Image model-based spare part defect detection method and system

Abstract

The present application relates to the technical field of intelligent manufacturing, and more particularly to a kind of spare parts defect detection method and system based on image model.The method comprises the following steps: obtaining spare parts image;According to the edge defect detection of spare parts image, edge defect data is obtained;According to the edge defect data, image defect model is constructed;According to the edge defect data, crack growth simulation is carried out, and crack data is obtained;Based on crack data, metal grain coarsening detection is carried out, and metal grain coarsening data is obtained;According to the metal grain coarsening data, the thermal stability of spare parts is evaluated, and the thermal stability data of spare parts is obtained;According to the data of spare parts thermal stability, muscle position area warping detection is carried out, and muscle position warping data is obtained;Based on muscle position warping data, warping morphology classification is carried out, and muscle position upwarping type data and muscle position concave type data are obtained.The present application improves the accuracy of multi-type defect recognition of spare parts and the adaptability of defect detection under complex working conditions based on intelligent manufacturing technology.

Description

Technical Field

[0001] This invention relates to the field of intelligent manufacturing technology, and in particular to a method and system for detecting defects in spare parts based on image models. Background Technology

[0002] Traditional component defect detection often relies solely on surface feature recognition from two-dimensional images, lacking comprehensive analysis of the internal material microstructure and mechanical properties. This makes it difficult to accurately determine the evolution of key defects such as potential crack propagation, grain coarsening, and thermal instability. The use of single image processing techniques limits the ability to identify complex defect morphologies (such as warping, impurity accumulation, and corrosion), easily leading to missed or false detections and affecting the accuracy and reliability of the results. The lack of dynamic feedback and optimization mechanisms for defect detection results prevents real-time adjustments to the detection model based on material performance degradation and changes in the operating environment, hindering the multi-dimensional comprehensive evaluation of components under various working conditions. Traditional detection systems often use fixed parameter settings, lacking simulation and adaptation to actual working conditions such as material fatigue performance, load variations, and ambient temperature, limiting their promotion and application in complex industrial scenarios. The lack of a systematic model integration strategy when processing multi-type defect information fusion results in a fragmented detection process, failing to achieve closed-loop management from defect identification to performance evaluation, and thus failing to meet the high-precision and high-efficiency detection requirements of modern intelligent manufacturing. Summary of the Invention

[0003] Therefore, it is necessary for the present invention to provide a method and system for detecting defects in spare parts based on image models, in order to solve at least one of the above-mentioned technical problems.

[0004] To achieve the above objectives, a method for detecting defects in spare parts based on image models includes the following steps:

[0005] Step S1: Acquire images of spare parts; perform edge defect detection on the spare parts based on the images to obtain edge defect data; construct an image defect model based on the edge defect data;

[0006] Step S2: Simulate crack growth based on edge defect data to obtain crack data; perform metal grain coarsening detection based on crack data to obtain metal grain coarsening data; evaluate the thermal stability of the component based on the metal grain coarsening data to obtain component thermal stability data.

[0007] Step S3: Detect warpage in the rib area based on the thermal stability data of the parts to obtain rib warpage data; classify the warpage morphology based on the rib warpage data to obtain rib up-curved type data and rib concave type data; perform assembly interference risk analysis based on the rib up-curved type data to obtain assembly interference risk data.

[0008] Step S4: Detect impurity accumulation based on the concave data of the rib position to obtain impurity accumulation data; assess the degree of passivation film corrosion based on impurity accumulation to obtain passivation film corrosion data; transmit the assembly interference risk data and passivation film corrosion data to the image defect model to obtain the image defect optimization model; perform defect detection on the component image based on the image defect optimization model to obtain component defect data.

[0009] This invention extracts edge defect information from component images and uses it to construct an image defect model, enabling preliminary defect detection and providing foundational data support for subsequent multi-dimensional feature analysis. Building upon this, it introduces crack growth simulation, incorporating real-world operating conditions such as material fatigue performance, load amplitude, and cycle count, while integrating key parameters like stress intensity factor and crack propagation rate. This allows for dynamic deduction of defect propagation trends, overcoming the limitations of traditional static image analysis. Furthermore, through microscopic grain structure analysis of the crack path region, combined with texture complexity and grain morphology indicators, it enables precise localization of coarse-grained areas, providing a reliable basis for material thermal stability evaluation. Under the guidance of thermal stability data, it detects rib warpage, considering not only temperature-induced thermal strain but also quantitatively calculating geometric offsets, clearly linking the degree of warpage deformation to thermal load and improving the interpretability of deformation detection. Further classification of warpage morphologies, dividing complex deformation behavior into upward and downward warping types, facilitates the separate construction of assembly interference analysis paths and impurity deposition channel prediction logic, improving... The system enhances the comprehensiveness of defect type identification. Specifically, interference risk assessment of upward-curving data utilizes interference space vector modeling to achieve precise interference fit simulation, improving the ability to predict assembly reliability. Impurity accumulation analysis of concave areas constructs a reasoning chain from concave geometry to impurity concentration to corrosion degree, using image quantification indicators such as concentration thresholds and concave area distribution to identify localized passivation film failures. Finally, assembly interference and corrosion information are fed back to the original image defect model, enabling dynamic updates of the detection model parameters and adaptability to changes in material state and operating conditions, thus enhancing the system's generalization and scenario adaptability. During the fusion of multiple defect information, a unified data flow channel and image space alignment strategy connect the entire process from image acquisition to feature recognition, evolution simulation, performance evaluation, and model feedback, achieving organic coupling between detection functions and performance analysis. This improves the system's closed-loop identification and dynamic management capabilities for component defects, thereby meeting the high-precision, high-robustness, and high-adaptability detection requirements in intelligent manufacturing scenarios.

[0010] Preferably, this specification also provides an image model-based component defect detection system for performing the image model-based component defect detection method described above. The image model-based component defect detection system includes:

[0011] The image defect model construction module is used to acquire images of spare parts; perform edge defect detection on the spare parts based on the images to obtain edge defect data; and construct an image defect model based on the edge defect data.

[0012] The component thermal stability assessment module is used to simulate crack growth based on edge defect data to obtain crack data; to detect metal grain coarsening based on crack data to obtain metal grain coarsening data; and to assess the thermal stability of the component based on the metal grain coarsening data to obtain component thermal stability data.

[0013] The assembly interference risk analysis module is used to detect warpage in the rib area based on the thermal stability data of the parts, and obtain rib warpage data; classify the warpage morphology based on the rib warpage data, and obtain rib upward warpage data and rib downward warpage data; and perform assembly interference risk analysis based on the rib upward warpage data to obtain assembly interference risk data.

[0014] The defect detection module is used to detect impurity accumulation based on the concave data of the rib position, and obtain impurity accumulation data; to assess the degree of passivation film corrosion based on the impurity accumulation, and obtain passivation film corrosion data; to transmit the assembly interference risk data and passivation film corrosion data to the image defect model, and obtain the image defect optimization model; and to perform defect detection on the component image based on the image defect optimization model, and obtain component defect data.

[0015] The present invention relates to a component defect detection system based on an image model. This system can implement any of the component defect detection methods based on the image model of the present invention. It serves as a medium for coordinating the operation and signal transmission between various modules to complete the component defect detection method based on the image model. The internal modules of the system cooperate with each other to improve the accuracy of multi-type defect identification of components and the adaptability of defect detection under complex working conditions. Attached Figure Description

[0016] Other features, objects, and advantages of the invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings:

[0017] Figure 1 This is a schematic diagram of the steps of a component defect detection method based on an image model according to the present invention;

[0018] Figure 2 This is a detailed flowchart of step S1 in the present invention;

[0019] Figure 3 This is a detailed flowchart of step S12 in the present invention;

[0020] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0021] The technical method of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without inventive effort are within the scope of protection of this invention.

[0022] Furthermore, the accompanying drawings are merely illustrative of the invention and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and therefore repeated descriptions of them will be omitted. Some block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor methods and / or microcontroller methods.

[0023] It should be understood that although the terms "first," "second," etc., may be used herein to describe various units, these units should not be limited by these terms. These terms are used merely to distinguish one unit from another. For example, without departing from the scope of the exemplary embodiments, a first unit may be referred to as a second unit, and similarly, a second unit may be referred to as a first unit. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0024] To achieve the above objectives, please refer to Figures 1 to 3 This invention provides a method for detecting defects in spare parts based on an image model, the method comprising the following steps:

[0025] Step S1: Acquire images of spare parts; perform edge defect detection on the spare parts based on the images to obtain edge defect data; construct an image defect model based on the edge defect data;

[0026] In this embodiment, a high-resolution industrial camera is used to acquire surface images of the parts, with an image resolution of no less than 5000×4000 pixels to ensure the visualization of minute defects. A standard light source illumination system is used, with the light source color temperature controlled within the range of 5500K±200K to avoid interference from reflected light spots and shadows. The acquired images are preprocessed, including Gaussian filtering for noise reduction, with the filter kernel size set to 5×5 and the standard deviation σ set to 1.0. Subsequently, an edge detection operator such as the Canny operator is used, with a low threshold of 50 and a high threshold of 150, to extract the edge information of the parts. Based on the edge extraction results, defect area identification is performed, using a closing operation (morphological closing operation) to connect edge discontinuities, with a 3×3 rectangular kernel as the structuring element. According to the defect edge contour, the geometric parameters such as the area, length, and width of the defect area are calculated, with areas above 50 pixels considered as valid defect areas. Based on the above detection results, an image defect model is constructed. This model uses a two-dimensional binary matrix to represent the defect area, with defective pixels marked as 1 and non-defective pixels marked as 0. The defect model is stored in a standard image file format (such as PNG) for subsequent processing. The entire process is implemented programmatically, using Python in conjunction with the OpenCV library for image processing to ensure both speed and accuracy.

