Appearance intelligent generation optimization method and system, computer device and storage medium

Abstract

The application discloses an appearance intelligent generation optimization method and system, relates to the technical field of computer-aided industrial design and generative artificial intelligence, and comprises the following steps: obtaining product internal core components and engineering constraints, and generating a three-dimensional safety envelope manifold; constructing a low-distortion manifold parameter domain based on a manifold spectrum operator or a conformal energy function; performing semantic analysis and frequency domain decoupling on creative materials or multi-modal creative input to obtain low-frequency appearance signals, medium-frequency structure signals and high-frequency texture signals; mapping and fusing multi-scale creative features to the three-dimensional safety envelope manifold, initializing a three-dimensional space representation field and generating a three-dimensional appearance model; generating physical residual errors or manufacturing constraint residual errors through a multi-physical field agent evaluation network, reversely positioning three-dimensional abnormal regions to the low-distortion manifold parameter domain and corresponding frequency domain signals, and regenerating after adjusting weights or field parameters; and outputting an appearance scheme and generating a fingerprint based on a comprehensive fitness function.

Description

Technical Field

[0001] This invention relates to the fields of computer-aided industrial design, computer graphics, generative artificial intelligence, manifold geometry, three-dimensional spatial representation, multiphysics proxy simulation and intelligent manufacturing technology, and more specifically, to an intelligent appearance generation optimization method and system. Background Technology

[0002] In the fields of computer-aided industrial design, computer graphics, generative artificial intelligence, manifold geometry, 3D spatial representation, multiphysics simulation, and intelligent manufacturing, the appearance of industrial products not only affects their visual recognition and user perception but also directly relates to their aerodynamic performance, heat dissipation, structural strength, assembly feasibility, and manufacturing costs. Traditional industrial design processes typically involve designers creating 2D sketches or conceptual renderings based on brand style, market trends, or biomimetic ideas. Structural engineers then perform 3D modeling and multiple rounds of modifications based on internal components, assembly gaps, heat dissipation space, structural stress areas, and manufacturing process requirements. This process suffers from low efficiency in connecting 2D ideas with 3D engineering structures, long design iteration cycles, delayed performance verification, and high costs associated with manual modifications.

[0003] In recent years, generative artificial intelligence and 3D representation technologies such as diffusion models, generative adversarial networks, flow matching models, autoregressive generative models, text-to-3D generation models, neural radiation fields, 3D Gaussian representations, and neural implicit fields have been increasingly applied to product appearance design. These technologies can rapidly generate diverse appearance schemes based on text, images, sketches, or multimodal inputs. However, existing technologies still face the following challenges when applied to the development of actual industrial product appearances: First, existing text- or image-driven 3D generation methods primarily focus on visual semantic consistency and multi-view rendering effects, lacking unified constraints on core components, assembly constraints, heat dissipation space, restricted areas, structural stress areas, and manufacturing safety clearances within the product. Consequently, the resulting appearance models may intrude into internal component spaces or fail to meet heat dissipation, assembly, and structural safety requirements.

[0004] Second, existing 3D spatial representation technologies, such as neural radiation fields, 3D Gaussian representations, neural symbolic distance fields, occupancy fields, or differentiable mesh fields, are typically used for the reconstruction, rendering, and representation optimization of existing scenes or objects, and have not formed an engineering constraint mapping mechanism for the generation of creative designs for industrial product appearances. They cannot directly map the overall contour, local structure, and surface texture in creative materials or multimodal creative inputs into controllable 3D appearance generation elements.

[0005] Third, while existing manifold parameterization, conformal mapping, and surface editing techniques can reduce the mapping distortion from three-dimensional surfaces to two-dimensional parameter domains, they are mostly used for texture mapping, surface flattening, or geometric processing, lacking a closed-loop generation mechanism that combines multimodal creative input, three-dimensional spatial representation fields, and feedback from physical residuals and manufacturing constraint residuals.

[0006] Fourth, existing physical information neural networks, neural operators, finite element proxy models, computational fluid dynamics proxy models, thermal simulation proxy models, or structural simulation proxy models are typically used to evaluate or predict the performance of candidate design schemes. They lack mechanisms to reverse-engineer abnormal regions such as flow separation, pressure surges, heat accumulation, stress concentration, insufficient wall thickness, draft infeasibility, or inaccessible processing, tracing them back to the source of the creative idea. In other words, existing methods can usually only score, filter, or post-process the generated results, making it difficult to reverse-engineer physical residuals or manufacturing constraint residuals onto creative frequency domain signals such as low-frequency shape, mid-frequency structure, and high-frequency texture.

[0007] Fifth, existing AI appearance generation results typically lack complete records of generation path, frequency domain weights, manifold mapping relationships, 3D representation parameters, physical residuals, and manufacturing constraint residuals, leading to difficulties in design asset tracing, version management, model comparison, and infringement analysis.

[0008] Therefore, there is an urgent need for an intelligent appearance generation optimization solution that can unify product engineering constraints, low-distortion manifold parameter domain, multimodal creative frequency domain decoupling, three-dimensional spatial characterization field, multiphysics field and manufacturing feasibility evaluation, residual reverse positioning correction, and fingerprint generation into the same closed-loop process. Summary of the Invention

[0009] To address the aforementioned issues, the present invention aims to provide a method and system for intelligent appearance generation and optimization in the fields of computer-aided industrial design, computer graphics, generative artificial intelligence, manifold geometry, 3D spatial representation, multiphysics proxy simulation, and intelligent manufacturing. This method transforms the product's internal core components, assembly constraints, heat dissipation constraints, restricted areas, structural stress areas, manufacturing safety clearances, and appearance design boundaries into a 3D safe envelope manifold. A low-distortion manifold parameter domain is constructed based on manifold spectrum operators or conformal energy functions. Semantic parsing and frequency domain decoupling are then performed on creative materials or multimodal creative inputs to obtain low-frequency shape signals, mid-frequency structural signals, and high-frequency texture signals. Multi-scale creative features are then mapped and fused into the 3D safe envelope manifold to initialize the 3D spatial representation field and generate a 3D appearance model. Physical residuals or manufacturing constraint residuals are obtained through a multiphysics proxy evaluation network, and abnormal regions are reverse-located from the 3D appearance model to the low-distortion manifold parameter domain and corresponding frequency domain signals. The frequency domain signal weights or local 3D spatial representation field parameters are adjusted before regeneration, thereby achieving a closed-loop appearance generation and optimization that balances creative expression, engineering constraints, physical performance, and manufacturing feasibility.

