum eigenvalue λ MAX and the minimum eigenvalue λ MIN of the covariance matrix C composed of the second-order central moments (μ 20 , μ 02 , μ 11 ), wherein the maximum eigenvalue λ MAX quantifies the extension degree of the shape in the maximum stretching direction, and the minimum eigenvalue λ MIN quantifies the extension degree of the shape in the minimum stretching direction. The final calculation formula is to realize normalization. The numerator represents the absolute difference between the two principal axis extension degrees, which is the most direct embodiment of "stretching degree"; and the denominator represents the total extension degree, which is the "total size" of the shape. Dividing the two, the interference of factors such as defect size and energy strength is completely eliminated, so that the final anisotropy index A becomes a pure, standardized index related only to the shape eccentricity. In summary, this deep insight of decoupling the roughness into a robust "amplitude" feature and a "directionality" feature with a clear physical principal axis enables the subsequent comprehensive evaluation to be based on more essential and less ambiguous data, greatly improving the interpretability, reliability and scientificity of the final detection results.

[0118] Example 4

[0119] This embodiment is an explanation and illustration in embodiment 3, please refer to Figure 1 and Figure 3 , specifically: S6 includes S61 and S62;

[0120] S61, based on the difference covariance scalar D and the anisotropy index A, the surface quality index algorithm is constructed, for the anisotropy index A, a nonlinear penalty function f(A) is designed and combined with the difference covariance scalar D to obtain the surface quality index Q by calculation;

[0121] Wherein, the surface quality index algorithm expression is as follows:

[0122] ;

[0123] ;

[0124] In the formula, ln represents the logarithmic function with natural constant e as the base;

[0125] S62, compare the surface quality index Q with the preset surface quality index threshold Q th ; th The preset surface quality index threshold Q th Is set by the staff in the field according to the actual generation demand;

[0126] If the surface quality index Q is less than the preset surface quality index threshold Q th , the detection evaluation result AIR is generated as the current workpiece roughness detection passes;

[0127] If the surface quality index Q is greater than or equal to the preset surface quality index threshold Q th , the detection evaluation result AIR is generated as the current workpiece roughness detection fails, and the operator is prompted to check the milling cutter and the operation program.

[0128] The specific example of generating the detection evaluation result AIR is as follows:

[0129] The original theoretical scattering spot gray image T R : ;

[0130] The original actual scattering spot gray image I R : ;

[0131] After normalization standard processing:

[0132] The theoretical scattering spot gray image T R : ;

[0133] The actual scattering spot gray image I R : ;

[0134] The differential covariance scalar D is calculated as follows:

[0135] ;

[0136] The residual image data G is calculated by subtracting the theoretical scattering spot gray scale image T R from the actual scattering spot gray scale image I R : ;

[0137] The zeroth moment μ 00 and the first moments (μ 10 , μ 01 ) are calculated as follows: 00 μ ≈1.6504, μ 10 ≈3.8518, and μ 01 ≈3.3008.

[0138] The centroid coordinates (x', y') are calculated as follows:

[0139] ;

[0140] The second central moments (μ 20 , μ 02 , μ 11 ) are calculated as follows: 20 μ ≈1.64, μ 02 ≈1.110, and μ 11 ≈1.12.

[0141] The maximum eigenvalue λ MAX : 2.55, and the minimum eigenvalue λ MIN : 0.19.

[0142] The anisotropy index A is calculated as follows:

[0143] ;

[0144] The surface quality index Q is calculated as follows:

[0145] ;

[0146] The preset surface quality index threshold Q th = 0.7.

[0147] Since the surface quality index Q is less than the preset surface quality index threshold Q th , the detection evaluation result AIR is generated as the current workpiece roughness detection fails, and a prompt to check the milling tool and the operation program is issued to the operator.

[0148] In this embodiment, the anisotropy index A is processed by designing a nonlinear penalty function f(A). The method is not simply linearly superimposed with amplitude and directivity, but according to the engineering practice experience, the strong directivity defect which is more harmful is applied with sharply increasing weight. This focused fusion method makes the final surface quality index Q more accurately reflect the comprehensive risk level of the workpiece, greatly improving the scientificity and practicality of the evaluation result. More importantly, by directly comparing the surface quality index Q with the preset surface quality index threshold Qth, the method converts the complex, multi-dimensional analysis result into a clear and explicit detection evaluation result AIR without manual interpretation. The subjectivity and inconsistency caused by human factors are eliminated, and the objective, comprehensive and automatic accurate judgment of the complex curved surface quality is realized.

