A method and system for surface quality inspection of metals

Abstract

This invention discloses a method and system for surface quality inspection of metals, belonging to the field of metal surface quality inspection technology. It addresses the problems of low accuracy in detecting micro-defects on highly reflective surfaces, weak correlation between process parameters, and lack of closed-loop control. The method involves sequentially irradiating the metal surface with a multi-angle coherent light field, analyzing the phase shift of interference fringes to generate three-dimensional topography data, and suppressing reflection noise interference. Based on process parameter constraints, dynamic segmentation extracts the defect depth gradient, and deconvolution is used to calculate and quantify deposition temperature and pressure deviations, constructing a process deviation feature distribution. Finite element simulation is used to generate a process-topography mapping library, and cross-domain invariant features are extracted through depth-constrained manifold alignment and contrastive learning to establish a causal relationship model between defect types and process parameters. Process parameters are dynamically adjusted according to weight gradients, and backflow data is used to update manifold rules. This achieves high-precision three-dimensional defect detection, process deviation tracing, and adaptive parameter optimization.

Description

Technical Field

[0001] This invention relates to the field of metal surface quality testing technology, specifically to a method and system for testing the surface quality of metals. Background Technology

[0002] In modern manufacturing, metallic materials, as key basic materials, are widely used in high-end equipment fields such as aerospace, precision manufacturing, and the automotive industry. Their surface quality directly impacts the service life and reliability of components. As high-end manufacturing moves towards intelligence and high precision, industry places higher demands on the detection of microscopic defects on metal surfaces. This requires not only highly sensitive identification of minute defects but also further exploration of the underlying causes to support quality traceability and process optimization. Therefore, developing metal surface quality detection methods with deep process correlation analysis capabilities has become a crucial supporting technology for improving the intelligence level of manufacturing processes.

[0003] Existing methods for metal surface inspection mainly include visual image-based defect recognition, laser contour scanning, and interferometry. While these methods have made some progress in feature extraction and surface morphology reconstruction, they generally suffer from several technical bottlenecks: First, most methods remain at the level of two-dimensional image recognition, lacking in-depth analysis of the three-dimensional structure of defects; second, there is a disconnect between detection and process parameters, making it difficult to achieve causal correlation analysis between defects and manufacturing processes; third, in complex environments, the detection system lacks robustness to illumination interference and material changes, leading to fluctuations in defect recognition accuracy; fourth, traditional detection methods lack data closure and feedback capabilities, making it difficult to guide the dynamic adjustment and iterative optimization of process parameters. Therefore, there is an urgent need for an intelligent detection method that integrates three-dimensional morphology analysis, frequency domain modeling, physical simulation, and causal reasoning to achieve a systematic upgrade from defect recognition to defect cause analysis and process optimization. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a method and system for surface quality testing of metals, thus solving the problems mentioned in the background.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a surface quality inspection method for metals, comprising the following steps: S1. Irradiating the metal surface with a multi-angle coherent light field sequence, collecting interference fringes caused by defect depth, and analyzing the phase shift of the defect region in the interference fringes based on the nonlinear mapping relationship between the metal dielectric properties and the optical path difference, thereby generating phase domain data containing the three-dimensional morphology of the defect; S2. Performing multi-scale frequency domain decomposition and dynamic segmentation constrained by process parameters on the phase domain data of the three-dimensional morphology of the defect, extracting the defect depth gradient distribution, and calculating the deconvolution of the metal processing history parameters and the defect depth gradient distribution. S3. Quantify deposition temperature deviation and pressure fluctuation, and generate defect topological feature distribution of quantified process deviation through multi-scale topological decomposition and defect geometric feature statistics; S4. Construct a physical simulation map library of process parameter-defect morphology mapping relationship based on finite element simulation, align the defect topological feature distribution with simulation data in the depth-constrained manifold space, extract cross-domain invariant features through comparative learning, and output defect type and its causal correlation weight distribution with process parameters; S5. Dynamically adjust process parameters according to the gradient direction of causal correlation weight distribution, and feed back the detection data under the new parameters to the simulation map library to update the depth-constrained manifold alignment rules.

[0006] Furthermore, based on the nonlinear mapping relationship between the dielectric properties of the metal and the optical path difference, the specific process for analyzing the phase shift of the defect region in the interference fringes is as follows: noise suppression and background correction are performed on the interference fringe images acquired from multiple angles, and fringe distortion caused by environmental interference is eliminated by phase shift coding method; based on the spatial variation characteristics of the dielectric constant of the metal, a nonlinear mapping model between the optical path difference and the defect depth is constructed, and the model is transformed into a nonlinear integral expression with constrained boundary conditions; the integral expression is numerically solved by multi-scale guided filtering and iterative optimization algorithm to extract the phase shift value.

[0007] Furthermore, the specific process for generating phase domain data containing the three-dimensional morphology of the defect is as follows: the phase offset is restored to a continuous phase distribution through a phase unwrapping algorithm; the three-dimensional height field of the defect surface is calculated by combining the wavelength of the light wave and the incident angle parameters through a three-dimensional coordinate transformation model; the height field data is smoothed by surface interpolation to eliminate local distortion caused by noise; the depth gradient field, local principal curvature distribution and normal vector direction of the defect region are calculated through geometric feature analysis to generate phase domain data containing the three-dimensional morphology, gradient features and geometric properties.

[0008] Furthermore, the specific process of extracting the depth gradient distribution of defects by performing multi-scale frequency domain decomposition and dynamic segmentation under process parameter constraints on the phase domain data of the three-dimensional morphology of defects is as follows: multi-scale frequency domain decomposition is performed on the phase domain data, and substrate reflection noise and high-frequency defect signals are separated by adaptive frequency band selection; a dynamic segmentation threshold is generated based on the statistical distribution of historical metal processing parameters, and the segmentation sensitivity is adjusted by combining the probability density function of the phase gradient amplitude; the consistency of the depth gradient direction is verified on the segmented defect region, and pseudo-defect signals that deviate from the material stress field direction by more than a preset angle are removed.

[0009] Furthermore, based on the deconvolution calculation of historical metal processing parameters and defect depth gradient distribution, the specific process for quantifying deposition temperature deviation and pressure fluctuation is as follows: establish the physical response relationship between process parameters and defect depth gradient, obtain the temperature deviation and pressure fluctuation by iteratively solving the deconvolution equation under constraints; verify the statistical significance of the deconvolution results using random sampling methods, screen out defect regions strongly correlated with process parameter deviations, and mark their spatial location and degree of deviation.

