Data-driven TPMS dot matrix mold structure thermal stress optimization design method

Abstract

This invention relates to the field of advanced manufacturing technology, specifically a data-driven TPMS lattice mold structure thermal stress optimization design method. It involves uniformly parameterizing the structural geometric parameters, material properties, and thermal load boundary conditions of complex structures to generate structural parameter vectors. Based on these parameter vectors, a finite element thermo-mechanical coupling sample dataset is constructed, and a surrogate model is trained to map the structural parameter vectors to the thermal stress response data. Then, the predicted thermal stress results output by the surrogate model are used to perform parameter optimization calculations. While ensuring consistency between the predicted results and the finite element thermal stress response in data structure and numerical dimensions, this method significantly reduces the computational cost of thermal stress prediction and parameter optimization, improves parameter space search efficiency and optimization stability, and enhances the engineering applicability of the method in the design of complex engineering structures.

Description

Technical Field

[0001] This invention relates to the field of advanced manufacturing, specifically to a data-driven method for optimizing the thermal stress of TPMS lattice mold structures. Background Technology

[0002] In the fields of advanced manufacturing, mold engineering, and high-performance structural design, complex engineering structures often experience significant thermal loads during service or manufacturing. The resulting thermal stress distribution within the structure directly affects its dimensional stability, service life, and safety reliability. With the increasing demands for lightweight design, multifunctional integration, and structural performance optimization, the geometry of complex structures is becoming increasingly refined. Material parameters and thermal load conditions exhibit multidimensional coupling characteristics, leading to a shift in structural thermal stress analysis from traditional empirical design to numerical simulation and computational optimization.

[0003] In existing technologies, for complex structural thermal stress problems, finite element analysis software is typically used to model the structural geometry, material properties, and thermal load boundary conditions. Steady-state or transient thermal analysis, along with thermo-structural coupling calculations, are then employed to obtain the thermal stress response of the structure under different operating conditions. Building upon this foundation, some studies have attempted to combine intelligent optimization algorithms to perform multiple iterative calculations of structural parameters, aiming to reduce maximum thermal stress or improve stress distribution.

[0004] However, existing techniques for analyzing and optimizing thermal stress in complex structures typically require repeatedly executing the complete finite element thermo-mechanical coupling solution process for each set of candidate structural parameters. When the number of structural geometric parameters is large, the coupling relationships between parameters are complex, and the mesh size is large, the cost of a single simulation is high, the computation cycle is long, and the parameter space search efficiency is low. Meanwhile, existing methods largely rely on discrete finite element simulation results to drive the optimization process, lacking the technical means to uniformly parameterize and model structural geometric parameters, material properties, and thermal load boundary conditions. This makes it difficult to establish a mapping relationship between structural parameters and thermal stress response, thus preventing rapid prediction of thermal stress without repeatedly executing finite element calculations. Furthermore, during parameter optimization, the lack of consistency between the predicted results and the finite element thermal stress response in terms of data structure and numerical dimensions further affects the stability and applicability of the optimization algorithm in high-dimensional parameter spaces. These technical problems limit the computational efficiency and application scalability of thermal stress prediction and parameter optimization methods for complex structures in practical engineering. Summary of the Invention

[0005] This application provides a data-driven thermal stress optimization design method for TPMS lattice mold structures, which solves the problems of scattered modeling of multiple source parameters and difficulty in establishing mapping relationships in the prior art.

[0006] To achieve the above objectives, the embodiments of this application disclose the following technical solutions:

[0007] This solution discloses a data-driven thermal stress optimization design method for TPMS lattice mold structures, including the following steps:

[0008] Step S1: Obtain structural geometric parameter data, material property parameter data, and thermal load boundary condition data of the complex structure; perform unified parameterization processing on the structural geometric parameter data, material property parameter data, and thermal load boundary condition data to generate a structural parameter vector; In step S1, the parameterization processing performed on the structural geometric parameter data includes: numerically encoding the thickness parameter, aperture parameter, and periodic characteristic parameter of the complex structure to generate a geometric feature sub-vector;

[0009] The numerical encoding is not an arbitrary assignment, but rather an explicit mapping of the geometric degrees of freedom in complex structures that have first-order or higher-order sensitivity to thermal stress distribution to the vector space, so that the influence of geometric nonlinearity on the thermo-mechanical coupling response can be continuously perceived in subsequent models.