[0010] To achieve the above technical objectives, this application provides an intelligent appearance generation and optimization method applied in the fields of computer-aided industrial design, computer graphics, generative artificial intelligence, manifold geometry, three-dimensional spatial representation, multiphysics proxy simulation, and intelligent manufacturing, comprising the following steps: Based on the 3D data of the internal core components and the constraints of assembly, heat dissipation, no-entry, stress, manufacturing safety clearance and appearance boundary, a 3D safety envelope manifold is generated. Based on the aforementioned three-dimensional secure envelope manifold, a low-distortion manifold parameter domain is constructed using manifold spectrum operators or conformal energy functions and distortion constraints; Semantic parsing and frequency domain decoupling are performed on creative materials or multimodal creative inputs to obtain multi-scale creative features containing low-frequency shape signals, mid-frequency structure signals and high-frequency texture signals. These features are then mapped to the parameter domain and fused into the manifold to form a cross-modal 3D generation basis. The three-dimensional spatial representation field is initialized using the aforementioned base to generate a multi-view appearance scheme and a three-dimensional appearance model. The model is input into a multiphysics proxy evaluation network to obtain physical and manufacturing performance indicators and residuals. When the residuals exceed the limits, the abnormal region is identified and reversed to locate the abnormal parameter sub-region. The contribution of the three frequency band residuals is calculated, and the frequency domain signal weights are adjusted accordingly, or the local field parameters are further adjusted and the model is regenerated. Based on the optimization of the comprehensive fitness function, the solution outputs the compliant, convergent or Pareto front solution and generates a fingerprint.

[0011] Preferably, when generating the three-dimensional secure envelope manifold, the three-dimensional data of the internal core components includes CAD solid models, mesh models, point cloud models, voxel models, implicit field models, or combinations thereof; the three-dimensional secure envelope manifold is obtained by performing at least one of the following methods on the internal core components, restricted areas, and manufacturing security gaps: distance field dilation, morphological dilation, offset surface generation, implicit isosurface extraction, convex hull, or α-shape envelope generation.

[0012] Preferably, the manifold spectrum operator includes at least one of the following: Laplace-Beltramian operator, discrete cochet Laplace operator, graph Laplace operator, thermal kernel operator, or curvature-weighted spectrum operator; the conformal energy function includes at least one of the following: harmonic energy, least squares conformal mapping energy, Ricci flow energy, or ARAP energy; and the distortion constraint includes at least one of the following: local angular distortion rate, area distortion rate, length distortion rate, harmonic energy, conformal energy, or ARAP energy.

[0013] Preferably, the frequency domain decoupling includes at least one of Fourier decomposition, wavelet decomposition, Laplace pyramid decomposition, diffusion model feature pyramid decomposition, surface spectrum decomposition, curvature spectrum decomposition, or manifold parameter domain spectrum decomposition; the low-frequency shape signal, mid-frequency structure signal, and high-frequency texture signal are divided according to the eigenvalue range of the manifold spectrum operator, the image frequency threshold, the wavelet scale, the Laplace pyramid level, or the local spatial scale.

[0014] Preferably, the creative materials or multimodal creative inputs include at least one of the following: natural biological images, brand logos, hand-drawn sketches, text prompts, user preference images, historical product images, competitor images, market trend images, 3D reference models, or multi-view reference images; when performing semantic parsing on the creative materials or multimodal creative inputs, at least one of the following is used: visual language model, image segmentation model, key point detection model, diffusion model encoder, multimodal feature encoding model, or visual basic model.

[0015] Preferably, when forming a cross-modal three-dimensional generative substrate, the low-frequency shape signal is used to control the overall contour, proportion, or macroscopic streamline; the mid-frequency structure signal is used to control the air guide groove, reinforcing rib, air inlet, heat dissipation gills, local concave and convex or functional structure; and the high-frequency texture signal is used to control the surface texture, microstructure, brand texture, decorative details, or tactile structure.

[0016] Preferably, the three-dimensional spatial characterization field includes at least one of a three-dimensional Gaussian field, a three-dimensional Gaussian sputtering field, a neural radiation field, a neural symbol distance field, an occupancy field, a point primitive field, a differentiable mesh field, or a voxel implicit field.

[0017] Preferably, the multiphysics proxy evaluation network includes at least one of the following: physical information neural network, neural operator, graph neural network proxy model, finite element proxy model, computational fluid dynamics proxy model, thermal simulation proxy model, structural simulation proxy model, or multifidelity proxy model.

[0018] Preferably, the physical performance indicators include at least one of drag coefficient, lift coefficient, pressure distribution, wake intensity, flow separation area, intake efficiency, hydrodynamic resistance, heat dissipation flux, peak temperature, temperature uniformity, thermal resistance, hot spot area, maximum stress, maximum deformation, stress concentration factor, stiffness, or fatigue risk; the manufacturing performance indicators include at least one of material volume, minimum wall thickness, draft angle, support volume, processing path complexity, radius of curvature, mold opening and closing direction, or assembly accessibility; the residuals include physical residuals and / or manufacturing constraint residuals.

[0019] Preferably, when determining abnormal regions, the abnormal regions include regions where the pressure residual exceeds a threshold, flow separation regions, temperature exceeds a threshold, heat flux is insufficient, stress concentration regions, deformation exceeds a threshold, wall thickness is insufficient, draft angle is insufficient, processing is inaccessible regions, or support volume exceeds a threshold.

[0020] Preferably, when adjusting the corresponding frequency domain signal weights, the contribution of the three-band residuals is calculated based on the local amplitude, gradient, energy proportion, curvature contribution, or residual gradient of the low-frequency shape signal, mid-frequency structure signal, and high-frequency texture signal within the abnormal parameter sub-region; and attenuation, enhancement, smoothing, migration, or resampling is performed on the corresponding frequency domain signal weights based on the contribution of the three-band residuals.

[0021] Preferably, the local field parameters of the three-dimensional spatial characterization field include at least one of the following: position, scale, density, opacity, rotation, covariance, or color parameters of the three-dimensional Gaussian field; density, color, or network feature parameters of the neural radiation field; symbol distance value, normal, or local feature parameters of the neural symbol distance field; occupancy probability of the occupancy field; and vertex position, normal, radius of curvature, or local subdivision parameters of the differentiable mesh field.

[0022] Preferably, the comprehensive fitness function includes physical performance indicators and manufacturing feasibility indicators, and further includes appearance consistency indicators or product differentiation indicators characterized by image feature vectors, three-dimensional shape feature vectors, curvature spectrum distance, contour distance, Hausdorff distance, semantic feature distance, material volume, processing path complexity, support volume, mold complexity, or number of assembly steps.

[0023] Preferably, the generated fingerprint includes at least one of the following: creative material feature summary, semantic label, frequency domain signal weight, manifold parameter domain mapping relationship, three-dimensional spatial characterization field parameter summary, physical residual distribution, manufacturing constraint residual distribution, optimization iteration record, final three-dimensional model hash value, curvature spectrum fingerprint, or local angular distortion rate record.