[0010] Furthermore, the specific process of generating the defect topological feature distribution of quantitative process deviation through multi-scale topological decomposition and defect geometric feature statistics is as follows: the defect area is spatially divided, and the pore density distribution, the consistency coefficient between the crack propagation direction and the principal stress direction, and the surface curvature abrupt change characteristics of each sub-region are statistically analyzed; the quantitative mapping relationship between defect geometric features and process parameter deviation is constructed through statistical correlation analysis, and a feature distribution map reflecting the degree of process deviation is generated.

[0011] Furthermore, based on the finite element simulation, a physical simulation map library of process parameters-defect morphology mapping relationships is constructed. The specific process of aligning the defect topological feature distribution with the simulation data in a depth-constrained manifold space is as follows: Based on the multi-physics coupled finite element simulation model, the defect generation process of the metal surface under different combinations of process parameters is simulated to generate a simulation dataset containing temperature field, stress field distribution, and defect morphology; the actual detected defect topological feature distribution and simulation data are projected onto the same manifold space through a depth-constrained manifold embedding algorithm, where the manifold distance calculation introduces the defect depth gradient similarity weight, and the alignment of feature distribution is achieved through geodesic optimization.

[0012] Furthermore, the specific process of extracting cross-domain invariant features through contrastive learning and outputting the defect type and its causal association weight distribution with process parameters is as follows: construct positive and negative sample pairs of actual defect features and simulation data in the manifold space, design a contrastive loss function for process parameter constraints, and extract cross-domain invariant features through gradient descent optimization; based on the optimized feature distribution, calculate the association weight between defect type and process parameters through causal inference, wherein the weight distribution is dynamically updated through deconvolution gradient backpropagation.

[0013] Furthermore, the process parameters are dynamically adjusted according to the gradient direction of the causal correlation weight distribution, and the detection data under the new parameters is fed back to the simulation graph library. The specific process of updating the manifold alignment rules of the depth constraint is as follows: the process parameter adjustment amount is generated according to the negative gradient direction of the causal correlation weight distribution, and the parameter adjustment is ensured to conform to the physical feasible region through the constraint optimization algorithm; the detection data under the new parameters is added to the simulation graph library, and the depth constraint weight matrix of the manifold space is optimized through the incremental parameter update method to realize the collaborative iteration of the detection model and the process parameters.

[0014] A surface quality inspection system for metals includes the following modules: a multi-angle coherent imaging module, a process dynamic quantization module, a simulation manifold alignment module, and a process adjustment module. The multi-angle coherent imaging module is used to sequentially illuminate the metal surface with multi-angle coherent light fields, acquire interference fringes caused by defect depth, and analyze the phase shift of the defect region in the interference fringes based on the nonlinear mapping relationship between the metal's dielectric properties and optical path difference, generating phase domain data containing the three-dimensional morphology of the defects. The process dynamic quantization module is used to perform multi-scale frequency domain decomposition and dynamic segmentation of the phase domain data of the three-dimensional morphology of the defects based on process parameter constraints, extract the defect depth gradient distribution, and adjust the phase domain data based on historical metal processing parameters and the defect depth gradient distribution. The deconvolution calculation of the cloth quantifies the deposition temperature deviation and pressure fluctuation. Through multi-scale topological decomposition and defect geometric feature statistics, a defect topological feature distribution with quantified process deviation is generated. The simulation manifold alignment module is used to construct a physical simulation map library of process parameters-defect morphology mapping relationship based on finite element simulation. It aligns the defect topological feature distribution with the simulation data in the depth-constrained manifold space. Cross-domain invariant features are extracted through comparative learning, and the defect type and its causal correlation weight distribution with process parameters are output. The process adjustment module is used to dynamically adjust the process parameters according to the gradient direction of the causal correlation weight distribution, and feed the detection data under the new parameters back to the simulation map library to update the depth-constrained manifold alignment rules.

[0015] The present invention has the following beneficial effects:

[0016] (1) A surface quality inspection method for metals, by introducing a multi-angle coherent optical field interference mechanism and a nonlinear phase solution method, can achieve high-precision three-dimensional morphology extraction of the defect region on the metal surface; combined with multi-scale frequency domain decomposition and dynamic region segmentation under process parameter constraints, it not only improves the separability of the defect depth gradient, but also realizes for the first time the quantitative inversion of deposition temperature deviation and pressure fluctuation based on morphology data, effectively enhancing the interpretability and process sensitivity of the inspection method. By integrating the process parameter-defect morphology map library generated by finite element simulation, and using the depth-constrained manifold alignment mechanism and contrastive learning algorithm, high-dimensional consistency alignment between the defect topology distribution and the simulation results is achieved. Furthermore, based on the causal correlation weight distribution, the process parameters are dynamically adjusted, and a closed-loop feedback mechanism for the inspection data is constructed, which can continuously optimize the simulation alignment rules, forming an intelligent inspection system with adaptive evolution capabilities, significantly improving the defect classification accuracy and process response capability of the system.

[0017] (2) A surface quality inspection system for metals integrates a multi-angle coherent imaging module, a process dynamic quantification module, a simulation manifold alignment module, and a process adjustment module to achieve full-process automation from data acquisition to process feedback, adapting to the real-time inspection needs of high-speed production lines. Through a physical simulation atlas library and manifold space alignment mechanism, it integrates actual inspection data with multi-physics simulation results, enhancing the system's generalization ability to complex working conditions. Based on causal correlation weighting and parameter optimization algorithms, it directly drives the dynamic adjustment of process units such as deposition equipment and heat treatment furnaces, reducing manual intervention and promoting the implementation of a closed-loop intelligent manufacturing system.

[0018] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0019] Figure 1 This is a flowchart of a surface quality testing method for metals according to the present invention.

[0020] Figure 2 This is a flowchart of a surface quality inspection system for metals according to the present invention. Detailed Implementation

[0021] This application provides a surface quality inspection method and system for metals, addressing the core problems of low accuracy in micro-defect detection under high-reflection interference, weak correlation between defects and process parameters, and insufficient generalization ability in small-sample scenarios in existing technologies. The overall approach is as follows:

[0022] By using multi-angle coherent light field serial illumination and phase domain analysis technology, the reflection noise of the metal surface is removed, and the interference fringes are transformed into three-dimensional topographic data containing defect depth and curvature, breaking through the limitations of traditional two-dimensional image detection.