[0149] Embodiment 5

[0150] A non-contact surface roughness detection system for free-form surface milling, please refer to Figure 2 , specifically: including simulation module, data acquisition module, differential covariance feature extraction module, residual image second moment calculation module, anisotropy index analysis module and comprehensive evaluation module;

[0151] The simulation module simulates the milling in the CAD simulation software based on the milling operation program, obtains the processed free-form surface workpiece model, and generates a set of surface normal vectors NVC of the free-form surface workpiece model in the three-dimensional coordinate system;

[0152] The data acquisition module generates a laser detector motion path LMP based on the set of surface normal vectors NVC, and uses the laser scattering method to collect a set of raw image data Raw and obtain theoretical scattering spot image data T and actual scattering spot image data I through standardization processing according to the laser detector motion path LMP for the free-form surface workpiece model and the actual free-form surface workpiece;

[0153] The differential covariance feature extraction module extracts the differential covariance feature based on the theoretical scattering spot image data T and the raw scattering spot image data I, and obtains the differential covariance scalar D;

[0154] The residual image second moment calculation module generates residual image data G and calculates residual energy centroid coordinates (x', y') based on the theoretical scattering spot image data T and the raw scattering spot image data I, and calculates the second central moments (μ 20 , μ 02 , μ 11 ) according to the residual image data G and the residual energy centroid coordinates (x', y');

[0155] The anisotropy index analysis module calculates the anisotropy index A based on the second central moments (μ20 , μ 02 , μ 11 ) is normalized to obtain an anisotropy index A, which evaluates the difference in the degree of stretching in each direction of the residual image data G;

[0156] The comprehensive evaluation module calculates a surface quality index Q based on the difference covariance scalar D and the anisotropy index A, and compares the surface quality index Q with a surface quality index threshold Q th to generate a detection evaluation result AIR.

[0157] Although the embodiments of the present application have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, substitutions and alterations can be made thereto without departing from the principles and spirit of the present application, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A non-contact surface roughness detection method for freeform surface milling, characterized by: The method comprises the following steps: S1, based on the milling operation program, milling simulation processing is performed in the CAD simulation software to obtain a processed free-form surface workpiece model, and a surface normal vector set NVC of the free-form surface workpiece model in a three-dimensional coordinate system is generated according to the free-form surface workpiece model; S2, based on the surface normal vector set NVC, a laser detector motion path LMP is generated, and the free-form surface workpiece model and an actual free-form surface workpiece are collected using a laser scattering method according to the laser detector motion path LMP to obtain a raw image data set Raw and perform standardization processing to obtain theoretical scattering spot image data T and actual scattering spot image data I; S3, based on the theoretical scattering spot image data T and the raw scattering spot image data I, difference covariance feature extraction is performed to obtain a difference covariance scalar D; S4, based on the theoretical scattered light spot image data T and the original scattered light spot image data I, generate the residual image data G and calculate the residual energy centroid coordinates (x', y'); calculate the second order central moment (μ 20 , μ 02 , μ 11 ) according to the residual image data G and the residual energy centroid coordinates (x', y'); S5, based on the second order central moment (μ 20 , μ 02 , μ 11 ) to obtain an anisotropy index A, to evaluate the difference in the stretching degree of the residual image data G in each direction; S6、based on the difference covariance scalar D and the anisotropy index A, calculate the surface quality index Q, and compare with the preset surface quality index threshold Q th comparison, generate the detection evaluation result AIR; S6 comprises S61 and S62; S61, based on the difference covariance scalar D and the anisotropy index A, a surface quality index algorithm is constructed, a nonlinear penalty function f(A) is designed for the anisotropy index A, and the surface quality index Q is obtained by calculation in combination with the difference covariance scalar D; The surface quality index algorithm expression is as follows: ； ； In the formula, ln represents a logarithmic function with a natural constant e as the base.

2. The non-contact surface roughness measurement method for freeform surface milling according to claim 1, wherein: S1 comprises S11; S11, based on the milling operation program, program codes for driving a numerical control machine tool and an original CAD model of a workpiece are input into the CAD simulation software to simulate a numerical control machine tool milling processing process, a free-form surface workpiece model is obtained, and a surface normal vector set NVC of the free-form surface workpiece model is calculated and generated in a three-dimensional coordinate system by surface scanning of the free-form surface workpiece model by the CAD simulation software.

3. The non-contact surface roughness measurement method for freeform surface milling according to claim 2, wherein: S2 comprises S21; S21, based on the set of surface normal vectors NVC, generating a laser probe motion path LMP perpendicular to the set of surface normal vectors NVC by the CAD simulation software, simulating the moving laser scattering probe in the CAD simulation software, and collecting an original theoretical scattering spot gray image T of M*N pixels of the freeform surface workpiece model R ; In the numerical control machine tool, input the laser detector motion path LMP, collect the M*N pixel original actual scattering spot gray image I of the actual free-form surface workpiece R , combine the original theoretical scattering spot image T R to generate the original image data set Raw; According to the original image data set Raw, the gray value of each pixel point of the original theoretical scattering light spot gray image T R and the original actual scattering light spot gray image I R is normalized and standardized to obtain the processed theoretical scattering light spot image data T and the actual scattering light spot image data I.