[0024] Based on the same inventive concept, this invention provides an intelligent appearance generation and optimization system, comprising: The safety envelope construction module is used to acquire the 3D data of the core components inside the product, assembly constraints, heat dissipation constraints, restricted areas, structural stress areas, manufacturing safety clearances and appearance design boundaries, and generate a 3D safety envelope manifold. The manifold parameter domain construction module is used to construct a low-distortion manifold parameter domain based on the manifold spectrum operator or conformal energy function calculated from the three-dimensional safe envelope manifold. The creative analysis and frequency domain decoupling module is used to acquire creative materials or multimodal creative inputs, perform semantic analysis and frequency domain decoupling on the creative materials or multimodal creative inputs, and obtain low-frequency shape signals, mid-frequency structure signals and high-frequency texture signals; The cross-modal mapping and fusion module is used to map the low-frequency shape signal, the mid-frequency structure signal and the high-frequency texture signal to the low-distortion manifold parameter domain and fuse them into the three-dimensional secure envelope manifold to form a cross-modal three-dimensional generation basis. The three-dimensional spatial representation generation module is used to initialize the three-dimensional spatial representation field with the cross-modal three-dimensional generation basis and generate a multi-view appearance scheme and a three-dimensional appearance model. The multiphysics evaluation module is used to input the three-dimensional appearance model into the multiphysics proxy evaluation network to obtain physical performance indicators, manufacturing performance indicators, physical residuals and / or manufacturing constraint residuals. The residual reverse correction module is used to determine the abnormal region and reverse locate it as an abnormal parameter sub-region when the physical residual and / or manufacturing constraint residual do not meet the preset conditions, calculate the contribution of low-frequency shape signal, mid-frequency structure signal and high-frequency texture signal to the residual, and adjust the corresponding frequency domain signal weight or local three-dimensional spatial characterization field parameters to regenerate the three-dimensional appearance model. The optimization and fingerprint output module is used to perform multi-objective optimization based on the comprehensive fitness function and output the appearance scheme and its generated fingerprint.

[0025] Based on the same inventive concept, this application also provides a computer device, including at least one processor and a memory communicatively connected to the at least one processor; wherein the memory stores a computer program executable by the at least one processor, and when the computer program is executed by the at least one processor, it implements the method described in any of the above-mentioned embodiments.

[0026] Based on the same inventive concept, this application also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the method described in any of the above-mentioned embodiments.

[0027] The present invention discloses the following technical effects: This invention defines the space that can be generated by the appearance through a three-dimensional safety envelope manifold, which can reduce the risk of the generated appearance intruding into internal components, assembly space, heat dissipation space or creating safety gaps.

[0028] This invention maps creative materials or multimodal creative inputs to a three-dimensional secure envelope manifold through a low-distortion manifold parameter domain, enabling the overall contour, functional structure, and surface texture of the creative materials to be controlled separately in the form of low-frequency shape signals, mid-frequency structure signals, and high-frequency texture signals.

[0029] This invention uses a three-dimensional spatial characterization field to support a cross-modal three-dimensional generation substrate, which can generate spatially consistent multi-view appearance schemes and three-dimensional appearance models, improving the three-dimensional consistency and editability of the generated appearance.

[0030] This invention obtains fluid, thermal, structural, and manufacturing constraint residuals through a multiphysics proxy evaluation network, and reverse-locates the abnormal region to the low-distortion manifold parameter domain and the corresponding frequency domain signal, thereby realizing closed-loop correction of the creative frequency domain signal by physical residuals or manufacturing constraint residuals.

[0031] This invention generates fingerprint records of creative features, frequency domain weights, manifold mapping relationships, three-dimensional spatial representation parameters, residual distributions, and model hash values, which can be used for design process traceability, model consistency verification, version management, and similarity comparison. Attached Figure Description

[0032] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly described below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0033] Figure 1 This is the overall flowchart of the intelligent appearance generation and optimization method described in this invention.

[0034] Figure 2 This is a schematic diagram of the three-dimensional secure envelope manifold, internal core components, forbidden regions, low-distortion manifold parameter domain mapping, and three-dimensional appearance model in an embodiment of the present invention.

[0035] Figure 3 This is a schematic diagram illustrating the semantic parsing and frequency domain decoupling principle of creative materials described in this invention.

[0036] Figure 4 This is a schematic diagram illustrating the principle of mapping multi-scale creative features to the low-distortion manifold parameter domain and fusing them into a three-dimensional secure envelope manifold, as described in this invention.

[0037] Figure 5 This is a schematic diagram illustrating the principle of the three-dimensional spatial representation field generation multi-view appearance scheme and three-dimensional appearance model described in this invention.

[0038] Figure 6 This is a schematic diagram illustrating the principle of the multiphysics proxy evaluation network described in this invention for generating physical residuals and creating constraint residuals.

[0039] Figure 7 This is a schematic diagram illustrating the principle of the present invention: the physical residual or manufacturing constraint residual is reverse-positioned to the low-distortion manifold parameter domain and frequency domain signal and then closed-loop corrected.

[0040] Figure 8 This is a schematic diagram illustrating the principles of fingerprint generation, curvature spectrum features, and model similarity comparison as described in this invention.

[0041] in, Figure 2 In the diagram, 201 is the three-dimensional safety envelope manifold; 202 is the assembly clearance area or functional opening area; 203 is the area affected by the mid-frequency structural signal; 204 is the parameter line formed by mapping the low-distortion manifold parameter domain; 205 is the area affected by the low-frequency shape signal; 206 is the internal core component; 207 is the manufacturing safety clearance or assembly safety clearance; 208 is the restricted area or engineering clearance area; and 209 is the three-dimensional appearance model.

[0042] Figure 3 In the diagram, 301 represents creative material or multimodal creative input; 302 represents the semantic parsing model; 303 represents the low-frequency shape signal; 304 represents the mid-frequency structure signal; 305 represents the high-frequency texture signal; and 306 represents the multi-scale creative features.

[0043] Figure 4 In the diagram, 401 represents the low-distortion manifold parameter domain; 402 represents the boundary anchor point; 403 represents the orientation field constraint; 404 represents the curvature continuity constraint; and 405 represents the fused cross-modal 3D generated basis.

[0044] Figure 5 In this diagram, 501 represents the cross-modal 3D generation basis; 502 represents the 3D spatial representation field; 503 represents the differentiable rendering module; 504 represents the multi-view appearance scheme; and 505 represents the 3D appearance model.

[0045] Figure 6 In the diagram, 601 represents the three-dimensional appearance model; 602 represents the multiphysics proxy evaluation network; 603 represents the fluid residual distribution; 604 represents the thermal residual distribution; 605 represents the structural residual distribution; and 606 represents the manufacturing constraint residual distribution.

