An aircraft lightweight biomimetic skeleton structure and a preparation and design method thereof

Abstract

This invention discloses a lightweight biomimetic skeleton structure for aircraft and its fabrication and design methods, belonging to the field of biomimetic aircraft technology. The lightweight biomimetic skeleton structure includes: a biomimetic skeleton structure composed of three periodic minimal curved surfaces; and a panel layer integrally connected to both sides of the biomimetic skeleton structure, the panel layer being a dense structure or a microporous structure. The biomimetic skeleton structure is constructed based on the topological features of the three-dimensional structure of human trabeculae, and a trabeculae-like microtopological structure is introduced into the local high-stress areas of the biomimetic skeleton structure. By mimicking the porous structure of human trabeculae, a highly efficient load-bearing lattice structure is used for the lightweight design and manufacturing of key load-bearing components of aircraft equipment. This achieves both structural lightweighting and high strength and toughness in the lightweight biomimetic skeleton structure for aircraft. It solves the problem that existing technologies lack in-depth reference to the efficient load-bearing mechanism of trabeculae, and fail to achieve structural self-adaptation and functional integration under extreme environments.

Description

Technical Field

[0001] This invention belongs to the field of biomimetic aircraft technology, specifically relating to a lightweight biomimetic skeleton structure for aircraft and its preparation and design methods. Background Technology

[0002] Existing aircraft equipment structures mostly utilize traditional metals or composite materials, presenting a challenge in balancing weight and strength. Bone-inspired structures, inspired by the human skeleton, possess lightweight, high strength, high toughness, and excellent fatigue and impact resistance, enabling them to support heavy loads while withstanding various external forces. This invention employs an innovative design method, utilizing bone-inspired design to construct a lattice structure that combines lightweight, high strength, and excellent fatigue and impact resistance. This addresses the problems of low structural efficiency, short lifespan, and poor adaptability in complex environments for aircraft equipment.

[0003] Currently, aircraft structural materials are mainly aluminum alloys and titanium alloys, which, while possessing certain strength, have high density and limited potential for weight reduction. Traditional lattice structures such as BCC and FCC can achieve lightweighting, but are prone to stress concentration and fatigue fracture under complex loads. International research has been conducted on thin-walled lattice structures such as three-period minimal surfaces (TPMS), and while China has accumulated some experience in engineering applications, significant gaps remain in biomimetic design and material-structure-function integration. Existing technologies lack in-depth understanding of the efficient load-bearing mechanisms of trabecular meshwork, failing to achieve structural self-adaptation and functional integration under extreme environments.

[0004] Based on this, the present invention proposes a lightweight biomimetic skeleton structure for aircraft and its preparation and design method, in order to solve the problems existing in the prior art. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention provides a lightweight biomimetic skeleton structure for aircraft and its fabrication and design methods, thereby solving the problem that existing technologies lack in-depth reference to the efficient load-bearing mechanism of trabecular bone and fail to achieve the self-adaptation and functional integration of the structure under extreme environments.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] The present invention provides a first solution:

[0008] A porous lattice core layer composed of three periodic minimal surfaces; and

[0009] A panel layer integrally connected to both sides of the porous lattice core layer, wherein the panel layer has a dense structure or a microporous structure;

[0010] The porous lattice core layer is constructed based on the topological features of the three-dimensional structure of human bone trabeculae, and the local high-stress areas of the biomimetic skeleton structure are introduced with biomimetic micro-topological structures, which are non-uniform reinforcing ribs arranged at the TPMS nodes according to the stress vector direction.

[0011] The porous lattice core layer adopts at least one of the Diamond structure or Double Diamond structure, and its geometric parameters are gradient distributed along the spatial coordinate direction to form a multi-level load-bearing system.

[0012] The micro-topology is achieved by adding non-uniformly distributed reinforcing ribs or variable-thickness shells at the nodes of the TPMS members, and the high-stress region is determined by finite element simulation analysis.

[0013] In a preferred embodiment of the present invention, the Diamond structure is defined by an implicit mathematical equation:

[0014] ;

[0015] The Double Diamond structure is defined by an implicit mathematical equation:

[0016] ;

[0017] Where x, y, and z are spatial Cartesian coordinates, k is a scale factor controlling the cell period, k = 2π / L, L is the cell edge length, and C is a threshold parameter controlling the offset of the isosurface. By adjusting C, gradient changes in porosity, curvature, and thickness are achieved.

[0018] In a preferred embodiment of the present invention, the biomimetic skeleton structure is constructed by a multi-level threshold entity difference method, including:

[0019] Set a set of decreasing threshold parameters C1, C2...C n This generates a series of nested TPMS entity fields;

[0020] By using Boolean subtraction operations on adjacent threshold entities, layered unit cells or hollow double-shell structures with hyperboloid shell characteristics are generated.

[0021] The wall thickness of the lamellar unit cell is controlled by the difference between two adjacent threshold parameters.

[0022] In a preferred embodiment of the present invention, the material of the skeleton structure is a lightweight, high-strength metal-based composite material, including any one of titanium-based composite materials, aluminum-based composite materials, or high-strength aluminum alloys.

[0023] This invention provides a second solution:

[0024] A design method for a lightweight biomimetic skeleton structure for aircraft includes the following steps:

[0025] S1: Bionic structure modeling: The three-dimensional structure of human bone trabeculae is obtained through Micro-CT scanning. Combined with topology optimization and parametric design, a bionic lattice unit library containing Diamond and Double Diamond structures is constructed.

[0026] S2: Multiphysics simulation optimization: Finite element software is used to perform multiphysics simulations of the skeleton structure, including statics, dynamics, and aerodynamic loads, to quantitatively evaluate specific stiffness and fatigue life.

[0027] S3: Machine Learning-Aided Design: Introduce machine learning models to establish a mapping relationship between "structural parameters and mechanical properties", and use multi-objective genetic algorithms to jointly optimize panel thickness, core porosity gradient and TPMS constant.

[0028] In a preferred embodiment of the present invention, the process of obtaining the three-dimensional structure of human trabecular bone in step S1 includes:

[0029] Micro-CT scans were performed on trabecular bone specimens to obtain CT images;

[0030] Three-dimensional models of trabecular bone were reconstructed using CT values ​​of 213–2938 Hu as the threshold range, and saved in STL format.

[0031] The spatial location of the femoral distance is used to automatically register the trabecular bone model with the complete femoral model, thereby achieving parametric modeling.

[0032] This invention provides a third solution:

[0033] A method for fabricating a lightweight biomimetic skeleton structure for an aircraft includes the following steps:

[0034] Material formulation optimization: Based on a machine learning model, a global search is performed on the reinforcing phase type, volume fraction, and particle size distribution of metal matrix composites to obtain the optimal material formulation;

[0035] Composite powder preparation: Uniform composite powder was prepared by high-energy ball milling, and the powder morphology and particle size distribution were monitored in real time using image analysis algorithms;

[0036] Additive manufacturing: Selective laser melting technology is used to form the skeleton structure. During the forming process, a molten pool monitoring system combined with a convolutional neural network is used to identify and adjust the laser process parameters in real time.

[0037] Post-processing: The formed skeleton is subjected to hot isostatic pressing to eliminate internal pores, and surface compressive stress is introduced in combination with sandblasting.

