Structure optimization design and debugging method and system based on image recognition

Abstract

The invention discloses a structure optimization design and debugging method and system based on image recognition, and the method comprises the steps: firstly extracting contour features from a component image, positioning a preliminary defect candidate region, carrying out the fine classification and severity evaluation of the size deviation in the region through a neural network, fusing the geometric and material attribute deviation information, and carrying out the precise classification of the size deviation. Analyzing a defect group distribution mode through clustering, triggering numerical simulation to predict performance stability when the distribution density exceeds a limit, reversely adjusting design and material parameters by utilizing an optimization algorithm according to stress and time domain response data obtained through simulation, and finally generating a virtual assembly model and an optimization scheme which accord with an assembly precision standard. According to the method, a whole-process closed loop from defect intelligent identification, group influence analysis to stability driving parameter optimization is realized, and the quality control and design optimization efficiency of a complex mechanical device is remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of industrial manufacturing and mechanical design technology, and in particular discloses a method and system for structural optimization design and debugging based on image recognition. Background Technology

[0002] In the fields of industrial manufacturing and mechanical design, structural optimization and debugging are crucial steps in ensuring equipment performance and production efficiency. This process directly impacts product quality and a company's core competitiveness. With the increasing complexity of industrial equipment, improving design accuracy and debugging efficiency through technological means has become an urgent need for industry development. Traditional structural design and debugging methods often rely on manual experience, which is insufficient to meet the high precision and efficiency requirements of modern industry. Therefore, the introduction of intelligent technologies is urgently needed to overcome existing bottlenecks.

[0003] Currently, most solutions, when dealing with complex mechanical structures, often lack the ability to dynamically capture and comprehensively analyze structural details, making it difficult to promptly identify and resolve design deviations and debugging problems. Especially when assembling multiple components and operating environments change, existing methods often fail to accurately identify subtle defects and struggle to systematically adjust and optimize design parameters. This limitation makes the design and debugging process time-consuming and costly, impacting the overall R&D cycle of industrial equipment.

[0004] A deeper technical challenge lies in the close relationship between identifying structural defects and generating optimization solutions. The first key factor is the accurate capture and analysis of structural images. Industrial equipment components vary in shape and size; without accurately extracting contour and dimensional information from images, hidden defects or deviations are difficult to detect. This problem further impacts the formulation of optimization solutions, because the lack of accurate deviation information prevents scientifically sound adjustments in direction and magnitude, potentially leading to optimization solutions that deviate from actual needs. For example, in the design of liquid distribution equipment, if the dimensional deviation at the connection of a critical component is not accurately identified, the optimization solution may not effectively improve assembly accuracy, ultimately affecting the stability of equipment operation.

[0005] Therefore, in complex industrial scenarios, how to accurately identify defects and generate scientific and reasonable optimization solutions through efficient analysis of structural images has become a key issue in improving design and debugging efficiency. Summary of the Invention

[0006] This invention provides a structural optimization design and debugging method and system based on image recognition, which aims to accurately identify defects and generate scientific and reasonable optimization schemes through efficient analysis of structural images.

[0007] One aspect of the present invention relates to a structural optimization design and debugging method based on image recognition, comprising the following steps: S100. Acquire component image data of the mechanical device, extract contour features from the component image data through image processing algorithm, and obtain preliminary defect candidate areas. The component image data includes multiple component structures, and the preliminary defect candidate areas correspond to potential abnormal parts. S200. A neural network model is used to classify dimensional deviations within the initial defect candidate region, and to determine the deviation type and severity. The deviation type involves geometric and material properties, and the severity is based on the deviation degree classification. S300. Obtain assembly accuracy influencing parameters from deviation type and severity, group similar defects using clustering method to obtain defect group distribution, where defect group distribution represents defect concentration pattern; S400. For the distribution of defect groups, determine whether the distribution density exceeds the threshold. If it does, use numerical simulation to simulate performance stability and obtain stability simulation results. The stability simulation results include stress and time-domain response data. S500: Obtain the basis for generating optimization schemes from the stability simulation results, adjust the design parameters through optimization algorithms, and determine the parameter optimization values, where the parameter optimization values ​​are adjusted for size and material. S600. Based on the optimized parameter values, generate a virtual assembly model, determine whether the assembly accuracy in the virtual assembly model meets the standard, and if it does, output the final optimized solution.

[0008] Further, step S100 includes: S110. Acquire component image data of the mechanical device, and perform median filtering on the component image data to obtain smooth image data. S120. Use edge detection operators to extract the contour features of component structures from smooth image data; S130. Perform a closed-loop search within the preset coordinate system based on the contour features to obtain a set of closed contours; S140. If the geometric parameter deviation between a specific region in the closed contour set and the standard component template exceeds a preset threshold, the specific region is determined as a preliminary defect candidate region.

[0009] Further, step S200 includes: S210. Extract multi-dimensional spatial geometric features and gray-level co-occurrence matrix texture features from the preliminary defect candidate region; S220. Input the multidimensional spatial geometric features and gray-level co-occurrence matrix texture features into the deep residual network model to obtain the probability distribution of geometric properties and the probability distribution of material properties. S230. Determine the deviation type of the preliminary defect candidate region based on the probability distribution of geometric properties and the probability distribution of material properties; S240. If the deviation type is a geometric attribute deviation, then calculate the numerical deviation. S250. If the deviation type belongs to material property deviation, then analyze the texture heterogeneity. S260. Determine the severity based on the amount of numerical deviation or the degree of textual heterogeneity. S270. Classify dimensional deviations within the initial defect candidate area based on deviation type and severity.

[0010] Further, step S300 includes: S310. Based on the type and severity of the deviation, retrieve the assembly clearance and geometric tolerance from the preset mapping matrix to determine the parameters affecting assembly accuracy. S320. Construct a high-dimensional feature vector using assembly accuracy influencing parameters, and divide the preliminary defect candidate region into spatial attributes using a density clustering algorithm to obtain a set of defects with similar geometric features and material properties. S330. Calculate the fluctuation range of contact stress and material hardness in local areas for defect sets, and determine the defect group by measuring the similarity between defect sets through Euclidean distance; S340. Extract the statistical features of surface roughness and load distribution within the defect group, and use the kernel density estimation method to calculate the spatial distribution density of defects on the workpiece surface to obtain the defect group distribution. S350. Identify abnormal clustering areas of thermal expansion and frictional resistance based on the distribution of defect groups, and determine the defect concentration mode by analyzing the coupling relationship between fit tolerance and structural stiffness.

[0011] Further, step S400 includes: S410. Extract local mesh coordinates and intrinsic material properties from the defect group distribution to determine the distribution density; S420. If the distribution density exceeds the preset critical evolution threshold, the finite element analysis model is invoked to obtain transient dynamic characteristics. S430. Based on the transient dynamic characteristics, the stiffness matrix and damping matrix are mapped, and the stability simulation results are obtained by iteratively solving the nonlinear motion equations. S440. Perform component decomposition on the stress tensor in the stability simulation results to determine the stress distribution state. S450: By matching the displacement vector and acceleration signal with the stress distribution state, the time-domain response data is obtained through spectrum transformation. S460: A performance evaluation matrix is ​​constructed by using stress distribution state and time-domain response data to achieve quantitative assessment of performance stability.

[0012] Further, step S500 includes: S510. Extract high-order eigenvectors from the performance evaluation matrix and map them to the multi-dimensional design space to determine the geometric topology variables to be corrected. S520. Perform gradient descent iterations on the geometric topological variables within the preset constraints to obtain the size evolution sequence; S530. Match the corresponding composite component ratios based on the extreme points in the size evolution sequence, and retrieve candidate reinforcing phase ratios from the material property library; S540. If the candidate reinforcing phase ratio meets the preset structural strength criterion threshold, the optimal configuration combination is obtained by coordinating the optimization of geometric topology variables and composite component ratios through a multi-objective particle swarm optimization algorithm. S550: Perform reconstruction processing on the original design model through the optimal configuration combination to determine the optimized parameter values ​​for size and material adjustments.

[0013] Further, step S600 includes: S610. Construct a geometric solid model in the 3D modeling engine based on the parameter optimization values. The geometric solid model is then assigned corresponding physical property materials. S620. Based on the geometric entity model, perform spatial pose alignment in the virtual simulation environment according to the preset topological connection relationship to generate a virtual assembly model. S630. Mesh out the contact surfaces in the virtual assembly model and extract the normal vector deviation of the mating surfaces, and determine whether the normal vector deviation is within the preset tolerance fluctuation range. S640. If the normal vector deviation is within the preset tolerance fluctuation range, extract the centroid offset between each component and calculate the assembly accuracy. S650: Determine whether the assembly accuracy is less than or equal to the preset error threshold. If the assembly accuracy is less than or equal to the preset error threshold, output the final optimized solution.

[0014] Another aspect of the present invention relates to an image recognition-based structural optimization design and debugging system, used to execute the above-described image recognition-based structural optimization design and debugging method, comprising: The preliminary defect candidate region acquisition module is used to acquire component image data of mechanical devices, extract contour features from component image data through image processing algorithms, and obtain preliminary defect candidate regions. The component image data includes multiple component structures, and the preliminary defect candidate regions correspond to potential abnormal parts. The deviation type and severity determination module is used to classify dimensional deviations within the initial defect candidate region using a neural network model, and determine the deviation type and severity. The deviation type involves geometric and material properties, and the severity is based on a deviation degree classification. The defect group distribution acquisition module is used to obtain assembly accuracy influence parameters from deviation type and severity, and group similar defects by clustering method to obtain defect group distribution, where defect group distribution represents defect concentration pattern; The stability simulation result acquisition module is used to determine whether the distribution density of the defect group exceeds the threshold. If it does, the performance stability is simulated using numerical simulation methods to obtain the stability simulation results, which include stress and time-domain response data. The parameter optimization value determination module is used to obtain the basis for generating optimization schemes from the stability simulation results, adjust the design parameters through optimization algorithms, and determine the parameter optimization values, where the parameter optimization values ​​are adjusted for size and material. The final optimization solution output module is used to generate a virtual assembly model based on the parameter optimization values, determine whether the assembly accuracy in the virtual assembly model meets the standard, and output the final optimization solution if it does.

[0015] The beneficial effects achieved by this invention are as follows: 1. The structural optimization design and debugging method and system based on image recognition provided by this invention are aimed at the complex business scenario problem of defect detection and assembly accuracy co-optimization in multi-component images of mechanical devices.

[0016] 2. This invention first extracts contour features from component images and locates preliminary defect candidate regions. Then, it uses a neural network to perform fine classification and severity assessment of dimensional deviations within the regions, integrates geometric and material property deviation information, and analyzes the distribution patterns of defect groups through clustering. When the distribution density exceeds the limit, numerical simulation is triggered to predict performance stability. Based on the stress and time-domain response data obtained from the simulation, the design and material parameters are adjusted in reverse using optimization algorithms. Finally, a virtual assembly model and optimization scheme that meet the assembly accuracy standards are generated.

[0017] 3. This invention realizes a closed-loop process from intelligent defect identification and group impact analysis to stability-driven parameter optimization, which significantly improves the efficiency of quality control and design optimization of complex mechanical devices. Attached Figure Description

[0018] Figure 1 This is a flowchart illustrating an embodiment of the image recognition-based structural optimization design and debugging method of the present invention; Figure 2 This is a functional block diagram of an embodiment of the image recognition-based structural optimization design and debugging system of the present invention.