[0023] By combining historical process parameters with multi-scale frequency domain decomposition, a dynamic segmentation and deconvolution model is constructed to directly invert process deviations such as temperature and pressure from defect morphology data, thereby achieving physical interpretability analysis of defect causes.

[0024] Based on the multiphysics finite element simulation generation process-morphology mapping library, feature transfer between actual detection data and simulation data is achieved through depth-constrained manifold alignment and contrastive learning, solving the problem of small sample defect classification.

[0025] The process parameters are dynamically adjusted according to the gradient direction of the causal weight distribution, and the adjusted new detection data is fed back to the simulation library to realize a closed-loop optimization process of detection-simulation-control.

[0026] Please see Figure 1 This invention provides a technical solution: a method for surface quality inspection of metals, comprising the following steps: S1. Irradiating the metal surface with a multi-angle coherent light field sequence, collecting interference fringes caused by defect depth, and analyzing the phase shift of the defect region in the interference fringes based on the nonlinear mapping relationship between the metal dielectric properties and the optical path difference, generating phase domain data containing the three-dimensional morphology of the defect; S2. Performing multi-scale frequency domain decomposition and dynamic segmentation constrained by process parameters on the phase domain data of the three-dimensional morphology of the defect, extracting the defect depth gradient distribution, and quantizing the deposition based on the deconvolution of the metal processing history parameters and the defect depth gradient distribution. S3. Based on the temperature deviation and pressure fluctuation, a physical simulation map library of process parameters-defect morphology mapping relationship is constructed. The defect topology feature distribution is aligned with the simulation data in the depth-constrained manifold space. Cross-domain invariant features are extracted through comparative learning, and the defect type and its causal correlation weight distribution with the process parameters are output. S4. The process parameters are dynamically adjusted according to the gradient direction of the causal correlation weight distribution, and the detection data under the new parameters is fed back to the simulation map library to update the depth-constrained manifold alignment rules.

[0027] In this implementation scheme, step S1: By irradiating the metal surface with a multi-angle coherent light field and acquiring interference fringes, high-precision three-dimensional morphology detection of the depth and distribution of micro-defects (such as scratches, dents, etc.) on the metal surface is achieved. The acquisition of phase domain data provides high-quality raw input for subsequent detailed defect analysis and process backtracking. Coherent light field: Laser irradiation with correlated phases can produce interference phenomena. Interference fringes: Fringe images formed by different optical path differences, reflecting the surface undulations. Phase offset: Reflecting local surface height changes, it is key to three-dimensional reconstruction. Phase domain data: The set of phase information at each point, which can be mapped to a three-dimensional surface model. Step S2: Multi-scale frequency domain decomposition of the phase domain data combined with dynamic segmentation based on process parameters can accurately identify the trend of defect morphology changes and infer potential process anomalies such as temperature and pressure causing defects during manufacturing. Furthermore, statistically significant spatial distribution characteristics of defects are generated through topological analysis. Multi-scale frequency domain decomposition: Such as wavelet transform, extracting morphological details at different scales. Dynamic segmentation: Flexibly adjusting the defect identification range according to the specific process background. Deconvolution calculation: A mathematical deduction method to deduce original process fluctuations from defects. Defect topological feature distribution: A set of features describing the morphology, structure, and spatial distribution of defects. Step S3: Construct a "process-defect" physical map based on finite element simulation, perform alignment matching with the detected topological features, extract common features between different data sources using a contrastive learning algorithm, and finally output the defect type and its quantitative causal weight with process parameters, which helps in source tracing and predictive control. Finite element simulation: Simulate the generation mechanism of defects in metal processing using numerical methods. Manifold space alignment: Embed simulation data and actual detection data in the same geometric space to make them comparable. Contrastive learning: Learn the essential features of data from "similar-dissimilar" pairs to enhance cross-domain adaptability. Causal association weight distribution: Characterize the causal strength and direction between a certain process variable and the defect type. Step S4: Based on the causal association weight between defects and process parameters, automatically identify key processes that affect them and perform parameter fine-tuning and optimization to improve product consistency. New detection data feeds back into the simulation model, continuously optimizing the alignment mechanism to achieve online model iteration and closed-loop process control. Dynamic gradient direction adjustment: Optimize process parameters based on the direction of maximum change in causal weights. Data feedback: Feed new detection data back to the model to enhance simulation prediction capabilities. Manifold alignment rule update: Dynamically calibrate mapping relationships to adapt to new process scenarios or material changes.

[0028] Specifically, based on the nonlinear mapping relationship between the dielectric properties of metals and optical path difference, the specific process for analyzing the phase shift of the defect region in the interference fringes is as follows: noise suppression and background correction are performed on the interference fringe images acquired from multiple angles, and fringe distortion caused by environmental interference is eliminated by phase shift coding method; based on the spatial variation characteristics of the dielectric constant of metals, a nonlinear mapping model between optical path difference and defect depth is constructed, and this model is transformed into a nonlinear integral expression with constrained boundary conditions; the integral expression is numerically solved by multi-scale guided filtering and iterative optimization algorithm to extract the phase shift value.

[0029] In this implementation scheme, interference fringe images are typically affected by environmental noise and illumination variations, impacting image accuracy. Noise suppression aims to remove irrelevant noise signals from the image, enhancing the recognizability of the interference fringes. Common noise suppression methods include Gaussian filtering, mean filtering, or median filtering. These methods smooth the image, reducing the interference of random noise on interference fringe resolution. Since metal surfaces may exhibit uneven illumination or surface reflection, background correction removes light intensity variations caused by these factors, ensuring the interference fringe morphology is correlated with metal surface defects. Background smoothing methods, such as low-pass filtering, correct background light intensity and eliminate these effects. Interference image representation model (frame k) during noise suppression and background correction stages: Parameter explanation: The intensity of the interference image at the k-th phase shift; Background light intensity (i.e., static component); : Modulated light intensity (i.e., amplitude of interference fringes); The initial phase distribution to be determined; The phase quantity of the k-th phase shift is usually... ; Noise term, including environmental interference and electronic noise; phase recovery formula (based on least squares method): Parameter explanation: The total number of phase-shifted frames, with other parameters remaining the same as in the previous formula. Nonlinear mapping model construction (metal dielectric properties and optical path difference), the relationship between the optical path difference distribution on the metal surface and the defect depth: Parameter explanation: Optical path difference caused by metal defects; Local depth at the defect; The equivalent dielectric function of the defect region varies along the depth. : Reference dielectric constant of the defect-free region of the metal. Constraints: , Parameter explanation: , The dielectric range allowed by the physical properties of the material; : Maximum possible defect depth. Discretization of the nonlinear integral expression and phase shift inversion discretize the integral model into a linear algebraic problem, constructing a linear mapping matrix: Parameter explanation: : A coefficient matrix generated from discrete dielectric differences; : Defect depth vector (unknown quantity); The vector corresponding to the optical path difference (which can be extracted from the interference fringes) is used to solve for the phase shift. A regularized iterative optimization objective function is introduced: Parameter explanation: The final phase offset value obtained; The wavelength of the laser used; Regularization weight coefficient; Gradient smoothing term (used to prevent excessive phase oscillation). Multi-scale guided filtering and iterative solution process; the multi-scale guided filter operates on scale... The guiding formula is: Parameter explanation: :scale The guided filter output phase; : Input phase estimate; Mean filter output; Guide image (grayscale) at scale Local mean under; : Guided filter response coefficients. Iterative optimization employs a successive approximation method: Parameter explanation: The phase solution of the nth iteration; Learning rate / step size coefficient; : The gradient of the loss function.

