A product quality whole life cycle management system based on digital twinning

Abstract

This invention relates to the fields of digital twin and computer graphics processing technology, specifically a product quality lifecycle management system based on digital twins. The system includes: a multi-dimensional data mapping module (acquiring a 3D mesh model and time-series operating condition data to construct a spatiotemporal correlation feature set); a physical perception evolution module (using a physical loss evolution algorithm to calculate the cumulative damage potential energy distribution of each region on the surface of the 3D mesh model); a generative texture synthesis module (driving a texture generation network to synthesize visual texture maps containing aging features based on the cumulative damage potential energy distribution); a topological deformation coupling module (nonlinearly offsetting vertices based on depth permutation information to generate a deformed 3D digital twin model); and a semantic consistency verification module. This invention transforms invisible historical operating condition data into a visual 3D aging model that conforms to physical laws, improving the intuitiveness and predictive accuracy of lifecycle management.

Description

Technical Field

[0001] This invention relates to the fields of digital twin and computer graphics processing technology, specifically to a product quality lifecycle management system based on digital twins. Background Technology

[0002] In the entire lifecycle management of industrial products, digital twin technology, by mapping real-time data and constructing virtual models to reflect the physical state, is a key means to ensure the safe operation of equipment; and the application effect of digital twins largely depends on the degree to which the virtual model restores the physical state of the entity. In traditional methods, digital twin models mainly focus on displaying numerical charts or rendering static geometric models. They can only provide feedback on discrete sensor readings such as temperature and rotation speed, and cannot intuitively present the appearance aging and deformation caused by long-term physical wear and tear of the product. This results in a separation between data and visual representation, making it difficult to meet the need for intuitively locating quality risks. Summary of the Invention

[0003] To address the aforementioned technical problems, this invention provides a product quality lifecycle management system based on digital twins. Specifically, the technical solution of this invention includes: Multidimensional data mapping module: used to acquire the three-dimensional mesh model data and time-series operating condition flow data of the target object, and map the time-series operating condition flow data to the surface vertices of the three-dimensional mesh model to construct a spatiotemporal correlation feature set; The physical perception evolution module is used to calculate the cumulative damage potential energy distribution of each region on the surface of a 3D mesh model based on a spatiotemporal correlation feature set and a physical loss evolution algorithm. Generative texture synthesis module: Based on the cumulative damage potential energy distribution, it drives the texture generation network to synthesize visual texture maps containing aging features in the corresponding texture coordinate space of the 3D mesh model. Topological deformation coupling module: Based on the depth displacement information in the visual texture map, it performs non-linear offset on the vertex geometric position of the 3D mesh model to generate a deformed 3D digital twin model; Semantic consistency verification module: used to calculate the semantic consistency score between visual texture maps and time-series operational data, and generate the final visualization rendering result based on the semantic consistency score.

[0004] Preferably, the modules are implemented using the following method: S1. Collect the 3D mesh model data of the target object and the time-series operating condition data throughout its entire life cycle. After preprocessing the data, map the time-series operating condition data to the surface vertices of the 3D mesh model to generate a spatiotemporal correlation feature set. S2. Input the spatiotemporal correlation feature set into the physical loss evolution algorithm to simulate the physical state changes of each region on the surface of the target object over time, and output the cumulative damage potential energy distribution matrix. S3. The cumulative damage potential energy distribution matrix is ​​used as a conditional constraint input to the texture generation network. Pixel-level prediction is performed in the UV texture space of the 3D mesh model to generate a visual texture map containing diffuse, normal, and displacement channels. S4. Analyze the displacement channel data in the visual texture map, calculate the normal offset of each vertex of the 3D mesh model, drive the 3D mesh model to undergo geometric deformation, and generate the deformed 3D digital twin model. S5. Extract the visual features of the deformed 3D digital twin model, perform logical matching with the time-series operating condition data, and calculate the semantic consistency score. S6. Based on semantic consistency scoring, the deformed 3D digital twin model is screened or its parameters are corrected, and the final visualization rendering result is output.

[0005] Preferably, S1 specifically includes: S11. Obtain the initial 3D mesh model of the target object, extract the spatial coordinates and normal vectors of all vertices in the 3D mesh model, and construct the geometric topology matrix; S12. Collect environmental load data and operating status data of the target object within a historical time window through a sensor array, perform time alignment and normalization processing on the environmental load data and operating status data, and construct standardized time-series operating condition stream data. S13. Using a spatial interpolation algorithm, the numerical intensity in the time-series operating condition stream data is mapped to the corresponding vertex coordinates in the geometric topology matrix; S14. For each vertex, establish a multi-dimensional feature vector containing spatial location, normal vector, and historical working condition sequence. Combine the multi-dimensional feature vectors of all vertices to output a spatiotemporal correlation feature set.

[0006] Preferably, S2 specifically includes: S21. Based on the spatiotemporal correlation feature set, construct a physical loss state variable for each vertex, and set the state decay coefficient and cumulative threshold of the physical loss state variable. S22. Using the stress transformation function, the historical working condition sequence of each vertex is converted into an instantaneous stress value, and the instantaneous stress value is input into the physical loss state variable for time-series numerical integration. S23. Introduce a preset hysteresis correction coefficient to dynamically adjust the integral result of the physical loss state variable in order to compensate for nonlinear superposition error and obtain the current damage potential energy value of each vertex. S24. Perform spatial smoothing on the current damage potential energy values ​​of all vertices according to the topological adjacency relationship of the three-dimensional mesh model to generate a cumulative damage potential energy distribution matrix that reflects the degree of surface damage.

[0007] Preferably, S3 specifically includes: S31. Construct a texture generation model based on conditional generative adversarial networks, and map the cumulative damage potential distribution matrix into a two-dimensional feature map as a guiding condition for the texture generation model. S32. Introduce a physical constraint loss function into the texture generation model. Based on the guiding conditions, predict the color value, roughness value and height value of the aged texture pixel by pixel on the UV unfolded map of the 3D mesh model. S33. Perform color space conversion on the predicted color values ​​to generate a diffuse texture map; perform gradient calculation on the predicted roughness values ​​to generate a normal texture map; perform grayscale quantization on the predicted height values ​​to generate a displacement texture map. S34. Merge the diffuse texture map, normal texture map and displacement texture map into channels to output a multi-channel visual texture map.

[0008] Preferably, S4 specifically includes: S61. Read the grayscale pixel values ​​in the displacement texture map and convert the grayscale pixel values ​​into geometric displacement scalars according to the preset pixel-to-physical depth conversion ratio. S62. Traverse all vertices of the 3D mesh model and index the corresponding geometric displacement scalar in the replacement texture map based on the texture coordinates of each vertex in the UV space. S63. Apply the corresponding geometric displacement scalar along the normal vector direction of each vertex, update the spatial coordinate position of the vertex, and complete the subdivision and reconstruction of the mesh surface. S64. Recalculate the vertex normals and tangent space of the reconstructed 3D mesh model, refresh the lighting and rendering properties of the model, and generate the deformed 3D digital twin model.