[0019] Explanation of icon numbers: 10. Preliminary defect candidate region acquisition module; 20. Deviation type and severity determination module; 30. Defect group distribution acquisition module; 40. Stability simulation result acquisition module; 50. Parameter optimization value determination module; 60. Final optimization scheme output module. Detailed Implementation

[0020] To better understand the above technical solutions, the following will provide a detailed explanation of the technical solutions in conjunction with the accompanying drawings and specific implementation methods.

[0021] like Figure 1 As shown, the first embodiment of the present invention proposes a structural optimization design and debugging method based on image recognition, including the following steps: Step S100: Obtain component image data of the mechanical device, extract contour features from the component image data through image processing algorithm to obtain preliminary defect candidate regions, wherein the component image data includes multiple component structures, and the preliminary defect candidate regions correspond to potential abnormal parts.

[0022] This step involves component image acquisition and defect candidate region extraction. Complete component image data of the mechanical device to be optimized and debugged is acquired. This data covers the structure of all components, including their appearance, connections, and key dimensions. High-definition industrial cameras and 3D scanning equipment are used to ensure the image data's clarity, completeness, and resolution meet recognition requirements. Image preprocessing and feature extraction algorithms (such as edge detection, contour extraction, and threshold segmentation) are employed to reduce noise, enhance, and grayscale the component image data. Contour features, dimensional information, and morphological details of each component are accurately extracted from the images. Based on preset anomaly criteria, areas with irregular contours, dimensions deviating from preset ranges, or abnormal shapes are selected as preliminary defect candidate regions. These regions correspond to potential abnormal parts of the mechanical device components, providing targeted objects for subsequent defect analysis.

[0023] Step S200: Use a neural network model to classify the dimensional deviations within the preliminary defect candidate area, and determine the deviation type and severity. The deviation type involves geometric and material properties, and the severity is based on the deviation degree classification.

[0024] This step involves defect deviation classification and severity determination. The preliminary defect candidate region data obtained in step S100 is input into a preset neural network model (such as a deep learning model like CNN or ResNet). After training with a large number of defect samples, this neural network model can accurately classify and quantify dimensional deviations within the candidate regions. Based on the design standards of the mechanical device, the deviation type is determined. This deviation type covers geometric attribute deviations (such as dimensional deviations, shape deviations, positional deviations, parallelism deviations, coaxiality deviations, etc.) and material attribute deviations (such as material surface defects, uneven material thickness, abnormal material density, etc.). Simultaneously, based on the difference between the actual deviation value and the design standard value, and the degree of influence of the deviation on the component's function, the severity of the deviation is graded according to preset levels (such as minor deviation, moderate deviation, severe deviation, fatal deviation). Finally, the deviation type and severity corresponding to each preliminary defect candidate region are output, providing a basis for subsequent assembly accuracy analysis.

[0025] Step S300: Obtain assembly accuracy influencing parameters from deviation type and severity, group similar defects by clustering method to obtain defect group distribution, where defect group distribution represents defect concentration pattern.

[0026] This step involves assembly accuracy impact analysis and defect grouping. Based on the deviation type and severity determined in step S200, core parameters affecting the assembly accuracy of the mechanical device (i.e., assembly accuracy impact parameters) are extracted, including deviation size, deviation location, the impact weight corresponding to the deviation type, and the assembly correlation of parts. Clustering algorithms (such as K-means, hierarchical clustering, etc.) are used, with the assembly accuracy impact parameters as the clustering basis, to group all candidate defect regions. Defects with similar deviation types, severity, and impact ranges are grouped into the same group, resulting in a defect group distribution. This defect group distribution visually represents the concentration pattern of defects, clearly presenting the clustering areas, clustering types, and distribution density of defects in the mechanical device, providing targeted directions for subsequent performance stability simulation and structural optimization.

[0027] Step S400: For the defect group distribution, determine whether the distribution density exceeds the threshold. If it does, use numerical simulation method to simulate performance stability and obtain stability simulation results, which include stress and time domain response data.

[0028] This step involves defect density determination and performance stability simulation. Based on the defect group distribution obtained in step S300, the distribution density of each defect group (i.e., the number of defects per unit area, the percentage of defect coverage area, etc.) is calculated, and the distribution density is compared with a preset density threshold to determine whether the degree of defect aggregation exceeds an acceptable range. If the defect group distribution density exceeds the preset threshold, it indicates that defects are concentrated in that area, affecting the performance stability of the mechanical device. In this case, numerical simulation methods (such as finite element analysis, dynamic simulation, etc.) are used to construct a local or overall model of the mechanical device containing the defect group, simulating the performance stability of the mechanical device under actual working conditions, and outputting the stability simulation results. These stability simulation results include stress distribution data, strain time-domain response data, vibration time-domain response data, fatigue life data, etc., used to assess the degree of impact of defect aggregation on device performance.

[0029] Step S500: Obtain the basis for generating the optimization scheme from the stability simulation results, adjust the design parameters through the optimization algorithm, and determine the parameter optimization values, wherein the parameter optimization values ​​are adjusted for size and material.

[0030] This step involves extracting the basis for the optimization scheme and adjusting the design parameters. From the stability simulation results obtained in step S400, key indicators affecting the performance stability of the mechanical device (such as maximum stress value, strain deviation, vibration amplitude, etc.) are extracted as the basis for generating the optimization scheme, clarifying the core objectives of structural optimization (such as reducing stress concentration, reducing dimensional deviations, and improving assembly accuracy). Intelligent optimization algorithms (such as genetic algorithms, particle swarm optimization algorithms, etc.) are used to iteratively adjust the design parameters of the mechanical device, guided by the optimization objectives. These design parameters include the dimensional parameters of the components (such as length, width, thickness, aperture, etc.) and material parameters (such as material type, material density, material hardness, etc.). Through multiple rounds of optimization calculations, optimal parameter values ​​that balance performance stability and assembly accuracy are determined, providing data support for virtual assembly and scheme verification.

[0031] Step S600: Generate a virtual assembly model based on the optimized parameter values, determine whether the assembly accuracy in the virtual assembly model meets the standard, and if it does, output the final optimized solution.

[0032] This step involves virtual assembly verification and the output of the final optimized solution. Based on the optimized parameter values ​​determined in step S500, a virtual assembly model of the mechanical device is constructed using 3D modeling software. This model recreates the assembly relationships, dimensional parameters, and material properties of the components, ensuring that the virtual model is completely consistent with the optimized design. The assembly accuracy of the virtual assembly model is verified by comparing it with the assembly standards and performance requirements of the mechanical device, checking whether the fit clearances, positional accuracy, and dimensional consistency between components meet the preset standards. If the assembly accuracy of the virtual assembly model meets the standards and the performance indicators meet the design requirements, the optimized parameter value is confirmed to be valid, and the final optimized solution, including the optimized design parameter values, defect rectification plan, assembly accuracy standards, and performance verification results, is output. If the design does not meet the standards, the process returns to step S500, and the design parameters are readjusted until the virtual assembly model meets the requirements.

[0033] Furthermore, the image recognition-based structural optimization design and debugging method proposed in this embodiment includes step S100 as follows: Step S110: Obtain component image data of the mechanical device, and perform median filtering on the component image data to obtain smooth image data.

[0034] Component image data is derived using the following formula: (1) In formula (1), Represents component image data, Indicates mechanical device components. The image data acquisition function encapsulates the control and data reading logic of image acquisition hardware (such as industrial cameras and scanners), and is responsible for converting the optical information of physical components into digital image data. The control logic of formula (1) achieves accurate mapping from mechanical components to digital images through standardized image acquisition functions, seamlessly integrating the structural information of physical entities into the digital design process. Compared with traditional manual measurement or drawing modeling methods, it can quickly and objectively obtain the real state of components, providing a high-fidelity data foundation for subsequent defect identification and structural optimization, and significantly improving the efficiency and accuracy of structural optimization design and debugging.

[0035] Smoothed image data is obtained using the following formula: (2) In formula (2), Indicates smooth image data at location The value at that location, Represents the pixel values ​​of the component image data. Indicates The center of the filtering window, The median function is represented. The control logic of formula (2) achieves precise smoothing of component images through median filtering, preserving geometric edge information to the maximum extent while removing noise, thus laying the foundation for subsequent high-precision structural feature extraction and defect detection. Compared with traditional linear filtering methods, it significantly improves data quality in the preprocessing stage, effectively reducing the false detection rate and false negative rate of subsequent algorithms, and providing more reliable data support for image recognition-based structural optimization design.

[0036] In the step of acquiring component image data for mechanical devices, consider a production line scenario for automotive engine parts. Assume a high-resolution industrial camera is used to photograph the engine block, obtaining a series of grayscale image data. This grayscale image data includes the surface texture and edge information of the block. In this way, the component image data not only captures the external shape of the component but also records potential surface defects such as scratches or dents, thus providing a basis for subsequent processing. This acquisition process typically involves lighting control and camera calibration to ensure data accuracy and avoid excessive noise due to environmental interference. In practical applications, the goal of this step is to establish reliable visual input, supporting improved accuracy throughout the defect detection process.

[0037] Median filtering is applied to component image data to obtain smooth image data, which can be detailed as follows. First, median filtering is a non-linear filtering technique. Its principle is to replace the center pixel value with the median value within the pixel's neighborhood, thereby effectively removing salt-and-pepper noise without blurring edge details. For example, when processing the aforementioned engine block image, a 3x3 filtering window is applied, and the values ​​of the nine pixels surrounding each pixel are sorted, with the median value selected as the new pixel value. After this processing, isolated noise points caused by shooting shake or dust are smoothed out, resulting in clearer image data. Specifically, this process brings significant benefits in business applications. For example, in precision machinery manufacturing, it improves image quality, reduces the risk of misjudging defects, and ensures more reliable input data for subsequent steps.

[0038] Step S120: Extract the contour features of the component structure from the smoothed image data using the edge detection operator.

[0039] The contour features of the component structure are derived using the following formula: (3) In formula (3), Describes the outline features of the component structure. Represents smoothed image data. express directional partial derivative, express directional partial derivative, The formula (3) represents the preset gradient magnitude threshold, which is used to extract contour features when the gradient magnitude exceeds the preset threshold. The control logic of formula (3) achieves accurate extraction of component contours through gradient thresholding. Based on the smooth image, it uses grayscale gradient changes to accurately locate structural edges, effectively avoiding interference from noise and texture. Compared with traditional edge detection methods, the extracted contour features are clearer and more representative, providing a solid foundation for subsequent high-precision dimensional measurement and defect identification, and significantly improving the accuracy and reliability of image recognition-based structural optimization design.

[0040] This technical topic, which utilizes edge detection operators to extract contour features of component structures from smoothed image data, requires an explanation of the role and implementation of edge detection operators. Edge detection operators, such as the Sobel operator, work by calculating image gradients to highlight areas of abrupt pixel value changes, representing object boundaries. For example, applying the Sobel operator to smoothed cylinder block image data involves first calculating the horizontal and vertical gradients, then synthesizing the gradient magnitudes, and finally thresholding to obtain the contour lines. In this way, the edges of holes and surface contours of the cylinder block are clearly extracted, forming a feature set. This extraction process connects to the smoothed output of the previous step, ensuring that the contours are not interfered with by noise. In business applications, this helps in the automated identification of the structural integrity of components.