[0027] Step S2: Simulate crack growth based on edge defect data to obtain crack data; perform metal grain coarsening detection based on crack data to obtain metal grain coarsening data; evaluate the thermal stability of the component based on the metal grain coarsening data to obtain component thermal stability data.

[0028] In this embodiment, the crack growth simulation platform uses professional software based on the finite element method (FEM), such as ABAQUS or COMSOL Multiphysics. The fatigue performance parameters of the material in the model are set as follows: fatigue strength coefficient σ_f' is 1200 MPa, fatigue strength index b is -0.12, stress intensity factor range is set from 5 MPa·m^0.5 to 30 MPa·m^0.5, and crack propagation rate constant C is set from 1 × 10^-10 mm / cycle to 1 × 10^-6 mm / cycle. The load amplitude is set from 100 N to 1000 N, the number of cycles is set from 10,000 to 1,000,000, and the ambient temperature is controlled from -40℃ to 80℃ to simulate the actual working environment. In the crack growth simulation step, the time step is set from 0.1 seconds to 1 second to ensure the temporal resolution of crack propagation, and the calculation accuracy is maintained between 0.01 mm and 0.1 mm. The output data of crack growth includes information such as crack length, crack propagation rate, and crack path. Subsequently, based on crack data, image segmentation technology and boundary recognition algorithms were used to identify metal grain boundaries and calculate grain diameter and grain uniformity indices. Grain diameter was calculated using average values, ranging from 50 to 200 micrometers, while grain uniformity was calculated using variance. Finally, based on the grain diameter and uniformity data, the material hardness was calibrated using a hardness tester (such as a Vickers hardness tester). Combined with a metal microstructure and property handbook, the thermal stability of the components was evaluated, and relevant thermal stability data were output.

[0029] Step S3: Detect warpage in the rib area based on the thermal stability data of the parts to obtain rib warpage data; classify the warpage morphology based on the rib warpage data to obtain rib up-curved type data and rib concave type data; perform assembly interference risk analysis based on the rib up-curved type data to obtain assembly interference risk data.

[0030] In this embodiment, a 3D laser scanning device is used to collect the surface morphology of the rib area of ​​the component, with a sampling accuracy of 0.01mm. A 3D morphology data of the rib area is constructed using a Digital Surface Model (DSM), and point cloud data processing technology is employed to remove noise and outliers. Based on the 3D morphology data, a curvature analysis algorithm is used to calculate the geometric offset of the rib, with an offset threshold set at ±0.05mm. The warpage amplitude is represented by the maximum geometric offset distance, calculated using Euclidean distance measurement. Warpage deformation direction analysis distinguishes between upward-curving and downward-curving ribs, with thresholds defined as ±0.05mm for each type. After classifying the warpage morphology, for the upward-curving data, finite element analysis software is used to simulate the assembly process and assess interference risk. The load condition is set to a standard assembly load of 500N, with a maximum allowable interference gap of 0.1mm. Interference risk data includes interference area, interference pressure, and interference position coordinates, stored using a two-dimensional mapping.

[0031] Step S4: Detect impurity accumulation based on the concave data of the rib position to obtain impurity accumulation data; assess the degree of passivation film corrosion based on impurity accumulation to obtain passivation film corrosion data; transmit the assembly interference risk data and passivation film corrosion data to the image defect model to obtain the image defect optimization model; perform defect detection on the component image based on the image defect optimization model to obtain component defect data.

[0032] In this embodiment, based on the concave data of the rib position obtained in step S3, the concavity depth is calculated. A three-dimensional measurement point cloud is compared with a reference plane, and the depth measurement accuracy is 0.01 mm. A geometric model of the rib concavity is constructed using the concavity depth data, and three-dimensional modeling is performed using CAD software (such as SolidWorks). Based on this model, surface hydrodynamic simulation is applied to identify surface retention areas. The fluid viscosity is set to 1.0 mPa·s, and the flow velocity is set to 0.1 m / s. Within the retention areas, image grayscale gradient analysis and local reflectivity change modeling are used to extract heterogeneous pixel clusters. The thresholds are set to grayscale changes greater than 20 and reflectivity changes greater than 15%. Based on the impurity accumulation areas, X-ray photoelectron spectroscopy (XPS) analysis is used to evaluate the passivation film corrosion degree. The analysis data includes corrosion depth and corrosion product concentration. Finally, assembly interference risk data and passivation film corrosion data are input into an image defect optimization algorithm. The optimization parameters include weight factors of 0.6 and 0.4. An iterative defect identification algorithm is used, with 100 iterations and a convergence error threshold of 1%. Based on the optimization results, defect detection is performed on the parts images, and the parts defect data is output. The data format includes the defect location coordinates, defect area and defect type.

[0033] Preferably, step S1 includes the following steps:

[0034] Step S11: Obtain the image of the spare parts and perform image preprocessing to obtain the image of the spare parts to be processed;

[0035] In this embodiment, an industrial-grade high-resolution camera is used to acquire images of the components. The image resolution is set to 6000×4000 pixels, and the color depth is 24 bits to ensure complete capture of the surface details of the components. Image acquisition employs a uniform ring LED lighting system with a fixed color temperature of 5500K and illuminance controlled at 1000 lux to avoid reflected light spots and shadows. After acquisition, the images undergo preprocessing, specifically including noise reduction, grayscale normalization, and contrast enhancement. Noise reduction uses Gaussian filtering with a 5×5 kernel size and a standard deviation σ of 1.2 to remove high-frequency noise. Grayscale normalization maps pixel values ​​to the range of 0 to 255 to ensure uniform grayscale. Contrast enhancement utilizes an adaptive histogram equalization algorithm, limiting the contrast factor to 2.0 to prevent over-enhancement and information loss. The preprocessed images are saved in a lossless format (TIFF) for subsequent analysis.

[0036] Step S12: Analyze gear tooth breakage based on the part images to obtain gear tooth breakage data;

[0037] In this embodiment, the gear contour is extracted using an edge detection algorithm with the Canny operator, a low threshold of 80, and a high threshold of 200 to accurately capture the edges of the gear teeth. Curvature analysis is then performed on the contour to calculate the radius of curvature of each tooth root and tip. Regions with a radius of curvature less than 0.5 mm are identified as potential tooth breakage defects. Morphological dilation is applied to the broken tooth region, with a 3×3 circle as the structuring element to eliminate edge fracture noise. Connected component analysis is used to extract the geometric features of the broken tooth notch, including notch width, depth, and area, with a width threshold of no less than 0.2 mm and a depth threshold of no less than 0.1 mm. The extracted broken tooth features are quantified to form a gear broken tooth dataset, stored in a JSON format, containing fields for broken tooth coordinates, dimensions, and defect type.

[0038] Step S13: Analyze the edge and corner damage of aluminum parts based on the part images to obtain edge and corner damage data;

[0039] In this embodiment, image segmentation technology is used to extract the corner regions of the aluminum parts. First, a region growing algorithm is used to locate the corner regions, with the starting point based on corner extraction from edge detection results. The Harris algorithm is used for corner detection, with a response threshold set to 0.01. Texture analysis is applied to the corner regions, using the Gray-Level Co-occurrence Matrix (GLCM) to calculate texture features and extract three indicators: contrast, energy, and uniformity. Thresholds are set for contrast > 0.3, energy < 0.4, and uniformity < 0.5, respectively. Damage identification is performed based on texture anomaly regions, with the damaged area not less than 30 pixels. Morphological opening operations are used to remove small-area noise, with a 3×3 rectangular kernel as the structural element. The final extracted geometric parameters of the damaged region include the length, width, and area of ​​the damage. The data format is unified as an XML structure for easy subsequent data integration.

[0040] Step S14: Integrate the gear tooth breakage data and the aluminum part edge damage data to obtain edge defect data;

[0041] In this embodiment, the gear tooth breakage data from step S12 and the aluminum part corner damage data from step S13 are aligned in spatial coordinates. Based on the coordinate system of the component design drawings, the two sets of data are mapped to a unified three-dimensional coordinate system, with the error range controlled within ±0.05mm. A weighted fusion algorithm is used to merge the defect information in the same area. The weight parameters are determined according to the severity of the defect and the detection accuracy; the weight for gear tooth breakage is 0.6, and the weight for aluminum part damage is 0.4. After fusion, the area, volume, and morphological parameters of the integrated defect area are calculated, with a defect area threshold of not less than 0.5mm. 2 The volume threshold is not less than 0.1 mm. 3 Generate a complete edge defect data file in standard CSV format, containing fields for defect type, location coordinates, size, and fusion weight.

[0042] Step S15: Construct an image defect model based on edge defect data.