[0009] Preferably, S5 specifically includes: S71. Extract visual defect feature regions from the deformed three-dimensional digital twin model, and calculate the geometric area and texture contrast of the visual defect feature regions; S72. Extract the corresponding cumulative load values ​​and environmental exposure duration from the time-series operating condition data to construct the theoretical values ​​of physical damage. S73. Calculate the normalized difference between the geometric area of ​​the visual defect feature region and the theoretical value of the physical damage to obtain the feature matching error; S74. Based on the feature matching error and the texture contrast, a weighted summation is performed using a preset weighting factor to obtain the semantic consistency score Q. S75. Compare the semantic consistency score Q with the preset qualification threshold T to generate a score comparison result.

[0010] Preferably, S6 specifically includes: S81. Obtain the semantic consistency score Q and the preset qualification threshold T; S82. In response to the semantic consistency score Q being greater than or equal to the qualified threshold T, determine that the currently generated visual effect conforms to physical laws, mark the deformed three-dimensional digital twin model as valid, and output high-precision rendering results. S83. In response to the semantic consistency score Q being less than the qualified threshold T, it is determined that the currently generated visual effect has semantic deviation. The difference between the score Q and the threshold T is calculated, and the weight of the physical constraint loss function in the texture generation model is adjusted in reverse based on the difference. The regeneration process of the texture generation network is triggered until the semantic consistency score Q meets the condition of being greater than or equal to the qualified threshold T, and the final visualization rendering result is output.

[0011] Compared with the prior art, the present invention has the following beneficial effects: 1. This system, through the collaborative work of a multi-dimensional data mapping module and a physical perception evolution module, achieves precise anchoring of time-series operating condition data to a three-dimensional spatial model, solving the technical problem that traditional digital twins can only display numerical charts and cannot intuitively present physical losses. By introducing a time integration algorithm including dissipation terms and a hysteresis effect correction mechanism, it can transform historical operating conditions into a cumulative damage potential energy distribution based on the material constitutive model. This makes the aging degree of the model surface no longer a simple image filter superposition, but a numerical simulation result based on physical laws. Through this physical-driven evolutionary calculation, invisible fatigue accumulation and stress state can be transformed into a visualized potential energy field, providing engineers with a scientific basis for intuitively locating potential quality risk areas and significantly improving the intuitiveness of life cycle management. 2. This system achieves dual-linked evolution of visual texture and geometric structure through the deep integration of generative texture synthesis module and topological deformation coupling module; by using the cumulative damage potential energy distribution as the guiding condition of deep learning network and introducing physical constraint loss function in texture generation process, it ensures that the generated aging features such as cracks and rust conform to the law of energy conservation; by parsing the permutation information in visual texture map and performing nonlinear offset on mesh vertices, the digital twin model not only changes in color, but also realistically reflects volume changes such as pits and peeling in geometric shape, thus ensuring that the self-shadow and occlusion relationship generated by light and shadow at the concave and convex points are completely in line with physical reality, greatly enhancing the realism of the virtual model in restoring the physical state of the entity; 3. This system effectively solves the problems of multi-source heterogeneous data fusion and the inability of sparse sensor data to cover the surface of high-precision models by adopting a normalization strategy that separates numerical values ​​from units and a spatial interpolation algorithm with dynamic radius expansion. By stripping physical units and extracting only pure numerical amplitudes for calculation, it avoids computational anomalies caused by confused dimensional definitions and ensures the mathematical rigor of data processing. By dynamically adjusting the inverse distance weighted interpolation of the search radius, it ensures that effective inference data can be obtained at any position on the grid. It establishes a multi-dimensional feature vector containing spatial position, normal vector and historical working condition sequence for each vertex, providing full-coverage and high-precision data support for subsequent refined physical calculations. 4. This system establishes a complete closed-loop verification system through a semantic consistency verification module and a closed-loop feedback control mechanism, which is driven by data, drives vision, and provides visual feedback data. By extracting the geometric visual features of the deformed model and comparing them with the theoretical values ​​of physical damage derived from working condition data, the system effectively identifies and intercepts visual illusions that violate physical facts using a scoring formula that includes a physical-visual conflict penalty term. By adjusting the physical constraint weights in the texture generation network based on the consistency score, the system achieves adaptive optimization of the generation process, ensuring that the generated visual effects must display details under the constraints of physical laws, thus guaranteeing the objectivity and rigor of the digital twin system in industrial applications. Attached Figure Description

[0012] The present invention will be further explained below with reference to the accompanying drawings and embodiments: Figure 1 This is a structural diagram of the system of the present invention. Detailed Implementation

[0013] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0014] Example 1: Please see Figure 1 A product quality lifecycle management system based on digital twins, comprising: Multidimensional data mapping module: used to acquire the three-dimensional mesh model data and time-series operating condition flow data of the target object, and map the time-series operating condition flow data to the surface vertices of the three-dimensional mesh model to construct a spatiotemporal correlation feature set; The physical perception evolution module is used to calculate the cumulative damage potential energy distribution of each region on the surface of a 3D mesh model based on a spatiotemporal correlation feature set and a physical loss evolution algorithm. Generative texture synthesis module: Based on the cumulative damage potential energy distribution, it drives the texture generation network to synthesize visual texture maps containing aging features in the corresponding texture coordinate space of the 3D mesh model. Topological deformation coupling module: Based on the depth displacement information in the visual texture map, it performs non-linear offset on the vertex geometric position of the 3D mesh model to generate a deformed 3D digital twin model; Semantic consistency verification module: used to calculate the semantic consistency score between visual texture maps and time-series operational data, and generate the final visualization rendering result based on the semantic consistency score.