[0038] In a preferred embodiment of the present invention, the machine learning optimization process includes:

[0039] Using matrix material and reinforcing phase parameters as input features, and tensile strength, yield strength and porosity of printed specimens as output targets, a random forest or deep neural network model is constructed.

[0040] A Bayesian optimization algorithm is used to search in the design variable space and output a set of Pareto optimal material formulations that satisfy density constraints.

[0041] In a preferred embodiment of the present invention, the process of adjusting the laser process parameters includes: dynamically adjusting the laser power P, scanning speed v, scanning spacing h and powder layer thickness t according to the diameter and tilt angle characteristics of the biomimetic skeleton structure, so that the volume energy density Ev is maintained within a preset process window.

[0042] In a preferred embodiment of the present invention, after the post-processing step, a performance verification and model correction step is further included:

[0043] Micro-CT was used to perform microscopic analysis on the failure fracture surface to determine the correlation between the crack initiation point and internal defects.

[0044] The three-dimensional solid model reconstructed by Micro-CT, which includes actual geometric deviations, is imported into the finite element software to modify the simulation boundary conditions and failure criteria of the initial design, forming a closed-loop optimization workflow.

[0045] Compared with the prior art, the present invention provides a lightweight biomimetic skeleton structure for aircraft and its preparation and design method, which has the following beneficial effects:

[0046] 1. Lightweight structure: Through a biomimetic porous trabecular structure, the weight of components is reduced by ≥25%, while the specific stiffness is increased by ≥20%;

[0047] 2. High strength and high toughness: Combining TPMS with trabecular bone structure significantly improves damage tolerance and impact resistance;

[0048] 3. Integrated Functionality: The dot matrix can be filled with damping material or embedded with sensors to achieve vibration reduction and real-time monitoring;

[0049] 4. High manufacturing precision: Using SLM technology, the error of key feature dimensions is controlled within ±8%;

[0050] This addresses the problem that existing technologies lack in-depth understanding of the efficient load-bearing mechanism of trabecular meshwork, thus failing to achieve structural self-adaptation and functional integration under extreme environments. Attached Figure Description

[0051] Figure 1This is a flowchart illustrating the fabrication process of the lightweight biomimetic skeleton structure for the aircraft of this invention.

[0052] Figure 2 The image shows a Diamond unit cell diagram of a lightweight biomimetic skeletal structure based on the Diamond Tri-Period Minimal Surface (TPMS).

[0053] Figure 3 The images show Diamond unit cells with different C values ​​for the lightweight biomimetic skeleton structure of the Diamond Tri-Period Minimal Surface (TPMS) of this invention.

[0054] Figure 4 This is a diagram of the lamellar unit cells of the lightweight biomimetic framework structure of the Diamond Tri-Period Minimal Surface (TPMS) of this invention.

[0055] Figure 5 The diagram shows the TPMS biomimetic skeleton (Diamond) of the lightweight biomimetic skeleton structure of the Diamond Three-Period Minimal Surface (TPMS) of this invention.

[0056] Figure 6 This is a unit cell diagram of the Double Diamond highly connected dual-network biomimetic skeleton structure generated based on a specific implicit function in this invention.

[0057] Figure 7 These are Double Diamond unit cell diagrams with different C values, generated based on specific implicit functions in this invention, representing a highly connected dual-network biomimetic skeleton structure of Double Diamond.

[0058] Figure 8 This is a diagram of the layered unit cells of the Double Diamond highly connected dual-network biomimetic framework structure generated based on a specific implicit function in this invention.

[0059] Figure 9 This is a diagram of the TPMS biomimetic skeleton (Double Diamond) of the Double Diamond high-connectivity dual-network biomimetic skeleton structure generated based on a specific implicit function according to the present invention. Detailed Implementation

[0060] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0061] As per the instruction manual Figure 1 -Appendix Figure 9 As shown, to address the lack of in-depth understanding of the efficient load-bearing mechanism of trabecular meshwork in existing technologies, which prevents structures from achieving self-adaptation and functional integration under extreme environments, this invention proposes a lightweight biomimetic skeleton structure for aircraft, along with its fabrication and design methods.

[0062] Example 1:

[0063] As per the instruction manual Figures 1-9 As shown, this embodiment provides a design and fabrication method for a lightweight biomimetic skeleton structure for aircraft. This method achieves a weight reduction of ≥25% and a specific stiffness increase of ≥20% through a biomimetic porous trabecular structure. Specific steps include:

[0064] S1: Perform biomimetic structural modeling;

[0065] The three-dimensional structure of human bone trabeculae was obtained by Micro-CT scanning, and its porous mesh model was reconstructed. A biomimetic lattice unit library was constructed by combining topology optimization and parametric design.

[0066] S1.1: Trabecular structure reconstruction analysis;

[0067] S1.1.1: Preparation of proximal femoral trabecular bone specimens using the scraping method;

[0068] A dried femoral specimen from an adult was selected. After visual inspection revealed no abnormalities in the bone structure and X-rays were taken to rule out internal tumors or other lesions, the surface of the proximal femoral specimen was sanded. Then, all surface cortical bone was removed using a bone file, bone scalpel, and bone forceps. After exposing the cancellous bone, the loose cancellous bone was scraped away with a curette, leaving only the trabeculae.

[0069] S1.1.2: Establishment of a three-dimensional digital model of the proximal femur containing trabeculae;

[0070] After X-ray imaging of the trabecular bone specimen, the distal cortex was removed, and a micro-CT scan was performed using the eXplore Locus User Guide (General Electric Company, USA). Based on the eXplore Locus, this system employs cone-beam flat panel detector technology, allowing imaging of the entire sample volume with one rotation. The system can simultaneously image single or multiple specimens up to 70 mm in diameter, with a maximum acquisition length of 100 mm in a single session, and a resolution <8 μm. CT images are saved in DICOM format.

[0071] CT images of the complete proximal femur and trabeculae were imported into Mimics 18.0 software (Materialise, Belgium). Using the standard adult bone threshold (CT value 213–2938 Hu) as the "Masks" range, the "Calculate 3D from Masks" option was selected from the segmentation menu. The "High Quality" calculation method was used to reconstruct the three-dimensional model of the proximal femur and trabeculae, which was then saved in STL format.

[0072] The spatial position of the femoral distance was used for registration, and a series of biomimetic lattice unit libraries were constructed by parametrically building a three-dimensional model of the proximal femur containing trabeculae. The specific steps are as follows: (1) Select 3 landmark points on the femoral surface of the trabeculae model; (2) Select 3 landmark points at the same position on the surface of the complete femoral model; (3) Use the automatic registration function of Mimics18.0 software to register the model with the landmark points.

[0073] S1.1.3: Registration is performed using the spatial location of the femoral distance to establish a 3D model of the proximal femur containing trabeculae, and a series of biomimetic lattice element libraries are constructed using parameterization. The specific steps are as follows:

[0074] (1) Selection of landmark points: On the femoral surface of the trabecular model, three locations with clear anatomical significance and stable spatial distribution, namely the upper, middle and lower femoral stature, are selected as landmark points; the same three landmark points are selected at the corresponding anatomical locations in the complete femoral model. As the main longitudinal load-bearing structure on the medial side of the proximal femur, the femoral stature has a spatial orientation that is highly consistent with the principal compressive stress trajectory, and is a key anatomical benchmark connecting the trabecular configuration and the external load transmission path.