4. The non-contact surface roughness measurement method for freeform surface milling according to claim 3, wherein: S3 comprises S31; S31, based on the theoretical scattering spot image T and the raw scattering spot image I, the difference covariance scalar D is obtained by calculating the logarithmic intensity covariance of the actual scattering spot image data I and the theoretical scattering spot image data T, and the scattering deviation caused by the real roughness is quantified by evaluating the difference degree between the actual scattering spot image data I and the theoretical scattering spot image data T; The difference covariance scalar D calculation formula is as follows: ； ； ； In the formula, I(x, y) represents a gray scale normalized value of a coordinate (x, y) point in the actual scattering spot image data I, T(x, y) represents a gray scale normalized value of a coordinate (x, y) point in the theoretical scattering spot image data T, LI(x, y) represents a logarithmic operation on the gray scale normalized value of the actual scattering spot image data I, LT(x, y) represents a logarithmic operation on the gray scale normalized value of the theoretical scattering spot image data T, I' represents an average value of the gray scale normalized values of all pixels in the actual scattering spot image data I, T' represents an average value of the gray scale normalized values of all pixels in the theoretical scattering spot image data T, ln represents a logarithmic function with a natural constant e as the base, and ∑ represents a traversal summation operation.

5. The non-contact surface roughness measurement method for freeform surface milling according to claim 4, wherein: S4 comprises S41 and S42; S41, based on the theoretical scattering spot image data T and the original scattering spot image data I, calculating the absolute value of the gray scale normalized value of each pixel point of the theoretical scattering spot image data T and the original scattering spot image data I, and generating residual image data G; calculating a zeroth moment μ 00 and a first moment (μ 10 , μ 01 ) of the residual image data G, and calculating a centroid coordinate (x', y') of the residual image data G from the zeroth moment μ 00 and the first moment (μ 10 , μ 01 ) of the residual image data G. where μ 10 represents the horizontal component of the centroid of the residual image data G, μ 01 represents the vertical component of the centroid of the residual image data G.

6. The non-contact surface roughness measurement method for freeform surface milling according to claim 5, wherein: S42、According to the residual image data G and the residual energy centroid coordinates (x', y'), the second order central moment (μ 20 , μ 02 , μ 11 ) of the residual image data G is calculated. 20 , μ 02 , μ 11 ) of the residual image data G is calculated. where μ 20 denotes the horizontal extent of the residual image data G, μ 02 denotes the vertical extent of the residual image data G, μ 11 denotes the diagonal correlation of the residual image data G.

7. The non-contact surface roughness measurement method for freeform surface milling according to claim 6, wherein: S5 includes S51; S51, based on the second order central moment (μ 20 , μ 02 , μ 11 ), by calculating the maximum eigenvalue λ 20 and the minimum eigenvalue λ 02 of the covariance matrix C composed of the second order central moment (μ 11 , μ MAX , μ MIN ), and combining the difference between the maximum eigenvalue λ MAX and the minimum eigenvalue λ MIN as the normalization numerator, and combining the sum of the maximum eigenvalue λ MAX and the minimum eigenvalue λ MIN as the normalization denominator, the anisotropy index A is calculated, to evaluate the difference in the stretching degree of the residual image data G in each direction; Wherein, the covariance matrix C expression is as follows: ； Wherein, the calculation formula of the anisotropy index A is as follows: ； ； 。 8. The non-contact surface roughness measurement method for freeform surface milling according to claim 7, wherein: S62, compare the surface quality index Q with a preset surface quality index threshold Q th comparison; If the surface quality index Q < a preset surface quality index threshold Q th The detection evaluation result AIR is generated as a current workpiece roughness detection pass. If the surface quality index Q is greater than or equal to a preset surface quality index threshold Q th The detection evaluation result AIR is generated as the current workpiece roughness detection fails, and a prompt of checking the milling cutter and the operation program is given to the operator.

9. A non-contact surface roughness detection system for freeform surface milling, applied to the non-contact surface roughness detection method for freeform surface milling in any one of claims 1-8, characterized in that: It includes a simulation module, a data acquisition module, a differential covariance feature extraction module, a residual image second moment calculation module, an anisotropy index analysis module and a comprehensive evaluation module. The simulation module simulates the milling process in the CAD simulation software based on the milling operation program, obtains the free-form surface workpiece model after processing, and generates the surface normal vector set NVC of the free-form surface workpiece model in the three-dimensional coordinate system according to the surface normal vector set NVC; The data acquisition module generates the laser detector motion path LMP based on the surface normal vector set NVC, uses the laser scattering method to collect the original image data set Raw and obtain the theoretical scattering spot image data T and the actual scattering spot image data I according to the laser detector motion path LMP for the free-form surface workpiece model and the actual free-form surface workpiece; The differential covariance feature extraction module extracts the differential covariance feature based on the theoretical scattering spot image data T and the original scattering spot image data I, and obtains the differential covariance scalar D. The second moment calculation module of the residual image generates residual image data G and calculates residual energy centroid coordinates (x', y') based on the theoretical scattering speckle image data T and the original scattering speckle image data I, and calculates the second central moment (μ 20 , μ 02 , μ 11 ) according to the residual image data G and the residual energy centroid coordinates (x', y'). The anisotropy index analysis module obtains an anisotropy index A by performing a normalization calculation based on second-order central moments (μ 20 , μ 02 , μ 11 ) to evaluate the difference in the degree of stretching in each direction of the residual image data G. The comprehensive evaluation module calculates a surface quality index Q based on the differential covariance scalar D and the anisotropy index A, and compares the surface quality index Q with a surface quality index threshold Q th The comparison generates an inspection evaluation result AIR.

Images