[0041] Step S130: Perform a closure search within a preset coordinate system based on the contour features to obtain a set of closed contours.

[0042] The closure retrieval score based on contour features is obtained using the following formula: (4) In formula (4), This represents the closed retrieval score based on contour features. Indicates the first The first outline The values ​​of the dimensional features in the preset coordinate system This indicates the standard closed feature of the preset coordinate system. Representing feature dimension, The scale parameter is used in this formula. A high score indicates a matching closure condition to form a set of closed contours. The control logic of formula (4) achieves intelligent retrieval of contour closure through multi-feature Gaussian product similarity, matching multi-dimensional geometric features with standard closure models instead of relying on a single endpoint coincidence check. Compared with traditional contour closure determination methods, it has higher tolerance and robustness to noise, breakpoints and slight deformations, and can extract effective closed contours from complex images more accurately, providing a reliable geometric basis for subsequent 3D reconstruction and dimensional measurement.

[0043] Based on contour features, a closed-loop retrieval is performed within a preset coordinate system to obtain a set of closed contours. This involves the application of contour tracking algorithms. Specifically, the principle of closed-loop retrieval is to trace the chain of contour points to determine whether a closed loop is formed. For example, using a chain code tracing method, starting from a starting point, adjacent points are traversed clockwise or counterclockwise until the starting point is returned, forming a closed curve. In the cylinder block example, the extracted contours are retrieved within a Cartesian coordinate system, filtering out circular closed contours such as cylinder bores and excluding incomplete line segments, thus obtaining a set containing multiple closed regions. This step is closely integrated with the aforementioned feature extraction, ensuring that only valid structural parts are processed. In defect detection operations, it can effectively isolate potential problem areas and improve overall efficiency.

[0044] Step S140: If the geometric parameter deviation between a specific region in the closed contour set and the standard component template exceeds a preset threshold, then the specific region is determined as a preliminary defect candidate region. Whether a specific region is a preliminary defect candidate region is determined by the following formula: (5) In formula (5), Indicates whether a specific area is a preliminary candidate for a defect. Geometric parameters representing a specific region within a set of closed contours. The geometric parameters represent the standard component template. The preset threshold is indicated. The control logic of formula (5) achieves automated initial screening of defect areas through geometric parameter threshold comparison. It directly and quantitatively compares the geometric information obtained from image recognition with the design standards to quickly locate the deviation area. Compared with the traditional manual visual inspection method, it is not only objective and consistent, but also can accurately lock in small deviations, providing a reliable starting point for subsequent defect analysis and structural optimization, and significantly improving the efficiency and accuracy of quality inspection.

[0045] If the geometric parameters of a specific region within the closed contour set deviate from those of the standard component template by more than a preset threshold, the specific region is identified as a preliminary candidate defect region. This determination process can be refined into parameter comparison. For example, the standard template defines geometric parameters such as a cylinder bore diameter of 80mm and a circumference of 251.3mm. The actual diameter and circumference of the closed contour are calculated. If the deviation exceeds a preset threshold of 5%, such as an actual diameter of 76mm, it is marked as a candidate defect. This process, based on the set from the previous step, logically extends the search results. In business operations, it achieves preliminary screening, reduces manual intervention, and provides a foundation for further verification, such as deep learning validation, thereby improving the quality control level of mechanical devices.

[0046] Preferably, the image recognition-based structural optimization design and debugging method proposed in this embodiment includes step S200: Step S210: Extract multidimensional spatial geometric features and gray-level co-occurrence matrix texture features from the preliminary defect candidate region.

[0047] Multidimensional spatial geometric features are extracted from the initial defect candidate region, including calculating parameters such as the region's area, perimeter, minimum bounding rectangle aspect ratio, and circularity. For example, when inspecting aero-engine turbine blades, a marked initial defect candidate region corresponds to a suspected defect on the blade edge. Specifically, the structural optimization design and debugging system calculates the pixel area of ​​this region and converts it into the actual physical area, while simultaneously measuring its perimeter. By calculating the circularity (4π multiplied by the area divided by the square of the perimeter), the deviation of the initial defect candidate region's shape from an ideal circle can be quantified. These geometric features together constitute a multidimensional vector describing the spatial morphology of the initial defect candidate region. It should be noted that the gray-level co-occurrence matrix texture feature is used to quantify the surface material properties of the candidate region. Taking a turbine blade as an example, its normal surface exhibits a uniform texture after precision machining, while cracked or corroded areas will produce abnormal texture patterns. Specifically, the gray-level co-occurrence matrix is ​​first calculated within the initial defect candidate region. This gray-level co-occurrence matrix statistically analyzes the probability of a pair of pixels with specific gray values ​​appearing simultaneously at a specific direction and distance. Subsequently, statistics such as contrast, correlation, energy, and homogeneity are extracted from this gray-level co-occurrence matrix. For example, a high contrast value suggests that there is a significant gray-level change in the area, such as a dark crack.

[0048] Step S220: Input the multidimensional spatial geometric features and gray-level co-occurrence matrix texture features into the deep residual network model to obtain the probability distribution of geometric properties and the probability distribution of material properties.

[0049] The probability distribution of geometric attributes is derived using the following formula: (6) In formula (6), Represents the probability distribution of geometric attributes. This represents a deep residual network model. Representing the geometric features of multidimensional space, The gray-level co-occurrence matrix texture features are represented. The control logic of formula (6) realizes intelligent recognition of geometric and material properties through a deep residual network, which deeply integrates multi-dimensional spatial geometric features with gray-level co-occurrence matrix texture features, breaking through the limitations of traditional methods that rely on a single feature. Compared with manual rules or shallow learning methods, the recognition accuracy and robustness under complex backgrounds and diverse defects are significantly improved, providing accurate attribute basis for subsequent defect classification and optimization design.

[0050] The probability distribution of material properties is derived using the following formula: (7) In formula (7), Represents the probability distribution of material properties. This represents a deep residual network model. Representing the geometric features of multidimensional space, The gray-level co-occurrence matrix texture features are represented. The control logic of formula (7) realizes intelligent recognition of material properties through a deep residual network, which deeply integrates multi-dimensional spatial geometric features with gray-level co-occurrence matrix texture features, breaking through the limitations of traditional methods that rely on a single feature. Compared with manual rules or shallow learning methods, the recognition accuracy and robustness under complex backgrounds and diverse defects are significantly improved, providing accurate material property basis for subsequent defect classification and optimization design.

[0051] The aforementioned multidimensional spatial geometric features and gray-level co-occurrence matrix texture features are input into a deep residual network model. This deep residual network model consists of multiple stacked residual blocks, each of which directly passes the input to the output through shortcut connections. This effectively alleviates the gradient vanishing problem in deep networks and facilitates the learning of more complex feature maps. The deep residual network model ultimately outputs two probability distributions: one for determining whether the defect belongs to a geometric property deviation, and the other for determining whether it belongs to a material property deviation. For example, the deep residual network model determines with 85% probability that the deviation of the current turbine blade candidate region originates from geometric properties.

[0052] S230. Determine the deviation type of the preliminary defect candidate region based on the probability distribution of geometric properties and the probability distribution of material properties.

[0053] The deviation type of the preliminary defect candidate region is determined using the following formula: (8) In formula (8), Indicates the first The deviation type of the initial defect candidate region is either geometric property deviation or material property deviation. The operator corresponding to the maximum value; Indicates the first The probability distribution of geometric attributes of a preliminary defect candidate region, i.e., the core probability value of the region predicted by the deep residual network model as a geometric attribute deviation. Indicates the first The material property probability distribution of the initial defect candidate region is the core probability value predicted by the deep residual network model as a material property deviation region. The control logic of formula (8) achieves accurate determination of the defect deviation type through the probability maximization criterion. It directly selects the category with higher confidence in the geometric and material property probability distributions output by the deep residual network as the deviation type, which is consistent with the physical meaning of the neural network's prediction of the target category - the higher the probability value, the stronger the confidence of the deep residual network model in the determination of the category. Compared with the traditional divergence analysis method, this determination logic is directly related to the physical essence of the model prediction. The mathematical derivation is concise and clear. Personnel in the relevant technical field can directly complete the classification based on the probability value, which greatly improves the understandability and reproducibility of the deviation type determination and provides accurate and clear decision-making basis for subsequent targeted geometric dimension correction or material property optimization.

[0054] Based on the maximum confidence values ​​of the geometric and material property probability distributions output by the deep residual network model, the deviation type of the preliminary defect candidate region is directly determined using the maximum value determination rule of formula (8): if If the deviation type of this region is determined to be geometric attribute deviation; if If the probability values ​​are equal, the area is marked as a material property deviation. If the probability values ​​are equal, the area is marked as a region to be reviewed. The final determination is then made in conjunction with manual screening to ensure the accuracy of the deviation type determination.

[0055] Step S240: If the deviation type belongs to geometric attribute deviation, then calculate the numerical deviation amount.

[0056] The numerical deviation is calculated using the following formula: (9) In formula (9), Indicates the numerical deviation. Represents the actual geometric attribute value. The standard geometric attribute value is represented. The control logic of formula (9) achieves precise quantification of geometric attribute deviation through the direct difference between the actual value and the standard value, and establishes a two-dimensional quantification standard of "direction + size" for geometric defects. Compared with the traditional evaluation method that only focuses on the absolute value of the deviation, it retains the deviation direction information, which can provide a more accurate adjustment basis for subsequent topology optimization (such as positive deviation guiding size reduction and negative deviation guiding size supplementation), significantly improving the pertinence and efficiency of structural optimization design.

[0057] If the probability of geometric property deviation is higher, the numerical deviation is further calculated. For example, for areas determined to be geometrically insufficient, the structural optimization design and commissioning system will compare its actual measured diameter or thickness with the standard design value and calculate the specific negative tolerance value, such as being 0.15 mm thinner than the standard.

[0058] Step S250: If the deviation type belongs to material property deviation, then analyze the texture heterogeneity.

[0059] Texture heterogeneity is calculated using the following formula: (10) In formula (10), Indicates the degree of heterogeneity in texture. Indicates the total number of sampling points. Indicates the first Texture attribute values ​​of each sampling point Indicates the first Texture gradient at each sampling point The formula (10) represents the gradient magnitude and calculates the standard deviation of texture attributes to measure the texture dispersion under material attribute deviation. The control logic of formula (10) achieves accurate quantification of material texture heterogeneity through the mean of texture gradient, and uses the intensity of local texture changes as the evaluation standard for material uniformity, breaking through the limitations of traditional methods that rely solely on grayscale statistics (such as standard deviation). Compared with traditional methods, it is more sensitive to small defects inside the material (such as microcracks and inclusions), and can identify material attribute deviations earlier and more accurately, providing a reliable quantitative basis for subsequent material ratio optimization and process adjustment.

[0060] If the probability of material property deviation is higher, then analyze the texture heterogeneity. This can be achieved by calculating the dispersion of each component in the texture feature vector, such as calculating the standard deviation of indicators like contrast and homogeneity. The larger the standard deviation, the worse the material uniformity within the region, and the more likely there may be inclusions or coating peeling.

[0061] Step S260: Determine the severity based on the numerical deviation or texture heterogeneity.