[0010] The elastic modulus, Poisson's ratio, coefficient of thermal expansion, and thermal conductivity in the material property parameter data are normalized to generate material property sub-vectors. The normalization process eliminates the dominant effect of different material property parameters in terms of numerical magnitude, avoids the gradient suppression caused by a single property parameter in the training of the surrogate model, and thus ensures that the synergistic effect of multiple properties is effectively learned.

[0011] The temperature loading path, loading amplitude, and loading time series in the thermal load boundary condition data are discretized to generate thermal load feature sub-vectors.

[0012] The discretization process transforms the continuous-time domain thermal load evolution process into a computable temporal characteristic, enabling the transient thermal stress accumulation effect to be captured by the surrogate model, rather than only reflecting the static or steady-state response.

[0013] The geometric feature sub-vectors, material property sub-vectors, and thermal load feature sub-vectors are concatenated in a preset order to form a structural parameter vector, which is then used as the unified input data for subsequent finite element analysis and surrogate model training.

[0014] This application does not simply involve numerically stitching together data from different sources. Instead, it constructs a unified structural parameter vector representation based on the physical mechanism that thermal stress response is jointly determined by geometric morphology, material constitutive properties, and the evolution path of thermal load. Without this unified parameterization, differences in numerical scale, physical dimensions, and response sensitivity between data from different dimensions will directly lead to inconsistencies in finite element analysis inputs, gradient oscillations or feature-dominated imbalances during surrogate model training, thus failing to stably characterize the mapping relationship between the structural parameter vector and the thermal stress response.

[0015] In existing technologies, geometric parameters, material parameters, and thermal load conditions are usually treated as isolated inputs, which leads to subsequent analysis relying on manual parameter tuning or empirical correction, making it difficult to support large-scale parameter search and optimization calculations.

[0016] By uniformly encoding thickness parameters, aperture parameters, periodic characteristic parameters, elastic modulus, Poisson's ratio, coefficient of thermal expansion, thermal conductivity, as well as temperature loading path, loading amplitude, and loading time sequence into a structural parameter vector, the influence of different physical fields during the thermo-mechanical coupling process is explicitly expressed in coordinate form in vector space, thus providing a consistent input topology for finite element analysis and surrogate models.

[0017] If a unified parameterization is not performed and the traditional modular input method is used, the surrogate model cannot identify the cross-sensitivity between different physical parameters, and the prediction results will degenerate into local fitting, which cannot be used for parameter optimization calculation.

[0018] This structural parameter vector serves as the common input basis for the finite element analysis in step S2 and the surrogate model training in step S4, enabling all subsequent calculations to unfold within the same parameter space and forming a continuous mapping chain across steps.

[0019] Step S2: Based on the structural parameter vector, call the finite element analysis program to perform thermo-mechanical coupled numerical calculations on the complex structure, obtain thermal stress response data corresponding one-to-one with the structural parameter vector, and construct a sample dataset composed of the structural parameter vector and thermal stress response data. In step S2, the finite element analysis program includes: automatically generating a finite element mesh model based on the structural parameter vector; assigning element material properties to the finite element mesh model based on material property parameter data; applying thermal boundary conditions to the finite element mesh model based on thermal load boundary condition data; performing thermo-mechanical coupled solution calculations to obtain temperature field data and stress field data corresponding to each element node in the finite element mesh model; extracting thermal stress response values ​​at at least one pre-selected node or set of nodes in the finite element mesh model from the stress field data to generate thermal stress response data, and associating and storing the thermal stress response data with the corresponding structural parameter vector; the thermal stress response data includes the maximum principal stress value, equivalent stress value, or stress value at a preset key node. The predicted thermal stress results output by the surrogate model are consistent with the thermal stress response data in terms of data structure and numerical dimension for unified processing in the parameter optimization calculation process.

[0020] This application does not treat finite element analysis merely as a single verification tool, but rather systematically embeds it into the sample construction process, making the finite element analysis program a mapping generator between structural parameter vectors and thermal stress response data. By generating one-to-one corresponding thermal stress response data for each structural parameter vector, a sample dataset with statistical coverage is formed, fundamentally changing the traditional engineering application method of finite element analysis of "calculate once, modify once".

[0021] Step S3: Perform data preprocessing on the sample dataset to generate a standardized sample dataset, and input the standardized sample dataset into the preset surrogate model training process; Step S3 includes: removing outlier samples in the sample dataset; performing same-scale standardization on the structural parameter vector and thermal stress response data; dividing the standardized sample dataset into a training subset and a validation subset according to a preset ratio; using the training subset to train the surrogate model, and using the validation subset to evaluate the prediction performance of the surrogate model, so as to generate data input for updating the parameters of the surrogate model.