[0046] Figure 7 In the diagram, 701 represents the three-dimensional anomaly region; 702 represents the anomaly parameter sub-region; 703 represents the contribution of low-frequency signals; 704 represents the contribution of mid-frequency signals; 705 represents the contribution of high-frequency signals; 706 represents the frequency domain weight adjustment module; 707 represents the local field parameter adjustment module; and 708 represents the regenerated three-dimensional appearance model.

[0047] Figure 8 In the diagram, 801 is the feature summary of the creative material; 802 is the frequency domain signal weight record; 803 is the manifold parameter domain mapping relationship; 804 is the three-dimensional spatial characterization field parameter summary; 805 is the residual distribution record; 806 is the curvature spectrum fingerprint; 807 is the final three-dimensional model hash value; and 808 is the generated fingerprint. Detailed Implementation

[0048] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0049] like Figures 1 to 8 As shown, this invention provides a method, system, device, and medium for intelligent generation and optimization of industrial product appearance design, applicable to the fields of computer-aided industrial design, computer graphics, generative artificial intelligence, manifold geometry, three-dimensional spatial representation, multiphysics proxy simulation, and intelligent manufacturing. It includes steps such as three-dimensional secure envelope manifold construction, low-distortion manifold parameter domain construction, creative material semantic parsing and frequency domain decoupling, cross-modal three-dimensional generation basis construction, three-dimensional spatial representation field generation, multiphysics and manufacturing feasibility evaluation, residual reverse positioning correction, multi-objective optimization, and fingerprint generation output.

[0050] The following describes the invention using the intelligent generation of the bionic shell of a racing drone as a specific example.

[0051] like Figure 2 As shown, the three-dimensional appearance model 209 is generated within the appearance-generable space defined by the three-dimensional safety envelope manifold 201. The three-dimensional safety envelope manifold 201 can be generated based on the internal core components 206, assembly constraints, heat dissipation constraints, restricted areas or engineering avoidance areas 208, structural stress areas, manufacturing safety clearances or assembly safety clearances 207, and appearance design boundaries. Figure 2The projection boundary of the three-dimensional safe envelope manifold 201 in the current view is indicated by a thick dashed line. The three-dimensional appearance model 209 is generated, deformed and optimized within the space defined by the three-dimensional safe envelope manifold 201.

[0052] Figure 2 The parameter line 204 represents the isoparametric lines, surface spectrum contour lines, or parametric mesh lines formed after the low-distortion manifold parameter domain is mapped onto the surface of the 3D safety envelope manifold 201 or the 3D appearance model 209. It is used to characterize the low-distortion mapping relationship of creative features on the 3D surface, and is not simply a decorative texture. The low-frequency shape signal action area 205 is used to control the overall contour, proportion, and macroscopic streamlines; the mid-frequency structure signal action area 203 is used to form guide channels, reinforcing ribs, heat dissipation openings, heat dissipation gills, or local concave-convex structures; the assembly avoidance area or functional opening area 202 is used to meet the requirements of camera exposure, sensor exposure, heat dissipation, fastener installation, assembly maintenance, or functional interface layout.

[0053] In one embodiment, the system first acquires a 3D CAD model of the core internal components of the racing drone. This 3D CAD model can be in STEP, IGES, OBJ, STL, or other data formats suitable for 3D geometric calculations. The core internal components may include a flight control motherboard, battery, motor mount, camera module, propeller protection area, antenna, heat sink, connecting bracket, and fasteners. The system also acquires assembly constraints, heat dissipation constraints, restricted areas, structural stress areas, manufacturing safety clearances, and appearance design boundaries. Manufacturing constraints may include minimum wall thickness thresholds, minimum draft angle thresholds, radius of curvature thresholds, machining path constraints, mold opening and closing directions, and assembly accessibility requirements.

[0054] Step S1: Construction of 3D Secure Envelope Manifold In one embodiment, step S1 is used to generate a three-dimensional safe envelope manifold of the appearance-generating space based on the product's internal structure and engineering constraints.

[0055] For example, the system places the three-dimensional data of the core internal components of the UAV in a unified coordinate system, wherein the core internal components can be as follows: Figure 2 The internal core component 206 is shown in the figure, and a distance field is generated based on the component's outer surface, assembly gap, heat dissipation space, restricted area or engineering avoidance area 208, and manufacturing safety gap or assembly safety gap 207.

[0056] For example, let the symbolic distance field corresponding to the surface of the internal component be SDF(x,y,z). The system determines the comprehensive safety distance D_safe based on the minimum wall thickness threshold D_wall, the heat dissipation gap threshold D_heat, the assembly gap threshold D_asm, and the manufacturing safety margin D_mfg. The system can traverse the sampling points in the computation space, determine the region that satisfies SDF(x,y,z)≤D_safe as the internal safety occupied region, and use the isosurface of SDF(x,y,z)=D_safe as the three-dimensional safety envelope manifold 201. Figure 2 The projection boundary of the three-dimensional safety envelope manifold 201 in the current viewpoint is indicated by a thick dashed line. The three-dimensional safety envelope manifold 201 is used to limit the space defined by the internal core component 206, the manufacturing safety clearance or assembly safety clearance 207, and the restricted area or engineering avoidance area 208, which are subsequently generated.

[0057] For example, the three-dimensional safety envelope manifold can also be obtained by performing morphological dilation, bias surface generation, implicit isosurface extraction, convex hull envelope, or α-shape envelope generation on the surfaces of internal components. For the UAV shell, the three-dimensional safety envelope manifold can also set local non-intrusive boundaries based on the motor mount, camera field of view, propeller safety area, and heat dissipation exhaust area.

[0058] Step S2: Construction of the low-distortion manifold parameter domain In one embodiment, step S2 is used to map the three-dimensional secure envelope manifold to a low-distortion manifold parameter domain suitable for carrying creative frequency domain signals.

[0059] For example, the system constructs a discrete triangular mesh based on the three-dimensional secure envelope manifold and calculates the manifold spectrum operator. The manifold spectrum operator can be the Laplace-Beltramian operator, the discrete cotangent Laplace operator, the graphical Laplace operator, the thermal kernel operator, or the curvature-weighted spectrum operator. The system can also calculate the conformal energy function, which can be the harmonic energy, the least-squares conformal mapping energy, the Ricci flow energy, or the ARAP energy.

[0060] For example, the system uses at least one of the following distortion constraints—local angular distortion rate, area distortion rate, length distortion rate, harmonic energy, conformal energy, or ARAP energy—to map the three-dimensional safe envelope manifold 201 into a two-dimensional or low-dimensional low-distortion manifold parameter domain. This low-distortion manifold parameter domain maintains the correspondence between the local structure of the three-dimensional surface and the parameter domain, facilitating the stable mapping of multi-scale features from creative materials or multimodal creative inputs to the three-dimensional appearance surface. For example... Figure 2 As shown, parameter line 204 is used to illustrate the mapping relationship between the low-distortion manifold parameter domain and the three-dimensional safe envelope manifold 201 or the three-dimensional appearance model 209.