[0062] In one embodiment, the severity is determined by the following formula: (11) In formula (11), Indicates the severity. The numerical deviation is represented by the control logic of formula (11), which achieves accurate quantification of the severity of geometric attribute defects through direct mapping, simplifying the complex geometric deviation assessment into a single numerical index, rather than relying on human experience judgment. Compared with traditional qualitative assessment methods, it can more objectively and precisely characterize the severity of geometric defects, significantly improving the intelligence and precision of industrial component quality control.

[0063] In another embodiment, the severity is determined by the following formula: (12) In formula (12), Indicate the severity, The formula (12) represents the degree of heterogeneity in material properties. The control logic of formula (12) achieves accurate quantification of the severity of material property defects through direct mapping, simplifying the complex material texture assessment into a single numerical index, rather than relying on human experience. Compared with traditional qualitative assessment methods, it can more objectively and precisely characterize the severity of material defects, significantly improving the intelligence and precision of industrial component quality control.

[0064] Ultimately, the structural optimization design and debugging system determines the severity of defects based on the calculated numerical deviation or material heterogeneity, referring to preset grading thresholds, such as classifying them as minor, moderate, or severe.

[0065] Step S270: Classify the dimensional deviations within the preliminary defect candidate area by deviation type and severity.

[0066] The classification results of dimensional deviations within the initial defect candidate region are obtained using the following formula: (13) In formula (13), This indicates the classification results of dimensional deviations within the initial defect candidate region. This indicates the weight of the deviation type, and is a coefficient that adjusts the proportion of the deviation type in the classification. It can be adjusted according to the project priority (such as prioritizing assembly accuracy). Indicates the type of deviation. This represents the severity weight, a coefficient that adjusts the proportion of severity in the classification, and can be adjusted according to quality control requirements. The formula (13) represents the severity level and completes the classification by linearly weighted fusion of type and severity. The control logic of formula (13) realizes intelligent classification of defects through linear weighted fusion, integrating the two key dimensions of deviation type and severity into a unified classification index, rather than relying on a single feature or human experience. Compared with traditional qualitative classification methods, it can more objectively and precisely depict defect characteristics, significantly improving the intelligence and precision of industrial component quality control.

[0067] By combining the identified deviation types and severity, precise classification of dimensional deviations or material defects within the candidate area can be achieved, providing a direct basis for subsequent maintenance decisions.

[0068] Furthermore, in the image recognition-based structural optimization design and debugging method proposed in this embodiment, step S300 includes: Step S310: Based on the type and severity of the deviation, retrieve the assembly clearance and geometric tolerance from the preset mapping matrix to determine the parameters affecting assembly accuracy.

[0069] The parameters affecting assembly accuracy are derived using the following formula: (14) In formula (14), This indicates the parameters that affect assembly accuracy. This represents the assembly clearance mapping value based on the type and severity of the deviation. This formula represents the geometric tolerance mapping value based on the deviation type and severity. It is used to determine the assembly accuracy influence parameter using the retrieved assembly clearance to geometric tolerance ratio. The control logic of formula (14) achieves a quantitative assessment of the impact of defect characteristics on assembly accuracy through a preset mapping matrix and ratio calculation, transforming qualitative deviation types and quantitative severity into a unified dimensionless index. Compared to traditional experience-based judgment methods, this approach can more objectively and accurately assess the impact of defects on assembly accuracy, significantly improving the targeting and effectiveness of subsequent defect clustering and optimization strategies.

[0070] Based on the type and severity of deviations, the system retrieves assembly clearances and geometric tolerances from a predefined mapping matrix. First, it's important to understand that the mapping matrix is ​​a predefined table structure that associates various deviation types, such as geometric or material deviations, with corresponding assembly parameters. Specifically, for a geometric deviation with a moderate severity, the structural optimization design and debugging system queries the corresponding rows and columns in the matrix to extract the assembly clearance value. For example, in the assembly of aero-engine turbine blades, if the deviation type is edge defect and the severity is moderate, the clearance might be set to 0.2 mm, while geometric tolerances such as parallelism are adjusted to 0.05 mm. These assembly accuracy influencing parameters directly affect the overall assembly accuracy, thus determining assembly accuracy influencing parameters such as clearance deviation rate and tolerance cumulative value. For instance, when inspecting automotive gearbox gears, assuming the deviation type is insufficient material hardness and the severity is severe, the mapping matrix would output that the assembly clearance needs to be increased to 0.3 mm to compensate for potential deformation, and the geometric tolerance would be set to roundness of 0.1 mm. This construction of assembly accuracy influencing parameters helps assess the stability during assembly.

[0071] Step S320: Construct a high-dimensional feature vector using assembly accuracy influencing parameters, and divide the preliminary defect candidate regions into spatial attributes using a density clustering algorithm to obtain a set of defects with similar geometric features and material properties.

[0072] High-dimensional feature vectors are constructed using the following formula: (15) In formula (15), Indicates the first High-dimensional feature vectors of preliminary defect candidate regions Indicates the first The first region One parameter that affects assembly accuracy. Represents the high-dimensional feature dimension. The transpose is indicated. The control logic of formula (15) realizes the spatial attribute mapping of defects by constructing high-dimensional feature vectors, integrating heterogeneous parameters affecting assembly accuracy into a unified feature space, and providing a precise similarity measurement basis for density clustering. Compared with the traditional single-dimensional grouping method, it can more comprehensively and finely characterize the similarity of defects, significantly improving the accuracy and rationality of defect set division, and providing a reliable basis for subsequent batch optimization and process improvement.

[0073] The set of defects with similar geometric features and material properties is derived by the following formula: (16) In formula (16), Indicates the first A set of defects with similar geometric features and material properties. This represents the defect candidate region index. Indicates the first Geometric features of each region Indicates the first Class center geometric features Indicates the first Material properties of each region Indicates the first Class-centered material properties and Let these represent the geometric similarity function and the material similarity function, respectively. and represents the geometric similarity threshold and the material similarity threshold, respectively, and ∧ represents the logical AND operator. The control logic of formula (16) achieves accurate clustering of defects through the joint threshold determination of dual feature similarity, coupling the similarity determination of geometric features and material properties, breaking through the limitations of traditional single-dimensional clustering or manual classification. Compared with existing technologies, it can automatically divide the defect set with highly homogeneous features in complex defect scenarios, so that subsequent structural optimization and material ratio adjustment can be implemented in batches for "same type of defects", significantly improving the efficiency and pertinence of optimization design, and providing core technical support for quality closed-loop control in intelligent manufacturing.

[0074] A high-dimensional feature vector is constructed using parameters affecting assembly accuracy. A density clustering algorithm is then used to spatially divide the initial candidate defect regions. Here, the density clustering algorithm refers to DBSCAN (Density-Based Spatial Clustering of Applications with Noise), which groups points based on density rather than a predetermined number of clusters. Specifically, parameters such as gap values ​​and tolerance values ​​are first combined into a high-dimensional feature vector, such as [0.2, 0.05, 0.1]. Then, the algorithm is applied to the point cloud data of the candidate regions, setting a radius parameter of 2 mm and a minimum number of points of 5, thereby dividing similar regions. For example, on the blade surface, multiple defect sets with similar geometric curvature and material homogeneity are clustered. These defect sets reflect the spatial distribution pattern of the defects. For instance, in wind turbine blade inspection, the high-dimensional vector includes curvature features under the influence of gaps. Density clustering will group the curved regions near the root into a single group, resulting in a defect set containing 10 defect points, facilitating subsequent analysis.

[0075] Step S330: Calculate the fluctuation range of contact stress and material hardness in local areas for the defect set, and determine the defect group by measuring the similarity between defect sets through Euclidean distance.

[0076] The range of contact stress fluctuations is derived using the following formula: (17) In formula (17), Indicates the range of contact stress fluctuation. Indicates a local area. Indicates the local region's first Point contact stress. The control logic of formula (17) achieves precise quantification of the contact stress fluctuation range through extreme value difference calculation, simplifying the complex stress distribution problem into a single fluctuation index, and providing a clear and comparable quantitative basis for the mechanical characteristic evaluation of defect sets. Compared with the traditional method that only focuses on average stress, it can more sensitively capture the problems of stress concentration and uneven distribution, significantly improving the accuracy of defect group division and the pertinence of subsequent optimization design.

[0077] The range of material hardness fluctuation is obtained by the following formula: (18) In formula (18), Indicates the range of material hardness fluctuation. Indicates a local area. Indicates the local region's first The control logic of formula (18) achieves precise quantification of the material hardness fluctuation range through extreme value difference calculation, simplifying the complex material performance distribution problem into a single fluctuation index, and providing a clear and comparable quantitative basis for the material characteristic evaluation of defect sets. Compared with the traditional method that only focuses on average hardness, it can more sensitively capture the inhomogeneity and potential defects inside the material, significantly improving the accuracy of defect group division and the pertinence of subsequent optimization design.

[0078] The following formula is used to measure similarity and determine defect groups: (19) In formula (19), Indicates the first The set of defects and the first Euclidean distance between sets of defects Indicates the number of feature dimensions. Indicates the first The defect set of the nth defect set 3D eigenvalues Indicates the first The defect set of the nth defect set Eigenvalues. The control logic of formula (19) achieves accurate measurement of the similarity between defect sets through Euclidean distance, transforming complex mechanical-material characteristic differences into calculable distance indicators, providing a solid mathematical foundation for the automatic division of defect groups. Compared with traditional manual grouping methods, it can more objectively and finely divide defect groups, enabling subsequent structural optimization and material ratio adjustment to be implemented in batches for "same type of mechanical risk", significantly improving the efficiency and pertinence of optimization design.

[0079] For each defect set, the fluctuation range of contact stress and material hardness in local areas is calculated. The similarity between these defect sets is measured using Euclidean distance, which is the linear distance between vectors used to quantify similarity. Specifically, for a defect set, local contact stress is first measured (e.g., average value 50 MPa, fluctuation range ±10 MPa), and hardness is assessed (e.g., Vickers hardness HV200 fluctuation ±20). Then, the distance between sets is calculated, for example, the square root of the sum of the squares of the differences in the eigenvectors of two sets. If it is less than a threshold of 5, they are grouped together. This method of grouping defects helps identify potential fault sources. For example, in the inspection of bridge steel structures, a defect set with stress fluctuation of ±15 MPa and hardness fluctuation of ±25, and a calculated distance of 3.5 from another defect set, is grouped together, revealing corrosion-related patterns.

[0080] Step S340: Extract the statistical features of surface roughness and load distribution within the defect group, and use the kernel density estimation method to calculate the spatial distribution density of defects on the workpiece surface to obtain the defect group distribution.

[0081] The spatial distribution density of defects on the workpiece surface is obtained by the following formula: (20) In formula (20), This indicates the total number of defects. This represents the bandwidth parameter, a parameter that controls the smoothness of the kernel function. The larger the value, the smoother the density estimation curve; The smaller the size, the more sensitive it is to capturing local details. Represents the kernel function. Indicates the first The spatial location of the defect Represents any point on the workpiece surface. The formula (20) represents the spatial distribution density of defects. The control logic of formula (20) achieves accurate characterization of the spatial distribution of defects through kernel density estimation, transforming discrete defect location data into a continuous density distribution heatmap, which intuitively reveals the clustering pattern of defects. Compared with traditional simple counting or grid statistical methods, it can capture the spatial distribution trend of defects more smoothly and accurately, significantly improving the depth of defect analysis and the spatial targeting of subsequent optimization design.