[0022] This application ensures that the surrogate model learns the intrinsic physical laws between the structural parameter vector and the thermal stress response, rather than finite element numerical noise or local outliers, through outlier removal, same-scale standardization, and the division of training and validation subsets. Omitting this step would lead to prediction drift or overfitting risks in the parameter optimization calculations of the surrogate model.

[0023] The preset ratio is a pre-defined proportion of training data to validation data set before model training, based on sample size, model complexity, and prediction accuracy requirements. This preset ratio is not arbitrarily chosen, but rather a balance between sample size, model complexity, and prediction accuracy requirements, ensuring the model has generalization ability rather than being effective only on training data.

[0024] Step 4: Train the surrogate model based on the standardized sample dataset to obtain a surrogate model that can represent the mapping relationship between the structural parameter vector and the thermal stress response data. In step S4, the surrogate model is a multi-layer feedforward neural network model, which includes an input layer, at least one hidden layer, and an output layer. The input layer is used to receive the structural parameter vector; the hidden layer is used to perform nonlinear mapping calculations; and the output layer is used to output the predicted thermal stress value corresponding to the structural parameter vector. By minimizing the error between the predicted thermal stress value output by the surrogate model and the thermal stress response data, the network parameters of the surrogate model are iteratively updated to obtain the trained surrogate model.

[0025] The multilayer feedforward neural network model, through the nonlinear mapping capability of the hidden layers, can approximate the high-dimensional nonlinear relationship between complex structural parameter vectors and thermal stress response, a relationship that cannot be effectively expressed in traditional linear regression or response surface models.

[0026] Compared to support vector machines or multinomial response surfaces, multilayer feedforward neural network models can maintain stable mapping capabilities even as the parameter dimension increases, making them an inevitable choice for handling high-dimensional structural parameter vectors.

[0027] Step S5: Obtain a set of candidate parameter vectors for the complex structure to be optimized, and input the set of candidate parameter vectors into the surrogate model to obtain the corresponding predicted thermal stress results. In step S5, obtaining the set of candidate parameter vectors includes: generating an initial set of candidate parameter vectors based on a preset parameter value range; performing feasibility constraint screening on the initial set of candidate parameter vectors to generate a set of feasible candidate parameter vectors; inputting the set of feasible candidate parameter vectors into the surrogate model one by one to obtain the corresponding predicted thermal stress results, and associating and storing the predicted thermal stress results with the corresponding feasible candidate parameter vectors.

[0028] The preset parameter value range is the allowable value range of each structural parameter that is preset before the parameter optimization calculation begins, based on the geometric constraints, material performance limitations, manufacturing process conditions, and actual service conditions of the complex structure to be optimized.

[0029] Step S6: Perform parameter optimization calculations based on the predicted thermal stress results, and output the target structural parameter vector that meets the preset constraints. Step S6 includes: inputting the predicted thermal stress results as an evaluation index into the optimization algorithm; performing parameter search and update operations on the candidate parameter vector set based on the optimization algorithm; in each round of parameter update, calling the surrogate model to generate new predicted thermal stress results; and outputting the target structural parameter vector when the preset termination condition is met. The optimization algorithm is a genetic algorithm, which includes: generating an initial population based on the candidate parameter vector set; performing selection, crossover, and mutation operations on the initial population to generate a new generation of parameter populations; inputting the new generation of parameter populations into the surrogate model to obtain the corresponding predicted thermal stress results; evaluating the fitness of the new generation of parameter populations based on the predicted thermal stress results, and using this evaluation for updating the next generation of populations. During the execution of the method, the processor stores the structural parameter vectors, thermal stress response data, surrogate model parameters, and target structural parameter vectors in memory, and performs subsequent surrogate model calls and parameter optimization calculations based on the data in memory.

[0030] This application does not directly perform parameter optimization calculations based on finite element analysis. Instead, it uses a surrogate model to quickly predict thermal stress results within the parameter space, freeing the genetic algorithm from the limitations of high-cost numerical solutions. A closed loop of "prediction-evaluation-update" is formed between the surrogate model and the genetic algorithm, transforming parameter optimization calculations from discrete trial-and-error to a continuous evolutionary process.