[0015] This embodiment provides a product quality lifecycle management system based on digital twins. This system aims to solve the technical problem in existing technologies where digital twin models can only display numerical charts and cannot intuitively represent the aging and deformation of a product due to long-term physical wear and tear, resulting in a separation between data and visual representation. The system mainly consists of the following core modules: The multidimensional data mapping module, as a data processing unit to address the heterogeneity of spatiotemporal data, accurately anchors one-dimensional time-series data onto a three-dimensional spatial model. In this embodiment, the module acquires high-precision three-dimensional mesh model data of target objects such as aero-engine blades or precision machine tool spindles, including vertex, facet, and UV coordinates, along with time-series operating condition data throughout the entire lifecycle, including sensor readings such as temperature, vibration, and stress. The module uses a spatial indexing algorithm to map the time-series operating condition data to the surface vertices of the three-dimensional mesh model, constructing a spatiotemporal correlation feature set. This feature set serves as the data foundation for all subsequent physical evolution calculations. The Physical Perception Evolution Module, as a computational engine based on the principles of materials mechanics and damage mechanics, introduces a physical loss evolution algorithm. This algorithm is not a simple image filter, but a numerical simulator based on physical laws. It calculates the cumulative damage potential energy distribution of each region, i.e., each vertex or facet, on the surface of a 3D mesh model based on a spatiotemporal correlated feature set. This distribution reflects the potential physical fatigue degree of each point on the model surface after experiencing historical working conditions, providing a physical basis for texture generation. Specifically, this module transforms multi-dimensional time-series working condition data into a stress tensor field through an embedded material constitutive model, and uses a time integration algorithm with dissipation terms to calculate the damage energy density, thereby accurately quantifying the cumulative physical degradation state of the material surface. The generative texture synthesis module uses deep learning technology to reconstruct visual features. In this embodiment, the module uses the cumulative damage potential energy distribution as a guide map or conditional mask to drive a pre-trained texture generation network. The network synthesizes visual texture maps containing aging features such as microcracks, corrosion, and oil deposits in the UV space of the corresponding texture coordinates of the three-dimensional mesh model. The topological deformation coupling module is used to realize the linkage between geometric structure and surface texture. Based on the depth displacement information in the visual texture map, it performs non-linear offset on the vertex geometric position of the 3D mesh model. This makes the model not only change in color, but also reflect volume changes such as pits and peeling in geometric shape, generating a deformed 3D digital twin model. The semantic consistency verification module, serving as the quality control unit of this system, is used to calculate the semantic consistency score between the visual texture map and the time-series operating condition data. Specifically, although the function of this module is defined as calculating the score of the visual texture map, in practice, it uses the deformed 3D digital twin model generated by the visual texture map as the direct calculation object. This is because the permutation information in the visual texture map has been transformed into the 3D geometric features of the model, namely crack depth and corrosion volume, through the topological deformation coupling module, making the model the precise carrier of the texture map in 3D physical space. Therefore, by extracting the geometric visual features of the deformed model and comparing them with the time-series operating condition data, the reverse verification of the semantic consistency between the visual texture map driving the deformation and the operating condition data is essentially achieved. This implementation method ensures that the verification process not only considers the pixel features of the texture image but also verifies the authenticity of the physical deformation expressed by the texture in 3D space. The system generates the final visualization rendering result based on this score, ensuring that the visual effect does not violate physical facts. This embodiment, through the collaborative work of the above modules, successfully transforms abstract and invisible historical operating condition data into a visualized three-dimensional aging model that conforms to physical laws. This not only overcomes the shortcomings of traditional digital twins that emphasize data but neglect visual aspects, but also provides engineers with an intuitive visual preview method. By observing the distribution of aging textures on the model surface, potential quality risk areas can be quickly located, greatly improving the intuitiveness and predictive accuracy of the entire life cycle management.

[0016] Example 2: The modules are connected in the following way: S1. Collect the 3D mesh model data of the target object and the time-series operating condition data throughout its entire life cycle. After preprocessing the data, map the time-series operating condition data to the surface vertices of the 3D mesh model to generate a spatiotemporal correlation feature set. S2. Input the spatiotemporal correlation feature set into the physical loss evolution algorithm to simulate the physical state changes of each region on the surface of the target object over time, and output the cumulative damage potential energy distribution matrix. S3. The cumulative damage potential energy distribution matrix is ​​used as a conditional constraint input to the texture generation network. Pixel-level prediction is performed in the UV texture space of the 3D mesh model to generate a visual texture map containing diffuse, normal, and displacement channels. S4. Analyze the displacement channel data in the visual texture map, calculate the normal offset of each vertex of the 3D mesh model, drive the 3D mesh model to undergo geometric deformation, and generate the deformed 3D digital twin model. S5. Extract the visual features of the deformed 3D digital twin model, perform logical matching with the time-series operating condition data, and calculate the semantic consistency score. S6. Based on semantic consistency scoring, the deformed 3D digital twin model is screened or its parameters are corrected, and the final visualization rendering result is output.

[0017] This embodiment further elaborates on the specific implementation methods and data flow logic between the various modules of the above system, forming a closed-loop generation and verification process: The process involves performing spatiotemporal mapping and feature set construction. This includes collecting 3D mesh model data of the target object, as well as time-series operating condition data throughout its entire lifecycle. After preprocessing the data, such as cleaning and denoising, spatial interpolation techniques are used to accurately map the time-series operating condition data to the surface vertices of the 3D mesh model, generating a spatiotemporal correlated feature set. This step achieves data conversion from the time domain to the spatial domain. Perform physical state simulation and potential energy calculation; input the spatiotemporal correlation feature set into the physical loss evolution algorithm; the algorithm simulates the physical state changes of each region on the surface of the target object over time, such as fatigue accumulation and corrosion penetration, and outputs the cumulative damage potential energy distribution matrix; this matrix is ​​a numerical field corresponding to the grid topology, and the value represents the probability of visual damage occurring in the region; The execution condition-driven texture generation steps are performed; the cumulative damage potential energy distribution matrix is ​​used as a strong constraint and input into the texture generation network; the network performs pixel-level prediction in the UV texture space of the 3D mesh model to generate a visual texture map containing a diffuse channel (representing color), a normal channel (representing micro-bumps), and a permutation channel (representing macro-deformation). Perform geometric deformation driving; analyze the displacement channel data in the visual texture map, calculate the normal offset of each vertex of the 3D mesh model; drive the 3D mesh model to undergo realistic geometric deformation, and generate the deformed 3D digital twin model; this step ensures the high realism of the model's outline and texture under lighting rendering; Perform visual and physical semantic verification; extract visual features from the deformed 3D digital twin model; the visual feature extraction here is a comprehensive representation based on the visual texture map generated in step S3 and mapped onto the model surface; the system obtains feature indicators such as crack length and corrosion area by recognizing the color and displacement details in the texture map and combining them with the geometric shape of the model, thereby indirectly verifying the quality of the texture map; perform logical matching with the original input time-series load data, i.e., cumulative load, and calculate the semantic consistency score; Implement closed-loop feedback and rendering output; screen the deformed 3D digital twin model based on semantic consistency scoring; if the score meets the standard, output the final visualization rendering result; if it does not meet the standard, trigger the parameter correction mechanism. This embodiment constructs a complete closed loop of data-driven physics, physics-driven vision, and vision feedback data. In particular, the introduction of verification and feedback steps ensures that the generated texture is no longer the subjective creation of an artist, but an objective deduction result verified by data logic, thus guaranteeing the rigor of digital twins in industrial applications.

[0018] S1 specifically includes: S11. Obtain the initial 3D mesh model of the target object, extract the spatial coordinates and normal vectors of all vertices in the 3D mesh model, and construct the geometric topology matrix; S12. Collect environmental load data and operating status data of the target object within a historical time window through a sensor array, perform time alignment and normalization processing on the environmental load data and operating status data, and construct standardized time-series operating condition stream data. S13. Using a spatial interpolation algorithm, the numerical intensity in the time-series operating condition stream data is mapped to the corresponding vertex coordinates in the geometric topology matrix; S14. For each vertex, establish a multi-dimensional feature vector containing spatial location, normal vector, and historical working condition sequence. Combine the multi-dimensional feature vectors of all vertices to output a spatiotemporal correlation feature set.