[0030] Specifically, the process of generating phase domain data containing the three-dimensional morphology of the defect is as follows: the phase offset is restored to a continuous phase distribution through a phase unwrapping algorithm; the three-dimensional height field of the defect surface is calculated by combining the wavelength of light and the incident angle parameters through a three-dimensional coordinate transformation model; the height field data is smoothed by surface interpolation to eliminate local distortion caused by noise; the depth gradient field, local principal curvature distribution and normal vector direction of the defect region are calculated through geometric feature analysis to generate phase domain data containing the three-dimensional morphology, gradient features and geometric properties.

[0031] In this implementation scheme, the purpose of phase unwrapping is to unwrap the wrapped phases obtained from the original fringe pattern. Restored to a physically continuous phase distribution Processing method: Mass-guided phase unwrapping is used. A three-dimensional coordinate transformation model is used to calculate the surface height field, resulting in a continuous phase. Converted to relative height values ​​on the defect surface The calculation model is as follows: Parameter explanation: : at pixel coordinates The corresponding relative three-dimensional height value at that location; The actual observed wavelength of the light wave (632.8 nm, commonly used in metal interferometry with HeNe laser); : The angle between the incident light and the normal to the metal surface; Continuous phase after unwrapping. Surface interpolation and smoothing are used to transform the height field. Interpolation (B-spline) and smoothing (bidirectional Gaussian filtering) operations are performed to reduce local distortions caused by stripe blur and image noise. Interpolation method: Constructing a polynomial spline surface. Maintaining the main trend through smoothing filtering: Anisotropic diffusion filtering or multi-scale wavelet denoising is employed to improve surface continuity. Gradient field calculation in the defect region is based on the smoothed height field. Calculate the local depth gradient field : Parameter explanation: : Depth gradient vector on a two-dimensional plane; :along Rate of change of elevation along the axial direction; :along The rate of change of elevation along the axial direction. The local principal curvatures are calculated by constructing the Hessian matrix using the second derivative, yielding the principal curvatures κ1 and κ2. Parameter explanation: : The Hessian matrix of the second derivative of local elevation; : Principal curvature (maximum, minimum) at (u,v); Eigenvalue operations on a Hessian matrix, solving for the two real eigenvalues ​​of the matrix; normal vector direction estimation, calculating the normal vector of a metal surface from the gradient field. : ,in, Parameter explanation: : Unit normal vector, representing the surface orientation; : These are the height fields at The first derivative in the direction; Normalization factor to ensure the normal vector is a unit vector. Phase domain data construction: Finally, the following information is integrated into a structured phase domain dataset: 3D height field: Local gradient field: Principal curvature: Normal vector direction: This phase domain data can be used as input for subsequent geometric identification of metal defects, material damage assessment, and defect source tracing analysis.

[0032] Specifically, the process of extracting the depth gradient distribution of defects by performing multi-scale frequency domain decomposition and dynamic segmentation constrained by process parameters on the phase domain data of the three-dimensional morphology of defects is as follows: multi-scale frequency domain decomposition is performed on the phase domain data, and substrate reflection noise and high-frequency defect signals are separated by adaptive frequency band selection; a dynamic segmentation threshold is generated based on the statistical distribution of historical metal processing parameters, and the segmentation sensitivity is adjusted by combining the probability density function of the phase gradient amplitude; the consistency of the depth gradient direction is verified on the segmented defect region, and pseudo-defect signals that deviate from the material stress field direction by more than a preset angle are removed.

[0033] In this implementation scheme, multi-scale frequency domain decomposition separates the substrate reflection noise (low-frequency component) from the defect information (high-frequency component) in the phase domain height field. Multi-scale frequency domain decomposition (such as wavelet transform or discrete cosine transform) is commonly expressed as follows: Parameter explanation: Original three-dimensional height field; : The frequency domain components of the s-th layer (including low-frequency substrate and high-frequency defects); Total number of decomposition scales; Indecomposable high-frequency residuals; adaptive frequency band selection automatically divides high and low frequency bands based on frequency domain amplitude statistics, selecting a set of frequency band indices that meet the high-frequency energy concentration conditions. Used to extract defect signals: Parameter explanation: : The average amplitude of the frequency domain components in the s-th layer; : The standard deviation of the amplitude of the s-th layer; The sensitivity coefficient is set empirically, usually between 1.5 and 3. The set of frequency bands identified as containing defect signals. A joint probability distribution is formed by dynamically setting a segmentation threshold (process parameter constraints) and combining it with historical metal processing parameters (such as depth of cut, feed rate, etc.). And dynamically set the defect segmentation threshold based on statistical quantiles. : Parameter explanation: : Depth gradient magnitude after bandgap synthesis; : Joint distribution function of metal processing parameters; : The p-quantile function of the distribution (e.g., the 85th percentile); Dynamic threshold for defect determination under process constraints; probability density function for obtaining gradient magnitude (PDF) for segmentation sensitivity adjustment (based on gradient probability density). And adjust the segmentation sensitivity coefficient by its skewness or kurtosis. The updated judgment threshold is: ,in, Parameter explanation: : The probability density function of the depth gradient magnitude; : The skewness or skewness of the PDF; Sensitivity amplification factor enhances the response to small defects; Adjusted threshold. Depth gradient direction consistency verification (removing false defects): For the initial segmentation results, calculate the normal vector direction in each defect region. With respect to the principal stress direction of the material The included angle : Remove regions that meet the following conditions: Parameter explanation: : Average normal vector within the defect area; The known stress field direction vector of the material; Angle between vectors; : Deviating from the threshold angle, generally ;like If the value exceeds the set range, it will be judged as processing noise or false detection of surface texture.