[0031] This application is not a simple parallel use of finite element analysis, surrogate models, or genetic algorithms. Instead, it establishes a systematic technical solution applicable to thermal stress prediction and parameter optimization of complex structures through the structural coupling between unified parameterization, sample construction, surrogate modeling, and optimization search. This fundamentally overcomes the problems of low computational efficiency and insufficient parameter space exploration capabilities of existing technologies.

[0032] This invention unifies the parameterization of structural geometric parameters, material properties, and thermal load boundary conditions of complex structures to construct structural parameter vectors. These vectors serve as the unified input for finite element analysis and surrogate model training, ensuring consistency between structural parameters and thermal stress response data in both data structure and numerical dimension. This addresses the problems of fragmented modeling of multiple parameters and difficulty in establishing mapping relationships in existing technologies. By constructing a sample dataset based on the structural parameter vectors and training a surrogate model that characterizes the mapping relationship between the structural parameter vectors and thermal stress response data, rapid prediction of thermal stress in complex structures is achieved without repeatedly performing finite element thermo-mechanical coupling calculations. This significantly reduces the cost per calculation and shortens the parameter optimization cycle. Furthermore, by combining parameter optimization calculations based on predicted thermal stress results, the search and update process of the candidate parameter vector set can be stably performed in a high-dimensional parameter space, improving parameter space search efficiency and the reliability of optimization results. This effectively overcomes the problems of low computational efficiency and limited applicability of existing methods for thermal stress analysis and optimization of complex structures, enhancing the method's application and scalability in engineering practice. Attached Figure Description

[0033] Figure 1 This is an overall flowchart of an embodiment of the present invention;

[0034] Figure 2 This is a schematic diagram illustrating the generation of structural parameter vectors in an embodiment of the present invention;

[0035] Figure 3 This is a flowchart illustrating the finite element analysis and sample dataset construction process of an embodiment of the present invention.

[0036] Figure 4 This is a closed-loop diagram of the proxy model training and parameter optimization calculation in an embodiment of the present invention.

[0037] Figure 5This is a schematic diagram of the Gyroid lattice structure under different parameter combinations in embodiments of the present invention. Detailed Implementation

[0038] 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 a part of the embodiments of the present invention, and not all of them. 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. In the following description, numerous specific details are set forth to provide a comprehensive understanding of the present invention. The present invention may be practiced without some or all of these specific details. In other instances, well-known processes have not been described in detail to avoid unnecessarily obscuring the present invention.

[0039] When used in conjunction with the terms "comprising," "method comprising," or similar language in this specification and appended claims, the singular forms "a," "some," and "the" include plural references unless the context clearly indicates otherwise. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0040] This invention proposes a data-driven thermal stress optimization design method for TPMS lattice mold structures. This method unifies the parameterization of structural geometric parameters, material property parameters, and thermal load boundary conditions, and combines finite element thermo-mechanical coupling numerical calculation with surrogate model learning mechanism to achieve rapid prediction of thermal stress response of complex structures, and on this basis, completes efficient optimization of structural parameters.

[0041] In this embodiment, the structural geometry parameters, material properties, and thermal load boundary conditions reflecting the service conditions of the complex structure to be analyzed are first obtained. These multi-source parameters are encoded using a unified parameterization method to form a structural parameter vector, which characterizes the comprehensive state of the complex structure under geometric morphology, material properties, and thermal load conditions. This structural parameter vector serves as the unified input data basis for subsequent finite element analysis and surrogate model training, facilitating collaborative modeling across different parameter dimensions.

[0042] After constructing the structural parameter vectors, a finite element analysis program is invoked based on these vectors to perform thermo-mechanical coupled numerical calculations on the complex structure, obtaining thermal stress response data corresponding one-to-one with each structural parameter vector. By associating and storing the structural parameter vectors with the thermal stress response data, a sample dataset for data-driven modeling is constructed. This sample dataset reflects the mapping relationship between structural parameter changes and thermal stress response, providing fundamental data support for the subsequent training of the surrogate model.

[0043] Furthermore, preprocessing operations are performed on the sample dataset to improve data quality and model training stability. These preprocessing operations include outlier removal, data scaling, and dataset partitioning, enabling effective modeling of parameter data with different scales within the same numerical space and providing standardized data input for the training and validation of the surrogate model.