[0019] This embodiment specifies the data processing procedure in step S1, focusing on solving the problem of fusing multi-source heterogeneous data: The system executes the geometric topology matrix construction step to obtain the initial 3D mesh model of the target object; extracts the spatial coordinates and normal vectors of all vertices in the 3D mesh model, and constructs the geometric topology matrix; Standardization of operating condition data is performed by collecting environmental load data (temperature and humidity) and operational status data (speed and torque) of the target object within a historical time window using a sensor array. This data is then time-aligned and normalized. Specifically, to address the issue of inconsistent sampling frequencies among different sensors, a linear interpolation algorithm is used to resample all data to a unified time reference. To address the issue of large differences in data dimensions, a Min-Max normalization method is used to map each physical quantity to a specific time base. The dimensionless interval; to avoid the confusion in the definition of the denominator dimension that may be caused by directly performing algebraic operations on physical quantities, this embodiment adopts a rigorous calculation strategy of separating numerical values ​​and units: stripping the physical units of each sensor data and extracting only its pure numerical amplitude. Apply the numerical normalization formula: in, Defined as a physical quantity Small compensation constants with the same physical dimensions, for example, taking values For example, if For temperature, then The unit is also Or specify before calculation All values ​​are pure numerical amplitudes after removing units; this constant is used to prevent the sensor reading from becoming constant. This prevents computational anomalies that result in a denominator of zero, thus ensuring the normalization result. It has a clear and rigorous dimensionless mathematical meaning, avoiding the error of mixing physical dimensions with compensation constants in calculation, thereby constructing standardized time-series operating condition flow data; Spatial interpolation mapping is performed. Since the number of sensors is much smaller than the number of model vertices, this embodiment employs a spatial interpolation algorithm, specifically the inverse distance weighting method. Based on the physical installation locations of the sensors on the model, the numerical intensity in the time-series load stream data is mapped to the corresponding vertex coordinates in the geometric topology matrix. To ensure the reproducibility of the algorithm, this embodiment defines the first... Interpolation values ​​of each grid vertex The calculation formula is: Among them, weight ;in, For the preset minimum distance smoothing term, for example This is used to prevent the sensor position from coinciding with the vertex. Computational overflow caused by the time weight approaching infinity; For the first The sampled values ​​of each sensor, For the sensor and vertex The Euclidean distance between them The distance decay exponent, set to 2 in this embodiment, is based on the inverse square law of energy decay with distance. The preset search radius with the vertex as the center of the sphere The number of effective sensors within; to avoid unclear parameter sources due to the term "preset", this embodiment explicitly specifies... The value of is not a fixed constant, but is defined as the length of the diagonal of the model bounding box. ,Right now ,in The maximum and minimum coordinates of the model vertices. This refers to the geometric coverage factor; specifically, it addresses the potential for localized coverage issues arising from sparse sensor distributions. This leads to interpolation distortion. This embodiment introduces a dynamic radius expansion mechanism: when a vertex is detected to be within the initial radius... Number of effective sensors inside When the iteration logic is triggered automatically. Repeat the search until or The entire model bounding box is covered to ensure that valid interpolation data can be obtained at any position in the mesh, so that each vertex has the calculated working condition data. Output a spatiotemporal correlation feature set, and establish a multi-dimensional feature vector for each vertex that includes spatial location, normal vector and historical working condition sequence; combine the multi-dimensional feature vectors of all vertices to output the spatiotemporal correlation feature set; This embodiment successfully solves the problem that sparse sensor data cannot cover the surface of a high-precision model by combining a geometric topology matrix with a spatial interpolation algorithm with dynamic radius expansion, providing comprehensive data support for subsequent refined physical calculations.

[0020] S2 specifically includes: S21. Based on the spatiotemporal correlation feature set, construct a physical loss state variable for each vertex and set the state decay coefficient and cumulative threshold of the physical loss state variable. S22. Using the stress transformation function, the historical working condition sequence of each vertex is transformed into instantaneous stress value, and the instantaneous stress value is input into the physical loss state variable for time-series numerical integration. S23. Introduce a preset hysteresis correction coefficient to dynamically adjust the integral result of the physical loss state variable in order to compensate for the nonlinear superposition error and obtain the current damage potential energy value of each vertex. S24. Perform spatial smoothing on the current damage potential energy values ​​of all vertices according to the topological adjacency relationship of the three-dimensional mesh model to generate a cumulative damage potential energy distribution matrix that reflects the degree of surface damage.

[0021] This embodiment details the specific implementation of the physical loss evolution algorithm in step S2, which is a crucial bridge connecting data and vision: Define physical loss state variables, construct physical loss state variables for each vertex based on spatiotemporal correlation feature sets; set the state decay coefficient and cumulative threshold of the variable; Instantaneous stress transformation and integration are performed, and the stress transformation function is used to construct a multidimensional linear weighted model: ,in, For a moment The Similar operating condition data includes temperature and vibration amplitude. For the corresponding stress transformation weighting coefficient, the weighting coefficient is... The method for determining the value is as follows: A finite element simulation model of the target object is established in advance, and the first unit intensity is applied to each object. Type of load, such as Temperature rise or Acceleration was used to calculate the average von Mises stress response value in the key region, and this response value was used as... Stored in the system parameter database; among which, This represents the total number of time-series data types collected. The reference stress constant is used to characterize the initial residual stress or environmental reference stress of the material under zero working condition. The historical working condition sequence of each vertex is converted into instantaneous stress value, and the instantaneous stress value is input into the physical loss state variable for time-series numerical integration. In order to accurately describe the damage accumulation process and reflect the role of the state decay coefficient set in step S21, this embodiment defines the calculation formula of the current damage potential energy value as follows: in, The source is a preset material damage model, and its physical meaning is a stress-damage transformation function, which in this embodiment specifically adopts a power-law form. ,in, The material damage constant, which includes a unit conversion factor, has its physical dimensions set to fit. The conversion to power units is necessary to ensure energy dimension balance. The fatigue index is determined by fitting the material stress-life curve of the target object using the least squares method. Corresponding fatigue strength coefficient, Corresponding to the fatigue ductility index; in order to strictly satisfy the dimensional consistency of the physical formula, this embodiment explicitly defines the constant. The unit ensures The output physical quantity is power, thus making the time unit s... Integration results It has the correct energy unit, avoiding the error of only performing numerical calculations while ignoring the physical meaning; The source is the result of processing the operating data using the above multidimensional linear weighted model; its physical meaning is the integration time. The instantaneous stress value is Unlike the points cap The unit is MPa; The source is a preset constant, and its physical meaning is a hysteresis effect correction coefficient, with a range of values. , obtained by material creep experiments, is used to characterize the delayed response of materials to loads, and is dimensionless; The origin is a mathematical constant, and its physical meaning is the base of the natural logarithm, with a value of approximately 2.71828; The source is set in step S21, and its physical meaning is the state decay coefficient, which is determined by material stress relaxation experiments. A typical value is... Used to simulate the microstructure recovery or stress relaxation effect of materials over long periods of time, with units of ; The source is the system clock, the physical meaning is the integral time variable, and the unit is seconds; To enable programmable computation of the aforementioned continuous integration process in a computer digital system, this embodiment discretizes it into a recursive difference equation: in, At the current sampling time, The sampling time interval is specifically set as the period for acquiring sensor data, for example... This ensures that the frequency of the input time-series operating condition data is consistent with the frequency of the input data; this recursive form ensures that the algorithm can process streaming data in real time without storing the full historical sequence. Nonlinear correction is performed by introducing a preset hysteresis effect correction coefficient to dynamically adjust the integral results; this simulates the material memory effect, that is, the damage depends not only on the current load, but also on the sequence pattern of historical loads, thereby compensating for the error caused by linear superposition. Spatial smoothing and matrix generation are performed, and the current damage potential values ​​of all vertices are spatially smoothed according to the topological adjacency relationship of the 3D mesh model. To address the uncertainty of spatial smoothing in code reproduction, this embodiment specifically uses the discrete Laplace smoothing operator, whose iterative formula is as follows: in, This is a smoothing coefficient, with a range of values. In this embodiment, the preferred embodiment is... To balance denoising effectiveness with feature preservation, As vertices The first-order neighborhood set, i.e. the set of vertices directly connected by the edges of the triangular mesh, is used to eliminate the high-frequency numerical noise generated by the integral calculation and generate the cumulative damage potential energy distribution matrix that reflects the degree of surface damage. This embodiment introduces an integral model with hysteresis correction and state decay, which can realistically simulate nonlinear physical processes such as material fatigue and aging. The generated cumulative damage potential energy distribution matrix is ​​no longer a simple heat map, but a physical field containing the mechanical properties of the material, laying a solid physical foundation for generating realistic aging textures.