[0034] Specifically, based on the deconvolution calculation of historical metal processing parameters and defect depth gradient distribution, the specific process for quantifying deposition temperature deviation and pressure fluctuation is as follows: establish the physical response relationship between process parameters and defect depth gradient, obtain the temperature deviation and pressure fluctuation by iteratively solving the deconvolution equation under constraints; verify the statistical significance of the deconvolution results using random sampling methods, screen out defect regions that are strongly correlated with process parameter deviations, and mark their spatial location and degree of deviation.

[0035] In this implementation scheme, the physical response relationship between defect depth gradient and processing temperature / pressure disturbance is established; the defect gradient distribution is assumed. For temperature field perturbation With pressure disturbance The convolutional superposition response is used to construct the following physical response model: ;in, : Depth gradient distribution in the defect region; Temperature response kernel function (thermal gradient-morphological response); : Pressure response kernel function (stress field-morphological response); *: Two-dimensional convolution operator; Localized depositional temperature deviation distribution; Localized processing pressure fluctuation distribution; Systematic error terms (Gaussian noise, measurement errors, etc.); iteratively solve the deconvolution problem under constraints to recover the perturbation source field; introduce Tikhonov regularization constraints (to suppress ill-posedness of the solution) and establish the deconvolution objective function: ;in, L2 norm, measures the fitting error; Regularization coefficient (controls smoothness); The gradient of the temperature / pressure perturbation field; numerical iterative solution using alternating least squares or gradient descent method, outputting the result. Statistical sampling was used to verify the significance of the deconvolution results and to quantify the correlation strength between defects and perturbation bias for each defect region. The set of perturbation values ​​after sampling and deconvolution Calculate its Z-score relative to the standard process value: , ; : The mean disturbance value of the j-th region; : Historical process standard perturbation mean; : Corresponding standard deviation; if (like If the significance level is 0.05, then the disturbance within the region is considered significantly abnormal. Significantly abnormal regions are then selected, and their spatial coordinates and degree of deviation are marked. The following marking function is used to output the coordinates and degree of abnormality: ; satisfy all The region outputs its center location With the maximum deviation amplitude .

[0036] Specifically, the process of generating the defect topological feature distribution of quantitative process deviation through multi-scale topological decomposition and defect geometric feature statistics is as follows: the defect area is spatially divided, and the pore density distribution, the consistency coefficient between the crack propagation direction and the principal stress direction, and the surface curvature abrupt change characteristics of each sub-region are statistically analyzed; the quantitative mapping relationship between defect geometric features and process parameter deviation is constructed through statistical correlation analysis, and a feature distribution map reflecting the degree of process deviation is generated.

[0037] In this implementation scheme, spatial region division and sub-region topological feature extraction are performed to identify defective regions. Divided into multiple sub-regions For each sub-region, the following three types of geometric topological features are extracted: Hole density distribution (number of defects / holes per unit area): ; : Subregion Pore ​​density; : The number of holes identified; : The area of ​​the crack. The cosine of the crack propagation direction and the principal stress direction (cosine of the direction angle): ;in, : Coefficient of directional consistency; : The principal direction vector of the crack in the i-th region (obtained by fitting the crack edge); : Principal stress direction vectors at the corresponding locations in the material (derived from stress field simulation or actual measurement); The angle between the two. Surface curvature abrupt change characteristics (ratio of maximum principal curvature to average principal curvature): ; Curvature mutation index; The maximum value of the principal curvature set at each point in the i-th region; The average principal curvature within the i-th region. A mapping model between defect geometric features and process parameter deviations is constructed using linear regression or multivariate statistical modeling to establish the response relationship between features and deviations. ;in, : The predicted process parameter deviation value (e.g., temperature deviation or load unevenness) for the corresponding sub-region. Regression coefficients reflect the sensitivity of each feature to the influence of the bias. The residual term, containing random perturbations and model errors, can be mapped to the optimal parameters using least-squares fitting. A topological feature distribution map reflecting the degree of process deviation is generated, encompassing all sub-regions. Mapping back to its original spatial coordinates forms a two-dimensional deviation feature map: ;in, : Indicates position On the strength of process deviation; Indicator function, if Belongs to the region If it is 1, then it is 1; otherwise, it is 0.

[0038] Specifically, the process of aligning the defect topological feature distribution with the simulation data within a depth-constrained manifold space is as follows: Based on the finite element simulation model, the defect generation process of the metal surface under different combinations of process parameters is simulated, generating a simulation dataset containing temperature field, stress field distribution, and defect morphology; the actual detected defect topological feature distribution and the simulation data are projected onto the same manifold space through a depth-constrained manifold embedding algorithm, where the manifold distance calculation introduces the defect depth gradient similarity weight, and the alignment of feature distribution is achieved through geodesic optimization.

[0039] In this implementation plan, a physical simulation library of the mapping relationship between process parameters and defect morphology is constructed using finite element simulation. First, a multi-physics coupled finite element simulation model is established. This model simulates the generation process of metal surface defects under different combinations of process parameters (such as temperature, stress, and pressure). The simulation process generates a physical simulation library including a temperature field (…). ), stress field ( ) and defect morphology ( The simulation dataset, in which, , , These are spatial coordinates. Temperature field: This represents the temperature distribution at various points on a metal surface. The temperature distribution is typically influenced by the material's thermal conductivity, the heating method, and boundary conditions. Stress field: It refers to the stress distribution on the surface and inside of a material, which depends on the applied external force and temperature fields, and characterizes the deformation process during metal processing. Defect morphology: The 3D morphology data of defects generated under different process parameters is a key output of the simulation model. Then, a depth-constrained manifold embedding algorithm is used to embed the actual detected defect topological features (such as...) ) and simulation data (such as They are aligned to the same manifold space. In this process, the manifold distance calculation introduces the defect depth gradient (…). Similarity weights for manifolds. Manifold distance calculation: manifold distance This involves calculating the geometric distance between actual defect data and simulated data using a manifold learning algorithm. Depth gradient weights: The depth gradient information of the defect serves as a weighting term, influencing the accuracy of the manifold alignment process. Depth gradient Used to measure local variations on the defect surface and enhance alignment in areas with significant depth variations. Geodesic optimization: Within the manifold space, geodesic optimization... This is used to minimize the deviation in the distribution of defect features and ensure alignment between actual defects and simulation data.