[0061] For example, the system stores the mapping relationship between the three-dimensional safe envelope manifold 201 and the parameter domain of the low-distortion manifold. This mapping relationship includes a one-to-one correspondence, an approximate correspondence, a UV mapping relationship, a barycenter coordinate relationship, or a nearest surface point mapping relationship between three-dimensional surface points and parameter domain coordinates. This mapping relationship can be generated by... Figure 2 The parameter line 204 is used as an illustration and is used to subsequently locate abnormal areas from the three-dimensional appearance model 209 to the parameter domain and frequency domain signals.

[0062] Step S3: Semantic analysis and frequency domain decoupling of creative materials In one embodiment, step S3 is used to acquire creative materials or multimodal creative inputs and decouple them into low-frequency shape signals, mid-frequency structure signals and high-frequency texture signals.

[0063] For example, in the scenario of drone exterior design, creative materials may include images of raptor wings, shark skin textures, racing car air intakes, brand logos, designer sketches, text prompts, multi-view reference images, or historical drone shell models. The system uses a visual language model, image segmentation model, key point detection model, diffusion model encoder, multimodal feature encoding model, or visual foundation model to perform semantic parsing on the creative materials, obtaining the overall outline, airfoil boundaries, air intake structures, heat dissipation openings, brand decoration areas, and surface micro-texture areas.

[0064] For example, the system performs frequency domain decoupling on creative materials in the image domain, feature domain, surface spectrum domain, or manifold parameter domain. The frequency domain decoupling may employ at least one of Fourier decomposition, wavelet decomposition, Laplace pyramid decomposition, diffusion model feature pyramid decomposition, surface spectrum decomposition, curvature spectrum decomposition, or manifold parameter domain spectrum decomposition.

[0065] For example, the system classifies the features of the overall shell outline, length-width-height ratio, head sharpness and macro-streamline of the control drone into low-frequency shape signals 303; classifies the features of the control air guide groove, air inlet, heat dissipation gills, reinforcing ribs, local concave-convex and functional openings into mid-frequency structure signals 304; and classifies the features of the control surface micro-texture, brand texture, bionic skin texture and tactile structure into high-frequency texture signals 305.

[0066] In another implementation, when the creative input is a text prompt, the system first encodes the text prompt into a semantic feature vector or conditional feature map, and then combines it with diffusion model priors, reference images, or shape priors to form decomposable multi-scale features. For a three-dimensional reference model, the system can perform low-frequency, mid-frequency, and high-frequency decomposition in the surface spectral domain, curvature spectral domain, or manifold parameter domain.

[0067] Step S4: Cross-modal mapping and 3D generative basis formation In one embodiment, step S4 is used to map multi-scale creative features to a low-distortion manifold parameter domain and fuse them into a three-dimensional secure envelope manifold.

[0068] For example, the system projects the low-frequency shape signal 303, the mid-frequency structure signal 304, and the high-frequency texture signal 305 onto the low-distortion manifold parameter domain 401. During the projection process, the system can set boundary anchor points 402 to keep the positions of the UAV battery compartment, camera opening, motor mount area, and heat dissipation opening stable; the system can also set directional field constraints 403 to control the flow channels to be arranged along a preset airflow direction; the system can also set curvature continuity constraints 404 to avoid the appearance of discontinuous peaks or abrupt structures that are not conducive to manufacturing on local surfaces.

[0069] For example, the system fuses low-frequency shape signals, mid-frequency structure signals and high-frequency texture signals into a three-dimensional secure envelope manifold by at least one of harmonic mapping, boundary anchor point constraints, orientation field constraints, Poisson constraints, angular distortion penalties or curvature continuity penalties, to form a cross-modal three-dimensional generative substrate 405.

[0070] The cross-modal 3D generation substrate retains the appearance style and multi-scale structure of the creative material, while being constrained by the engineering boundary of the 3D safe envelope manifold. Therefore, it can be used as the initialization condition for the subsequent generation of the 3D spatial representation field.

[0071] Step S5: Generation of three-dimensional spatial representation field In one embodiment, step S5 is used to initialize the three-dimensional spatial representation field with a cross-modal three-dimensional generation basis and generate a multi-view appearance scheme and a three-dimensional appearance model.

[0072] For example, the system initializes the three-dimensional spatial representation field 502 with a cross-modal three-dimensional generation substrate 501. The three-dimensional spatial representation field can be at least one of a three-dimensional Gaussian field, a three-dimensional Gaussian sputtering field, a neural radiation field, a neural symbol distance field, an occupancy field, a point primitive field, a differentiable mesh field, or a voxel implicit field.

[0073] In one embodiment employing a three-dimensional Gaussian field, the system samples the three-dimensional secure envelope manifold and its outer candidate appearance regions as multiple Gaussian elements, and encodes the low-frequency shape signal, mid-frequency structure signal, and high-frequency texture signal as initialization conditions for the position, scale, covariance, opacity, color, normal, or local density parameters of the Gaussian elements.

[0074] In one embodiment employing a neural symbolic distance field, the system encodes the cross-modal 3D generation basis as a local feature vector of the symbolic distance function and generates a 3D appearance surface by isosurface extraction.

[0075] In one embodiment employing a differentiable mesh field, the system uses a three-dimensional secure envelope manifold as the initial mesh and drives changes in vertex position, normal, radius of curvature, and local subdivision parameters based on multi-scale creative features.

[0076] For example, the system generates a multi-view appearance scheme 504 and a three-dimensional appearance model 505 constrained by the same three-dimensional spatial representation field through a differentiable rendering module 503 or differentiable geometry optimization. Figure 2 In the illustrated embodiment, the three-dimensional appearance model corresponds to the three-dimensional appearance model 209, and its generation process is jointly constrained by the three-dimensional safety envelope manifold 201, the internal core component 206, the manufacturing safety clearance or assembly safety clearance 207, and the restricted area or engineering avoidance area 208. Since the multi-view images are all rendered or optimized from the same three-dimensional spatial representation field, contour misalignment, local component drift, and texture inconsistency between different viewpoints can be reduced.

[0077] Step S6: Multiphysics and Manufacturing Feasibility Assessment In one embodiment, step S6 is used to input the three-dimensional appearance model into the multiphysics proxy evaluation network and generate physical residuals or manufacturing constraint residuals.

[0078] For example, the system converts the 3D appearance model 601 into at least one input feature selected from point cloud, mesh, voxel, implicit field sampling, curvature distribution, normal distribution, thickness distribution, air inlet location, heat dissipation area location, or material properties, and inputs it into a multiphysics proxy evaluation network 602. The multiphysics proxy evaluation network may include a physical information neural network, a neural operator, a graph neural network proxy model, a finite element proxy model, a computational fluid dynamics proxy model, a thermal simulation proxy model, a structural simulation proxy model, or a multifidelity proxy model.