[0022] Example 3: S3 specifically includes: S31. Construct a texture generation model based on conditional generative adversarial networks, and map the cumulative damage potential distribution matrix into a two-dimensional feature map as a guiding condition for the texture generation model. S32. Introduce a physical constraint loss function into the texture generation model. Based on the guiding conditions, predict the color value, roughness value and height value of the aged texture pixel by pixel on the UV unfolded map of the 3D mesh model. S33. Perform color space conversion on the predicted color values ​​to generate a diffuse texture map; perform gradient calculation on the predicted roughness values ​​to generate a normal texture map; perform grayscale quantization on the predicted height values ​​to generate a displacement texture map. S34. Merge the diffuse texture map, normal texture map and displacement texture map into channels to output a multi-channel visual texture map.

[0023] This embodiment provides a deep learning-based specification for the texture generation process in step S3: Feature mapping and conditional guidance are performed to construct a texture generation model based on conditional generative adversarial networks; the cumulative damage potential distribution matrix is ​​mapped into a two-dimensional feature map; the mapping process specifically adopts texture space rasterization technology: in order to solve the problem of physical feature mapping conflict caused by UV overlap in industrial models, the system detects the UV layout of the model. If there is overlap, the least squares conformal mapping algorithm is called to reduce texture distortion as the optimization goal, and the texture coordinates of the mesh vertices are recalculated to generate a set of non-overlapping second UV channels; A two-dimensional frame buffer with the same input resolution as the texture generation model is established. All triangular faces of the three-dimensional mesh model are traversed, and the cumulative damage potential energy value and non-overlapping UV coordinates of the face vertex are read. Using the scan line algorithm and centroid coordinate interpolation, the potential energy values ​​on the vertex are smoothly filled into the corresponding pixel coordinates of the two-dimensional frame buffer to generate a two-dimensional feature map with dense pixel distribution and no spatial ambiguity, which serves as the guiding condition for the texture generation model. Pixel prediction under physical constraints is performed, and a physical constraint loss function is introduced into the texture generation model. To ensure the controllability of the generation process and support closed-loop feedback in subsequent S6, this embodiment designs the generation process as a gradient-based iterative optimization process. That is, during the inference phase, the total loss function is minimized by freezing network parameters and fine-tuning the input latent code, or a conditionally parameterized network structure is adopted to adjust the weights. The feature vectors are mapped to the input network; this loss function is used to force the output of the generator network to conform to the law of conservation of physical energy, specifically defining the total loss function. as follows: in, Traditional adversarial loss is used to ensure texture fidelity; however, to clarify the specific meaning of "traditional" and improve training stability, this embodiment specifically selects the loss form of the Least Squares Generative Adversarial Network (LSGAN), namely... Compared to the standard Sigmoid cross-entropy loss, this form can provide a stronger gradient signal for the generated samples and avoid the gradient vanishing problem. For physical constraint loss; to address the shortcomings of existing technologies that directly normalize and compare energy and depth, resulting in inconsistencies in dimensions and unclear physical logic, this embodiment constructs an energy-geometry mapping loss function based on material constitutive relations: ;in, This indicates that the physical depth is normalized to... Range operations ensure consistency with network output. Dimensions are consistent; The predicted displacement channel depth; The pre-defined damage potential-theoretical depth conversion operator is defined as follows: in, The cumulative damage potential energy input is... , For example, the yield strength of a material, such as alloy steel. This can be obtained by consulting the material properties handbook. To incorporate unit conversion factors, and to offset the dimensional effects of the stress term in the denominator and ensure the output is in units of length, the physical dimensions are set to [value missing]. , value take , The damage evolution index has a value of [value missing]. The operator was determined by fatigue crack propagation rate experiments; Its function is to map the scalar energy field into a theoretical damage depth field with definite length dimensions, thereby enabling the prediction of geometric depth. The difference between the theoretical depth of the physical deduction and the physical deduction is compared under completely consistent physical dimensions, which ensures the rigor and rationality of the physical consistency constraints. The preset physical weight coefficients are initially set to 10.0, which are determined based on experimental experience of model training convergence. They will be fine-tuned by observing the convergence curve of the validation set loss. Based on the optimization of the guiding conditions and the composite loss function, the network predicts the properties of the aging texture pixel by pixel on the UV unfolded map of the 3D mesh model. In order to achieve decoupled prediction of different physical properties and support accurate physical constraints, this embodiment explicitly specifies that the decoder output layer architecture of the texture generation network is a multi-channel convolutional layer. Channels Ch1-Ch3 correspond to the Lab color space. The Lab color space is chosen here instead of RGB space because it leverages the uniformity of the Lab space in human visual perception. This makes the generated texture color difference more consistent with the human eye's perception of minute signs of aging. The Tanh activation function is used to map to... The normalized range is then remapped back to the Lab color space during the subsequent rendering stage. The actual value range, channel Ch4 corresponds to the roughness scalar, which is mapped to using the Sigmoid activation function. Channel Ch5 corresponds to the permutation height value, which is mapped to using the Sigmoid activation function. This specific channel isolation design makes the physical constraint loss function... It can perform backpropagation calculations only for Ch5, thereby precisely controlling the geometric deformation parameters without interfering with the artistic generation process of color and texture in Ch1-Ch4; Generate multi-channel textures; perform color space conversion on the predicted color values, such as from Lab to sRGB, to generate diffuse texture maps and simulate the albedo changes of the material surface; for the step of generating normal texture maps by calculating gradients from the predicted roughness values, to avoid gradient calculations becoming a black box description, this embodiment explicitly uses the Sobel operator for discrete differentiation: defining horizontal convolution kernels. With vertical convolution kernel ; predict the roughness map respectively with Perform convolution operations to obtain gradient components. and ; Reconstructing formula using tangent space normal Generate the normal vector, where, The preset normal intensity coefficient is used, and the vector components are mapped to... For integer ranges, generate normal texture maps; perform grayscale quantization on the predicted height values ​​to generate displacement texture maps representing surface bumps and undulations; Complete the channel merging, merge the three types of textures mentioned above, and output a multi-channel visual texture map; This embodiment generates not only the color changes of the surface through multi-channel parallel generation and a clear definition of the network output layer, but also the normals required for lighting and the displacement information required for geometry, so that the final rendered aging effect has a strong sense of three-dimensionality and material realism.