[0040] Specifically, the process of extracting cross-domain invariant features through contrastive learning and outputting the defect type and its causal correlation weight distribution with process parameters is as follows: construct positive and negative sample pairs of actual defect features and simulation data in the manifold space, design a contrastive loss function for process parameter constraints, and extract cross-domain invariant features through gradient descent optimization; based on the optimized feature distribution, calculate the correlation weight between defect type and process parameters through causal inference, wherein the weight distribution is dynamically updated through backpropagation of deconvolution gradient.

[0041] In this implementation scheme, cross-domain invariant features are extracted within the manifold space through contrastive learning. This facilitates the analysis of the correlation between defect types and process parameters. The specific steps are as follows: First, based on actual defect features and simulation data, positive and negative sample pairs are constructed. Positive sample pairs consist of similar defect topological features and corresponding process parameters, while negative sample pairs consist of samples with different defect types or process conditions. Positive sample pairs: ( When the defect topology features and process parameters are consistent. Negative sample pairs: ( When the defect topology characteristics and process parameter combinations are inconsistent, design a contrastive loss function. This is used to optimize the feature embeddings of positive and negative sample pairs. The contrastive loss function measures whether the distance between similar sample pairs in the embedding space should be small, and the distance between dissimilar sample pairs should be large. Common contrastive loss functions include: ;in, It is the embedding of positive sample pairs. It is the embedding of negative sample pairs. It represents the Euclidean distance in the embedding space. The contrastive loss is optimized by minimizing the contrastive loss function using the gradient descent optimization algorithm (Adam optimizer). This allows for the extraction of cross-domain invariant features. These features are applicable to both actual detection data and simulation data, preserving crucial information related to process parameters. Causal inference and the calculation of the association weights between defect types and process parameters are then performed. The optimized feature distribution can be used to calculate the causal relationship between defect types and process parameters through causal inference methods. This process dynamically updates the weight distribution based on backpropagation of deconvolutional gradients. Causal inference model: The causal inference model... The model infers the causal relationship between process parameters and defect types. It uses feature representation (...). This is used to quantify the impact of different process parameters on defect types. Weight calculation: Calculate the causal weight distribution of process parameters. This weight represents the contribution of each process parameter to the defect generation process.

[0042] Specifically, the process parameters are dynamically adjusted according to the gradient direction of the causal correlation weight distribution, and the detection data under the new parameters is fed back to the simulation graph library. The specific process of updating the manifold alignment rules of the depth constraint is as follows: the process parameter adjustment amount is generated according to the negative gradient direction of the causal correlation weight distribution, and the parameter adjustment is ensured to conform to the physical feasible region through the constraint optimization algorithm; the detection data under the new parameters is added to the simulation graph library, and the depth constraint weight matrix of the manifold space is optimized through the incremental parameter update method to realize the collaborative iteration of the detection model and the process parameters.

[0043] In this implementation scheme, the process parameter adjustment amount is generated based on the negative gradient direction of the causal weight distribution. First, the adjustment direction of the process parameters is determined using the previously calculated negative gradient direction of the causal weight distribution. The negative gradient direction represents the direction required for process parameter adjustment to reduce inconsistencies between defect generation and process parameters. Negative gradient direction: The negative gradient direction of the causal weight distribution Wcausal is calculated, which is the adjustment direction of the process parameters. Let the process parameter be p, and the negative gradient direction be: Where p = (p1, p2, ..., pn) represents the process parameter vector, and Wcausal is the causal weight distribution between process parameters and defects. Adjustment amount generation: Based on the negative gradient direction, the adjustment amount Δp of the process parameters is generated. This adjustment amount characterizes the corrections needed for each process parameter. Generally, the calculation form of the adjustment amount is as follows: Where α is the step size coefficient, used to control the magnitude of each adjustment. Constraint optimization algorithms ensure that parameter adjustments conform to the physical feasible region. To ensure that the results of process parameter adjustments remain within the physical feasible region, constraint optimization algorithms are used to correct the adjustment amounts. Feasible region constraints include physical limitations such as temperature, stress, and pressure. Constraint optimization problem: The constraint optimization problem for process parameter adjustment can be expressed as: Where f(p) is the objective function, which can be the error function of the process parameter adjustment amount, and Pfeasible is the physical feasible region of the process parameter. Optimization process: Using constrained optimization algorithms (such as interior point method, quasi-Newton method, etc.), the adjustment amount Δp is optimized to ensure that the adjusted process parameter pnew = p + Δp is within the feasible region. Once the new process parameter is determined to be within the physical feasible region, the next step is to conduct production testing and feed the testing data under the new parameter back to the simulation graph library to further improve the accuracy of the simulation graph. Testing data generation: Using the new process parameters... Inspect the actual production process and collect new defect datasets. Data reflow: This involves feeding the new detection dataset back into the system. Adding to the existing simulation atlas library enriches the diversity and accuracy of simulation data. After feeding new detection data back to the simulation atlas library using an incremental parameter update method to optimize the depth constraint weight matrix of the manifold space, the manifold alignment rules and depth constraint weight matrix need to be updated. (Depth constraint weight matrix in manifold space) This is used to accurately align features between actual detection data and simulation data. Incremental update: An incremental parameter update method is used to update the depth constraint weight matrix in the manifold space. The update process can be represented as: ;in, It's about updating the step size. This update, calculated based on new data, represents the new alignment requirement between the defect feature distribution and the simulation data. Optimization objective: To optimize the depth constraint weight matrix of the manifold space. This ensures a more precise alignment between actual and simulated data within the manifold space, thereby improving the accuracy of subsequent detection models. Finally, based on the updated depth-constrained manifold alignment rules, the detection model and process parameters begin collaborative iteration to further refine the relationship between process parameters and defect features. This iterative process continues until the matching degree between process parameters and defect generation reaches the predetermined accuracy requirements. Collaborative Iteration Process: In each iteration, the adjustment amount of process parameters interacts with the feedback from detection data, gradually reducing errors in the defect generation process by adjusting the process parameters. Ultimate Goal: To achieve precise control of the defect generation process through iterative optimization based on feedback from process parameters and detection data, thereby improving production efficiency and product quality.