[0043] In this embodiment, the model adopts a two-dimensional matrix mapping structure. The matrix size corresponds to the pixel size of the component image, the defective area corresponds to the element with a value of 1 in the matrix, and the non-defective area corresponds to 0. An interpolation algorithm (bilinear interpolation) is used to smooth the defective edges, with a smoothing radius set to 2 pixels to ensure the continuity of the model edges. During the model construction process, a defect severity index is defined based on the geometric feature parameters in the edge defect data. The index calculation formula is: Severity = Defective Area × Defective Depth / Maximum Defective Size. This index is used to classify and label the defective areas, with severity thresholds divided into mild (<0.3), moderate (0.3-0.7), and severe (>0.7). The final generated image defect model is stored in the standard HDF5 file format for easy subsequent analysis and model iteration updates.

[0044] Preferably, step S12 includes the following steps:

[0045] Step S121: Segment the gear region based on the part image to obtain gear region data;

[0046] In this embodiment, a high-resolution color image of the component is acquired using an industrial high-definition digital camera. The image resolution is set to 6000×4000 pixels to ensure that gear details are clearly discernible. Subsequently, image grayscale conversion technology is used to convert the color image into a single-channel grayscale image with a grayscale value range of 0 to 255. Based on the grayscale image, the optimal segmentation threshold is automatically calculated using the Otsu thresholding method. The threshold range is generally between 100 and 150 to ensure clear separation between the gear and the background. The grayscale image is then binarized, with the foreground corresponding to the gear region and the background set to zero. The binarized image is then filled with small holes within the gear contour using morphological closing operations (structural element is a 5×5 square), while opening operations are applied to remove minor noise. Next, connected component analysis is used to extract the largest connected region as the gear region. Contour tracking is performed on the boundary of this region to obtain the coordinate data of the gear contour, which is stored as a structured data file (such as JSON format), referred to as the gear region data.

[0047] Step S122: Detect poor lubrication based on gear area data to obtain poor lubrication data;

[0048] In this embodiment, local grayscale statistics are performed on the gear region using a sliding window of 50×50 pixels. The average grayscale and standard deviation within each window are calculated. Regions with good lubrication exhibit uniform grayscale with a standard deviation less than 10, while regions with poor lubrication show significant grayscale fluctuations with a standard deviation greater than 25. Furthermore, texture analysis is used to calculate the contrast characteristics in the gray-level co-occurrence matrix (GLCM) of the local region. Regions with a contrast ratio higher than 0.4 are identified as poorly lubricated areas. By aggregating these regions, poor lubrication data is generated, including location coordinates, area size, and contrast index, and stored in CSV format.

[0049] Step S123: Predict gear surface wear based on poor lubrication data to obtain gear surface wear data;

[0050] In this embodiment, based on the data on poor lubrication, detailed features are extracted from the surface image of the corresponding area. A differential operator (Sobel operator, with 3×3 convolution kernels in both the horizontal and vertical directions) is used to calculate the gradient magnitude. Gradient magnitudes greater than 50 (grayscale) are identified as wear edges. Combining the positional relationship between the wear edges and the poorly lubricated area, the area and depth of the wear region are calculated. The wear depth estimation is based on the high light intensity attenuation rate under multi-angle illumination, using camera exposure time and light intensity calibration, with a depth range from 0.01mm to 0.5mm. The predicted wear data is stored in the form of a numerical matrix, including the coordinates, wear depth, and gradient information of each wear point.

[0051] Step S124: Perform grain structure analysis based on gear surface wear data to obtain grain structure data;

[0052] In this embodiment, the surface region located based on wear data is used to acquire a high-magnification image of that region using a scanning electron microscope (SEM). The magnification is set to 5000x, and the image resolution reaches 2048×2048 pixels. Grain boundaries are extracted through image processing, and an edge detection operator (Laplacian operator, kernel size 3×3) is applied to enhance grain boundary contrast. Based on the grain boundary distribution, grain size and shape parameters are calculated. The grain diameter ranges from 0.5μm to 50μm, and the shape indicators include roundness and aspect ratio. All grain parameters are stored in structured data form, which is the grain structure data.

[0053] Step S125: Perform grain boundary slip analysis based on grain structure data to obtain grain boundary slip data;

[0054] In this embodiment, based on grain structure data, the formation and expansion of grain boundary slip bands are analyzed in detail. Electron backscatter diffraction (EBSD) is used to obtain the dislocation density and slip band distribution at grain boundaries, with a scanning area size of 100 μm × 100 μm and a step size of 0.1 μm. Image reconstruction is performed on the EBSD data, and high dislocation density regions are extracted as slip band locations. A dislocation density threshold of 1 × 10^14 m^-2 is set, and regions exceeding this threshold are marked as slip bands. The length, width, and distribution density of slip bands are statistically analyzed to generate grain boundary slip data, which is stored in a dedicated data format for subsequent tooth fracture analysis.

[0055] Step S126: Perform gear tooth breakage analysis based on grain boundary slip data to obtain gear tooth breakage data.

[0056] In this embodiment, based on grain boundary slip data and gear profile deformation data, a laser scanning confocal microscope (LSCM) was used to scan the micro-cracks and broken tooth areas on the gear surface. The scanning range covered the suspected broken tooth area, achieving a resolution of 1 μm. Three-dimensional reconstruction technology was used to obtain the spatial morphology of the broken tooth notch, and the depth, width, and fracture morphology parameters were measured. The notch depth threshold was set to be no less than 0.1 mm, and the width threshold was set to be no less than 0.2 mm. Combined with the distribution of grain boundary slip bands, the tooth formation mechanism was analyzed, ultimately generating gear tooth fracture data including the location, size, and fracture characteristics of the broken tooth. The data was saved in XML format for easy subsequent defect tracking and management.

[0057] Preferably, step S13 includes the following steps:

[0058] Step S131: Segment the corner area of ​​the aluminum part according to the part image to obtain the corner area data of the aluminum part;

[0059] In this embodiment, an industrial-grade high-resolution digital camera is used to acquire multi-angle images of the parts. The image resolution is set to 5000×3500 pixels to ensure clear corner details. The acquired images undergo color space conversion, converting RGB images to Lab color space to enhance color contrast in edge areas. Using the Canny edge detection algorithm, with a low threshold of 50 and a high threshold of 150, edge information is extracted from the images. Morphological erosion and dilation operations (with a 3×3 rectangle as the structuring element) are combined to clean up edge noise and connect edge breaks, obtaining a complete corner contour. A contour extraction algorithm is used to obtain the coordinate data of the corner region contour. Combined with a pre-defined geometric range of the corner region (the four edges with a side length not exceeding 30mm), the region is restricted to complete the segmentation of the aluminum part's corner region, generating aluminum part corner region data. This data includes the coordinate point set and geometric parameters of the corner region.

[0060] Step S132: Perform plastic deformation detection based on the data of the aluminum part's corner area to obtain plastic deformation data;

[0061] In this embodiment, digital image correlation (DIC) technology is used to detect plastic deformation in the corner areas of the aluminum part. Specifically, a randomly distributed micron-sized high-contrast speckle pattern is first sprayed onto the surface of the aluminum part. The speckles are approximately 30 μm in size and 10 μm apart. High-resolution images are acquired before and after loading, maintaining a resolution of 6000×4000 pixels. The speckle displacement vector field is calculated using an image registration algorithm at a resolution of 5 μm, yielding deformation field data. Local strain values ​​are calculated using the deformation gradient, with a strain threshold of 0.02 used as the criterion for judging plastic deformation. The degree of plastic deformation at each pixel is calculated, ultimately forming plastic deformation data. The data is stored in the form of a strain value matrix, indicating the deformation area and its strain magnitude.

[0062] Step S133: Evaluate the hardness of the aluminum part's edge and corner material based on the plastic deformation data to obtain the hardness data of the aluminum part's edge and corner material;

[0063] In this embodiment, a mapping relationship between plastic deformation and material hardness is established based on the hardness variation law corresponding to plastic deformation. A nanoindentation testing device is used to measure the hardness of multiple representative corner areas, with the measured hardness range between HV100 and HV250 and a sampling point spacing of 1 mm. Based on the fitted curve of the nanoindentation data, the material hardness corresponding to the plastic deformation data is calculated using the least squares fitting formula: H = H0 - k × ε_p, where H is the hardness value, H0 is the initial hardness, k is the hardening coefficient, and ε_p is the plastic strain. Specifically, H0 is set to 210 HV, and k is set to 120 HV. The mapping result is applied to the plastic deformation data, converting it into hardness data of the aluminum part's corner material, in the form of a hardness value distribution matrix.

[0064] Step S134: Identify low-hardness aluminum part corner areas based on the hardness data of the aluminum part corner material;

[0065] In this embodiment, low-hardness regions are identified based on the hardness data of the aluminum component's edge and corner materials. A hardness threshold of 180 HV is set; regions with hardness below this threshold are defined as low-hardness aluminum component edge and corner regions. The hardness data is binarized, with a value of 1 assigned to a hardness value less than 180 HV and 0 to a value greater than 180 HV. Connected component analysis is used to extract continuous low-hardness regions, and the area, boundary length, and shape characteristics of each region are calculated. The area threshold for low-hardness regions is set to be no less than 5 mm. 2 Low-hardness areas that meet the criteria are selected. The spatial coordinate information of these areas is extracted to form low-hardness aluminum part corner area data, which is represented by area labels and corresponding geometric parameters.