[0082] Statistical features of surface roughness and load distribution are extracted within the defect groups. The spatial distribution density of defects on the workpiece surface is calculated using kernel density estimation, a non-parametric method that uses a Gaussian kernel function to smooth data points and estimate the density. Specifically, the average roughness Ra value (1.6 μm) and the standard deviation of the load distribution (20N) are first calculated. Then, a kernel function is applied to the group points with a bandwidth of 1 mm to obtain a density map. High-density areas indicate concentrated defects, thus revealing the defect group distribution and providing a basis for optimized design. For example, in the inspection of precision instrument housings, after extracting roughness statistics, kernel density estimation shows a surface center density peak of 0.8, indicating uneven defect group distribution.

[0083] Step S350: Identify abnormal clustering areas of thermal expansion and frictional resistance based on the distribution of defect groups, and determine the defect concentration mode by analyzing the coupling relationship between fit tolerance and structural stiffness.

[0084] The defect set pattern is determined using the following formula: (twenty one) In formula (21), Indicates the defect concentration pattern index. Indicates the candidate clustering area. Indicates the first The number of defect groups Indicates candidate clustering area The formula (21) determines the defect concentration pattern by maximizing the defect density within the area limit. The control logic of formula (21) achieves intelligent identification of defect concentration patterns by maximizing defect density, combining multi-physics coupling analysis with spatial distribution statistics to accurately locate the typical pattern where defects are most likely to aggregate. Compared with traditional single-point defect analysis methods, it can reveal the formation mechanism and distribution law of defects at the system level, significantly improving the pertinence and effectiveness of structural optimization design, and providing core technical support for quality root cause analysis in intelligent manufacturing.

[0085] Based on the distribution of the defect groups, abnormal clusters of thermal expansion and frictional resistance are identified. The coupling relationship between fit tolerances and structural stiffness is analyzed; here, the coupling relationship refers to how tolerance changes affect stiffness. Specifically, abnormal thermal expansion areas (e.g., expansion of 0.4 mm above average) are first identified from the distribution, and a peak frictional resistance of 15 N is calculated. Then, the interaction between a tolerance of 0.1 mm and a stiffness modulus of 200 GPa is analyzed. If tolerance relaxation leads to a 10% decrease in stiffness, it is identified as a concentrated pattern, such as edge clustering. This can improve maintenance efficiency and prevent overall failure.

[0086] Preferably, the image recognition-based structural optimization design and debugging method proposed in this embodiment includes step S400 as follows: Step S410: Extract local mesh coordinates and intrinsic material properties from the defect group distribution to determine the distribution density.

[0087] The distribution density is obtained using the following formula: (twenty two) In formula (22), Indicates the distribution density. Indicates the number of points within the grid. The grid volume is represented by the control logic of formula (22). The control logic realizes the quantitative analysis of defect distribution through grid density calculation, discretizing the continuous spatial distribution of defects into calculable grid density values, providing an objective and comparable quantitative basis for defect clustering assessment. Compared with the traditional kernel density estimation method, it has higher computational efficiency, is more suitable for engineering applications, and can quickly generate defect distribution heatmaps, significantly improving the efficiency of defect analysis and optimization design.

[0088] The distribution density is determined by extracting local grid coordinates and intrinsic material properties from the defect cluster distribution. Local grid coordinates refer to dividing the workpiece surface into fine grid cells, each recording x, y, and z coordinate values. Intrinsic material properties include inherent characteristics such as elastic modulus and Poisson's ratio. Specifically, in aero-engine blade inspection, the defect clusters are first sampled by coordinates. For example, the grid coordinates at the blade root are taken as (10, 5, 2) mm, and the material elastic modulus is extracted as 200 GPa. The distribution density is obtained by calculating the number of defect points within each grid and the average value of the properties. For example, the distribution density formula is based on the number of points divided by the grid volume. If the result is 0.5 points per cubic millimeter, it represents the distribution density, helping to identify potential fatigue sources. This method is based on the fact that defect clusters have already been obtained from previous clustering, and the density value is used for subsequent threshold judgment, improving early warning of high-risk areas in business operations.

[0089] Step S420: If the distribution density exceeds the preset critical evolution threshold, the finite element analysis model is invoked to obtain transient dynamic characteristics.

[0090] The transient dynamic characteristics are derived using the following formula: (twenty three) In formula (23), Indicates transient dynamic characteristics, Indicates the first First-order modal shape vector, Indicates the first First modal response coordinates, The modal order is represented by the modal superposition expression of the transient dynamic characteristics output by the finite element model. The control logic of formula (23) achieves efficient calculation of transient dynamic characteristics through modal superposition, reducing the high-dimensional finite element analysis problem to modal space, and significantly improving analysis efficiency while ensuring calculation accuracy. Compared with the traditional direct integration method, it is more suitable for the dynamic analysis of structures with complex defects, and can quickly and accurately assess the impact of defects on the dynamic characteristics of the structure, providing core technical support for structural optimization and reliability assessment in intelligent manufacturing.

[0091] If the distribution density exceeds a preset critical evolution threshold, a finite element analysis (FEM) model is invoked to obtain transient dynamic characteristics. The FEM is a numerical simulation tool that discretizes the structure into a finite number of elements and simulates dynamic behavior by solving differential equations. The transient dynamic characteristics include the time-domain responses of displacement, velocity, and acceleration, used to extract structural mode shapes, natural frequencies, and modal damping ratios, providing a basis for subsequent stiffness / damping matrix correction and optimization. Specifically, in bridge steel beam inspection, the assumed critical evolution threshold is 0.3 points per cubic millimeter. If the distribution density reaches 0.4, the FEM model is activated, inputting mesh data and load conditions, such as applying a 10kN transient force. The FEM model calculates characteristics such as a peak acceleration of 5 m / s². This invocation simulates the defect evolution process, enabling dynamic assessment of structural safety in practical applications and preventing sudden failures.

[0092] Step S430: Based on the transient dynamic characteristics, map the stiffness matrix and damping matrix, and obtain the stability simulation results by iteratively solving the nonlinear motion equations.

[0093] The corrected stiffness matrix is ​​obtained using the following formula: (twenty four) In formula (24), This represents the corrected stiffness matrix (n×n order, where n is the number of degrees of freedom of the structural finite element mesh), which is the final stiffness matrix used to solve the nonlinear motion equations. The modal mapping matrix (n×m, where n is the number of degrees of freedom, m is the extracted modal order, and m≤n) represents the mapping from modal space to physical space. The initial stiffness matrix (m×m) is derived from the material constitutive tensor. The core inherent property of structural stiffness is its direct construction with geometric topology. ( For geometric function matrices, (Unit volume). This represents the transpose of the modality mapping matrix; The stiffness correction matrix (n×n order) represents the mode shapes extracted from transient dynamic characteristics. and natural frequency The calculated value is used to correct the deviation between the initial stiffness matrix and the actual response. The modal shape matrix (n×m order) represents the extracted transient dynamic features. The intrinsic frequency vector (m×1 order) represents the extracted transient dynamic features. The control logic of formula (24) is "intrinsic property construction of initial matrix + dynamic feature correction and optimization", which fully conforms to the basic principles of structural mechanics: 1. First, the material constitutive tensor is used to construct the initial matrix. and geometric topology (shape functions) Unit volume Construct the initial stiffness matrix 1. Ensure that the core of the stiffness matrix is ​​determined by the inherent properties of the structure; 2. Extract mode shapes and natural frequencies through transient dynamic features to construct a mode mapping matrix. 3. To achieve the transformation from modal space to physical space; 4. To use the modal mapping matrix to convert the initial stiffness matrix. Mapped to physical space, and then processed by a stiffness correction matrix. By correcting the deviation between the initial matrix and the actual dynamic response, a precise stiffness matrix that matches the actual dynamic characteristics of the structure is finally obtained. 4. Throughout the process, transient dynamic characteristics are used only as a basis for verification and correction, rather than as the source for constructing the stiffness matrix, thus completely avoiding the logical contradiction of "constructing inherent properties from the response results".

[0094] The corrected damping matrix is ​​obtained using the following formula: (25) In formula (25), This represents the modified damping matrix (n×n order, where n is the number of degrees of freedom). Let n represent the modal mapping matrix (n×m order) consistent with formula (24). The initial damping matrix (m×m) is constructed using the Rayleigh damping model. , , The Rayleigh damping coefficient is... (where is the initial mass matrix), and is an inherent property of damping; The damping correction matrix (n×n order) is derived from the mode shapes. And modal damping ratio (Extracted from transient dynamic characteristics) Calculated; This represents the transpose of the modal mapping matrix. Formula (25) control logic: Consistent with the stiffness matrix mapping logic, the initial damping matrix is ​​first constructed using the Rayleigh damping model and the inherent properties of the structure. Then, it is mapped to physical space through a modal mapping matrix, and the modal damping ratio is extracted using transient dynamic features. Constructing the correction matrix The deviation between the initial damping matrix and the actual response is corrected to obtain the accurate damping matrix. .

[0095] First, the initial stiffness matrix is ​​constructed from the material constitutive tensor and the structural geometry and topology. and initial damping matrix Both are inherent properties of the structure, with the initial stiffness matrix being... ( For geometric function matrices, This is the transpose of the geometric function matrix. For the material constitutive tensor, (e.g., unit volume), initial damping matrix Constructed from Rayleigh damping model ( Secondly, modal identification is performed on transient dynamic characteristics (displacement, velocity, and acceleration response) to extract the preceding parameters. Mode matrix of the first principal modes Natural frequency And modal damping ratio , modal matrix After orthogonalization and normalization, the mode mapping matrix is ​​obtained. The dimensional matching rule is defined as follows: the row dimension equals the number of degrees of freedom of the finite element mesh of the structure. The column dimension is equal to the extracted modal order. ( Then, according to formulas (24) and (25), the initial stiffness matrix and initial damping matrix are mapped to the physical space through the modal mapping matrix, and the stiffness correction matrix is ​​constructed using the parameters extracted from the transient dynamic features. and damping correction matrix The initial matrix is ​​corrected to obtain a corrected stiffness matrix that matches the actual dynamic response. and the corrected damping matrix Finally, the corrected stiffness matrix is... Damping matrix With structural mass matrix Substituting into the nonlinear equation of motion ( For acceleration, For speed, For displacement, (For external loads), the equations are solved using the Newton-Raphson iterative method to obtain stability simulation results including stress, displacement, and time-domain response.

[0096] Based on the transient dynamic characteristics, the stiffness matrix and damping matrix are mapped, and the stability simulation results are obtained by iteratively solving the nonlinear equations of motion. Here, the structural stiffness matrix represents the rigid connection between elements, the damping matrix describes energy dissipation, and the nonlinear equations of motion involve quadratic terms of displacement. Iterative solutions are obtained by using the Newton-Raphson method to approximate the solution step by step. Specifically, in wind turbine tower applications, stiffness matrix elements are mapped from the characteristics, such as a diagonal value of 1000 N / m and a damping matrix of 0.5 Ns / m. Then, the equations are solved iteratively 10 times. If the vibration amplitude decreases to 0.1 mm, it indicates stability. The cause of this mapping and solving process is transient data input, and the consequence is to provide a quantitative basis for stability.

[0097] Step S440: Perform component decomposition on the stress tensor in the stability simulation results to determine the stress distribution state.