[0075] (2) Coordinate System 1: Import the two types of models (trabecular bone model and complete femur model) into Mimics 18.0 software, and establish the initial spatial correspondence between the two models by manually picking the above three sets of corresponding markers. This step is essentially to establish a preliminary rigid body transformation matrix between the anatomical coordinate system and the engineering reference coordinate system, providing initial values ​​for subsequent automatic fine registration.

[0076] (3) Automatic fine registration based on point sets: The "automatic registration" function in Mimics software is used. This function is based on the iterative nearest point algorithm. With three sets of marker points as the initial correspondence, it further uses large-scale point cloud data of the two model surfaces for global iterative optimization to minimize the root mean square distance between the two sets of point clouds, thereby achieving high-precision matching of the trabecular bone model and the complete femur model in terms of spatial position and orientation. After registration, the three-dimensional spatial distribution of the trabecular bone is accurately mapped to the anatomical coordinate system of the complete femur.

[0077] (4) Extraction of parametric modeling references: Under the registered unified coordinate system, the spatial orientation line of the femoral distance, the principal stress trajectory surface, and the trabecular density distribution field are extracted and used as the parametric control references for the biomimetic lattice unit library. Specifically, this includes:

[0078] The direction of the femoral neck is used as the main load-bearing direction of the lattice structure;

[0079] The spatial gradient distribution of trabecular bone density distribution field is mapped to lattice unit (such as porosity and wall thickness).

[0080] The transition region between the "panel and core" in a sandwich structure is defined by the relative positional relationship between the femoral neck and the cortical bone.

[0081] Through the above registration and parameterized mapping process, the efficient load-bearing topological features formed by human trabeculae under long-term mechanical adaptation are systematically transformed into quantifiable structural parameters for mechanical design, thereby constructing a biomimetic lattice unit library with optimal biomechanical properties.

[0082] S1.2: Design the lattice structure of the established biomimetic structural model;

[0083] S1.2.1: Design Software: 3-Matic, Within, and Rhino, along with their plugins, are primarily used as the core design tools. Referring to the existing biomimetic lattice element library, the entire workflow from concept generation to model preparation is implemented.

[0084] S1.2.2: TPMS biomimetic skeleton structure design;

[0085] TPMS structures have smooth curved surfaces, which can reduce stress concentration and significantly improve fatigue life and impact resistance. They not only have excellent mechanical properties, but also have the advantages of high specific stiffness, lightweight, and highly interconnected pores. At the same time, their continuous curved surface structure has a small step effect in SLM forming and high forming stability, making them suitable for additive manufacturing of complex components and exhibiting excellent manufacturing friendliness.

[0086] (1) Structural type selection: Diamond and Double Diamond TPMS structures are selected as the core biomimetic skeleton. The Diamond structure is characterized by its fourfold symmetry, high node connectivity and excellent mechanical stiffness, forming an efficient force transmission path; Double Diamond forms two highly intertwined and independent channel systems on the basis of a single network, which significantly improves the specific surface area, permeability and material exchange efficiency while maintaining high stiffness, and is suitable for scenarios with higher requirements for load-bearing efficiency, multiphase transmission or multifunctional integration.

[0087] (2) Parametric control: By adjusting the structural constant C in the TPMS equation, the gradient changes of porosity, curvature and thickness can be achieved, thereby adapting to the mechanical and physical environment requirements of the component in different regions;

[0088] S1.2.3: Sandwich lattice structure design;

[0089] Sandwich lattice structures are widely used in materials design. They possess a number of advantages:

[0090] (1) Lightweight and high stiffness: Sandwich structure achieves extremely high bending stiffness with extremely low weight, and is suitable for large-area components with high stiffness requirements such as wings and bulkheads.

[0091] (2) Impact resistance and energy absorption: The TPMS core layer has good plastic deformation capacity and energy dissipation characteristics, while the panel layer prevents local penetration and damage;

[0092] (3) Multifunctional integrated design: The internal space of the mezzanine can be used to deploy sensor networks, embedded damping units or fluid cooling channels to achieve integrated design of structure-function-material;

[0093] (4) Strong compatibility of manufacturing process: The panel and core layer can be printed with the same or different materials in an integrated SLM process, avoiding the interface weakening problem caused by adhesive bonding or mechanical connection.

[0094] A sandwich configuration of "dense / microporous panel + TPMS core layer + dense / microporous panel" is adopted. Based on the TPMS sandwich structure, a bone-like trabecular microtopology (the microtopology is a non-uniform reinforcing rib arranged according to the stress vector direction at the TPMS nodes) is introduced in the high-stress areas of the core layer or locally in the panel. The TPMS, as the intermediate core layer, bears the shear and compressive loads, while the upper and lower panels provide bending stiffness and surface integrity, forming a multi-level load-bearing system.

[0095] S1.2.4: Simulation analysis and performance verification;

[0096] Using general-purpose finite element analysis software such as ANSYS, ABAQUS, and COMSOL, the designed lattice structure was simulated in multiple physics fields, including statics, dynamics, heat transfer, and typical UAV operating conditions (such as aerodynamic loads and maneuvering overloads).

[0097] Through simulation results, key performance indicators such as stiffness, strength, fatigue life, and energy absorption of the structure are quantitatively evaluated to verify the rationality and reliability of the design. Unsatisfactory schemes are rapidly iterated and optimized. At the same time, key parameters such as panel thickness, core layer pore gradient, TPMS type, and constant c are synergistically optimized to achieve optimal performance of the component under multiple physical fields such as static and dynamic loads and vibration fatigue. Finally, a closed-loop process of "design-simulation-optimization" is formed, which significantly reduces the cost of physical prototyping and development risks.

[0098] S2: Material design and fabrication for biomimetic structural models;

[0099] S2.1: Target material system and performance requirements;

[0100] S2.1.1: System selection;

[0101] Lightweight, high-strength metal matrix composites suitable for SLM forming are selected. The materials must have good shape stability to ensure the complete forming of the micro-topology of the trabecular bone, such as titanium matrix composites (Ti6Al4V+TiB2), aluminum matrix composites (AlSi10Mg+SiC), or high-strength aluminum alloys (Scalmalloy®).

[0102] S2.1.2: Target performance;

[0103] Density < 2.0 g / cm³ to achieve lightweighting; tensile strength > 500 MPa, yield strength > 400 MPa; elongation > 8% to ensure certain toughness; high cycle fatigue limit (R=0.1) > 250 MPa to meet the long life requirements of aerospace structures; good laser absorption rate and molten pool stability to adapt to SLM process.

[0104] S2.2: Optimize the material ratio;

[0105] Material proportion optimization is performed using a machine learning-based multi-component material proportion optimization method. The specific process is as follows:

[0106] S2.2.1: Data collection and feature engineering;

[0107] (1) Input characteristics (design variables);

[0108] Matrix material type (e.g., Ti6Al4V, AlSi10Mg); reinforcing phase type (e.g., TiB2, SiC, graphene, etc.); volume fraction of reinforcing phase (Vf, range: 0.5%~5%); particle size distribution of reinforcing phase (D50, e.g., 0.1~10 μm); process pretreatment parameters (e.g., ball milling time, drying temperature).

[0109] (2) Output target (performance response);

[0110] Tensile strength, yield strength, elongation; microhardness, elastic modulus; melt pool stability score (based on online monitoring image analysis); porosity of printed specimens (based on Micro-CT).