[0066] Step S135: Perform aluminum alloy surface peeling analysis based on the corner area of ​​the low-hardness aluminum part to obtain aluminum alloy surface peeling data; evaluate the degree of oxide film peeling based on the aluminum alloy surface peeling data to obtain oxide film peeling data.

[0067] In this embodiment, a high-magnification optical microscope was used to scan the surface morphology of the low-hardness region at a magnification of 200x and a resolution of 0.5μm. Surface texture features were extracted using a gray-level co-occurrence matrix (GLCM), with a focus on calculating contrast and entropy values. The peeled-off regions exhibited high contrast and high entropy values. Thresholds were set at a contrast of 0.6 and an entropy of 0.8; regions exceeding these thresholds were identified as peeled-off regions. The area and depth of the peeled-off regions were measured using a three-dimensional laser scanning microscope, with a depth range of 0.01mm to 0.2mm. The boundary and morphological feature data of the peeled-off regions were compiled into aluminum alloy surface peeling data. Subsequently, the degree of oxide film peeling was calculated based on the peeling data. X-ray photoelectron spectroscopy (XPS) was used to determine the oxide film thickness in the peeled-off regions, with a thickness threshold range of 10nm to 50nm; a thickness reduction exceeding 20% ​​was considered peeling. Based on the differences in peeling thickness in different regions, oxide film peeling data was generated and stored as a thickness distribution matrix.

[0068] Step S136: Analyze the edge and corner damage of aluminum parts based on the oxide film peeling data to obtain the edge and corner damage data of aluminum parts.

[0069] In this embodiment, the criteria for determining damaged areas are defined by combining the geometric parameters and hardness distribution information of the peeled-off region. A damaged area must meet the following conditions: the oxide film peeling thickness exceeds 30 nm and the material hardness is below 170 HV, accompanied by cracks or peeling marks with a depth of not less than 0.05 mm on the surface morphology. Using multispectral image fusion technology, optical images, XPS thickness maps, and hardness distribution maps are combined, and image registration methods are used to complete the spatial alignment of multiple data layers. By comprehensively analyzing multiple data layers, threshold filtering and edge detection are used to extract the accurate boundaries of the damaged areas, and their area, depth, and location coordinates are calculated. Finally, the aluminum part corner damage data is output in a multidimensional array format, containing the spatial coordinates, geometric parameters, and corresponding physical properties of the damaged areas.

[0070] Preferably, step S2, which involves simulating crack growth based on edge defect data, includes:

[0071] Import edge defect data into the crack growth simulation platform;

[0072] In this embodiment, the aforementioned edge defect data is exported in a digital format, which is a two-dimensional or three-dimensional coordinate point set containing information on the crack initiation point, length, and width. The edge defect data is imported into the crack growth simulation platform using a standard CAD data exchange format (such as STEP or IGES) or a custom structured data format. During the import process, the data parsing module performs format verification on the input data to ensure the geometric information is complete and error-free, especially checking that the spatial positioning accuracy of the crack initiation point is within ±0.01mm and the accuracy of the crack size parameters is not less than 0.05mm. The imported data undergoes spatial coordinate system transformation to unify it into the global coordinate system of the simulation platform, ensuring spatial consistency of the data during the simulation. After import, the preprocessing module performs mesh generation on the edge defect data, with the mesh cell size controlled between 0.01mm and 0.05mm to ensure the accuracy requirements of subsequent crack propagation calculations.

[0073] Material fatigue performance parameters were set in the crack growth simulation platform, with the stress intensity factor ranging from 5 MPa·m. 0.5 -30MPa·m 0.5 The crack propagation rate constant is 1×10 -10 Up to 1×10 -6 mm / cycle;

[0074] In this embodiment, fatigue performance parameters are input to the crack growth simulation platform based on the material properties of the component being tested. Specific parameters include the stress intensity factor range and the crack propagation rate constant. The stress intensity factor K_IC is set to 5 MPa·m. 0.5 -30MPa·m 0.5 The crack propagation rate constant C is determined based on mechanical testing data of metallic materials, and the values ​​are derived from a fatigue test database. C is taken as 1 × 10⁻⁶. -10 Up to 1×10 -6 mm / cycle, selected based on experimental results of the Paris Law for the material. These two parameters are input into the simulation software's parameter configuration module in numerical range format, ensuring the values ​​are within the permissible range of the material's physical properties. After parameter input, a parameter consistency check is performed to verify the logical relationship between parameter values ​​and the correctness of units, preventing simulation failure due to abnormal input.

[0075] In the crack growth simulation platform, the load amplitude range is set to 100N-1000N, the number of cycles ranges from 10,000 to 1,000,000, and the ambient temperature ranges from -40℃ to 80℃.

[0076] In this embodiment, the load amplitude range is set to 100N to 1000N, based on the maximum and minimum load values ​​measured in the actual working environment of the component. Data acquisition is completed by a force sensor with an accuracy of ±1N. The cycle count range is set to 10,000 to 1,000,000 times, reflecting different stages of the component's service life. The ambient temperature range is set to -40℃ to 80℃, covering the temperature variation range under common operating conditions. These parameters are set through the environmental condition input module of the simulation platform to ensure accurate parameter mapping. The load waveform is defined as a sinusoidal load with a frequency range of 1Hz to 100Hz, and the time dependence of the load is handled using numerical integration. After the parameters are set, the load curve is verified to ensure that the input waveform is continuous and meets the numerical stability requirements.

[0077] In the crack growth simulation platform, the simulation time step is set to 0.1s-1s, and the crack propagation calculation accuracy is 0.01mm-0.1mm;

[0078] In this embodiment, the simulation time step is set to 0.1 to 1 second, adjusted according to the crack growth timescale to ensure temporal resolution during the calculation. The crack propagation calculation accuracy is set to 0.01 mm to 0.1 mm, controlling the spatial step size of crack propagation to ensure the precision of the simulation results. The time step and spatial step size parameters are input to the numerical calculation module of the simulation platform, and incremental calculations of the crack propagation path are performed in conjunction with finite element analysis technology. Pre-calculation tests are conducted to ensure that the step size parameters strike a balance between ensuring computational stability and efficiency, preventing simulation jumps caused by excessively large time steps or wasted computational resources due to excessively small time steps.

[0079] Run the crack growth simulation module and output crack data.

[0080] In this embodiment, after initiating the simulation, the platform iteratively calculates the stress field at the crack tip using the finite element method based on the input initial data of the edge defect, material parameters, load, and environmental conditions. This is combined with the Paris-Erdogan crack propagation formula to achieve numerical simulation of crack growth. The crack growth path and rate are updated with the time step, and the data is stored in the form of time-series crack length and crack morphology evolution. The output crack data includes crack length, width, propagation rate, crack tip location coordinates, and the number of cycles corresponding to crack growth. The data is exported in CSV or HDF5 format, stored in a structured manner for easy subsequent analysis. The output file includes simulation parameter records to ensure the traceability of crack growth data and to meet verification requirements.

[0081] Preferably, step S2, which involves detecting coarse metal grains based on crack data, includes:

[0082] Grain boundaries are identified based on crack data to obtain grain boundary data.

[0083] In this embodiment, crack data is imported into metal microstructure images acquired by high-resolution scanning electron microscopy (SEM) or focused ion beam microscopy (FIB-SEM). The image resolution is required to be 0.1 micrometers or higher to ensure clear boundaries of fine grains. Image segmentation technology is used, and edge detection algorithms (such as the Sobel operator and Canny operator) are employed to extract the gray-level gradient changes around the crack region. The edge detection threshold is set to 30%-40% of the gray-level value, and the threshold is dynamically adjusted based on gray-level histogram analysis to ensure complete and continuous boundary lines. The extracted boundary lines are closed using morphological closing operations, with a 3×3 pixel square convolution kernel selected as the structuring element. The resulting continuous boundary lines are the grain boundary data, containing two-dimensional coordinate information of the boundary positions, and stored in a structured point set data format.

[0084] The uniformity of the microstructure is detected based on the grain boundary data, and the uniformity of the microstructure is obtained.

[0085] In this embodiment, the grain boundary data is converted into a polygonal mesh structure, with each polygon representing a grain. The area and perimeter of each grain are calculated, and the uniformity of the microstructure is evaluated using the coefficient of variation (CV_A) and the shape factor (roundness). Specifically, the coefficient of variation (CV_A) is defined as the standard deviation (grain area) / average grain area, and roundness is defined as 4π × area / perimeter². The threshold values ​​for the coefficient of variation (CV_A) and roundness are set to 0.3 and 0.7, respectively. Grain groups with uniform area distribution and high roundness are identified as uniform microstructure regions, while those with uneven distribution are identified as non-uniform regions. The results are represented as a two-dimensional heatmap, where the coordinates correspond to the spatial location of the grains, and the heatmap values ​​correspond to the uniformity index. The output is the grain microstructure uniformity data, including the uniformity score of each grain and its spatial distribution.

[0086] Identify oxide inclusion-rich regions based on grain structure uniformity data;

[0087] In this embodiment, non-uniform regions in tissue homogeneity data are used as potential oxide inclusion regions, which are further confirmed through image grayscale analysis. High grayscale pixels within the non-uniform regions are extracted, with a threshold set to be above the 80th percentile of the overall image grayscale histogram. A region growing algorithm is used to perform connected component analysis on the high grayscale pixels, and the area of ​​the connected components must be greater than 5 μm. 2 Only then can it be identified as an oxide inclusion. The location and size of the detected oxide inclusions are statistically analyzed to generate oxide inclusion enrichment region data, which includes the two-dimensional spatial coordinates, area, and mean gray value of each inclusion.