[0024] S4 specifically includes: S61. Read the grayscale pixel values ​​in the displacement texture map and convert the grayscale pixel values ​​into geometric displacement scalars according to the preset pixel-to-physical depth conversion ratio. S62. Traverse all vertices of the 3D mesh model and index the corresponding geometric displacement scalar in the replacement texture map based on the texture coordinates of each vertex in the UV space. S63. Apply the corresponding geometric displacement scalar along the normal vector direction of each vertex, update the spatial coordinate position of the vertex, and complete the subdivision and reconstruction of the mesh surface. S64. Recalculate the vertex normals and tangent space of the reconstructed 3D mesh model, refresh the lighting and rendering properties of the model, and generate the deformed 3D digital twin model.

[0025] This embodiment describes in detail how texture-driven model geometric deformation is used in step S4: Perform pixel-to-depth conversion and read the grayscale pixel values ​​from the displacement texture map; to ensure the accuracy of physical units, this embodiment normalizes the pixel values ​​to... The floating-point number within the range is used to convert grayscale pixel values ​​into geometric displacement scalars based on a preset pixel-to-physical depth conversion ratio; the calculation formula is as follows: in, The source is the pixel grayscale values ​​of the displacement texture map after Min-Max normalization. Its physical meaning is the relative height ratio, dimensionless, and its value range is... ; The source is a preset value, and its physical meaning is the zero-displacement reference plane, representing the original surface position of the model. It is usually set to... , corresponding to neutral gray; The source is set based on the actual maximum permissible deformation depth of the target object. Its physical meaning is the maximum physical replacement depth, in mm. The setting is based on the maximum permissible corrosion or crack depth defined by the component's scrapping standard. For example... ; Perform UV indexing, traverse all vertices of the 3D mesh model, and based on the texture coordinates of each vertex in UV space, index the corresponding geometric displacement scalar in the texture map. Update the vertex position by applying the corresponding geometric displacement scalar along the normal vector direction of each vertex and updating the spatial coordinate position of the vertex. Perform normal reconstruction, recalculating the vertex normals and tangents of the reconstructed 3D mesh model. To eliminate ambiguity from the recalculation, this embodiment uses a weighted average method of face normals for specific geometric updates: traversing all reconstructed triangular faces. Extract its three vertices Calculate the surface normal vector Initialize all vertex normals to zero vectors, then iterate through all faces again, and calculate the results. The summation is applied to the three vertices of the face; the summation of the normal vector for each vertex is also applied. L2 normalization processing This allows us to obtain accurate normal vectors that conform to the deformed geometric surface, refresh the model's lighting and rendering properties, and generate a deformed 3D digital twin model. This embodiment achieves true geometric-level damage simulation; unlike texture mapping techniques that only change color, this solution actually changes the mesh structure of the model, so that the self-shadowing and occlusion relationships generated by light and shadow on the uneven parts are completely in line with physical reality.

[0026] S5 specifically includes: S71. Extract visual defect feature regions from the deformed 3D digital twin model, and calculate the geometric area and texture contrast of the visual defect feature regions. S72. Extract the corresponding cumulative load values ​​and environmental exposure duration from the time-series operating condition data to construct the theoretical values ​​of physical damage. S73. Calculate the normalized difference between the geometric area of ​​the visual defect feature region and the theoretical value of physical damage to obtain the feature matching error; S74. Based on feature matching error and texture contrast, a weighted summation is performed using preset weighting factors to obtain the semantic consistency score Q. S75. Compare the semantic consistency score Q with the preset pass / fail threshold T to generate a score comparison result.