[0111] S2.2.2: Machine learning model construction and training;

[0112] (1) Model selection;

[0113] Random Forest or XGBoost can be used to establish the mapping relationship between "material formulation → performance". For high-dimensional nonlinear relationships, deep neural networks (DNNs) can be introduced, with ReLU as the activation function and multiple regression as the output layer.

[0114] (2) Training process;

[0115] An initial dataset is constructed based on historical experimental data or literature data; 5-fold cross-validation is used to prevent overfitting; feature importance analysis is performed using SHAP (SHapley Additive exPlanations) to identify key influencing factors.

[0116] S2.2.3: Optimization algorithm and recipe recommendation;

[0117] Optimization objective: To minimize material density and cost while satisfying constraints on strength, toughness, and fatigue performance.

[0118] Algorithm flow: Use a trained machine learning model as a surrogate model; employ Bayesian optimization or NSGA-II multi-objective genetic algorithm to perform a global search in the design variable space;

[0119] Output a set of Pareto optimal solutions, i.e., multiple alternative material formulations with balanced performance; and select the final formulation based on actual process feasibility (such as powder flowability and SLM formability).

[0120] S2.3: Material preparation process and intelligent control;

[0121] S2.3.1: Powder pretreatment and composite powder preparation;

[0122] Uniform composite powders are prepared using high-energy ball milling or plasma spheroidization. The powder morphology and particle size distribution are monitored in real time using image analysis algorithms (such as OpenCV edge detection), and the ball milling parameters are dynamically adjusted.

[0123] S2.3.2: Material-process co-control in SLM forming;

[0124] Based on the different regional characteristics of the lattice structure (such as rod diameter and tilt angle), the laser process parameters are dynamically adjusted to achieve regional optimization of the material microstructure (such as grain size and texture); the molten pool monitoring system + convolutional neural network (CNN) is used to identify molten pool anomalies (such as spatter and porosity) in real time, and the laser power and scanning speed are adjusted accordingly.

[0125] S2.3.3: Improvement of material properties through post-processing;

[0126] Hot isostatic pressing (HIP) is used to eliminate internal porosity, and sandblasting / polishing is combined to improve surface integrity. By using residual stress testing and X-ray diffraction analysis, the post-processing parameters (temperature, pressure, time) are optimized to ensure that the fatigue performance of the material is improved by ≥200%.

[0127] S3: Structural Design and Simulation;

[0128] Static, modal, and fatigue analyses of the lattice structure of the biomimetic structure are performed using finite element software (such as ANSYS); machine learning algorithms are introduced to quickly predict structural performance and optimize element distribution and gradient design.

[0129] S3.1: Enhancement of Machine Learning Models:

[0130] S3.1.1: Model selection;

[0131] (1) Basic task: For fast input-output mapping, gradient boosting trees (such as XGBoost, LightGBM) usually perform well on small and medium datasets, combining accuracy and speed.

[0132] (2) Complex tasks: If the design space is extremely large or there is strong nonlinearity, deep neural networks can be considered. For gridded field outputs (such as stress cloud maps), convolutional neural networks or graph neural networks can be introduced to directly learn the mapping between the structure and the performance field.

[0133] S3.1.2: Multi-task learning;

[0134] Statics (stiffness / strength), modal dynamics (frequency), and fatigue (life) are often related. Constructing a neural network with multiple output heads that share underlying features and can simultaneously predict multiple performance parameters can improve data utilization and prediction consistency.

[0135] S3.1.3: Quantification of uncertainty;

[0136] Introducing Bayesian optimization or using models that provide predictive uncertainty (such as Gaussian process regression or Bayesian neural networks) can not only find the optimal solution but also assess the reliability of the optimization results and guide subsequent validation.

[0137] It should be noted that the machine learning model is trained based on no less than 500 sets of finite element simulation data, and the simulation covers the design space with panel thickness of 0.5-3 mm, C-value gradient of 0.8-1.5, and porosity of 30%-70%. The model adopts 5-fold cross-validation, and the average absolute error of the predicted tensile strength is less than 15 MPa before it can be used for subsequent optimization.

[0138] S3.2: Integration and upgrading of optimization algorithms;

[0139] Algorithm flow:

[0140] 1. Define the design space;

[0141] Determine the optimization variables (panel thickness, core porosity gradient, TPMS type and C value gradient, etc.) and their value ranges.

[0142] 2. Design of Experiments (DOE);

[0143] Latin hypercube sampling is used to generate N sets (N≥500) of design sample points within the design space.

[0144] 3. Construction of finite element simulation dataset;

[0145] For each sample group, multi-physics simulations such as statics, modalities, and fatigue are performed to extract response values ​​such as specific stiffness, fatigue life, and mass, forming a "structural parameter-performance" dataset.

[0146] 4. Training of neural network proxy models;

[0147] A deep neural network is constructed, with structural parameters as input and simulation performance as output. The high-dimensional nonlinear mapping relationship between input and output is fitted, and the model accuracy is evaluated by 5-fold cross-validation.

[0148] 5. Validation of the surrogate model's accuracy;

[0149] If the prediction error (e.g., tensile strength MAE > 15 MPa) exceeds the threshold, return to step 2 to add samples or adjust the network structure.

[0150] 6. Multi-objective optimization;

[0151] Based on the trained surrogate model, the NSGA-II multi-objective genetic algorithm is used for global optimization, with the objectives of minimizing mass, maximizing specific stiffness, and maximizing fatigue life, and outputting the Pareto front solution set.

[0152] 7. Optimal solution verification;

[0153] Several candidate schemes are selected from the Pareto front and finite element simulation is performed again for verification. If the verification results deviate significantly from the predictions of the surrogate model, new samples are added to the dataset to update the surrogate model and achieve closed-loop correction.

[0154] S3.2.1: Input feature definition (structural parameters);

[0155] The input features of the optimization model are the key geometric and topological parameters that affect the mechanical properties of the biomimetic skeleton. Specifically, these include:

[0156] Table 1: Key Geometric and Topological Parameters

[0157] Serial Number Input features Symbols / Units Physical meaning Range of values ​​(example) 1 Panel thickness tp / mm Thickness of the upper and lower panels of the sandwich structure 0.5 ~ 3.0 mm 2 Core porosity φ / % Average porosity of TPMS core layer 30% ~ 70% 3 Porosity gradient direction — Gradient change direction Enumeration: X, Y, Z, Radial 4 Porosity gradient magnitude Δφ / % Change in porosity along the gradient direction 0% ~ 30% 5 TPMS type — Lattice topology Diamond, Double Diamond, Gyroid 6 TPMS scaling factor k To control the cell period, k = 2π / L (where L is the cell edge length). 0.5 ~ 2.0 7 TPMS threshold parameter C Controlling the isosurface offset determines the wall thickness and relative density. Diamond: 0.8 ~ 1.5Double Diamond: 2.5 ~ 3.5 8 Wall thickness distribution coefficient α The linear proportionality coefficient between wall thickness and C value is t≈α⋅C 0.1 ~ 0.5 9 Parameters of stiffeners in local high-stress areas — Thickness, width, and distribution density of reinforcing ribs Determined based on stress field dynamics

[0158] The above features comprise a total of 9 to 12 input variables, forming the design space.