[0088] Amorphous phase barrier points were determined based on oxide inclusion enrichment regions, and amorphous phase barrier point data were obtained.

[0089] In this embodiment, the spatial distribution of oxide inclusions is used as candidate locations for amorphous phase barrier points. Combined with local grain orientation data, the grain orientation angle field is obtained through electron backscatter diffraction (EBSD) measurements with a resolution of at least 0.2 μm. The grain orientation gradient near the oxide inclusions is calculated, with a gradient threshold set at 10 degrees / μm. Locations exceeding this threshold are identified as amorphous phase barrier points. The spatial coordinates of the oxide inclusions and the grain orientation gradient results are integrated to calibrate the final coordinates of the amorphous phase barrier points. The amorphous phase barrier point data, including location coordinates and the magnitude of the orientation gradient, is output.

[0090] Metal grain coarseness detection was performed based on amorphous phase barrier point data to obtain metal grain coarseness data.

[0091] In this embodiment, a coarse grain region is defined based on amorphous phase barrier point data, and the grain area distribution within a 10 μm radius around the amorphous phase barrier point is statistically analyzed. If the average grain area within the region exceeds 150 μm... 2 Furthermore, the density of amorphous phase barrier points is greater than 5 points / 100μm. 2 The region was identified as having coarse grains. The degree of grain coarsness was recorded by calculating the average grain diameter within the region. The coarse grain data included the spatial extent of the coarse region, the average grain area, the grain diameter, and the density of amorphous phase barrier points. The data was output in vector format for easy use in subsequent thermal stability analysis.

[0092] Preferably, step S2, which assesses the thermal stability of components based on the data of metal grain coarseness, includes:

[0093] The grain boundary area is calculated based on the coarseness data of the metal grains to obtain the grain boundary area data;

[0094] In this embodiment, coarse data of metal grains is imported into grain images acquired by a high-resolution scanning electron microscope (SEM) or electron backscatter diffraction (EBSD) device, with the image resolution controlled within 0.1 micrometers. Grain boundaries are binarized using image processing software, with a threshold set to 50% of the average grayscale value of the image. Grain boundary edges are extracted using the Canny edge detection algorithm, with the boundary line width limited to 1-2 pixels. A morphological thinning algorithm is then used to adjust the boundary lines to a single-pixel width, ensuring clear grain boundary contours. Based on the extracted boundary contours, the perimeter of each grain is calculated, and the actual grain boundary length is obtained by converting the pixel size. The grain boundary area is calculated by integrating the perimeters of all grains by the grain boundary thickness, with the grain boundary thickness set to 0.1 micrometers according to material physical thickness standards. Finally, the grain boundary area data is output in structured data form, including the total grain boundary area, local grain boundary area distribution, and related spatial coordinates.

[0095] The material hardness decay analysis was performed based on the grain boundary area data to obtain the material hardness decay data;

[0096] In this embodiment, a nanoindenter is used to perform hardness testing on sample areas with relatively large grain boundary areas. The maximum load is set to 100 mN, the loading time is set to 10 seconds, the loading speed is set to 10 mN / s, and the hardness test interval is controlled at 5 micrometers. Based on the correspondence between grain boundary area and nanoindentation test position, the hardness data and grain boundary area data are spatially mapped. The empirical formula H = H_0 - k × A_gb is used, where H is the tested hardness value, H_0 is the matrix hardness unaffected by grain boundaries (e.g., 250 HV), and k is the hardness attenuation coefficient, which is taken as 0.005 HV / μm. 2 A_gb represents the grain boundary area of ​​the corresponding region. The hardness attenuation coefficient was obtained through experimental fitting of multiple sets of samples. Hardness attenuation values ​​were calculated for the entire sample region, generating a hardness attenuation dataset containing spatial location, hardness attenuation value, and corresponding grain boundary area. The data format is a three-dimensional matrix for easy subsequent thermal stability analysis.

[0097] Material stress data is obtained by using large amounts of coarse metal grain data;

[0098] In this embodiment, the geometric information of the coarse-grained region is input into a digital image correlation (DIC) system to acquire surface deformation images of the component under loading conditions, with an image resolution of at least 1 μm. The surface strain field is calculated using an image difference algorithm, and a sub-pixel interpolation algorithm is used to improve the strain calculation accuracy to 0.001%. The stress value is calculated using the material's elastic modulus (valued at 210 GPa) and the calculated strain data, employing Hooke's law σ = E × ε, where σ is stress, E is the elastic modulus, and ε is strain. The material stress data is output in the form of a two-dimensional stress distribution map, including principal stress, shear stress, and stress direction information, with a resolution of at least 10 μm. The data storage format is a stress tensor matrix, facilitating the identification of stress concentration regions.

[0099] Identify stress concentration areas in materials based on material stress data;

[0100] In this embodiment, a threshold segmentation algorithm is applied to the output stress distribution map. The threshold is set to 1.5 times the average stress value; regions exceeding this value are considered stress concentration areas. Connected component analysis is used to extract all continuous high-stress regions, with each connected component having an area of ​​at least 100 μm. 2 The stress concentration regions are retained as valid stress concentration areas. The boundaries of these regions are contoured and simplified using a polygon approximation algorithm, with vertex errors controlled within ±2μm. The stress concentration region data includes the region's location, area, maximum stress value, and corresponding stress direction, and is saved in a structured vector data format.

[0101] Based on the analysis of the reduction of grain boundary strengthening effect in the stress concentration region of the material, data on the reduction of grain boundary strengthening effect are obtained.

[0102] In this embodiment, the attenuation of the strengthening effect at grain boundaries due to local stress concentration is analyzed by combining data on the material's microstructure and stress concentration regions. According to the grain boundary strengthening attenuation model defined in the literature, the grain boundary strengthening attenuation coefficient K_gb varies linearly with the stress concentration coefficient K_t, with the relationship K_gb=K_gb0-α×(K_t-1), where K_gb0 is the initial grain boundary strengthening coefficient, taken as 0.3, α is the attenuation coefficient, taken as 0.15, and K_t is the stress concentration factor, calculated by dividing the maximum stress value of the region by the overall average stress value. Using the calculated K_t, the reduction in grain boundary strengthening effect is calculated region by region. The results are presented in tabular form, including region coordinates, stress concentration factors, and the reduction in grain boundary strengthening effect, with data accuracy controlled within 0.01.

[0103] The thermal stability of the components is evaluated based on the data on material hardness decay and the data on the reduction of grain boundary strengthening effect, and the thermal stability data of the components is obtained.

[0104] In this embodiment, based on the hardness attenuation value H_d and the grain boundary strengthening attenuation value K_gb, and combined with the material thermal stability index formula T_s=β_1×(1-H_d / H_0)+β_2×(1-K_gb / K_gb0), where β_1 and β_2 are weighting coefficients, respectively, taking values ​​of 0.6 and 0.4, and H_0 and K_gb0 are the reference hardness and reference strengthening coefficient, respectively. The H_d and K_gb values ​​corresponding to each region are substituted into the formula to calculate the thermal stability index T_s, with a value range of 0 to 1; a lower value indicates poorer thermal stability. The calculation results are output in the form of a two-dimensional heat map, with a resolution of no less than 20μm, indicating the thermal stability level of each region. The thermal stability data is stored in matrix form, including location coordinates and corresponding T_s values, for subsequent overall performance evaluation of components.

[0105] Preferably, step S3 includes the following steps:

[0106] Step S31: Identify heat-sensitive areas based on the thermal stability data of the components;

[0107] In this embodiment, the thermal stability data of the components is imported into a two-dimensional spatial coordinate system, with the resolution of the data points controlled to be no less than 20 μm. A threshold segmentation method is used to define regions with thermal stability indices below 0.5 as heat-sensitive areas. The threshold of 0.5 is determined based on historical material performance experimental data, corresponding to the critical value at which the thermal performance of the material significantly decreases under high-temperature conditions. The data is scanned point-by-point using a two-dimensional thermal stability matrix to determine whether the thermal stability index of each data point is below the threshold. A connected component analysis algorithm is used to spatially cluster all data points below the threshold, retaining areas greater than 0.01 mm². 2Connected regions are used as effective heat-sensitive regions. Polygon fitting is performed on the boundaries of the identified heat-sensitive regions, with the fitting error controlled within ±5μm. The spatial boundary data and corresponding coordinate information of the heat-sensitive regions are output.

[0108] Step S32: Calculate the geometric offset of the ribs based on the heat-sensitive area to obtain the geometric offset data of the ribs;

[0109] In this embodiment, the three-dimensional geometric coordinates of the components during the design phase are used as a reference. A high-precision three-dimensional laser scanning device is employed to collect the actual morphological data of the components, with a point cloud density of no less than 1000 points per square millimeter. A difference comparison algorithm is used to compare the design coordinates with the actual coordinates, focusing on calculating the geometric offset of the corresponding rib area in the heat-sensitive region. The offset is represented by the displacement vector of the spatial coordinate point, with the vector component accuracy controlled within 0.01 mm. During the calculation process, outliers are removed through spatial neighborhood filtering, with the filtering radius set to 0.05 mm to ensure data smoothness. The output is a set of three-dimensional offset vectors for each sampling point in the rib area, constituting the rib geometric offset data.