[0044] Based on this, a surrogate model is constructed and trained using a preprocessed sample dataset. This surrogate model learns the nonlinear mapping relationship between structural parameter vectors and thermal stress response data, thereby enabling rapid prediction of the thermal stress response corresponding to new combinations of structural parameters without frequently calling the finite element analysis program. By continuously iteratively updating the model parameters, the surrogate model can accurately characterize the thermal stress variation patterns of complex structures. The surrogate model refers to an approximate prediction model that learns the mapping relationship between input parameters and thermal stress response based on sample data, used to replace computationally expensive numerical simulations.

[0045] After training the surrogate model, the candidate parameter vector set of the structure to be optimized is input into the surrogate model to obtain the corresponding predicted thermal stress results. Based on the predicted thermal stress results, a parameter optimization algorithm is introduced to search and update the candidate parameter vector set. Under the condition of satisfying preset constraints and termination conditions, the target structure parameter vector is output. Through the above process, the prediction of thermal stress and parameter optimization of complex structures under thermal load are achieved in a coordinated manner, thereby significantly reducing the overall computational cost and improving the optimization efficiency.

[0046] The preset constraints include, but are not limited to, geometric constraints, material property constraints, boundary condition constraints, and allowable range constraints for structural strength or deformation related to complex structures, in order to ensure the engineering feasibility of the structural scheme during the optimization process.

[0047] To further illustrate the application process of the method of the present invention in specific engineering scenarios, the technical solution of the present invention will be described in detail below with reference to specific embodiments.

[0048] Example: Thermal stress prediction and parameter optimization based on TPMS lattice mold structure

[0049] In this embodiment, a mold structure subject to significant thermal loads is used as the specific application object to illustrate the method of the present invention. Specifically, a shoe sole mold is used as an example. A three-period minimal surface (TPMS) lattice structure is introduced inside the mold, and combined with thermal-structural coupling simulation, surrogate modeling, and parameter optimization methods, lightweight design and thermal stress performance optimization of the mold structure are achieved. The method described in this embodiment is also applicable to other mold structures subject to thermal loads, such as injection molds, die-casting molds, and hot-pressing molds.

[0050] In this embodiment, a three-dimensional geometric model of the shoe sole mold is first obtained. This geometric model can be derived from existing CAD design files or 3D scanned models. To facilitate the parametric generation and continuous modeling of the lattice structure, the mold geometric model is imported into a parametric modeling platform that supports implicit modeling. The solid geometric model is then converted into an implicit solid structure using an implicit volume conversion method. This allows the internal structure of the mold to be continuously modeled based on a function-driven approach, avoiding the geometric discontinuity problems caused by traditional Boolean operations.

[0051] After the implicit processing is completed, a TPMS lattice structure is introduced into the internal space of the mold as the filling structure. In this embodiment, a Gyroid-type three-period minimal surface structure is selected as the lattice topology. This Gyroid lattice structure extends continuously in space with a smooth transition, without sharp corners or abrupt nodes, which helps to reduce stress concentration. At the same time, it forms a completely interconnected three-dimensional channel inside, making the material distribution more uniform and exhibiting good mechanical isotropic properties in three orthogonal directions. In addition, this type of lattice structure is suitable for integral forming in additive manufacturing processes and has good manufacturing feasibility.

[0052] Based on this, in order to achieve controllable generation of lattice structures and subsequent parameter optimization analysis, it is necessary to parametrically describe the geometric features of the lattice structures.

[0053] This embodiment uses a Gyroid lattice structure as the internal filling structure to illustrate the lattice parameterization design method proposed in this invention. By selecting different parameter combinations, various internal filling structure morphologies are generated to verify the applicability and stability of the proposed solution under different parameter conditions.

[0054] In this embodiment, the lattice mold structure is modeled parametrically, and key geometric parameters of the lattice structure are set as design variables. These structural parameters include unit cell size parameters (x, y, z) to control the periodic dimensions of the lattice structure in three dimensions; lattice wall thickness (Approximate thickness) parameters to control the equivalent wall thickness of the TPMS surface; and mold shell thickness (Shell thickness) parameters to control the solid thickness of the mold shell. The value range of these parameters can be set according to the mold size, manufacturing process, and actual operating conditions.

[0055] In this embodiment, to facilitate the explanation of the specific implementation of the technical solution of the present invention, some key parameters of the lattice structure are set by way of example. It should be noted that the parameter ranges below are only used to illustrate this embodiment and do not constitute a limitation on the scope of protection of the present invention. The value ranges of each structural parameter used in this embodiment are shown in Table 1 below.