[0098] Component decomposition is performed on the stress tensor in the stability simulation results to determine the stress distribution. The stress tensor is a second-order tensor, including normal stress and shear stress. Component decomposition separates it into principal stress directions. Specifically, in the inspection of automotive chassis parts, tensors such as σ are extracted from the results. xx =50MPa、σ xy =20MPa, the principal stresses 50MPa and -10MPa are obtained through eigenvalue decomposition, and the distribution state such as the tensile dominant zone is determined. This decomposition helps to reveal the stress concentration points.

[0099] Step S450: Match the displacement vector and acceleration signal using the stress distribution state, and obtain the time-domain response data through spectrum transformation.

[0100] The time-domain response data is obtained using the following formula: (26) In formula (26), Represents time-domain response data. Indicates the spectrum after matching. Represents the signal spectrum. The inverse spectral transform is represented by the control logic of formula (26). This logic achieves precise reconstruction of the time-domain response through spectral matching and inverse transform, fusing the stress distribution state with the displacement / acceleration signal in the frequency domain, thus overcoming the limitations of traditional time-domain analysis that relies on a single signal. Compared to the direct time-domain integration method, this approach more accurately captures the dynamic response of the structure under complex stress states, significantly improving the accuracy of dynamic performance evaluation and fatigue life prediction, and providing core technical support for structural optimization and reliability assessment in intelligent manufacturing.

[0101] By matching the stress distribution state with the displacement vector and acceleration signal, time-domain response data is obtained through spectral transformation, such as Fourier transform, which converts the time-domain signal into the frequency domain. Specifically, in precision machine tool spindle operations, the stress distribution state is matched with the displacement vector (0.2, 0.1, 0) mm and the acceleration signal, and then transformed to obtain a spectral peak of 2 m / s² at 10 Hz. The time-domain response data reflects the vibration mode.

[0102] Step S460: Construct a performance evaluation matrix using stress distribution state and time-domain response data to achieve quantitative evaluation of performance stability.

[0103] The performance evaluation matrix is ​​constructed using the following formula: (27) In formula (27), Represents the performance evaluation matrix. Indicates the stress distribution state. Represents time-domain response data. The formula (27) represents element-wise multiplication, which constructs a performance evaluation matrix through element-wise multiplication. The control logic of formula (27) achieves the coupling of stress state and time-domain response through element-wise multiplication, constructing a performance evaluation matrix containing spatiotemporal information, breaking through the limitations of traditional static or single time-domain evaluation. Compared with traditional single-index evaluation methods, it can more comprehensively and accurately characterize the performance stability of the structure under complex working conditions, significantly improving the depth of performance evaluation and the pertinence of subsequent optimization design, and providing core technical support for structural optimization and reliability evaluation in intelligent manufacturing.

[0104] A performance evaluation matrix is ​​constructed by combining the stress distribution state with the time-domain response data to achieve a quantitative assessment of performance stability. The performance evaluation matrix is ​​a multi-dimensional array that integrates the state and data for scoring. Specifically, in the inspection of nuclear power equipment pipelines, the matrix is ​​constructed such that rows correspond to stress components and columns correspond to spectral values. An average score of 85 points indicates high stability. This construction can guide maintenance strategies and improve overall reliability in business operations.

[0105] Furthermore, the image recognition-based structural optimization design and debugging method proposed in this embodiment includes step S500: Step S510: Extract high-order eigenvectors from the performance evaluation matrix and map them to the multidimensional design space to determine the geometric topology variables to be corrected.

[0106] The higher-order feature vector is extracted using the following formula: (28) In formula (28), Represents a higher-order eigenvector. Indicates higher order from 2 to , The performance evaluation matrix represents the first... Singularity, This represents the corresponding left singular vector. Let represent the rank of the matrix. This formula is used to extract higher-order features from the performance evaluation matrix. The control logic of formula (28) achieves accurate extraction of higher-order features from the performance evaluation matrix through singular value decomposition and summation of higher-order terms. It focuses on higher-order patterns that reflect local performance fluctuations and defects, breaking through the limitation of traditional principal component analysis (PCA) which only focuses on the overall trend. Compared with traditional experience-driven optimization methods, it can automatically discover local defects and optimization directions from the data, significantly improving the intelligence and accuracy of structural optimization design, and providing core technical support for structural optimization and reliability assessment in intelligent manufacturing.

[0107] The geometric topological variables to be corrected are determined using the following formula: (29) In formula (29), Represents geometric topological variables. Represents a single geometric topological variable. Represents the performance function. Indicates the design space point, The preset threshold is indicated. The control logic of formula (29) achieves precise positioning of the geometric topology variables to be corrected through sensitivity analysis and threshold screening, narrowing the optimization problem from "full-space search" to "key variable correction", breaking through the limitations of traditional experience-driven optimization. Compared with the traditional blind trial-and-error optimization method, it can accurately locate the optimization target, significantly improving the efficiency and success rate of structural optimization design, and providing core technical support for structural optimization and reliability assessment in intelligent manufacturing.

[0108] Higher-order eigenvectors are extracted from the performance evaluation matrix and mapped to a multidimensional design space to determine the geometric topological variables to be corrected. Here, the performance evaluation matrix is ​​a multidimensional array that integrates stress and time-domain response data. The higher-order eigenvectors are abstract representations extracted from the matrix through principal component analysis. The mapping process involves projecting these vectors onto the coordinate system of the design space. For example, in the optimization of aero-engine blades, the performance evaluation matrix is ​​first decomposed into singular values ​​to extract the first three higher-order vectors, such as the features representing the main vibration modes. Then, these vectors are mapped to a three-dimensional design space to identify the blade curvature and thickness as geometric topological variables to be corrected. The reason for this extraction and mapping is that the matrix has been obtained from previous simulations. The consequence is that the design elements that need to be optimized are accurately located, which can guide subsequent structural adjustments in business operations.

[0109] Step S520: Perform gradient descent iterations on the geometric topology variables within the preset constraint range to obtain the size evolution sequence.

[0110] The size evolution value is obtained using the following formula: (30) In formula (30), Indicates the first Each size evolution value, Represents geometric topological variables. Represents the size mapping function. Indicates the preset constraint range. The formula (30) represents the number of iterations and defines the dimensional evolution sequence obtained through iteration. The control logic of formula (30) achieves the controllable evolution of the geometric topological variable dimensions through gradient descent iteration and constraint projection, transforming the optimization process into a traceable dimensional sequence and overcoming the limitations of traditional one-time optimization. Compared to traditional trial-and-error methods, it can efficiently find the optimal dimensional scheme under strict engineering constraints, significantly improving the efficiency and reliability of structural optimization design, and providing core technical support for structural optimization and reliability assessment in intelligent manufacturing.

[0111] Gradient descent iteration is performed on the geometric topology variables within a preset constraint range to obtain the size evolution sequence. Gradient descent iteration is an optimization algorithm that gradually adjusts the variable values ​​by calculating the gradient of the objective function. The constraint range is such that the thickness must not be less than 2 mm. In the design of bridge steel beams, the variable is assumed to be the beam cross-section width, and the preset constraint is that the width is between 100 and 200 mm. The iteration process starts from the initial value of 150 mm, and the gradient is calculated, such as decreasing by 5 mm in each step. After 5 iterations, the sequence is obtained, such as 150, 145, 140 mm, etc. The size evolution sequence records the trajectory of size change.

[0112] Step S530: Match the corresponding composite component ratio according to the extreme points in the size evolution sequence, and retrieve the candidate reinforcing phase ratio from the material property library.

[0113] The candidate enhancement phase ratio is obtained using the following formula: (31) In formula (31), This represents the vector of extreme points in the size evolution sequence. Represents the material property library, Indicates the first in the library Item scale to size mapping Indicates the first in the library Item composite ratio, Indicates temperature parameter, The formula (31) represents the candidate enhancement ratio, and the optimal candidate enhancement ratio is obtained by averaging all items in the Gaussian kernel weighted average library. The control logic of formula (31) realizes the intelligent matching of geometric extrema and material ratio through Gaussian kernel weighted average, transforming the discrete ratio data in the material property library into a continuous candidate ratio highly coupled with the current geometric features, breaking through the limitation of traditional material selection relying on experience. Compared with the traditional single material matching method, it can find the enhancement ratio that best matches the current structural features in the global material space, significantly improving the efficiency and performance of structure-material co-optimization, and providing core technical support for lightweight design and high-performance material development in intelligent manufacturing.

[0114] The extreme points in the size evolution sequence are matched with the corresponding composite component ratios, and candidate reinforcing phase ratios are obtained by searching the material property library. The extreme points are the maximum or minimum values ​​in the sequence. The matching process involves finding the corresponding ratios. The material property library is a database that stores the properties of various composite materials. In the wind turbine tower business, extreme values ​​such as the minimum thickness of 3 mm are found from the size evolution sequence, and the ratio of carbon fiber to epoxy resin is matched to 60:40. Then, similar ratios are searched in the library to obtain candidates such as a ratio of 70:30 with added glass fiber reinforcing phase. This matching and searching can expand the range of material selection.

[0115] Step S540: If the candidate reinforcing phase ratio meets the preset structural strength criterion threshold, then the optimal configuration combination is obtained by coordinating the optimization of geometric topology variables and composite component ratios through a multi-objective particle swarm optimization algorithm.

[0116] The optimal configuration combination can be obtained using the following formula: (32) In formula (32), and This represents the optimal geometric topological variables and the optimal proportion of composite components. Represents geometric topological variables. Indicates the proportion of composite components. and This represents a multi-objective function, and the formula applies the geometric topological variables using a multi-objective particle swarm optimization algorithm. Ratio of composite components The optimal configuration combination is obtained through collaborative optimization. The control logic of formula (32) realizes the collaborative optimization of geometric topology and material ratio through multi-objective particle swarm optimization, integrating structural optimization and material optimization, and breaking through the limitations of traditional serial optimization. Compared with the traditional single-objective optimization method, it can find the best balance point among multiple performance indicators, significantly improving the overall performance and economy of the structure-material system, and providing core technical support for lightweight design and high-performance material development in intelligent manufacturing.

[0117] If the candidate reinforcing phase ratio meets the preset structural strength criterion, the geometric topology variables and the ratio of the composite components are collaboratively optimized using a multi-objective particle swarm optimization algorithm to obtain the optimal configuration combination. The multi-objective particle swarm optimization algorithm simulates the foraging behavior of bird flocks and optimizes multiple objectives such as strength and weight by updating particle positions. The structural strength criterion is that the tensile strength exceeds 500 MPa. In the optimization of automotive chassis parts, it checks whether the ratio meets the criterion. If so, the multi-objective particle swarm optimization algorithm initializes 20 particles and iteratively updates variables such as thickness and ratio to obtain the optimal combination such as a thickness of 4 mm and a ratio of 65:35.

[0118] Step S550: Perform reconstruction processing on the original design model through the optimal configuration combination to determine the parameter optimization values ​​for size and material adjustments.

[0119] The optimized parameter values ​​for size and material adjustments are determined using the following formula: (33) In formula (33), This represents the optimized parameter value. Represents the original design model. This represents the optimal configuration combination. Indicates the parameters to be optimized. The formula (33) represents the reconstruction loss function. It determines the parameter optimization value by minimizing the loss function through the reconstruction of the original design model using the optimal configuration combination. The control logic of formula (33) is to achieve the precise transformation of the optimal configuration combination into engineering parameters through loss minimization reconstruction. It transforms the abstract optimization result into an executable size and material adjustment scheme, breaking through the bottleneck of traditional optimization results being difficult to engineer. Compared with the traditional manual parameter adjustment method, it can automatically and accurately determine the optimal parameters, significantly improving the efficiency and engineering feasibility of structural optimization design, and providing core technical support for product development and performance verification in intelligent manufacturing.