[0159] S3.2.2: Neural Network Agent Model Architecture;

[0160] This method uses a deep neural network (DNN) as a surrogate model, and the specific architecture is as follows:

[0161] (1) Input layer: Dimension = Number of input features (n_features = 12), including all the above structural parameters;

[0162] (2) Hidden layers: A 4-layer fully connected network is used, with the number of neurons in each layer as follows:

[0163] Hidden layer 1: 128 neurons, activation function ReLU;

[0164] Second hidden layer: 64 neurons, activation function ReLU;

[0165] The third hidden layer consists of 32 neurons, with the ReLU activation function.

[0166] Fourth hidden layer: 16 neurons, activation function ReLU;

[0167] (3) Output layer: Dimension = Number of output targets (n_outputs=3), which are: specific stiffness (E / ρ) / MPa·cm³ / g;

[0168] Fatigue life (N_f) / number of cycles (logarithmic field);

[0169] Structural mass (m) / g;

[0170] The output layer has no activation function and directly outputs the regression values.

[0171] (4) Regularization: Add a Dropout layer after each hidden layer with a dropout rate of 0.2 to prevent overfitting.

[0172] (5) Loss function: mean squared error (MSE) + L2 regularization term (coefficient λ=0.001).

[0173] (6) Optimizer: Adam, initial learning rate = 0.001, learning rate decay factor = 0.9, decay step size = 50 epochs.

[0174] (7) Number of training rounds: max_epochs=500, early stopping mechanism (patience=50).

[0175] S3.3.3: Dataset construction and training closed loop;

[0176] (1) Data sources and scale;

[0177] Generate no fewer than 500 design samples using finite element simulation (ANSYS / ABAQUS) to cover the aforementioned design space.

[0178] Each sample set contains a complete input feature vector and the corresponding simulation output results (specific stiffness, fatigue life, mass).

[0179] Simulation settings: The equivalent properties of the TPMS core layer are calculated using the homogenization method. The macroscopic lattice structure is modeled using solid elements. Boundary conditions simulate typical aircraft operating conditions (such as wing surface aerodynamic loads and maneuvering overloads).

[0180] (2) Data preprocessing;

[0181] Input feature normalization: Min-Max normalization is used to normalize to the [0, 1] interval.

[0182] The fatigue life in the output target is taken as the logarithm to the base 10 to mitigate the difference in magnitude.

[0183] (3) Model training and validation;

[0184] We employ 5-fold cross-validation, randomly dividing the dataset into 5 subsets. Each time, we use 4 subsets for training and 1 subset for validation, and repeat this process 5 times to take the average performance.

[0185] Model accuracy evaluation metrics:

[0186] Specific stiffness: Mean absolute error (MAE) ≤ 5 GPa·cm³ / g;

[0187] Fatigue life (logarithm): MAE ≤ 0.2;

[0188] Structural mass: MAE ≤ 2 g;

[0189] If any indicator exceeds the threshold, then: increase the number of simulation samples (incremental sampling), adjust the network structure (such as increasing the width of the hidden layer), or introduce a more complex model (as shown in the figure, neural network processing spatial field distribution).

[0190] (4) Optimize closed-loop and model iteration;

[0191] The optimization process does not end with a single optimization attempt, but rather establishes a closed-loop mechanism for continuous iteration:

[0192] 1. Initial optimization: The surrogate model is trained based on the initial 500 sets of simulation data, and the Pareto front is generated through NSGA-II;

[0193] 2. Scheme verification: Select representative schemes from the Pareto front (e.g., 3-5 groups) and perform high-fidelity finite element simulation verification;

[0194] 3. Error feedback: Compare the predicted values ​​of the surrogate model with the high-fidelity simulation results. If the error exceeds the threshold, add new samples to the training set.

[0195] 4. Model Update: Retrain or fine-tune the surrogate model to improve local accuracy;

[0196] 5. Secondary optimization: Based on the updated surrogate model, perform a local fine-grained search near the Pareto front and output the final optimized solution;

[0197] This closed-loop mechanism ensures the prediction accuracy of the surrogate model in critical design areas, while realizing fully automated iteration of the "simulation-surrogate-optimization-verification-correction" process.

[0198] S3.4.4: Configuration of multi-objective optimization algorithm;

[0199] 1. Algorithm: NSGA-II (Non-dominated sorting genetic algorithm);

[0200] 2. Population size: 100;

[0201] 3. Number of generations: 200;

[0202] 4. Crossover probability: 0.9;

[0203] 5. Mutation probability: 1 / n_features;

[0204] 6. Objective function (based on surrogate model prediction): Minimize structural mass m; maximize specific stiffness E / ρ; maximize fatigue life N_f (logarithmic domain);

[0205] 7. Constraints: Porosity range: 30%~70%; Panel thickness range: 0.5 mm~3.0 mm; Manufacturability constraint: Minimum wall thickness ≥ 0.2 mm (meeting SLM process requirements);

[0206] S4: Additive Manufacturing and Fabrication Process Control of Bionic Structures:

[0207] Selective laser melting technology is used to form a lattice structure, and parameters such as laser power and scanning speed are optimized; post-treatments such as hot isostatic pressing and sandblasting are implemented to improve fatigue performance and surface quality.

[0208] S4.1: Parametric design;

[0209] S4.1.1: Key parameters (input variables);

[0210] (1) Laser process parameters: laser power P, scanning speed v, spot diameter, scanning spacing h, powder layer thickness t. Volume energy density E is commonly used. v = P / (v·h·t) is used as a comprehensive indicator.

[0211] (2) Scanning strategies: stripe, checkerboard, island scanning and their rotation angles;

[0212] (3) Substrate preheating temperature: affects residual stress and forming.

[0213] S4.1.2: DOE method: Use response surface methodology, Taguchi method or Latin hypercube sampling to efficiently arrange experimental points in the parameter space, covering the process window from under-melting to over-melting.

[0214] S4.2: Multi-scale, multi-dimensional data acquisition and feature engineering;

[0215] S4.2.1: Online monitoring data: Use the molten pool monitoring system to obtain image features such as brightness, size, and stability of the molten pool;

[0216] S4.2.2: Offline representation data (core):

[0217] (1) Geometric quality: Porosity, pore size / morphology distribution, and rod size deviation were quantified by micro-CT scanning;

[0218] (2) Surface quality: Surface roughness (Sa, Sz) is obtained by white light interferometer or roughness meter, especially the characteristics of the surface of nodes and rods;

[0219] (3) Microstructure: Metallographic analysis is used to obtain the morphology of the molten pool, grain structure, and defects (cracks, lack of fusion);

[0220] (4) Mechanical properties: Perform rod-shaped specimens or micro-compression / tension tests to obtain micro-area hardness, elastic modulus and yield strength;

[0221] (5) Fatigue performance (gold standard): Perform fatigue tests on the printed lattice structure or representative specimens to obtain SN curves or fatigue limit data. This is the most expensive but most critical data.

[0222] (6) Post-processing impact data: Compare the changes in porosity, surface roughness, residual stress and fatigue life before and after HIP treatment.

[0223] S5: Performance verification and model correction;

[0224] The structural performance was verified through quasi-static tensile, compression, and vibration fatigue tests; the actual geometric model was reconstructed based on Micro-CT, and the simulation boundary conditions and failure criteria were corrected.

[0225] S5.1: Basic materials and unit cell performance testing;

[0226] Objective: To obtain the inherent properties of the lattice rod material and provide validation data for the homogenization method.