[0027] This embodiment elaborates on the calculation logic of semantic consistency score in step S5, which is the core of ensuring the scientific nature of the system: Visual defect features are extracted from the deformed 3D digital twin model using an image segmentation algorithm. To avoid the uncertainty caused by the image segmentation algorithm being a black box and to ensure that the extracted regions have clear physical meaning, this embodiment specifically performs the following operations: retrieve the displacement texture map generated in step S3 and calculate its pixel grayscale histogram; adaptively calculate the segmentation threshold using the maximum inter-class variance method. ; Replace the grayscale values ​​in the displacement map The pixels are marked as defect areas with Mask=1, and the rest are marked as background with Mask=0; this logic ensures that the so-called visual defects are based on the physical deformation depth definition, rather than simple surface color differences, thus guaranteeing physical homogeneity when comparing the area with the theoretical value of physical damage in the future. The specific implementation of this step relies on the parsing of visual texture maps; the system reads the visual texture map data mapped to the model surface and identifies high-contrast aging texture areas, i.e., visual defects; in order to solve the cross-dimensional calculation problem from 2D texture pixels to 3D physical area, this embodiment explicitly adopts the UV-physical area ratio integral method to calculate the geometric area. : Traverse all triangles of the model Calculate their Euclidean areas in 3D space. And the number of pixels occupied in the UV texture space Thus, the unit of the face is obtained. physical area coefficient per unit pixel ; Analyze the pixel set in the visual defect feature region Mask generated based on the permutation threshold, and identify each defect pixel. The corresponding face index , will the corresponding To accumulate, that is This allows for the accurate determination of the true physical area of ​​the defect region on the curved three-dimensional surface; simultaneously, the texture contrast is calculated. The gray-level root mean square (RMS) formula is used to define: in, This represents the normalized grayscale value of the pixels within the defect area. The average gray level of the region; This represents the total number of pixels in the visual defect feature region Mask. Physical theoretical values ​​are constructed by extracting the corresponding cumulative load values ​​and environmental exposure durations from time-series operating condition data, and combining them with material lifetime decay curves to construct theoretical values ​​for physical damage. This embodiment introduces a reference power density constant. Damage rate coefficient per unit area Make corrections and clarify the theoretical value of physical damage. The calculation formula is: in, Theoretical value of physical damage, in square millimeters. ; The critical power density constant that leads to accelerated material damage is determined through standard material fatigue failure tests, and its unit is 1. Units and Consistency ensures that the exponent term is dimensionless; : Geometric topology correction factor, a dimensionless ratio used to correct plane projection errors based on curvature; Damage propagation rate coefficient, derived from material aging experiments, is measured in square millimeters per second. , used to convert the time dimension of the integral result into the area dimension; Integral-differential operators, unit: seconds ; The result is the cumulative damage area, thus ensuring that the dimensions of both sides of the equation are strictly balanced. ; This formula corrects the mathematical error in the original formula of directly adding the area dimension to the dimensionless scalar, and clearly expresses that the damaged area is the result of cumulative energy dissipation after geometric correction. in, The normalized instantaneous load, The preset reference power density, The total duration of environmental exposure is represented by the exponential term, which characterizes the nonlinear acceleration effect of the load on the damage rate. The integral result represents the equivalent aging time after considering the load weighting. The damage propagation coefficient; The geometric topology correction factor is calculated using the following formula: ,in, This represents the actual three-dimensional surface area of ​​the defect region. Its projected area on the tangent plane is used to correct the nonlinear effect of curvature on the damage propagation rate; through this dimensional correction, it is ensured that the calculation of the theoretical value of physical damage conforms to the law of conservation of energy. Calculate feature matching error The relative error calculation logic is adopted: in, For example, a preset non-zero minimum value. This is only used to prevent damage when the theoretical damage value is... To prevent calculation errors caused by division by zero, we ensure the stability of numerical calculations. The source is calculated, and its physical meaning is feature matching error, which is dimensionless. The source is calculated using the UV-physical area ratio integral method mentioned above. The physical meaning is the geometric area of ​​the visual defect feature region, with the unit being square millimeters. The source is derived from the above formula based on working condition data, and the physical meaning is the theoretical value of physical damage, with the unit being square millimeters; The semantic consistency score is calculated based on feature matching error and texture contrast, using a weighted summation with preset weighting factors. To overcome the logical flaws of traditional linear weighting formulas in handling high-contrast illusory textures, i.e., when physical errors occur... Extremely high but texture contrast At extremely high levels, a simple weighted summation may incorrectly award high scores, causing the system to output sharp artifacts that contradict physical facts. This embodiment constructs an improved weighted summation formula that includes a physical-visual conflict penalty term: in, The source is calculated, and the physical meaning is the final score. If the calculation result is less than 0, it is set to 0. The physical meaning refers to the physical matching degree term, when the error... When the value is extremely small, it approaches 1; The source is a preset positive weight; in this embodiment, it is taken as... This contributes to balancing physical accuracy and visual clarity. The source is a preset penalty weight; in this embodiment, it is taken as... The above weights are determined based on the optimal operating point of the ROC curve from the historical defect sample library. The physical meaning is the conflict penalty term, which characterizes the degree of coupling between physical bias and visual saliency; The innovative logic of this formula lies in: when visual features highly match physical theories, that is... At that time, the penalty item approaches zero, and the score... Primarily determined by physical matching, it allows low-contrast, genuine latent defects to pass with high scores; however, when visual features severely violate physical theory... Large yet exhibiting high contrast When the value is very high, the penalty will increase dramatically, forcibly lowering the overall score. Up to the qualified threshold The following method effectively identifies and blocks such visual deception, triggering the subsequent regeneration process; The scoring comparison is performed by comparing the semantic consistency score with a preset pass / fail threshold to generate a scoring comparison result; This embodiment establishes an objective standard for evaluating the quality of generated content by introducing a scoring mechanism with penalties. It mandates that the generated visual effects must display details under the constraints of physical laws, thus avoiding the situation where physical realism is sacrificed in the pursuit of visual impact.

[0028] S6 specifically includes: S81. Obtain the semantic consistency score Q and the preset qualified threshold T; S82. In response to the semantic consistency score Q being greater than or equal to the qualified threshold T, determine that the currently generated visual effect conforms to physical laws, mark the deformed 3D digital twin model as valid, and output high-precision rendering results. S83. In response to the semantic consistency score Q being less than the qualified threshold T, it is determined that the currently generated visual effect has semantic deviation. The difference between the score Q and the threshold T is calculated, and the weight of the physical constraint loss function in the texture generation model is adjusted in reverse based on the difference. This triggers the regeneration process of the texture generation network until the semantic consistency score Q meets the condition of being greater than or equal to the qualified threshold T. Finally, the final visualization rendering result is output.

[0029] This embodiment describes the closed-loop control logic in step S6: Perform a positive output judgment to obtain the semantic consistency score and the preset qualified threshold; in response to the semantic consistency score being greater than or equal to the qualified threshold, determine that the currently generated visual effect conforms to physical laws, mark the deformed 3D digital twin model as valid, and output high-precision rendering results; Perform negative feedback and regeneration. In response to a semantic consistency score that is less than the acceptable threshold, determine that the currently generated visual effect has semantic deviation, such as the generated crack being too large or too small; calculate the difference between the score and the threshold. The weights of the physical constraint loss function in the texture generation model are adjusted based on the difference; the adjustment formula is as follows: in, The source is calculated, and the physical meaning is the adjusted weight of the physical constraint loss function, which is dimensionless. For example, the preset weight limit To prevent due to Excessive size can cause network training to diverge or the model to collapse. The source is the value from the previous iteration; its physical meaning is the original weight, which is dimensionless. The source is a preset parameter, and its physical meaning is to adjust the step size coefficient. For example, set it to... This is used to control the sensitivity of weight adjustments; The source is calculated, and the calculation formula is: The physical meaning is the difference between the score and the threshold; After triggering regeneration and adjusting the weights, the texture generation network is regenerated until the semantic consistency score meets or exceeds the passing threshold, at which point the final visualization rendering result is output. It should be noted that to avoid infinite loops caused by parameter oscillations or failure to converge, the system presets a maximum number of iterations. For example, set to This is to meet the system's real-time requirements; In the regeneration logic, if the number of iterations... and Still smaller When the loop terminates, the program will forcibly terminate the loop and output the current iteration history. The rendering result with the largest value is used as the degraded output, thereby ensuring the termination and stability of the algorithm in engineering applications; This embodiment implements an adaptive generation optimization mechanism; when the generation result is inaccurate, the system can automatically increase the weight of physical constraints, forcing the neural network to more strictly follow physical laws to generate, thereby gradually approaching the most realistic physical visual mapping state without human intervention.