[0120] The original design model is reconstructed by the optimal configuration combination to determine the optimized values ​​of parameters for size and material adjustments. The reconstruction process involves updating the geometric and material parameters of the model. In the precision machine tool spindle business, the combination is applied to the CAD model to adjust the size, such as the diameter from 50 mm to 52 mm, and the material from steel to composite ratio. The output optimized value is such as the new elastic modulus of 210 GPa. This reconstruction enables design iteration in business and improves durability.

[0121] Preferably, in the image recognition-based structural optimization design and debugging method proposed in this embodiment, step S600 includes: Step S610: Construct a geometric solid model in the 3D modeling engine based on the parameter optimization values. The geometric solid model is then assigned the corresponding physical property material.

[0122] Construct a geometric solid model using the following formula: (34) In formula (34), A geometric solid model, generated by a 3D modeling engine, contains complete geometric and topological information and forms the basis for subsequent simulation and manufacturing. This refers to the 3D modeling engine, the core engine that performs geometric modeling operations, such as a CAD kernel (Parasolid, ACIS) or parametric modeling tools. Represents the vertex set, the set of coordinates of all vertices that make up the geometric model, whose coordinates are determined by the construction parameters. (i.e., parameter optimization value) )drive. A face set is the collection of all faces (patches) that make up a geometric model, whose topological relationships are determined by the vertex set and construction parameters. A joint decision. This refers to the construction parameters, i.e., the optimized parameter values ​​obtained in step S550. This includes geometric dimensions, positions, etc., used to drive the generation of vertices and faces. This formula is used to generate geometric solid models based on parameter optimization values ​​in a 3D modeling engine. The control logic of formula (34) is achieved through a parameter-driven 3D modeling engine, realizing the automatic generation of optimized parameters into geometric solids, opening up the digital link from performance optimization to geometric design, and breaking through the limitations of traditional CAD modeling that relies on manual operation. Compared with traditional manual modeling methods, it can automatically and accurately generate geometric models that meet the optimal performance requirements, significantly improving the efficiency and consistency of product development, and providing core technical support for digital twins and concurrent engineering in intelligent manufacturing.

[0123] The optimized parameters are input into the 3D modeling engine, a software system integrating a geometric modeling kernel and parametric driving functions. This engine can automatically generate 3D solids based on the input size and shape parameters. For example, in the optimized design of a robot arm, the optimized parameters include the bore diameter, thickness, and chamfer radius of the connecting flange. Upon receiving these values, the 3D modeling engine invokes its internal Boolean operations and surface generation modules. Starting from a basic cylinder and cuboid, it precisely constructs a complex flange solid model with bolt holes and reinforcing ribs through union, difference operations, and fillet transitions. After construction, the structural optimization design and debugging system uses a pre-set material library and, based on the composite material mix specified in the optimization scheme (e.g., carbon fiber reinforced nylon), assigns corresponding physical properties such as density and elastic modulus to the geometric solid model, thus forming a digital entity that possesses both geometric shape and material properties. Specifically, multiple such geometric solid models are imported into a virtual simulation environment, a digital platform capable of simulating physical space and the relationships between components.

[0124] Step S620: Based on the geometric entity model, perform spatial pose alignment in the virtual simulation environment according to the preset topological connection relationship to generate a virtual assembly model.

[0125] The virtual assembly model is generated using the following formula: (35) In formula (35), The virtual assembly model represents a complete digital prototype assembled from multiple geometric solid models according to design requirements, and is the core object for subsequent simulation and verification. This refers to topological union operations performed in a virtual simulation environment, including spatial pose alignment, constraint definition, and Boolean operations, which combine independent geometric entities into a complete assembly. Indicates the aligned first A geometric solid model, The total number of geometric entities is represented. The control logic of formula (35) realizes the automated generation of virtual assembly through topological joint operations, integrating the pose alignment of geometric entities with the definition of topological constraints, breaking through the limitations of traditional virtual assembly that relies on manual constraints. Compared with the traditional manual assembly method, it can automatically and accurately generate digital prototypes that meet the design intent, significantly improving the efficiency and accuracy of product development, and providing core technical support for digital twins and concurrent engineering in intelligent manufacturing.

[0126] Aligned Geometric solid model This can be derived from the following formula: (36) In formula (36), Indicates the aligned first A geometric solid model, This represents a topology-based pose transformation function. Represents the original geometric entity. A geometric solid model. Indicates the first The topological neighborhood of the geometric entity, and the . The formula (36) is used to align the spatial poses of a single geometric entity by aligning all adjacent components and their constraints (such as reference planes, holes, shafts, mating surfaces, etc.) that have an assembly relationship. The control logic of formula (36) is to realize the automatic spatial alignment of a single geometric entity through topology-driven pose transformation, and to convert the assembly constraint information into an executable transformation matrix, breaking through the limitation of traditional virtual assembly relying on manual pose specification. Compared with the traditional manual alignment method, it can automatically and accurately complete the pose adjustment of parts, significantly improving the efficiency and accuracy of virtual assembly, and providing core technical support for digital twins and concurrent engineering in intelligent manufacturing.

[0127] Based on the preset topological connections, i.e., the assembly constraints defined between components, the structural optimization design and debugging system automatically performs spatial pose alignment. Taking the virtual assembly of an industrial gearbox as an example, the topological relationships define the meshing relationships between the input shaft gear, intermediate idler gear, and output shaft gear, as well as the fixed positions of the bearing housings. The alignment algorithm in the simulation environment calculates the optimal position and orientation of each part based on these constraints. For example, by minimizing the distance deviation between the axis centers and ensuring the alignment of the tooth surface contact points, it precisely "places" all parts in the correct positions, thereby generating a complete virtual assembly model that can be used for subsequent analysis.

[0128] Step S630: Mesh the contact surfaces in the virtual assembly model and extract the normal vector deviation of the mating surfaces, and determine whether the normal vector deviation is within the preset tolerance fluctuation range.

[0129] The normal vector deviation is obtained by the following formula: (37) In formula (37), Indicates the deviation of the normal vector. The normal vector of the contact surface is the unit normal vector of the actual contact surface (such as hole, shaft, and mating surface) in the virtual assembly model, which is extracted after mesh generation. The normal vector of the mating surface is represented by the unit normal vector of the ideal mating surface in the design intent, which is obtained from the geometric definition of the CAD model. The control logic of formula (37) realizes the automated quantitative evaluation of assembly accuracy through the calculation of normal vector deviation, transforming the geometric information of the virtual assembly model into verifiable tolerance indicators, breaking through the limitations of traditional assembly accuracy evaluation relying on physical prototypes and manual inspection. Compared with traditional experience-based judgment methods, it can quickly and accurately diagnose assembly errors, significantly improving the efficiency and quality of product development, and providing core technical support for digital twins and concurrent engineering in intelligent manufacturing.

[0130] For the generated virtual assembly model, a detailed analysis of its key contact surfaces is required. Meshing is the process of discretizing continuous contact surfaces into a large number of tiny units to facilitate numerical calculations. For example, when analyzing the fit between the wing and fuselage of an aircraft, the mating surfaces of these two components are meshed into triangular grids. After meshing, the structural optimization design and debugging system extracts the normal vector at each grid node, i.e., the direction vector perpendicular to the surface at that point, and calculates the angular deviation between the normal vectors of corresponding nodes on the two mating surfaces. This angular deviation reflects the degree of fit between the contact surfaces.

[0131] Step S640: If the normal vector deviation is within the preset tolerance fluctuation range, extract the centroid offset between each component and calculate the assembly accuracy.

[0132] The centroid offset between each component is extracted using the following formula: (38) In formula (38), Representation Component With components The centroid offset between them Representation Component The center of mass, Representation Component The centroid is calculated from its geometry and material density. The control logic of formula (38) realizes the global resolution of assembly accuracy through the calculation of centroid offset, transforming the component-level centroid position difference into a quantifiable assembly accuracy index, breaking through the limitation of traditional assembly accuracy assessment that only focuses on local surfaces. Compared with traditional experience-based judgment methods, it can quickly and accurately assess the overall assembly accuracy, significantly improving the efficiency and quality of product development, and providing core technical support for digital twins and concurrent engineering in intelligent manufacturing.

[0133] The assembly accuracy is calculated using the following formula: (39) In formula (39), Indicates assembly precision. Indicates the number of component pairs. Indicates the first The centroid offset of the component. The control logic of formula (39) realizes the automated global evaluation of assembly accuracy through the average centroid offset, transforming local assembly errors into a unified accuracy index, breaking through the limitation of traditional assembly accuracy evaluation relying on human experience. Compared with the traditional single-point detection method, it can quickly and comprehensively evaluate assembly accuracy, significantly improving the efficiency and quality of product development, and providing core technical support for digital twins and concurrent engineering in intelligent manufacturing.

[0134] The judgment process involves comparing all calculated deviation values ​​with a preset tolerance fluctuation range, such as ±0.5 degrees. If all deviations fall within this range, it indicates that the contact surfaces are aligned well, and the next step, overall accuracy assessment, can proceed. Once the contact surfaces are aligned correctly, the macroscopic accuracy of the entire assembly needs to be evaluated. At this point, the structural optimization design and debugging system extracts the centroid coordinates of each component after assembly and compares them with their theoretical centroid coordinates under ideal design conditions to calculate the centroid offset. Calculating assembly accuracy involves using a specific algorithm, such as root mean square error calculation, to synthesize the offsets of all components and derive a quantitative index characterizing the overall assembly deviation. This process is crucial in the assembly of lens assemblies for precision optical instruments.

[0135] S650: Determine whether the assembly accuracy is less than or equal to the preset error threshold. If the assembly accuracy is less than or equal to the preset error threshold, output the final optimized solution. The verification output is obtained using the following formula: (40) In formula (40), This indicates the verification output. Indicates assembly precision. This represents the preset error threshold, the maximum allowable assembly error required by the design, and is the criterion for determining whether the accuracy is qualified. This represents the final optimized solution. This formula indicates rejection; it is used when... Less than or equal to Time output Otherwise, it is rejected. The control logic of formula (40) realizes the automated decision-making of assembly accuracy verification through threshold comparison, and constructs a precise closed loop from virtual optimization to physical realization, breaking through the limitation of traditional product development relying on repeated iterations of physical prototypes. Compared with the traditional experience-driven verification method, it can quickly and objectively determine the feasibility of the optimization scheme, significantly improve the efficiency and success rate of product development, and provide core technical support for digital twins and concurrent engineering in intelligent manufacturing.

[0136] The structural optimization design and debugging system calculates the actual centroid position of each lens after assembly and compares it with the theoretical position of the optical axis. If the calculated assembly accuracy index, such as the comprehensive eccentricity error, is less than the preset error threshold of 5 micrometers, and meets the design standard, it proves that the model built and assembled based on the parameter optimization value meets the requirements. At this time, a complete optimization scheme containing all final dimensions, materials and assembly relationships can be output to guide actual production.