[0079] For example, in the case of a drone, the system outputs fluid performance indicators, including drag coefficient, lift coefficient, pressure distribution, wake intensity, flow separation area, and intake efficiency; the system also outputs thermal performance indicators, including heat dissipation flux, peak temperature, temperature uniformity, thermal resistance, and hot spot area; the system further outputs structural performance indicators, including maximum stress, maximum deformation, stress concentration factor, stiffness, and fatigue risk; the system also outputs manufacturing performance indicators, including material volume, minimum wall thickness, draft angle, support volume, processing path complexity, radius of curvature, mold opening and closing direction, and assembly accessibility.

[0080] For example, when a sudden increase in pressure or flow separation occurs in the windward area, the system generates a fluid residual distribution 603; when there is insufficient heat dissipation on the surface near the battery or motherboard, the system generates a thermal residual distribution 604; when stress concentration occurs near the motor mount or connecting bracket, the system generates a structural residual distribution 605; when there is insufficient wall thickness, insufficient draft angle, or unreachable machining in a local area, the system generates a manufacturing constraint residual distribution 606.

[0081] Step S7: Residual reverse positioning and frequency domain weight correction In one embodiment, step S7 is used to reverse the location of the abnormal region to the low distortion manifold parameter domain and the corresponding frequency domain signal when the physical residual or manufacturing constraint residual does not meet the preset conditions, and to perform closed-loop correction.

[0082] For example, the system determines a three-dimensional abnormal region 701 based on surface points, mesh cells, voxel cells, or spatial characterization field sampling points where the residual exceeds a threshold. The abnormal region may include regions where the pressure residual exceeds a threshold, flow separation regions, temperature regions exceeding a threshold, insufficient heat flux regions, stress concentration regions, deformation regions exceeding a threshold, insufficient wall thickness regions, insufficient draft angle regions, unprocessable regions, or support volume regions exceeding a threshold.

[0083] For example, the system utilizes the mapping relationship between the three-dimensional secure envelope manifold 201 and the low-distortion manifold parameter domain to map the three-dimensional anomalous region 701 to an anomalous parameter sub-region 702 in the low-distortion manifold parameter domain. The mapping relationship can be derived from... Figure 2 As shown in parameter line 204, the mapping method may include at least one of UV mapping, barycentric coordinate mapping, nearest surface point mapping, differentiable rendering inverse mapping, or implicit field gradient projection.

[0084] For example, the system calculates the contribution of low-frequency shape signal, mid-frequency structure signal, and high-frequency texture signal to physical residual or manufacturing constraint residual within the abnormal parameter sub-region 702. The contribution can be calculated based on local amplitude, gradient, energy proportion, curvature contribution, or residual gradient.

[0085] In one implementation, let the weights of the low-frequency, mid-frequency, and high-frequency signals in the t-th iteration be w_L(t), w_M(t), and w_H(t), respectively; let the sub-region of the outlier parameter be Ω_a; let the comprehensive residual be R(u); and let the low-frequency, mid-frequency, and high-frequency signals be F_L(u), F_M(u), and F_H(u), respectively. For frequency band k∈{L,M,H}, the system can calculate the contribution as follows: C_k = ∫Ω_a | F_k(u)|·R(u)du / Σ_j∫Ω_a | F_j(u)|·R(u)du Where C_k represents the contribution of the k-th frequency band to the abnormal residual. The system can update the frequency domain signal weights as follows: w_k(t+1)=clip[w_k(t)-η·C_k·R_mean,w_k,min,w_k,max] Where η is the update step size, R_mean is the average residual within the outlier parameter sub-region, and clip indicates that the weights are limited to a preset range.

[0086] For example, when a high-frequency biomimetic texture in the high-pressure area facing the drone causes a sudden increase in local pressure, the system can reduce the weight of the high-frequency texture signal, or adopt the following attenuation method: w_H(t+1)=w_H(t)·exp(-β·R_p) Where R_p is the pressure residual and β is the attenuation coefficient. This method allows for the automatic attenuation of high-frequency textures in the high-pressure, windward region, resulting in a smoother guiding surface.

[0087] For example, when flow separation is primarily caused by improper placement of the intermediate frequency (IF) channel, the system can smooth, migrate, or resample the IF structural signal to adjust the channel position or flow direction. When insufficient heat dissipation occurs near the motherboard or battery area, the system can enhance the IF heat dissipation opening signal or migrate the heat dissipation gill position. When stress concentration occurs near the connecting bracket, the system can adjust the low-frequency profile, increase the local radius of curvature, or enhance the IF reinforcing rib signal. When manufacturing constraint residuals indicate insufficient local wall thickness or draft angle, the system can reduce the high-frequency texture amplitude, smooth the IF structure, or adjust the position of the differential mesh vertices.

[0088] For example, the system can further adjust the local field parameters corresponding to the anomalous region in the three-dimensional spatial representation field. For a three-dimensional Gaussian field, the local field parameters may include position, scale, density, opacity, rotation, covariance, or color; for a neural symbol distance field, the local field parameters may include symbol distance value, normal, or local features; for an occupancy field, the local field parameters may include occupancy probability; for a differentiable mesh field, the local field parameters may include vertex position, normal, radius of curvature, or local subdivision parameters.

[0089] After adjusting the frequency domain signal weights or local field parameters as described above, the system regenerates the three-dimensional appearance model and performs multiphysics and manufacturing feasibility evaluation again until the preset conditions are met, the number of iterations is reached, or the comprehensive fitness function converges. Figure 2In the structure shown, when there are abnormal regions on the surface of the 3D appearance model 209 where physical residuals or manufacturing constraint residuals exceed a threshold, the system can, based on the mapping relationship represented by parameter line 204, reverse-locate the abnormal regions to abnormal parameter sub-regions in the low-distortion manifold parameter domain, and determine the contribution of the low-frequency shape signal action region 205, the mid-frequency structure signal action region 203, and the high-frequency texture signal to the residuals. After adjusting the corresponding frequency domain signal weights according to the contribution, the system regenerates the 3D appearance model 209 within the space defined by the 3D safety envelope manifold 201.

[0090] Step S8: Multi-objective optimization and fingerprint generation output In one embodiment, step S8 is used to perform multi-objective optimization based on the comprehensive fitness function and output the appearance scheme and generate fingerprint.

[0091] For example, the system constructs a comprehensive fitness function. The comprehensive fitness function includes at least physical performance indicators and manufacturing feasibility indicators, and may further include appearance consistency indicators or product differentiation indicators characterized by image feature vectors, three-dimensional shape feature vectors, curvature spectrum distance, contour distance, Hausdorff distance, semantic feature distance, material volume, processing path complexity, support volume, mold complexity, or number of assembly steps.