[0227] S5.1.1: Tensile test of solid rods;

[0228] (1) Method: Use solid small-diameter rod-shaped specimens (e.g., consistent with the diameter of the lattice rod) printed in the same batch and with the same process parameters as the lattice structure.

[0229] (2) Measurement: Obtain the elastic modulus (E) and yield strength (σ). y ), tensile strength (σ) uts ), elongation (δ) and fracture toughness (K) IC ).

[0230] (3) Purpose: This is the basic input for the material model in the simulation. By comparing it with the standard manual value, the influence of the SLM process on the intrinsic properties of the material can be evaluated.

[0231] S5.1.2: Quasi-static compression test of single-cell structure;

[0232] (1) Method: A compression test was conducted on a small array consisting of several basic unit cells (e.g., 3×3×3), and full-field strain was measured using digital image correlation (DIC) technology;

[0233] (2) Measurement: Obtain the equivalent elastic modulus (E_eff), equivalent yield strength, and collapse mode (buckling, plastic hinge, fracture) of the unit cell.

[0234] (3) Application: Directly verify the accuracy of homogenization theory or unit cell finite element model based on solid material properties; DIC data can be used to calibrate local deformation in simulation.

[0235] S5.2: Macroscopic lattice structure performance test;

[0236] Objective: To verify the overall structural performance, especially the size effect, boundary effect, and design functionality.

[0237] S5.2.1: Macroscopic quasi-static compression / tension test;

[0238] (1) Method: Test lattice structural blocks or components with engineering application dimensions. DIC or numerous strain gauges must be used;

[0239] (2) Measurement: Overall load-displacement curve, local strain field at specific locations, failure initiation location and path. Record initial stiffness, peak load, and energy absorption efficiency;

[0240] (3) Purpose: to verify the accuracy of the overall simulation prediction; the failure initiation point is the gold standard for judging whether the simulation failure criterion is correct.

[0241] S5.2.2: Three-point / four-point bending test;

[0242] (1) Method: Bending test was performed on the elongated lattice structure;

[0243] (2) Measurement: bending stiffness, ultimate bending moment, and neutral axis position;

[0244] (3) Application: It is particularly suitable for verifying the performance of gradient lattice or non-uniform design, examining its bending resistance, and comparing it with simulation.

[0245] S5.2.3: Modal impact test;

[0246] (1) Method: Experimental modal analysis was performed using a force hammer and an accelerometer;

[0247] (2) Measurement: the first few natural frequencies, damping ratios, and mode shapes of the structure;

[0248] (3) Application: Directly verify the simulation results of modal analysis. Frequency deviation can be used to indirectly infer and calibrate boundary conditions (such as actual clamping stiffness) and material dynamic modulus in the simulation.

[0249] S5.3: Fatigue and Durability Testing;

[0250] Objective: To acquire fatigue life data and observe fatigue failure mechanisms in order to improve fatigue simulation models.

[0251] S5.3.1: Stress-controlled / strain-controlled fatigue test;

[0252] (1) Method: Tension-tension or compression-compression fatigue tests were performed on macroscopic lattice specimens on a servo hydraulic fatigue testing machine; it is recommended to use the step method or small sample method to improve testing efficiency;

[0253] (2) Measurement: Record the SN curve (stress amplitude-life curve). Use an infrared thermal imager to monitor the temperature rise hotspots (potential damage accumulation zone) during fatigue. Use acoustic emission to monitor real-time signals of crack initiation and propagation;

[0254] (3) Purpose: To provide baseline data for fatigue life. Failure location and crack initiation point (usually at surface roughness or internal defects) provide a basis for correcting the initial defect settings in fatigue simulation.

[0255] S5.3.2: Vibration fatigue test;

[0256] (1) Method: The lattice structure was installed on an electromagnetic vibration table, and a fixed-frequency or swept-frequency vibration test was performed near the resonant frequency;

[0257] (2) Measurement: Monitor the strain time history at key locations and record the time / number of cycles until structural failure (the appearance of visible cracks or a significant decrease in stiffness);

[0258] (3) Application: To simulate actual vibration environments and verify the durability of structures under dynamic loads. The data is used to calibrate frequency-domain-based fatigue simulations.

[0259] S5.4: Special test on the effect of post-processing technology;

[0260] Objective: To quantify the contribution of post-processing to performance improvement and provide key data for the "process-performance" model.

[0261] S5.4.1: Before and after comparison test;

[0262] (1) Method: For the same batch of printed twin specimens, one group was subjected to hot isostatic pressing (HIP) and / or sandblasting, while the other group was kept in its original state and the above-mentioned tensile, compression and fatigue tests were performed in parallel.

[0263] (2) Measurement: Percentage improvement in key performance indicators (such as fatigue limit);

[0264] (3) Purpose: To accurately quantify the value of post-processing and provide data for cost-benefit analysis.

[0265] S5.4.2: Residual stress test;

[0266] (1) Method: The residual stress distribution on the surface and subsurface of the HIP lattice structure before and after the HIP was measured by X-ray diffraction or profilometry.

[0267] (2) Application: To correct the initial stress field boundary conditions in the simulation. The compressive stress introduced by sandblasting is a key factor in improving fatigue performance and must be quantified.

[0268] S5.5: In-situ / destructive analysis combined with Micro-CT depth;

[0269] Objective: To establish a deterministic correlation between defects / geometry and performance, which is the most direct input for correcting simulations.

[0270] S5.5.1: Microscopic analysis of the failure cross-section;

[0271] (1) Method: Micro-CT scanning and scanning electron microscopy were performed on the failure section immediately after fatigue or static testing;

[0272] (2) Measurement: Determine the precise crack initiation point (whether it is a surface rough peak, subsurface porosity, or internal unfused defect), and observe the fracture morphology (ductile fracture, brittle fracture).

[0273] (3) Application: Directly modify failure criteria. For example, if 90% of fatigue cracks originate from subsurface pores of a specific size, then equivalent defects should be implanted at the corresponding locations in the simulation.

[0274] S5.5.2: In-situ loading of Micro-CT experiment (if conditions permit);

[0275] (1) Method: The lattice sample was placed in the Micro-CT on the micro loading stage and scanned while loading was performed;

[0276] (2) Measurement: to obtain the dynamic evolution process of internal defects (pores, cracks) under load, and the sequence of buckling or fracture of the rod;

[0277] (3) Application: This is the most powerful verification and correction tool. It can compare the matching degree between the stress field predicted by simulation and the actual failure process frame by frame, and directly calibrate the damage evolution model of the material.

[0278] S6: Perform data integration and simulation correction workflow;

[0279] S6.1: Geometric correction;

[0280] (1) The actual three-dimensional model reconstructed by Micro-CT (including surface roughness, rod diameter variation, and internal pores) is directly imported into the finite element software for mesh generation, replacing the ideal CAD model;

[0281] (2) Key operations: In the simulation, based on the pore size and location distribution statistically obtained by CT, the material properties are weakened by explicitly modeling at the corresponding locations or by assigning field variables.

[0282] S6.2: Boundary condition correction;

[0283] (1) Compare the experimental DIC full-field strain with the simulated strain cloud diagram, and adjust the load application method and contact conditions in the simulation to make the two match;

[0284] (2) Use the frequency and mode shape obtained from the modal test to back-calculate and correct the constraint stiffness in the simulation.