[0044] Please see Figure 2 A surface quality inspection system for metals includes the following modules: a multi-angle coherent imaging module, a process dynamic quantization module, a simulation manifold alignment module, and a process adjustment module. The multi-angle coherent imaging module is used to sequentially illuminate the metal surface with multi-angle coherent light fields, acquire interference fringes caused by defect depth, and analyze the phase shift of the defect region in the interference fringes based on the nonlinear mapping relationship between the metal's dielectric properties and optical path difference, generating phase domain data containing the three-dimensional morphology of the defects. The process dynamic quantization module is used to perform multi-scale frequency domain decomposition and dynamic segmentation of the phase domain data of the three-dimensional morphology of the defects based on process parameter constraints, extract the defect depth gradient distribution, and adjust the phase domain data based on historical metal processing parameters and the defect depth gradient. The deconvolution calculation of the cloth quantifies the deposition temperature deviation and pressure fluctuation. Through multi-scale topological decomposition and defect geometric feature statistics, it generates a defect topological feature distribution that quantifies process deviations. The simulation manifold alignment module is used to construct a physical simulation map library of process parameters-defect morphology mapping relationship based on finite element simulation. It aligns the defect topological feature distribution with the simulation data in the depth-constrained manifold space. Through contrastive learning, it extracts cross-domain invariant features and outputs the defect type and its causal correlation weight distribution with process parameters. The process adjustment module is used to dynamically adjust the process parameters according to the gradient direction of the causal correlation weight distribution and feed the detection data under the new parameters back to the simulation map library to update the depth-constrained manifold alignment rules.

[0045] In this implementation scheme, a multi-angle coherent imaging module is included. This module employs a multi-angle coherent light field sequential illumination method, combined with a nonlinear mapping model between the dielectric properties of metallic materials and optical path difference, to analyze the phase shift of interference fringes caused by defects on the metal surface, thereby generating phase domain data with three-dimensional depth information. Compared with existing single-angle or fixed-viewpoint coherent detection methods, this module significantly improves spatial resolution and depth recognition capabilities, and is particularly suitable for fine imaging of complex morphologies or microstructure defects. A process dynamic quantification module is also included. This module introduces a multi-scale frequency domain decomposition + process-constrained dynamic segmentation method to separate high-frequency defect signals from low-frequency background noise in the defect morphology data. Simultaneously, a dynamic segmentation threshold is constructed by combining the statistical distribution of historical metal processing parameters, enabling the segmentation process to adapt to different process conditions. Furthermore, the deposition temperature and pressure fluctuations are quantified using a defect depth gradient deconvolution method. This path of inferring process perturbations from morphology is significantly superior to traditional empirical inference methods, enabling quantifiable, localizable, and traceable identification of abnormal process states, possessing high industrial practical value and technological breakthrough. Simulation Manifold Alignment Module: This module, based on a multi-physics coupled finite element simulation library, is the first to embed the topological features of defects obtained from actual detection into the simulation manifold space. It introduces defect depth gradient similarity weights and a geodesic optimization mechanism to improve the accuracy of feature alignment. Through depth-constrained manifold embedding and a contrastive learning mechanism, it can extract cross-domain (actual detection vs. simulation) invariant features, overcoming the identification bias problem caused by inconsistencies between actual working conditions and simulation data in traditional detection systems. This provides a more stable feature foundation for defect identification and causal tracing, demonstrating significant technological advancement. Process Adjustment Module: This module utilizes the causal correlation weight distribution between defect types and process parameters generated in the previous step. It employs a gradient direction optimization method to achieve directional and quantitative adjustments to process parameters. Simultaneously, it dynamically feeds back detection data under new conditions to the simulation library and synchronously optimizes the manifold embedding rules through an incremental update mechanism, forming a complete closed-loop iterative mechanism of "detection-cause determination-adjustment-re-detection-re-optimization". Unlike traditional process feedback methods that are based on rules or experience, this module has data-driven, adaptive adjustment, and autonomous learning and evolution capabilities, making it a key supporting component for realizing closed-loop control in intelligent manufacturing.

[0046] In summary, this application has at least the following effects:

[0047] A surface quality inspection method and system for metals is proposed. It establishes a mapping relationship between process parameters and defect morphology through multi-physics coupled finite element simulation, making the defect generation process predictable and controllable under different process conditions. A depth-constrained manifold embedding algorithm, integrating defect depth gradient similarity and geodesic optimization strategies, aligns actual detected defects with simulated defects in the feature space, improving defect identification accuracy. A contrastive learning mechanism under process parameter constraints effectively extracts cross-domain invariant features, providing a stable representation basis for defect identification and classification. Causal inference methods are used to dynamically calculate the correlation weights between defect types and process parameters, quantifying the impact of process factors on defect formation and supporting decision optimization for quality control. Process parameters are dynamically adjusted based on the gradient direction of the causal correlation weight distribution, and closed-loop optimization of the defect detection model and process design is achieved through data feedback and incremental updates of manifold weights. Continuous feedback of new detection data and updates to the simulation map ensure the timeliness and representativeness of the simulation database, enhancing the system's adaptability under complex operating conditions.

[0048] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0049] This invention is described with reference to flowchart illustrations and / or block diagrams of systems, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0050] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1The function specified in one or more boxes.