[0092] For example, the system may employ at least one of Pareto front optimization, Bayesian optimization, reinforcement learning, genetic algorithms, simulated annealing, multi-agent search, or gradient descent to perform optimization. In the UAV embodiment, the system may perform multi-objective trade-offs among drag coefficient, heat dissipation efficiency, maximum stress, material volume, minimum wall thickness, draft angle, and appearance feature retention, outputting an appearance scheme that satisfies preset constraints and whose overall fitness value reaches a preset threshold, meets iterative convergence conditions, or is located on the Pareto front.

[0093] For example, in one embodiment, after multiple rounds of closed-loop correction, the system can reduce the high-frequency texture amplitude in the high-pressure area facing the wind on the drone's shell, make the guide channel position closer to the low-pressure guide area, increase the area of ​​the heat dissipation gills near the battery compartment, increase the radius of curvature near the connecting bracket, and ensure that the local wall thickness and draft angle meet the manufacturing threshold. Compared to the initial generation scheme, the system can reduce the number of manual modifications to the prototype and improve the usability of the appearance model to enter the engineering review stage.

[0094] For example, the generated fingerprint 808 output by the system may include at least one of the following: creative material feature summary 801, semantic tag, frequency domain signal weight record 802, manifold parameter domain mapping relationship 803, three-dimensional spatial characterization field parameter summary 804, physical residual distribution, manufacturing constraint residual distribution 805, optimization iteration record, final three-dimensional model hash value 807, curvature spectrum fingerprint 806, and local angular distortion rate record. The generated fingerprint can be used for model consistency verification, version management, design process traceability, and similarity comparison.

[0095] In another embodiment, the present invention can be applied to the generation of the front appearance of new energy vehicles. The system acquires three-dimensional data of the front bumper, radar, camera, air intake channel, light assembly, radiator, and body connection structure, and generates a three-dimensional safe envelope manifold. The system decouples the brand logo, competitor front appearance image, and aerodynamic guide image into a low-frequency front appearance proportion signal, a mid-frequency air intake grille and guide structure signal, and a high-frequency decorative texture signal. The system generates a three-dimensional appearance model of the front face through a three-dimensional spatial representation field, and evaluates wind resistance, air intake efficiency, and pressure distribution through a multi-physics proxy evaluation network. When the local high-frequency texture of the grille causes an increase in pressure residual, the system reverses the location of the abnormal area to the parameter domain and the high-frequency texture signal, reduces the corresponding weight, and regenerates it, thereby obtaining an appearance scheme that meets brand characteristics, air intake efficiency, and manufacturing constraints.

[0096] In another embodiment, the present invention can be applied to the generation of consumer electronics casings. The system acquires three-dimensional data of the battery, motherboard, camera module, screen module, speaker, antenna, and heat sink, and generates a three-dimensional secure envelope manifold. The system decouples brand patterns, user preference images, and historical product images into low-frequency shape rounded corner ratio signals, mid-frequency button and opening structure signals, and high-frequency surface texture signals. The system evaluates temperature peaks, hot spot areas, minimum wall thickness, undercut areas, and processing accessibility through a multi-physics proxy evaluation network. When the heat source corresponding area has insufficient heat dissipation, the system enhances the mid-frequency heat dissipation structure signal or relocates the heat dissipation opening position. When the high-frequency texture causes insufficient local wall thickness, the system reduces the high-frequency texture amplitude and regenerates the casing model.

[0097] It should be understood that although the steps in the flowcharts of this application's embodiments are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some of the steps in the figures may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the sub-steps or stages of other steps. Obviously, those skilled in the art can make various modifications and variations to this invention without departing from the spirit and scope of this invention. Thus, if these modifications and variations of this invention fall within the scope of the claims of this invention and their equivalents, then this invention also intends to include these modifications and variations.

Claims

1. A method for intelligent generation and optimization of appearance, characterized in that, include: Based on the 3D data of the internal core components and the constraints of assembly, heat dissipation, no-entry, stress, manufacturing safety clearance and appearance boundary, a 3D safety envelope manifold is generated. Based on the aforementioned three-dimensional secure envelope manifold, a low-distortion manifold parameter domain is constructed using manifold spectrum operators or conformal energy functions and distortion constraints; Semantic parsing and frequency domain decoupling are performed on creative materials or multimodal creative inputs to obtain multi-scale creative features containing low-frequency shape signals, mid-frequency structure signals and high-frequency texture signals. These features are then mapped to the parameter domain and fused into the manifold to form a cross-modal 3D generation basis. The three-dimensional spatial representation field is initialized using the aforementioned base to generate a multi-view appearance scheme and a three-dimensional appearance model. The model is input into a multiphysics proxy evaluation network to obtain physical and manufacturing performance indicators and residuals. When the residuals exceed the limits, the abnormal region is identified and reversed to locate the abnormal parameter sub-region. The contribution of the three frequency band residuals is calculated, and the frequency domain signal weights are adjusted accordingly, or the local field parameters are further adjusted and the model is regenerated. Based on the optimization of the comprehensive fitness function, the solution outputs the compliant, convergent or Pareto front solution and generates a fingerprint.

2. The appearance intelligent generation and optimization method according to claim 1, characterized in that: When generating a three-dimensional secure envelope manifold, the three-dimensional data of the internal core components include CAD solid models, mesh models, point cloud models, voxel models, implicit field models, or combinations thereof; the three-dimensional secure envelope manifold is obtained by performing at least one of the following methods on the internal core components, restricted areas, and manufacturing security gaps: distance field dilation, morphological dilation, offset surface generation, implicit isosurface extraction, convex hull, or α-shape envelope generation.

3. The intelligent appearance generation and optimization method according to claim 1, characterized in that: The manifold spectrum operator includes at least one of the following: Laplace-Beltramian operator, discrete cochet Laplace operator, graph Laplace operator, thermal kernel operator, or curvature-weighted spectrum operator; the conformal energy function includes at least one of the following: harmonic energy, least squares conformal mapping energy, Ricci flow energy, or ARAP energy; the distortion constraint includes at least one of the following: local angular distortion rate, area distortion rate, length distortion rate, harmonic energy, conformal energy, or ARAP energy.

4. The intelligent appearance generation and optimization method according to claim 1, characterized in that: The frequency domain decoupling includes at least one of Fourier decomposition, wavelet decomposition, Laplacian pyramid decomposition, diffusion model feature pyramid decomposition, surface spectrum decomposition, curvature spectrum decomposition, or manifold parameter domain spectrum decomposition; the low-frequency shape signal, mid-frequency structure signal, and high-frequency texture signal are divided according to the eigenvalue range of the manifold spectrum operator, the image frequency threshold, the wavelet scale, the Laplacian pyramid level, or the local spatial scale.