[0110] Step S33: Calculate the warping amplitude based on the rib geometric offset data to obtain the rib warping data;

[0111] In this embodiment, the geometric offset vector of the rib is decomposed into a normal vector component and a tangent vector component. The normal vector corresponds to the normal offset of the rib surface and is used to calculate the warpage amplitude. The warpage amplitude is defined as the maximum absolute value of the normal offset, and the calculation range is limited to the thermally sensitive area of ​​the rib. The normal offset data is smoothed using a surface fitting algorithm, with the fitting error controlled within 0.02 mm to eliminate the influence of noise. The peak and valley values ​​of the maximum normal offset are calculated for the fitted surface. The peak value represents the upward warpage amplitude of the rib, and the valley value represents the downward warpage amplitude of the rib. The spatial distribution of the peak and valley values ​​and their corresponding values ​​are summarized to form the rib warpage data. The data format includes three-dimensional spatial coordinates and the corresponding warpage amplitude.

[0112] Step S34: Identify the deformation direction based on the rib warping data to obtain deformation direction data; classify the rib warping data into warping morphology based on the deformation direction data to obtain rib warping data with upward warping and rib warping data with downward warping.

[0113] In this embodiment, the sign of the normal offset in the rib warping data is used to determine the deformation direction. A positive value is defined as upward deformation (upward warping), and a negative value is defined as downward deformation (downward concavity). The sign of the normal offset is calculated for all sampling points within the rib region, and combined with neighborhood analysis, the continuous deformation direction region is determined. The neighborhood is defined as the set of points within a circular area with a radius of 0.1 mm centered on the sampling point. Boundary extraction is performed on the continuous deformation direction region, with the error controlled within ±0.01 mm. Based on the spatial distribution of the deformation direction region, the rib warping data is segmented into rib upward warping data and rib downward concavity data, with the data format being a vector set containing spatial coordinates and deformation direction, respectively.

[0114] Step S35: Perform assembly interference risk analysis based on the rib upturned type data to obtain assembly interference risk data.

[0115] In this embodiment, the data of the upward-curving ribs is 3D aligned with the data of the assembly reference surface of the parts, and the point cloud resolution of the reference surface is controlled at 1000 points per square millimeter. A collision detection algorithm is used to calculate the gap between the upward-curving portion of the rib and the assembled parts point by point. The distance threshold between calculated points is set to 0.1 mm, and areas with a distance less than this value are marked as interference risk areas. Cluster analysis is performed on the interference risk areas, retaining areas with a minimum area of ​​0.02 mm. 2 Clustering is used to identify effective risk regions. For each risk region, the maximum interference depth and contact area are calculated. The maximum interference depth threshold is set to 0.05 mm, and the contact area is expressed in mm². 2 Measurement. Output assembly interference risk data, including the spatial location of the risk area, interference depth, contact area, and risk level. The data is stored in a structured table format for easy subsequent risk management.

[0116] Of particular importance is that step S35 includes the following steps;

[0117] Step S351: Determine the assembly reference surface of the rib area based on the rib upturn type data, and obtain the assembly reference surface data;

[0118] In this embodiment, after acquiring the data of the upturned rib, the local normal vector set is first extracted based on the distribution characteristics of its 3D point cloud data. Principal Component Analysis (PCA) is then used to calculate the main distribution plane of the rib region. The PCA input point set is limited by the rib surface contour boundary, and the number is controlled within 5000 points to reduce computational resource consumption. The first principal direction and its corresponding normal vector are obtained by sorting the eigenvalues ​​of the covariance matrix. The angle θ between this normal vector and the standard assembly direction (Z-axis) is then calculated. If θ is less than 15°, the extracted plane is considered a reasonable assembly reference plane. Finally, the standard plane equation Ax + By + Cz + D = 0 is constructed using this plane normal vector and its origin coordinates, and the output is the assembly reference plane data.

[0119] Step S352: Analyze the installation posture of the target part based on the assembly reference surface data to obtain installation posture data;

[0120] In this embodiment, after obtaining the assembly reference plane, a Hough transform is used to fit a straight line to the reference edge of the part to be mated, and the attitude deviation is determined by combining the angle α between the fitted line and the normal vector of the reference plane. Specifically, Euler angles (pitch, yaw, roll) are used as attitude parameters. The assembly reference plane is defined as the coordinate system reference plane, and the edge direction vector of the target part is transformed to this reference system to calculate the attitude change angle. The least squares method is used to fit the transformation matrix, and the attitude offset parameters (Δθx, Δθy, Δθz) are extracted. The offset of each dimension is retained to two decimal places. If the attitude offset angle exceeds ±2.5° (mechanical assembly tolerance standard), it is marked as an abnormal installation attitude, and the installation attitude data is output.

[0121] Step S353: Based on the installation attitude data, perform rib interference space positioning to obtain interference space boundary data;

[0122] In this embodiment, based on the attitude data output in step S352, a three-dimensional Boolean operation function is called to calculate the difference between the volume of the installed part and the original design spatial volume. The Boolean algorithm used is based on a half-edge structure, with an input model accuracy of 0.01 mm and a mesh triangular facet count of no less than 200,000 faces to ensure calculation accuracy. During the calculation process, the rib area is designated as the interference-sensitive area, and the boundary is extended outward by 1 mm as a buffer layer. The calculated difference is the spatial interference region. The coordinate range of the six faces of the difference bounding box is extracted, and the interference spatial boundary data (format [minX, maxX, minY, maxY, minZ, maxZ]) is output based on its maximum boundary range.

[0123] Step S354: Solve for the interference vectors of the mating parts based on the interference space boundary data to obtain the interference vector data of the mating parts. In this embodiment, a spatial voxel mesh is constructed using the interference space boundary data, and the target mating part model is spatially sampled. The sampling accuracy is set to 0.2 mm, and whether each voxel contains multiple part mesh points is used to determine whether it is an interference voxel. Subsequently, the normal vector direction is calculated at the center point of each interference voxel. The normal vector value is the unit vector of the shortest path from the mating part to the center of the interference boundary. A three-dimensional vector Vi = (vx, vy, vz) is constructed at each interference point, and the frequency distribution and amplitude statistics of this type of vector direction are statistically analyzed. The interference vector data is output in the format of a vector set {Vi} and its statistical histogram.

[0124] Step S355: Perform assembly interference risk analysis based on the interference vector data of mating parts to obtain assembly interference risk data.

[0125] In this embodiment, based on the interference vector data of mating parts, a risk level classification function R(v,θ) is established, where v is the vector amplitude (unit: mm) and θ is the angle between the vector and the assembly direction (unit: °). The function form is R(v,θ) = kvv + kθ|θ|, where kv = 1 and kθ = 0.1. By traversing all interference vectors, the corresponding R values are calculated, and the risk levels are classified according to the threshold criteria. R ≤ 2 is determined as low risk, 2 < R ≤ 5 is medium risk, and R > 5 is high risk. At the same time, the positions of high-risk vectors are reflected in the image coordinates of the rib position area, and the area is colored and marked, and the assembly interference risk data is output. The output content includes the risk level distribution map and the list of high-risk point coordinates.

[0126] Preferably, step S4 includes the following steps:

[0127] Step S41: Calculate the depression depth according to the concave data of the rib position to obtain the depression depth data;

[0128] In this embodiment, the surface of the rib position area is reconstructed, and the weighted B-spline surface fitting algorithm is used. During the fitting process, a local control point grid is used, and the control point spacing is 0.1 mm to ensure that the reconstructed surface has sufficient smoothness and local adaptability. Then, the ideal design reference plane data is extracted as the reference plane for depression detection. The difference in the normal distance between the surface and the reference plane is compared, and the point-by-point vertical projection calculation method is used. The normal distances of all sampling points need to be calculated, and the output is the initial matrix of the depression depth. The depth values of this matrix are screened, and the areas greater than 0.02 mm are retained. This threshold is set based on the lower limit of the acceptable depression limit for the aluminum alloy surface in the national standard GB / T24641. After screening, the effective depression area is interpolated and reconstructed, and the maximum depth, average depth, and depression area are calculated, which are recorded as d_max, d_avg, and S_r respectively. Finally, the depression depth data is output in the structure of three-dimensional coordinates plus depth values for the next step of depression model construction.

[0129] Step S42: Construct a rib position depression model based on the depression depth data; identify the surface retention area according to the rib position depression model; detect impurity accumulation according to the surface retention area to obtain impurity accumulation data;

[0130] In this embodiment, a three-dimensional model of the rib depression is constructed using a mesh generation method. The depression area is divided into equilateral triangular meshes with a side length of 0.05 mm to ensure the resolution of the surface geometry. An explicit depression contour extraction algorithm is used, and the depression boundary line is filtered by setting a depression boundary curvature change threshold of 0.3 (unit: 1 / mm), and the morphological relationship between the depression bottom and the boundary is digitally encoded. Subsequently, based on the gravitational potential energy theory and surface tension model, the aggregation trend of liquid impurities under the guidance of gravity and curvature is simulated. The local minimum height point is identified in the depression model as the liquid retention point, and a spatial region with a neighborhood radius of 0.1 mm is selected for grayscale image mapping. The corresponding image grayscale value range is set to [180, 255] as the threshold standard for signs of contamination on a smooth reflective surface. Combining image reflectivity data and light spot diffusion contours, the surface retention area is detected, and the impurity distribution center, distribution radius, and area are recorded and summarized as impurity accumulation data.