[0051] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0052] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0053] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for surface quality inspection of metals, characterized in that, Includes the following steps: S1. By sequentially irradiating the metal surface with multi-angle coherent light fields, interference fringes caused by the defect depth are collected. Based on the nonlinear mapping relationship between the metal dielectric properties and the optical path difference, the phase shift of the defect region in the interference fringes is analyzed, and phase domain data containing the three-dimensional morphology of the defect is generated. S2. Perform multi-scale frequency domain decomposition and dynamic segmentation of process parameter constraints on the phase domain data of the three-dimensional morphology of defects, extract the defect depth gradient distribution, and quantify the deposition temperature deviation and pressure fluctuation based on the deconvolution calculation of the metal processing history parameters and the defect depth gradient distribution. Generate the defect topological feature distribution with quantified process deviation through multi-scale topological decomposition and defect geometric feature statistics. S3. Construct a physical simulation map library of process parameters-defect morphology mapping relationship based on finite element simulation, align the defect topology feature distribution with the simulation data in the depth-constrained manifold space, extract cross-domain invariant features through comparative learning, and output the defect type and its causal correlation weight distribution with process parameters. S4. Dynamically adjust the process parameters according to the gradient direction of the causal correlation weight distribution, and feed the detection data under the new parameters back to the simulation graph library to update the manifold alignment rules of the depth constraint; The specific process for quantifying deposition temperature deviation and pressure fluctuation based on the deconvolution calculation of historical metal processing parameters and defect depth gradient distribution is as follows: Establish the physical response relationship between process parameters and defect depth gradient, and obtain the temperature deviation and pressure fluctuation by iteratively solving the deconvolution equation under constraints. The statistical significance of the deconvolution results was verified by random sampling, and defect regions that are strongly correlated with process parameter deviations were screened out and their spatial locations and deviation degrees were marked. The specific process of aligning the distribution of defect topological features with the simulation data within a depth-constrained manifold space, based on the physical simulation library that constructs the mapping relationship between process parameters and defect morphology using finite element simulation, is as follows: Based on the multiphysics coupled finite element simulation model, the defect generation process of metal surface under different combinations of process parameters is simulated, and a simulation dataset containing temperature field, stress field distribution and defect morphology is generated. The depth-constrained manifold embedding algorithm projects the actual detected defect topological feature distribution and simulation data onto the same manifold space. The manifold distance calculation introduces the defect depth gradient similarity weight, and the alignment of feature distribution is achieved through geodesic optimization. The specific process of extracting cross-domain invariant features through contrastive learning and outputting the defect type and its causal relationship weight distribution with process parameters is as follows: In the manifold space, positive and negative sample pairs of actual defect features and simulation data are constructed. A contrast loss function constrained by process parameters is designed, and cross-domain invariant features are extracted through gradient descent optimization. Based on the optimized feature distribution, the correlation weights between defect types and process parameters are calculated through causal reasoning, where the weight distribution is dynamically updated through backpropagation of deconvolution gradients.

2. The surface quality testing method for metals according to claim 1, characterized in that: Based on the nonlinear mapping relationship between the dielectric properties of metals and optical path difference, the specific process for analyzing the phase shift of the defect region in the interference fringes is as follows: Noise suppression and background correction are performed on interference fringe images acquired from multiple angles, and fringe distortion caused by environmental interference is eliminated by phase shift coding method; Based on the spatial variation characteristics of the dielectric constant of metals, a nonlinear mapping model between optical path difference and defect depth is constructed, and the model is transformed into a nonlinear integral expression with constrained boundary conditions. The integral expression is numerically solved using a multi-scale guided filtering and iterative optimization algorithm to extract the phase offset value.

3. The surface quality detection method for metals according to claim 2, characterized in that: The specific process for generating phase domain data containing the three-dimensional morphology of the defect is as follows: The phase offset is restored to a continuous phase distribution through a phase unwrapping algorithm. Combined with the light wavelength and incident angle parameters, the three-dimensional height field of the defect surface is calculated through a three-dimensional coordinate transformation model. The height field data is smoothed by surface interpolation to eliminate local distortion caused by noise. The depth gradient field, local principal curvature distribution and normal vector direction of the defect area are calculated by geometric feature analysis to generate phase domain data containing three-dimensional morphology, gradient features and geometric properties.

4. The surface quality detection method for metals according to claim 3, characterized in that: The specific process of extracting the defect depth gradient distribution by performing multi-scale frequency domain decomposition and dynamic segmentation constrained by process parameters on the phase domain data of the three-dimensional morphology of the defect is as follows: Multi-scale frequency domain decomposition is performed on the phase domain data, and substrate reflection noise and high-frequency defect signals are separated by adaptive frequency band selection; A dynamic segmentation threshold is generated based on the statistical distribution of historical parameters of metal processing, and the segmentation sensitivity is adjusted by combining the probability density function of the phase gradient magnitude. The depth gradient direction consistency of the segmented defect region is verified, and false defect signals that deviate from the material stress field direction by more than a preset angle are eliminated.

5. A method for surface quality testing of metals according to claim 4, characterized in that: The specific process of generating the defect topological feature distribution for quantifying process deviations through multi-scale topological decomposition and defect geometric feature statistics is as follows: The defect area is divided into spatial regions, and the porosity distribution, the consistency coefficient between the crack propagation direction and the principal stress direction, and the surface curvature abrupt change characteristics of each sub-region are statistically analyzed. By using statistical correlation analysis, a quantitative mapping relationship between defect geometric features and process parameter deviations is constructed, generating a feature distribution map that reflects the degree of process deviation.

6. A method for surface quality testing of metals according to claim 5, characterized in that: The specific process of dynamically adjusting process parameters based on the gradient direction of the causal correlation weight distribution, and then feeding back the detection data under the new parameters to the simulation graph library to update the manifold alignment rules of the depth constraint is as follows: The process parameter adjustment amount is generated based on the negative gradient direction of the causal correlation weight distribution, and the parameter adjustment is ensured to conform to the physical feasible region through a constraint optimization algorithm. The detection data under the new parameters is added to the simulation graph library, and the depth constraint weight matrix of the manifold space is optimized by the incremental parameter update method to achieve the coordinated iteration of the detection model and process parameters.

7. A surface quality inspection system for metals, applied to the surface quality inspection method for metals according to any one of claims 1-6, characterized in that, It includes the following modules: multi-angle coherent imaging module, process dynamic quantization module, simulation manifold alignment module, and process adjustment module; The multi-angle coherent imaging module is used to irradiate the metal surface with a multi-angle coherent light field sequence, collect interference fringes caused by the defect depth, and analyze the phase shift of the defect region in the interference fringes according to the nonlinear mapping relationship between the metal dielectric properties and the optical path difference, thereby generating phase domain data containing the three-dimensional morphology of the defect. The process dynamic quantization module is used to perform multi-scale frequency domain decomposition and dynamic segmentation of process parameter constraints on the phase domain data of the three-dimensional morphology of defects, extract the defect depth gradient distribution, and quantify the deposition temperature deviation and pressure fluctuation based on the deconvolution calculation of the metal processing history parameters and the defect depth gradient distribution. Through multi-scale topological decomposition and defect geometric feature statistics, the defect topological feature distribution of quantified process deviation is generated. The simulation manifold alignment module is used to construct a physical simulation map library of process parameters-defect morphology mapping relationship based on finite element simulation, align the defect topological feature distribution with the simulation data in the depth-constrained manifold space, extract cross-domain invariant features through comparative learning, and output the defect type and its causal correlation weight distribution with process parameters. The process adjustment module is used to dynamically adjust the process parameters according to the gradient direction of the causal correlation weight distribution, and to feed the detection data under the new parameters back to the simulation graph library to update the manifold alignment rules of the depth constraint.

Images