5. The intelligent appearance generation and optimization method according to claim 1, characterized in that: The creative materials or multimodal creative inputs include at least one of the following: natural biological images, brand logos, hand-drawn sketches, text prompts, user preference images, historical product images, competitor images, market trend images, 3D reference models, or multi-view reference images; when performing semantic parsing on the creative materials or multimodal creative inputs, at least one of the following is used: visual language model, image segmentation model, key point detection model, diffusion model encoder, multimodal feature encoding model, or visual basic model.

6. The intelligent appearance generation and optimization method according to claim 1, characterized in that: When forming a cross-modal 3D generative substrate, the low-frequency shape signal is used to control the overall outline, proportions, or macroscopic streamlines; the mid-frequency structure signal is used to control the air guide groove, reinforcing ribs, air inlet, heat dissipation gills, local concave and convex or functional structures; and the high-frequency texture signal is used to control the surface texture, microstructure, brand texture, decorative details, or tactile structure.

7. The intelligent appearance generation and optimization method according to claim 1, characterized in that: The three-dimensional spatial characterization field includes at least one of the following: a three-dimensional Gaussian field, a three-dimensional Gaussian sputtering field, a neural radiation field, a neural symbol distance field, an occupancy field, a point primitive field, a differentiable mesh field, or a voxel implicit field.

8. The intelligent appearance generation and optimization method according to claim 1, characterized in that: The multiphysics proxy evaluation network includes at least one of the following: physical information neural network, neural operator, graph neural network proxy model, finite element proxy model, computational fluid dynamics proxy model, thermal simulation proxy model, structural simulation proxy model, or multifidelity proxy model.

9. The intelligent appearance generation and optimization method according to claim 1, characterized in that: The physical performance indicators include at least one of the following: drag coefficient, lift coefficient, pressure distribution, wake intensity, flow separation area, intake efficiency, hydrodynamic resistance, heat dissipation flux, peak temperature, temperature uniformity, thermal resistance, hot spot area, maximum stress, maximum deformation, stress concentration factor, stiffness, or fatigue risk; the manufacturing performance indicators include at least one of the following: material volume, minimum wall thickness, draft angle, support volume, processing path complexity, radius of curvature, mold opening and closing direction, or assembly accessibility; the residuals include physical residuals and / or manufacturing constraint residuals.

10. The intelligent appearance generation and optimization method according to claim 1, characterized in that: When identifying abnormal regions, the abnormal regions include regions where the pressure residual exceeds a threshold, flow separation regions, temperature exceeding a threshold, insufficient heat flux, stress concentration regions, deformation exceeding a threshold, insufficient wall thickness, insufficient draft angle, unreachable processing regions, or support volume exceeding a threshold.

11. The intelligent appearance generation and optimization method according to claim 1, characterized in that: When adjusting the corresponding frequency domain signal weights, the contribution of the three-band residuals is calculated based on the local amplitude, gradient, energy proportion, curvature contribution, or residual gradient of the low-frequency shape signal, mid-frequency structure signal, and high-frequency texture signal within the abnormal parameter sub-region; and attenuation, enhancement, smoothing, migration, or resampling is performed on the corresponding frequency domain signal weights based on the contribution of the three-band residuals.

12. The intelligent appearance generation and optimization method according to claim 1, characterized in that: The local field parameters of the three-dimensional spatial characterization field include the position, scale, density, opacity, rotation, covariance, or color parameters of the three-dimensional Gaussian field; the density, color, or network feature parameters of the neural radiation field; the symbol distance value, normal, or local feature parameters of the neural symbol distance field; the occupancy probability of the occupancy field; and at least one of the vertex position, normal, radius of curvature, or local subdivision parameters of the differentiable mesh field.

13. The intelligent appearance generation and optimization method according to claim 1, characterized in that: The comprehensive fitness function includes physical performance indicators and manufacturing feasibility indicators, and further includes appearance consistency indicators or product differentiation indicators characterized by image feature vectors, three-dimensional shape feature vectors, curvature spectrum distance, contour distance, Hausdorff distance, semantic feature distance, material volume, processing path complexity, support volume, mold complexity, or number of assembly steps.

14. The intelligent appearance generation and optimization method according to claim 1, characterized in that: The generated fingerprint includes at least one of the following: creative material feature summary, semantic label, frequency domain signal weight, manifold parameter domain mapping relationship, three-dimensional spatial characterization field parameter summary, physical residual distribution, manufacturing constraint residual distribution, optimization iteration record, final three-dimensional model hash value, curvature spectrum fingerprint, or local angular distortion rate record.

15. An intelligent appearance generation and optimization system, characterized in that, The method for implementing the intelligent appearance generation and optimization method as described in any one of claims 1 to 14 includes: The safety envelope construction module is used to acquire the 3D data of the core components inside the product, assembly constraints, heat dissipation constraints, restricted areas, structural stress areas, manufacturing safety clearances and appearance design boundaries, and generate a 3D safety envelope manifold. The manifold parameter domain construction module is used to construct a low-distortion manifold parameter domain based on the manifold spectrum operator or conformal energy function calculated from the three-dimensional safe envelope manifold. The creative analysis and frequency domain decoupling module is used to acquire creative materials or multimodal creative inputs, perform semantic analysis and frequency domain decoupling on the creative materials or multimodal creative inputs, and obtain low-frequency shape signals, mid-frequency structure signals and high-frequency texture signals; The cross-modal mapping and fusion module is used to map the low-frequency shape signal, the mid-frequency structure signal and the high-frequency texture signal to the low-distortion manifold parameter domain and fuse them into the three-dimensional secure envelope manifold to form a cross-modal three-dimensional generation basis. The three-dimensional spatial representation generation module is used to initialize the three-dimensional spatial representation field with the cross-modal three-dimensional generation basis and generate a multi-view appearance scheme and a three-dimensional appearance model. The multiphysics evaluation module is used to input the three-dimensional appearance model into the multiphysics proxy evaluation network to obtain physical performance indicators, manufacturing performance indicators, physical residuals and / or manufacturing constraint residuals. The residual reverse correction module is used to determine the abnormal region and reverse locate it as an abnormal parameter sub-region when the physical residual and / or manufacturing constraint residual do not meet the preset conditions, calculate the contribution of low-frequency shape signal, mid-frequency structure signal and high-frequency texture signal to the residual, and adjust the corresponding frequency domain signal weight or local three-dimensional spatial characterization field parameters to regenerate the three-dimensional appearance model. The optimization and fingerprint output module is used to perform multi-objective optimization based on the comprehensive fitness function and output the appearance scheme and its generated fingerprint.

16. A computer device, characterized in that, include: At least one processor; And a memory communicatively connected to the at least one processor; wherein the memory stores a computer program executable by the at least one processor, which, when executed by the at least one processor, implements the method as described in any one of claims 1 to 14.

17. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program that, when executed by a processor, implements the method as described in any one of claims 1 to 14.

Images