[0137] Please see Figure 2This embodiment also provides an image recognition-based structural optimization design and debugging system for executing the above-described image recognition-based structural optimization design and debugging method. The system includes a preliminary defect candidate region acquisition module 10, a deviation type and severity determination module 20, a defect group distribution acquisition module 30, a stability simulation result acquisition module 40, a parameter optimization value determination module 50, and a final optimization scheme output module 60. The preliminary defect candidate region acquisition module 10 acquires component image data of the mechanical device and extracts contour features from the component image data using an image processing algorithm to obtain preliminary defect candidate regions. The component image data includes multiple component structures, and the preliminary defect candidate regions correspond to potential abnormal parts. The deviation type and severity determination module 20 uses a neural network model to classify dimensional deviations within the preliminary defect candidate regions and determine the deviation type and severity. The deviation type involves geometric and material properties. The severity is graded based on the degree of deviation; the defect group distribution acquisition module 30 is used to obtain assembly accuracy influence parameters from the deviation type and severity, and group similar defects by clustering to obtain the defect group distribution, where the defect group distribution represents the defect concentration pattern; the stability simulation result acquisition module 40 is used to determine whether the distribution density of the defect group distribution exceeds the threshold. If it exceeds the threshold, the performance stability is simulated by numerical simulation method to obtain the stability simulation result, where the stability simulation result includes stress and time domain response data; the parameter optimization value determination module 50 is used to obtain the basis for generating optimization schemes from the stability simulation results, and adjust the design parameters through optimization algorithm to determine the parameter optimization value, where the parameter optimization value is adjusted for size and material; the final optimization scheme output module 60 is used to generate a virtual assembly model based on the parameter optimization value, and determine whether the assembly accuracy in the virtual assembly model meets the standard. If it does, the final optimization scheme is output.

[0138] The image recognition-based structural optimization design and debugging method and system provided in this embodiment, compared with existing technologies, first extracts contour features from component images and locates preliminary defect candidate regions. Then, a neural network is used to perform fine classification and severity assessment of dimensional deviations within the regions, integrates geometric and material property deviation information, and analyzes defect group distribution patterns through clustering. When the distribution density exceeds the limit, numerical simulation is triggered to predict performance stability. Based on the stress and time-domain response data obtained from the simulation, optimization algorithms are used to inversely adjust design and material parameters, ultimately generating a virtual assembly model and optimization scheme that meets assembly accuracy standards. This embodiment realizes a closed-loop process from intelligent defect identification and group influence analysis to stability-driven parameter optimization, significantly improving the efficiency of quality control and design optimization of complex mechanical devices.

[0139] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention. Clearly, those skilled in the art can make various alterations and modifications to the invention without departing from its spirit and scope. Thus, if these modifications and modifications of the invention fall within the scope of the claims and their equivalents, the invention is also intended to include these modifications and modifications.

Claims

1. A structural optimization design and debugging method based on image recognition, characterized in that, Includes the following steps: S100. Acquire component image data of the mechanical device, extract contour features from the component image data using an image processing algorithm, and obtain preliminary defect candidate regions, wherein the component image data includes multiple component structures, and the preliminary defect candidate regions correspond to potential abnormal parts. S200. A neural network model is used to classify the dimensional deviations within the preliminary defect candidate region to determine the deviation type and severity, wherein the deviation type involves geometric and material properties, and the severity is based on a deviation degree classification. S300. Obtain assembly accuracy influence parameters from the deviation type and severity, group similar defects by clustering method to obtain defect group distribution, wherein the defect group distribution represents defect concentration pattern; S400. For the distribution of the defect group, determine whether the distribution density exceeds the threshold. If it does, use a numerical simulation method to simulate the performance stability and obtain the stability simulation results. The stability simulation results include stress and time-domain response data. S500. Obtain the basis for generating the optimization scheme from the stability simulation results, adjust the design parameters through the optimization algorithm, and determine the parameter optimization values, wherein the parameter optimization values ​​are adjusted for size and material. S600. Based on the optimized parameter values, generate a virtual assembly model, determine whether the assembly accuracy in the virtual assembly model meets the standard, and if it does, output the final optimized solution.

2. The structural optimization design and debugging method based on image recognition according to claim 1, characterized in that, Step S100 includes: S110. Acquire component image data of the mechanical device, and perform median filtering on the component image data to obtain smooth image data; S120. Extract the contour features of the component structure from the smoothed image data using an edge detection operator; S130. Based on the contour features, perform a closed-loop search within a preset coordinate system to obtain a set of closed contours; S140. If the geometric parameters of a specific region in the closed contour set deviate from those of the standard component template by more than a preset threshold, then the specific region is determined as a preliminary defect candidate region.

3. The structural optimization design and debugging method based on image recognition according to claim 1, characterized in that, Step S200 includes: S210. Extract multi-dimensional spatial geometric features and gray-level co-occurrence matrix texture features from the preliminary defect candidate region; S220. Input the multidimensional spatial geometric features and gray-level co-occurrence matrix texture features into the deep residual network model to obtain the geometric property probability distribution and the material property probability distribution; S230. Determine the deviation type of the preliminary defect candidate region based on the probability distribution of the geometric properties and the probability distribution of the material properties; S240. If the deviation type is a geometric attribute deviation, then calculate the numerical deviation. S250. If the deviation type belongs to material property deviation, then analyze the texture heterogeneity. S260. Determine the severity based on the numerical deviation or the texture heterogeneity. S270. The classification of dimensional deviations within the preliminary defect candidate area is completed based on the deviation type and the severity.

4. The structural optimization design and debugging method based on image recognition according to claim 1, characterized in that, Step S300 includes: S310. Based on the type and severity of the deviation, retrieve the assembly clearance and geometric tolerance from the preset mapping matrix to determine the parameters affecting assembly accuracy. S320. Construct a high-dimensional feature vector using the assembly accuracy influencing parameters, and divide the preliminary defect candidate region into spatial attributes using a density clustering algorithm to obtain a set of defects with similar geometric features and material properties. S330. Calculate the fluctuation range of contact stress and material hardness in the local area for the defect set, and determine the defect group by measuring the similarity between the defect sets through Euclidean distance; S340. Extract the statistical features of surface roughness and load distribution within the defect group, and calculate the spatial distribution density of defects on the workpiece surface using the kernel density estimation method to obtain the defect group distribution. S350. Identify abnormal clustering areas of thermal expansion and frictional resistance based on the distribution of the defect groups, and determine the defect concentration pattern by analyzing the coupling relationship between fit tolerance and structural stiffness.

5. The structural optimization design and debugging method based on image recognition according to claim 4, characterized in that, Step S400 includes: S410. Extract local mesh coordinates and intrinsic material properties from the defect group distribution to determine the distribution density; S420. If the distribution density exceeds the preset critical evolution threshold, the finite element analysis model is invoked to obtain transient dynamic characteristics. S430. Based on the transient dynamic characteristic mapping stiffness matrix and damping matrix, the stability simulation results are obtained by iteratively solving the nonlinear motion equations. S440. Perform component decomposition on the stress tensor in the stability simulation results to determine the stress distribution state. S450. Using the stress distribution state to match the displacement vector and acceleration signal, time-domain response data is obtained through spectrum transformation processing; S460. A performance evaluation matrix is ​​constructed using the stress distribution state and the time-domain response data to achieve a quantitative assessment of performance stability.

6. The structural optimization design and debugging method based on image recognition according to claim 1, characterized in that, Step S500 includes: S510. Extract high-order eigenvectors from the performance evaluation matrix and map them to the multi-dimensional design space to determine the geometric topology variables to be corrected. S520. Perform gradient descent iterations on the geometric topology variables within a preset constraint range to obtain the size evolution sequence; S530. Match the corresponding composite component ratio according to the extreme points in the size evolution sequence, and retrieve the candidate reinforcing phase ratio from the material property library; S540. If the candidate reinforcing phase ratio meets the preset structural strength criterion threshold, the optimal configuration combination is obtained by coordinating the optimization of the geometric topology variables and the ratio of the composite components through a multi-objective particle swarm optimization algorithm. S550. Perform a reconstruction process on the original design model using the optimal configuration combination to determine the optimized parameter values ​​for size and material adjustments.

7. The structural optimization design and debugging method based on image recognition according to claim 1, characterized in that, Step S600 includes: S610. Construct a geometric solid model in the 3D modeling engine according to the parameter optimization value, and assign the corresponding physical property material to the geometric solid model. Construct a geometric solid model using the following formula: ； in, Represents a geometric solid model. This refers to a 3D modeling engine. Represents a vertex set. Represents a face set. Indicates the construction parameters; S620. Based on the geometric entity model, perform spatial pose alignment in the virtual simulation environment according to the preset topological connection relationship to generate a virtual assembly model. S630. Mesh out the contact surfaces in the virtual assembly model and extract the normal vector deviation of the mating surfaces, and determine whether the normal vector deviation is within the preset tolerance fluctuation range. S640. If the normal vector deviation is within the preset tolerance fluctuation range, then extract the centroid offset between each component and calculate the assembly accuracy. S650. Determine whether the assembly accuracy is less than or equal to a preset error threshold. If the assembly accuracy is less than or equal to the preset error threshold, output the final optimization scheme.

8. The structural optimization design and debugging method based on image recognition according to claim 7, characterized in that, In step S620, the virtual assembly model is generated using the following formula: ； in, Represents a virtual assembly model. This represents a topological union operation in a virtual simulation environment. Indicates the aligned first A geometric solid model, Indicates the total number of geometric entities; Aligned Geometric solid model This can be derived from the following formula: ； in, Indicates the aligned first A geometric solid model, This represents a topology-based pose transformation function. Represents the original geometric entity. A geometric solid model, Indicates the first The topological neighborhood of a geometric entity.

9. The structural optimization design and debugging method based on image recognition according to claim 8, characterized in that, In step S630, the normal vector deviation is obtained using the following formula: ； in, Indicates the deviation of the normal vector. This represents the normal vector of the contact surface. This represents the normal vector of the mating surface.

10. A structural optimization design and debugging system based on image recognition, used to execute the structural optimization design and debugging method based on image recognition as described in any one of claims 1 to 9, characterized in that, include: The preliminary defect candidate region acquisition module (10) is used to acquire component image data of mechanical device, extract contour features from the component image data through image processing algorithm, and obtain preliminary defect candidate regions, wherein the component image data includes multiple component structures, and the preliminary defect candidate regions correspond to potential abnormal parts. A deviation type and severity determination module (20) is used to classify dimensional deviations in the preliminary defect candidate region using a neural network model, and determine the deviation type and severity, wherein the deviation type involves geometric and material properties, and the severity is based on a deviation degree grading. The defect group distribution acquisition module (30) is used to acquire assembly accuracy influence parameters from the deviation type and severity, group similar defects by clustering method, and obtain defect group distribution, wherein the defect group distribution represents defect concentration pattern; The stability simulation result acquisition module (40) is used to determine whether the distribution density of the defect group exceeds the threshold. If it does, the performance stability is simulated by numerical simulation method to obtain the stability simulation result, wherein the stability simulation result includes stress and time domain response data. The parameter optimization value determination module (50) is used to obtain the basis for generating the optimization scheme from the stability simulation results, adjust the design parameters through the optimization algorithm, and determine the parameter optimization value, wherein the parameter optimization value is adjusted for size and material. The final optimization scheme output module (60) is used to generate a virtual assembly model based on the parameter optimization value, determine whether the assembly accuracy in the virtual assembly model meets the standard, and output the final optimization scheme if it does.

Images