[0131] Step S43: Perform concentration detection based on impurity accumulation data to obtain impurity concentration data;

[0132] In this embodiment, a grayscale image block is extracted for each retention area, with the image block size set to 100×100 pixels, using image data with a resolution of 5μm / pixel. The histogram grayscale statistical method is applied to calculate the average grayscale value μ and standard deviation σ within the area, and an impurity reflectance grayscale calibration value μ0 = 210 is introduced. An impurity concentration evaluation index C is constructed according to the formula C = (μ - μ0) / σ. Areas with a C value greater than 1.5 are defined as high-concentration impurity areas. To improve detection accuracy, a multispectral image verification step is used: images are acquired in the visible light (550nm) and near-infrared (850nm) bands respectively, and the reflectance change ΔR of impurities at different wavelengths is compared. ΔR > 15% is set as the confirmation criterion for the presence of high-concentration impurities. Finally, impurity concentration data is output, including concentration level (high, medium, low), area distribution, center coordinates, and multispectral response ratio.

[0133] Step S44: Based on the impurity concentration data, perform passivation film surface corrosion detection to obtain passivation film corrosion data;

[0134] In this embodiment, a combination of plasma interferometry imaging and multispectral reflectance imaging technology is used to detect passivation film corrosion. First, the high-concentration region in the impurity concentration data is selected as the detection target. Local reflectance R1, R2, and R3 are measured at three wavelengths: 550nm, 650nm, and 850nm, respectively, and the spectral reflectance ΔR is calculated. 12 =R1-R2, ΔR 23 =R2-R3. Based on the reflectivity change curve of aluminum passivation film after corrosion in the material literature, ΔR is set as follows: 12 >10%, ΔR 23>8% is the film corrosion threshold. By combining the changes in interference fringes in the interference image, regions of varying film thickness are identified. Areas with an interference fringe spacing of less than 5 μm are considered regions of sharp film thickness decreases and are identified as corrosion points. Corrosion data includes corrosion area A_c, corrosion depth d_c (calculated by fringe counting and phase shift method), corrosion location coordinates (X, Y), and corrosion morphology type (pitting corrosion, surface corrosion). Finally, a passivation film corrosion data table is generated for subsequent image model updates.

[0135] Of particular importance, step S44 includes the following steps:

[0136] Step S441: Reconstruct the surface concentration distribution based on the impurity concentration data to obtain the impurity concentration distribution map;

[0137] In this embodiment, during the surface concentration distribution reconstruction based on impurity concentration data, the acquired impurity accumulation image is first used to extract the mean and standard deviation of the grayscale values ​​of pixel block regions using a local grayscale statistical method. The image size is set to 1920×1080 pixels, using an 8-bit grayscale image format, with the grayscale value of each pixel ranging from 0 to 255. The image is divided into 32×32 pixel non-overlapping blocks, and a two-dimensional impurity concentration heatmap is constructed by bilinear interpolation of the mean grayscale value within each block. The interpolation method used is bilinear interpolation. The color levels of the heatmap are mapped to concentration units (e.g., mg / cm³). 2 The conversion between grayscale values ​​and concentration values ​​is performed using a standard curve (derived from known impurity standard samples through spectral grayscale fitting). The output is an impurity concentration distribution map with a spatial resolution of 100 sampling points per square millimeter. The data structure is a two-dimensional matrix, and the unit is mg / cm³. 2 .

[0138] Step S442: Identify corrosion-sensitive areas based on the impurity concentration distribution map;

[0139] In this embodiment, during the process of identifying corrosion-sensitive areas based on the impurity concentration distribution map, image segmentation and region growing methods are used to identify high-concentration abnormal areas. Specifically, the initial corrosion-sensitive concentration threshold in the impurity concentration distribution map is first set to 15 mg / cm³. 2 This threshold was obtained statistically from multiple batches of historical corrosion samples. All pixels in the image with a value greater than or equal to this threshold were labeled as connected components, using an 8-neighbor pixel connection method. A morphological opening operation (structural element is a 3×3 matrix) was performed on each connected region, removing pixels with an area smaller than 0.2 mm. 2 Isolated noise areas are identified. The remaining high-concentration areas are retained as corrosion-sensitive areas, and the output format is a binary mask image, where white areas represent corrosion-sensitive areas and black areas represent non-sensitive areas.

[0140] Step S443: Perform electrochemical reaction detection on the corrosion-sensitive area to obtain electrochemical reaction data;

[0141] In this embodiment, when detecting electrochemical reactions based on corrosion-sensitive areas, the electrochemical activity texture recognition method is first used. High-frequency texture features are extracted from the image blocks corresponding to the corrosion-sensitive areas using wavelet transform. The Daubechies-4 wavelet basis is selected, and a three-level wavelet decomposition is performed to extract the energy density of the high-frequency components. Areas with a high-frequency energy density exceeding 0.035 (in normalized energy) are identified as having a tendency for active electrochemical reactions. Simultaneously, local color reflectance analysis is performed on these areas using the RGB channel reflectance ratio method. The R:G:B ratio difference standard is set to 1.5:1:1. If the proportion of reddish areas exceeds 30% of the image area, the area is considered to have a local anodic reaction. Combining these two criteria, all electrochemical reaction areas are labeled, and electrochemical reaction data is generated in the format of location coordinates + reaction level label.

[0142] Step S444: Identify the active region of the electrochemical reaction based on the electrochemical reaction data;

[0143] In this embodiment, during the process of labeling the active regions of electrochemical reactions based on electrochemical reaction data, regions of level 2 and 3 are extracted according to the electrochemical reaction level labels obtained in S443 (levels are 0-none, 1-slight, 2-moderate, 3-vigorous). Area filtering is then applied to these regions to remove those smaller than 0.1 mm. 2 The regions were selected for edge fitting using the Canny edge detection algorithm (high threshold of 150, low threshold of 50), and closed curve interpolation was applied to the edges. Each closed region was defined as an electrochemically active region. These regions were numbered and packaged with corresponding impurity concentration and area information to form an electrochemically active region calibration data table. The table's fields included region number, location coordinates, and area (in mm). 2 ), reaction level, maximum concentration value, etc.

[0144] Step S445: Conduct passivation film surface corrosion detection based on the active electrochemical reaction region to obtain passivation film corrosion data.

[0145] In this embodiment, when detecting passivation film surface corrosion based on electrochemically active regions, a combined texture change and morphological depression identification method is adopted. First, the gray-level co-occurrence matrix (GLCM) is extracted from the active region image, and texture feature values ​​(including energy, contrast, and homogeneity) are calculated. A contrast threshold of 12 and a homogeneity threshold of 0.25 are set; values ​​exceeding these thresholds indicate significant texture degradation. Next, morphological reconstruction is performed on the image. Surface normal distribution and height map are calculated using Shape-from-Shading technology based on the illumination image, and a depression depth threshold of 5 μm is set. If a point cloud with a depression depth greater than this threshold exists within the region, passivation film corrosion is identified. Finally, the detection results are output as corrosion region images and a passivation film corrosion data list, with fields including corrosion region number and corrosion area (in mm). 2 ), maximum indentation depth, average texture contrast, etc.

[0146] Step S45: Transmit the assembly interference risk data and passivation film corrosion data to the image defect model to obtain the image defect optimization model;

[0147] In this embodiment, the assembly interference risk data obtained in step S35 and the passivation film corrosion data obtained in step S44 are converted into a defect label map in a unified format. The defect label map uses a Boolean matrix with the same resolution as the original image, where a value of 1 represents a defective area. The defect label map is mapped to the original image coordinate system through matrix overlay and merged with existing defect labels such as edge defects, broken teeth, and damage. A logical "OR" operation is used to integrate all defective areas. Subsequently, image pyramid processing is performed on the integrated defect distribution map to generate a multi-scale image pyramid with a scaling ratio of 0.5 for each layer and a maximum of 5 layers, used for multi-scale defect alignment and identification. The image defect optimization model constructs a convolutional region priority mapping matrix based on this multi-layer image and defect labels, with priority levels divided into five categories, P1-P5, to guide subsequent defect detection strategies.

[0148] Step S46: Perform defect detection on the part image based on the image defect optimization model to obtain part defect data.

[0149] In this embodiment, a joint detection algorithm based on image gradient edge features and texture changes is used to perform the final defect detection. Based on the optimized image defect model, image edge information is extracted using the Histogram of Oriented Gradients (HOG), with image blocks set to 16×16 pixels and a stride of 8 pixels. For each priority region, the threshold strategy is adjusted: a threshold of 30 is used for the edge gradient magnitude in the P5 region (highest priority), and a threshold of 70 is used in the P1 region (lowest priority). Texture changes are calculated using Local Binary Pattern (LBP), with a window size of 3×3 and rotation-invariant encoding. After merging all edge and texture change regions, a morphological closing operation (with a 5×5 cross-shaped structuring element) is performed to remove isolated pixels and small-area noise. The final component defect data is output, including attributes such as defect type, spatial location, area, and average grayscale. The results are provided in the form of a structured defect report for subsequent quality control and rework recommendations.