[0030] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A product quality lifecycle management system based on digital twins, characterized in that, include: Multidimensional data mapping module: used to acquire the three-dimensional mesh model data and time-series operating condition flow data of the target object, and map the time-series operating condition flow data to the surface vertices of the three-dimensional mesh model to construct a spatiotemporal correlation feature set; The physical perception evolution module is used to calculate the cumulative damage potential energy distribution of each region on the surface of a 3D mesh model based on a spatiotemporal correlation feature set and a physical loss evolution algorithm. Generative texture synthesis module: Based on the cumulative damage potential energy distribution, it drives the texture generation network to synthesize visual texture maps containing aging features in the corresponding texture coordinate space of the 3D mesh model. Topological deformation coupling module: Based on the depth displacement information in the visual texture map, it performs non-linear offset on the vertex geometric position of the 3D mesh model to generate a deformed 3D digital twin model; Semantic consistency verification module: used to calculate the semantic consistency score between visual texture maps and time-series operational data, and generate the final visualization rendering result based on the semantic consistency score.

2. The product quality lifecycle management system based on digital twins according to claim 1, characterized in that, The modules are connected in the following way: S1. Collect the 3D mesh model data of the target object and the time-series operating condition data throughout its entire life cycle. After preprocessing the data, map the time-series operating condition data to the surface vertices of the 3D mesh model to generate a spatiotemporal correlation feature set. S2. Input the spatiotemporal correlation feature set into the physical loss evolution algorithm to simulate the physical state changes of each region on the surface of the target object over time, and output the cumulative damage potential energy distribution matrix. S3. The cumulative damage potential energy distribution matrix is ​​used as a conditional constraint input to the texture generation network. Pixel-level prediction is performed in the UV texture space of the 3D mesh model to generate a visual texture map containing diffuse, normal, and displacement channels. S4. Analyze the displacement channel data in the visual texture map, calculate the normal offset of each vertex of the 3D mesh model, drive the 3D mesh model to undergo geometric deformation, and generate the deformed 3D digital twin model. S5. Extract the visual features of the deformed 3D digital twin model, perform logical matching with the time-series operating condition data, and calculate the semantic consistency score. S6. Based on semantic consistency scoring, the deformed 3D digital twin model is screened or its parameters are corrected, and the final visualization rendering result is output.

3. A product quality lifecycle management system based on digital twins according to claim 2, characterized in that, S1 specifically includes: S11. Obtain the initial 3D mesh model of the target object, extract the spatial coordinates and normal vectors of all vertices in the 3D mesh model, and construct the geometric topology matrix; S12. Collect environmental load data and operating status data of the target object within a historical time window through a sensor array, perform time alignment and normalization processing on the environmental load data and operating status data, and construct standardized time-series operating condition stream data. S13. Using a spatial interpolation algorithm, the numerical intensity in the time-series operating condition stream data is mapped to the corresponding vertex coordinates in the geometric topology matrix; S14. For each vertex, establish a multi-dimensional feature vector containing spatial location, normal vector, and historical working condition sequence. Combine the multi-dimensional feature vectors of all vertices to output a spatiotemporal correlation feature set.

4. A product quality lifecycle management system based on digital twins according to claim 3, characterized in that, S2 specifically includes: S21. Based on the spatiotemporal correlation feature set, construct a physical loss state variable for each vertex, and set the state decay coefficient and cumulative threshold of the physical loss state variable. S22. Using the stress transformation function, the historical working condition sequence of each vertex is converted into an instantaneous stress value, and the instantaneous stress value is input into the physical loss state variable for time-series numerical integration. S23. Introduce a preset hysteresis correction coefficient to dynamically adjust the integral result of the physical loss state variable in order to compensate for nonlinear superposition error and obtain the current damage potential energy value of each vertex. S24. Perform spatial smoothing on the current damage potential energy values ​​of all vertices according to the topological adjacency relationship of the three-dimensional mesh model to generate a cumulative damage potential energy distribution matrix that reflects the degree of surface damage.

5. A product quality lifecycle management system based on digital twins according to claim 4, characterized in that, S3 specifically includes: S31. Construct a texture generation model based on conditional generative adversarial networks, and map the cumulative damage potential distribution matrix into a two-dimensional feature map as a guiding condition for the texture generation model. S32. Introduce a physical constraint loss function into the texture generation model. Based on the guiding conditions, predict the color value, roughness value and height value of the aged texture pixel by pixel on the UV unfolded map of the 3D mesh model. S33. Perform color space conversion on the predicted color values ​​to generate a diffuse texture map; perform gradient calculation on the predicted roughness values ​​to generate a normal texture map; perform grayscale quantization on the predicted height values ​​to generate a displacement texture map. S34. Merge the diffuse texture map, normal texture map and displacement texture map into channels to output a multi-channel visual texture map.

6. A product quality lifecycle management system based on digital twins according to claim 5, characterized in that, S4 specifically includes: S61. Read the grayscale pixel values ​​in the displacement texture map and convert the grayscale pixel values ​​into geometric displacement scalars according to the preset pixel-to-physical depth conversion ratio. S62. Traverse all vertices of the 3D mesh model and index the corresponding geometric displacement scalar in the replacement texture map based on the texture coordinates of each vertex in the UV space. S63. Apply the corresponding geometric displacement scalar along the normal vector direction of each vertex, update the spatial coordinate position of the vertex, and complete the subdivision and reconstruction of the mesh surface. S64. Recalculate the vertex normals and tangent space of the reconstructed 3D mesh model, refresh the lighting and rendering properties of the model, and generate the deformed 3D digital twin model.

7. A product quality lifecycle management system based on digital twins according to claim 6, characterized in that, S5 specifically includes: S71. Extract visual defect feature regions from the deformed three-dimensional digital twin model, and calculate the geometric area and texture contrast of the visual defect feature regions; S72. Extract the corresponding cumulative load values ​​and environmental exposure duration from the time-series operating condition data to construct the theoretical values ​​of physical damage. S73. Calculate the normalized difference between the geometric area of ​​the visual defect feature region and the theoretical value of the physical damage to obtain the feature matching error; S74. Based on the feature matching error and the texture contrast, a weighted summation is performed using a preset weighting factor to obtain the semantic consistency score Q. S75. Compare the semantic consistency score Q with the preset qualification threshold T to generate a score comparison result.

8. A product quality lifecycle management system based on digital twins according to claim 7, characterized in that, S6 specifically includes: S81. Obtain the semantic consistency score Q and the preset qualification threshold T; S82. In response to the semantic consistency score Q being greater than or equal to the qualified threshold T, determine that the currently generated visual effect conforms to physical laws, mark the deformed three-dimensional digital twin model as valid, and output high-precision rendering results. S83. In response to the semantic consistency score Q being less than the qualified threshold T, it is determined that the currently generated visual effect has semantic deviation. The difference between the score Q and the threshold T is calculated, and the weight of the physical constraint loss function in the texture generation model is adjusted in reverse based on the difference. The regeneration process of the texture generation network is triggered until the semantic consistency score Q meets the condition of being greater than or equal to the qualified threshold T, and the final visualization rendering result is output.

Images