[0285] S6.3: Revision of Material Model and Failure Criteria;

[0286] (1) Use solid rods and unit cell test data to calibrate the parameters of the elastoplastic constitutive model;

[0287] (2) Based on the failure section analysis results, revise the failure criteria in the simulation (such as the maximum principal stress criterion and strain energy density criterion), and determine the threshold for considering the influence of defects;

[0288] (3) Use fatigue SN curve data and crack initiation location information to calibrate stress-based (such as SWT model) or strain energy-based fatigue life prediction models.

[0289] The lightweight biomimetic skeleton structure for aircraft prepared using the above method is as follows: Figure 9 As shown.

[0290] Example 2:

[0291] As per the instruction manual Figures 2-9 As shown, unlike Embodiment 1 above, this embodiment provides a lightweight biomimetic skeleton for an aircraft obtained using the design method described in Embodiment 1 above. The lightweight biomimetic skeleton for an aircraft includes a lightweight biomimetic skeleton structure based on Diamond Triple Period Minimal Surface (TPMS) and a Double Diamond high-connectivity dual-network biomimetic skeleton structure (biomimetic skeleton structure) generated based on a specific implicit function.

[0292] A lightweight biomimetic skeleton structure based on the Diamond Tri-Period Minimal Surface (TPMS) is disclosed. This structure is governed by implicit mathematical equations, and the fundamental surface of the Diamond biomimetic skeleton is defined by the following mathematical formula:

[0293]

[0294] Where x, y, and z are the Cartesian coordinates in space, k is the scale factor controlling the cell period (k = 2π / L), L is the cell edge length, and C is the threshold parameter controlling the offset of the isosurface. By changing the value of C, the solid volume enclosed by the surface in space can be precisely defined.

[0295] Specifically, such as Figure 2 As shown, the above-mentioned skeleton structure is controlled by implicit level set equations. Its geometric surface has an average curvature of zero everywhere in space, and has smooth, seamless, and non-self-intersecting topological features, which can effectively avoid the stress concentration problem at the nodes of traditional lattice structures.

[0296] Specifically, such as Figure 3 As shown, the above-mentioned bionic skeleton is constructed based on the multi-level threshold entity difference method: first, a set of decreasing threshold parameters C is set to generate a series of nested entity domains.

[0297] In this embodiment, parameter C1=1.4 corresponds to structure a, parameter C2=1.3 corresponds to structure b, parameter C3=1.2 corresponds to structure c, and parameter C4=1.1 corresponds to structure d.

[0298] It should be noted that the threshold parameter C and the physical wall thickness t satisfy the empirical relationship t≈α·C.

[0299] Specifically, such as Figure 3 As shown, there are containment relationships among the above entity structures. As the absolute value of parameter C decreases, the volume of the generated TPMS entity domain gradually increases.

[0300] In this embodiment, the solid volume of structure d contains structure c, structure c contains structure b, and structure b contains structure a. This nesting characteristic provides the geometric basis for subsequently constructing hollow thin-walled structures using Boolean operations.

[0301] Specifically, such as Figure 4 As shown, this embodiment uses Boolean subtraction operations between adjacent threshold entities to generate layered unit cells. The first type of unit cell (1) is obtained by subtracting entity a (i.e., ba) from entity b; the second type of unit cell (2) is obtained by subtracting entity b (i.e., cb) from entity c; and the third type of unit cell (3) is obtained by subtracting entity c (i.e., dc) from entity d.

[0302] Specifically, the solid wall thickness of the aforementioned unit cell is determined by the difference ΔC between two adjacent thresholds. For example, the wall thickness of unit cell (1) is controlled by ΔC = |C2-C1| = 0.1. When ΔC remains constant, the generated biomimetic skeleton has a uniform wall thickness distribution; when ΔC is adjusted, the local wall thickness of the skeleton can be fine-tuned, thereby changing the relative density and mechanical stiffness of the structure while maintaining the overall topology unchanged.

[0303] Specifically, the lamellar cell units (1), (2), and (3) have hyperboloid shell characteristics. Compared with traditional beam-based skeletons, this continuous shell structure provides a larger specific surface area and can distribute stress more evenly on the curved surface when subjected to external loads, thereby improving the energy absorption efficiency and fatigue resistance of the structure.

[0304] Specifically, such as Figure 5As shown, the final TPMS biomimetic framework is composed of multiple unit cells stacked and arrayed along the Z-axis. Due to the periodicity of the Diamond surface, the tangential directions of adjacent unit cells at the interface are continuous, achieving a smooth transition between layers without the need for additional connectors, thus ensuring the integrity of the overall framework structure.

[0305] Specifically, this biomimetic framework supports gradient structure design. When stacking the array, different threshold-generated unit cells can be used sequentially along the Z-axis (e.g., ba unit cells for the bottom layer and dc unit cells for the top layer). This design allows the porosity and pore size of the framework to exhibit gradient changes in the height direction, which can simulate the density change pattern of natural bone tissue from cortical bone to cancellous bone, meeting the requirements of biomechanical adaptation.

[0306] Specifically, for the overall structure after arraying, Boolean intersection operations can be performed based on the boundary shape of the actual parts to remove excess parts, thereby obtaining a personalized bionic implant skeleton that perfectly matches the anatomical shape of the implantation site.

[0307] A highly connected dual-network biomimetic skeleton structure based on a specific implicit function, the Double Diamond structure, compared to the standard single Diamond structure, naturally forms two interwoven but non-intersecting continuous networks in space. Its more complex topology significantly improves the specific stiffness and fracture toughness of the structure. The Double Diamond skeleton surface is defined by the following specific mathematical equations:

[0308] ;

[0309] In the formula, x, y, z are spatial Cartesian coordinates, and C is a threshold parameter controlling the offset of the isosurface. The introduction of the product of the second harmonic term (2kx, 2ky, 2kz) and the sine term (sin(2kx)sin(2ky)sin(2kz)) in the formula accurately constructs a three-period minimum surface with dual network characteristics.

[0310] Specifically, the Double Diamond structure constructed by the above mathematical formula has a high degree of symmetry. The first three terms (cosine product term) construct the basic lattice framework, while the fourth term (sine product term) modifies and connects the lattice nodes in a specific shape, making the generated skeleton fuller at the nodes and effectively enhancing the overall stiffness of the structure.

[0311] Specifically, such as Figure 7As shown, based on the above formula, this embodiment uses a multi-level threshold control method to construct the skeleton entity. By setting a series of different C values, entity regions with different volume proportions in the scalar field can be extracted. For example, setting the parameter C to 3.4, 3.2, 3.0, and 2.8 respectively generates entity structures e, f, g, and h.

[0312] It should be noted that the above-mentioned entity structures exhibit an isotropic expansion trend as the threshold C decreases. Entities generated at high thresholds (such as structure e) closely surround the central line of the skeleton, forming a fine network; entities generated at low thresholds (such as structure h) expand outwards on this basis, forming a robust support network.

[0313] Specifically, such as Figure 8 As shown, in this embodiment, a hollow double-network cell is constructed by interlayer Boolean subtraction. The first-level cell structure (4) is obtained by subtracting entity e (i.e., fe) from entity f; the second-level cell structure (5) is obtained by subtracting entity f (i.e., gf) from entity g; and the third-level cell structure (6) is obtained by subtracting entity g (i.e., hg) from entity h.

[0314] Specifically, the structures (4) to (6) generated by Boolean operations are a surrounding hollow double-shell structure. This design removes the central material of the low-stress zone (i.e., subtracts solid a) and retains the outer shell with a high moment of inertia, thereby significantly reducing the structural weight while maintaining the excellent bending and torsional resistance of the Double Diamond structure.