[0150] Preferably, this specification also provides an image model-based component defect detection system for performing the image model-based component defect detection method described above. The image model-based component defect detection system includes:

[0151] The image defect model construction module is used to acquire images of spare parts; perform edge defect detection on the spare parts based on the images to obtain edge defect data; and construct an image defect model based on the edge defect data.

[0152] The component thermal stability assessment module is used to simulate crack growth based on edge defect data to obtain crack data; to detect metal grain coarsening based on crack data to obtain metal grain coarsening data; and to assess the thermal stability of the component based on the metal grain coarsening data to obtain component thermal stability data.

[0153] The assembly interference risk analysis module is used to detect warpage in the rib area based on the thermal stability data of the parts, and obtain rib warpage data; classify the warpage morphology based on the rib warpage data, and obtain rib upward warpage data and rib downward warpage data; and perform assembly interference risk analysis based on the rib upward warpage data to obtain assembly interference risk data.

[0154] The defect detection module is used to detect impurity accumulation based on the concave data of the rib position, and obtain impurity accumulation data; to assess the degree of passivation film corrosion based on the impurity accumulation, and obtain passivation film corrosion data; to transmit the assembly interference risk data and passivation film corrosion data to the image defect model, and obtain the image defect optimization model; and to perform defect detection on the component image based on the image defect optimization model, and obtain component defect data.

[0155] Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of the equivalents of the application are intended to be included within the invention.

[0156] The above description is merely a specific embodiment of the present invention, enabling those skilled in the art to understand or implement the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features of the invention herein.

Claims

1. A method for detecting defects in spare parts based on image models, characterized in that, Includes the following steps: Step S1: Obtain images of spare parts; Edge defect detection is performed on the parts images to obtain edge defect data; Constructing an image defect model based on edge defect data, step S1 includes the following steps: Step S11: Obtain the image of the spare parts and perform image preprocessing to obtain the image of the spare parts to be processed; Step S12: Analyze gear tooth breakage based on the part images to obtain gear tooth breakage data; Step S13: Analyze the edge and corner damage of aluminum parts based on the part images to obtain edge and corner damage data; Step S14: Integrate the gear tooth breakage data and the aluminum part edge damage data to obtain edge defect data; Step S15: Construct an image defect model based on edge defect data; Step S2: Simulate crack growth based on edge defect data to obtain crack data; perform metal grain coarsening detection based on crack data to obtain metal grain coarsening data; evaluate the thermal stability of the component based on the metal grain coarsening data to obtain component thermal stability data. Step S3: Based on the thermal stability data of the parts, perform warpage detection on the rib area to obtain rib warpage data; classify the warpage morphology based on the rib warpage data to obtain rib upward warpage data and rib downward warpage data; perform assembly interference risk analysis based on the rib upward warpage data to obtain assembly interference risk data. Step S3 includes the following steps: Step S31: Identify heat-sensitive areas based on the thermal stability data of the components; Step S32: Calculate the geometric offset of the ribs based on the heat-sensitive area to obtain the geometric offset data of the ribs; Step S33: Calculate the warping amplitude based on the rib geometric offset data to obtain the rib warping data; Step S34: Identify the deformation direction based on the rib warping data to obtain deformation direction data; classify the rib warping data into warping morphology based on the deformation direction data to obtain rib warping data with upward warping and rib warping data with downward warping. Step S35: Perform assembly interference risk analysis based on the rib upturn type data to obtain assembly interference risk data; Step S4: Based on the concave data of the rib, perform impurity accumulation detection to obtain impurity accumulation data; assess the passivation film corrosion degree based on impurity accumulation to obtain passivation film corrosion data; transmit the assembly interference risk data and passivation film corrosion data to the image defect model to obtain the image defect optimization model; perform defect detection on the component image based on the image defect optimization model to obtain component defect data. Step S4 includes the following steps: Step S41: Calculate the depression depth based on the concave shape data of the rib to obtain the depression depth data; Step S42: Construct a rib depression model based on the depression depth data; identify the surface retention area based on the rib depression model; perform impurity accumulation detection based on the surface retention area to obtain impurity accumulation data; Step S43: Perform concentration detection based on impurity accumulation data to obtain impurity concentration data; Step S44: Based on the impurity concentration data, perform passivation film surface corrosion detection to obtain passivation film corrosion data; Step S45: Transmit the assembly interference risk data and passivation film corrosion data to the image defect model to obtain the image defect optimization model; Step S46: Perform defect detection on the part image based on the image defect optimization model to obtain part defect data.

2. The image model-based defect detection method for spare parts according to claim 1, characterized in that, Step S12 includes the following steps: Step S121: Segment the gear region based on the part image to obtain gear region data; Step S122: Detect poor lubrication based on gear area data to obtain poor lubrication data; Step S123: Predict gear surface wear based on poor lubrication data to obtain gear surface wear data; Step S124: Perform grain structure analysis based on gear surface wear data to obtain grain structure data; Step S125: Perform grain boundary slip analysis based on grain structure data to obtain grain boundary slip data; Step S126: Perform gear tooth breakage analysis based on grain boundary slip data to obtain gear tooth breakage data.

3. The image model-based defect detection method for spare parts according to claim 1, characterized in that, Step S13 includes the following steps: Step S131: Segment the corner area of ​​the aluminum part according to the part image to obtain the corner area data of the aluminum part; Step S132: Perform plastic deformation detection based on the data of the aluminum part's corner area to obtain plastic deformation data; Step S133: Evaluate the hardness of the aluminum part's edge and corner material based on the plastic deformation data to obtain the hardness data of the aluminum part's edge and corner material; Step S134: Identify low-hardness aluminum part corner areas based on the hardness data of the aluminum part corner material; Step S135: Perform aluminum alloy surface peeling analysis based on the corner area of ​​the low-hardness aluminum part to obtain aluminum alloy surface peeling data; evaluate the degree of oxide film peeling based on the aluminum alloy surface peeling data to obtain oxide film peeling data. Step S136: Analyze the edge and corner damage of aluminum parts based on the oxide film peeling data to obtain the edge and corner damage data of aluminum parts.

4. The image model-based defect detection method for spare parts according to claim 1, characterized in that, Step S2, which involves simulating crack growth based on edge defect data, includes: Import edge defect data into the crack growth simulation platform; In the crack growth simulation platform, the material fatigue performance parameters are set, with the stress intensity factor ranging from [value missing]. - The crack propagation rate constant is to mm / cycle; In the crack growth simulation platform, the load amplitude range is set to 100N-1000N, the number of cycles ranges from 10,000 to 1,000,000, and the ambient temperature ranges from -40℃ to 80℃. In the crack growth simulation platform, the simulation time step is set to 0.1s-1s, and the crack propagation calculation accuracy is 0.01mm-0.1mm; Run the crack growth simulation module and output crack data.

5. The image model-based defect detection method for spare parts according to claim 1, characterized in that, Step S2, which involves detecting coarse metal grains based on crack data, includes: Grain boundaries are identified based on crack data to obtain grain boundary data. The uniformity of the microstructure is detected based on the grain boundary data, and the uniformity of the microstructure is obtained. Identify oxide inclusion-rich regions based on grain structure uniformity data; Amorphous phase barrier points were determined based on oxide inclusion enrichment regions, and amorphous phase barrier point data were obtained. Metal grain coarseness detection was performed based on amorphous phase barrier point data to obtain metal grain coarseness data.

6. The image model-based defect detection method for spare parts according to claim 1, characterized in that, Step S2, which assesses the thermal stability of components based on data on metal grain size, includes: The grain boundary area is calculated based on the coarseness data of the metal grains to obtain the grain boundary area data; The material hardness decay analysis was performed based on the grain boundary area data to obtain the material hardness decay data; Material stress data is obtained by using large amounts of coarse metal grain data; Identify stress concentration areas in materials based on material stress data; Based on the analysis of the reduction of grain boundary strengthening effect in the stress concentration region of the material, data on the reduction of grain boundary strengthening effect are obtained. The thermal stability of the components is evaluated based on the data on material hardness decay and the data on the reduction of grain boundary strengthening effect, and the thermal stability data of the components is obtained.

7. A component defect detection system based on an image model, characterized in that, For performing the image model-based component defect detection method as described in claim 1, the image model-based component defect detection system comprises: The image defect model construction module is used to acquire images of spare parts; perform edge defect detection on the spare parts based on the images to obtain edge defect data; and construct an image defect model based on the edge defect data. The component thermal stability assessment module is used to simulate crack growth based on edge defect data to obtain crack data; to detect metal grain coarsening based on crack data to obtain metal grain coarsening data; and to assess the thermal stability of the component based on the metal grain coarsening data to obtain component thermal stability data. The assembly interference risk analysis module is used to detect warpage in the rib area based on the thermal stability data of the parts, and obtain rib warpage data; classify the warpage morphology based on the rib warpage data, and obtain rib upward warpage data and rib downward warpage data; and perform assembly interference risk analysis based on the rib upward warpage data to obtain assembly interference risk data. The defect detection module is used to detect impurity accumulation based on the concave data of the rib position, and obtain impurity accumulation data; to assess the degree of passivation film corrosion based on the impurity accumulation, and obtain passivation film corrosion data; to transmit the assembly interference risk data and passivation film corrosion data to the image defect model, and obtain the image defect optimization model; and to perform defect detection on the component image based on the image defect optimization model, and obtain component defect data.

Images