[0315] Specifically, such as Figure 9 As shown, the complete three-dimensional biomimetic framework is composed of the aforementioned unit cells arranged in a periodic array along the X, Y, and Z directions. Due to the periodicity of terms such as cos(2kx) in the formula, the arrayed structure automatically satisfies the continuity condition at the boundary, requiring no additional connecting structures, thus forming a monolithic, seamlessly connected three-dimensional porous body.

[0316] It should be noted that this Double Diamond structure is particularly suitable for metal additive manufacturing (3D printing). Its unique curved surface geometry gives the structure excellent self-support, minimizing the need for support material during the printing process and reducing the difficulty of removing the support in post-processing.

[0317] It should be noted that the skeleton structure described in this embodiment supports gradient functionalization design based on spatial fields. Through parametric modeling, a mapping function C = f(x, y, z) between the threshold parameter C and spatial coordinates (x, y, z) is constructed, which enables continuous spatial adjustment of the skeleton's relative density. For example, when designing orthopedic implants or impact-resistant components, a radial gradient function can be used to gradually decrease parameter C from the center to the outer surface (i.e., the threshold decreases, and the solid domain expands outward). This design creates a lightweight region with high C value and high porosity in the core of the part, which is conducive to cell ingrowth or low-resistance flow of fluids; at the same time, a dense protective layer with low C value and near-solid structure is formed on the surface of the part, significantly enhancing the surface hardness and wear resistance of the structure. This continuous gradient configuration of "sparse inside and dense outside" effectively eliminates the macroscopic interface in traditional multilayer structures, avoids the risk of interlayer delamination caused by abrupt changes in stiffness, and realizes an optimized design that integrates structure and function.

[0318] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A lightweight biomimetic skeleton structure for aircraft, characterized in that, include: A porous lattice core layer composed of three periodic minimal surfaces; as well as A panel layer integrally connected to both sides of the porous lattice core layer, wherein the panel layer has a dense structure or a microporous structure; The porous lattice core layer is constructed based on the topological features of the three-dimensional structure of human bone trabeculae, and the local high-stress areas of the biomimetic skeleton structure are introduced with biomimetic micro-topological structures, which are non-uniform reinforcing ribs arranged at the TPMS nodes according to the stress vector direction. The porous lattice core layer adopts at least one of the Diamond structure or Double Diamond structure, and its geometric parameters are gradient distributed along the spatial coordinate direction to form a multi-level load-bearing system. The micro-topology is achieved by adding non-uniformly distributed reinforcing ribs or variable-thickness shells at the nodes of the TPMS members, and the high-stress region is determined by finite element simulation analysis.

2. The lightweight biomimetic skeleton structure for aircraft according to claim 1, characterized in that, The Diamond structure is defined by an implicit mathematical equation: ； The Double Diamond structure is defined by an implicit mathematical equation: ； Where x, y, and z are spatial Cartesian coordinates, k is a scale factor controlling the cell period, k = 2π / L, L is the cell edge length, and C is a threshold parameter controlling the offset of the isosurface. By adjusting C, gradient changes in porosity, curvature, and thickness are achieved.

3. The lightweight biomimetic skeleton structure for aircraft according to claim 2, characterized in that, The biomimetic skeleton structure is constructed using a multi-level threshold entity difference method, including: Set a set of decreasing threshold parameters C1, C2...C n This generates a series of nested TPMS entity fields; By using Boolean subtraction operations on adjacent threshold entities, layered unit cells or hollow double-shell structures with hyperboloid shell characteristics are generated. The wall thickness of the lamellar unit cell is controlled by the difference between two adjacent threshold parameters.

4. The lightweight biomimetic skeleton structure for aircraft according to claim 1, characterized in that, The material of the skeleton structure is a lightweight, high-strength metal matrix composite material, including any one of titanium matrix composites, aluminum matrix composites, or high-strength aluminum alloys.

5. A design method for a lightweight biomimetic skeleton structure for an aircraft as described in claim 1, characterized in that, Includes the following steps: S1: Bionic structure modeling: The three-dimensional structure of human bone trabeculae is obtained through Micro-CT scanning. Combined with topology optimization and parametric design, a bionic lattice unit library containing Diamond and Double Diamond structures is constructed. S2: Multiphysics simulation optimization: Finite element software is used to perform multiphysics simulations of the skeleton structure, including statics, dynamics, and aerodynamic loads, to quantitatively evaluate specific stiffness and fatigue life. S3: Machine Learning-Aided Design: Introduce machine learning models to establish the mapping relationship between "structural parameters and mechanical properties", and use multi-objective genetic algorithms to jointly optimize panel thickness, core porosity gradient and TPMS constant.

6. The design method for a lightweight biomimetic skeleton structure for aircraft according to claim 5, characterized in that, The process of obtaining the three-dimensional structure of human trabecular bone in step S1 includes: Micro-CT scans were performed on trabecular bone specimens to obtain CT images; Three-dimensional models of trabecular bone were reconstructed using CT values ​​of 213–2938 Hu as the threshold range, and saved in STL format. The spatial location of the femoral distance is used to automatically register the trabecular bone model with the complete femoral model, thereby achieving parametric modeling.

7. A method for preparing a lightweight biomimetic skeleton structure for an aircraft as described in claim 1, characterized in that, Includes the following steps: Material formulation optimization: Based on a machine learning model, a global search is performed on the reinforcing phase type, volume fraction, and particle size distribution of metal matrix composites to obtain the optimal material formulation; Composite powder preparation: Uniform composite powder was prepared by high-energy ball milling, and the powder morphology and particle size distribution were monitored in real time using image analysis algorithms; Additive manufacturing: Selective laser melting technology is used to form the skeleton structure. During the forming process, a melt pool monitoring system combined with a convolutional neural network is used to identify and adjust the laser process parameters in real time. Post-processing: The formed skeleton is subjected to hot isostatic pressing to eliminate internal pores, and surface compressive stress is introduced in combination with sandblasting.

8. The method for preparing a lightweight biomimetic skeleton structure for an aircraft according to claim 7, characterized in that, The machine learning optimization process includes: Using matrix material and reinforcing phase parameters as input features, and tensile strength, yield strength and porosity of printed specimens as output targets, a random forest or deep neural network model is constructed. A Bayesian optimization algorithm is used to search in the design variable space and output a set of Pareto optimal material formulations that satisfy density constraints.

9. The method for preparing a lightweight biomimetic skeleton structure for an aircraft according to claim 7, characterized in that, The adjustment process of the laser process parameters includes: dynamically adjusting the laser power P, scanning speed v, scanning spacing h and powder layer thickness t according to the diameter and tilt angle characteristics of the biomimetic skeleton structure, so that the volume energy density Ev is kept within the preset process window.

10. The method for preparing a lightweight biomimetic skeleton structure for an aircraft according to claim 7, characterized in that, Following the post-processing steps, performance verification and model correction steps are also included: Micro-CT was used to perform microscopic analysis on the failure fracture surface to determine the correlation between the crack initiation point and internal defects. The three-dimensional solid model reconstructed by Micro-CT, which includes actual geometric deviations, is imported into the finite element software to modify the simulation boundary conditions and failure criteria of the initial design, forming a closed-loop optimization workflow.

Images