[0056] Table 1:

[0057]

[0058] Among them, the unit cell size is used to characterize the scale of the lattice unit in three spatial directions, and its value range can be adjusted according to the overall structural dimensions and mechanical performance requirements; the lattice thickness affects the local stiffness and load-bearing capacity of the lattice structure; and the shell thickness is used to ensure the stability and forming strength of the overall structure. By reasonably combining and setting the above parameters, the goal of lightweight design can be achieved while meeting the structural performance requirements.

[0059] Based on the above parameter range, in order to verify the influence of different parameter combinations on the morphology and performance of the lattice structure, this embodiment selects several representative parameter combinations as examples for illustration.

[0060] Within the aforementioned parameter range, this embodiment further selects three representative parameter combinations to generate three different Gyroid internal filling lattice structures, as illustrated below. Figure 5 As shown.

[0061] in, Figure 5 The parameter combinations corresponding to the Gyroid lattice structure shown are as shown in Table 2 for the unit cell size:

[0062] Table 2:

[0063]

[0064] The above parameter combinations are merely exemplary parameters selected in this embodiment and do not constitute a limitation on the range of parameters.

[0065] After modeling the lattice mold structure, the model is meshed. To ensure the computational stability and simulation accuracy of the finite element analysis, surface reconstruction and volume meshing are performed on the lattice mold structure in this embodiment. By reasonably setting the feature size of the mesh elements, the geometric features of the lattice can be effectively represented while also considering computational efficiency. In a preferred embodiment, the feature size of the mesh elements can be set to approximately 10 mm, but is not limited to this value. After meshing, the lattice mold model is exported into a mesh file format recognizable by the finite element analysis software for subsequent thermal-structural coupling simulation analysis.

[0066] In one implementation, to calculate the thermal stress response of the lattice mold structure under thermal load, the meshed lattice mold model is exported in a mesh file format supported by the finite element analysis software and imported into the finite element analysis platform to construct a thermal-structural sequential coupling analysis process that combines steady-state thermal analysis and static structural analysis.

[0067] During the thermal analysis phase, thermal boundary conditions are set according to the actual service conditions of the mold. This includes applying a preset temperature load to the upper surface of the mold and setting convective heat transfer boundary conditions on the sidewalls and bottom surface of the mold to simulate the heat transfer environment of the mold during operation. The temperature field distribution of each node of the lattice mold structure is obtained through steady-state thermal analysis.

[0068] Subsequently, the temperature field distribution results are input as thermal loads into the static structural analysis module, and necessary structural constraints are applied to the mold during the structural analysis phase to simulate the stress conditions of the mold in the clamping state. Without applying additional mechanical loads, structural calculations are performed to obtain the stress field distribution of the mold structure under thermal loads.

[0069] Based on the above simulation results, performance indicators reflecting the thermal stress characteristics of the structure can be further extracted to construct a data mapping relationship between structural parameters and thermal stress response.

[0070] In one embodiment, the thermal boundary conditions may include ambient temperature conditions, constant temperature load conditions, and convective heat transfer conditions.

[0071] For example, the ambient temperature can be set to normal, a constant temperature load can be applied to the upper surface of the mold, and convective heat transfer boundaries can be set on the sidewalls and bottom surface of the mold to simulate the heat transfer environment of the mold during actual operation. In one example setting, the ambient temperature can be set to approximately 25°C, the upper surface temperature load can be set to approximately 200°C, and the convective heat transfer coefficients of the sidewalls and bottom surface can be set to preset values.

[0072] It should be noted that the above-mentioned thermal boundary conditions and their parameter values ​​are only an exemplary setting in this embodiment. The specific values ​​can be adjusted according to the mold type, material properties and actual working conditions, and do not constitute a limitation on the technical solution of the present invention.

[0073] After solving the temperature field, the obtained temperature field results are used as thermal load input to the static structural analysis module, and the stress distribution of the mold under thermal load is solved in combination with the necessary structural constraints.

[0074] In this embodiment, thermal stress performance indicators are extracted from the thermal-structural coupling simulation results as structural performance evaluation parameters. These thermal stress performance indicators include maximum thermal stress, minimum thermal stress, and the thermal stress difference obtained from the difference between the two, used to characterize the uniformity of stress distribution within the mold. By performing multiple thermal-structural coupling simulation analyses on the lattice mold structure under different combinations of structural parameters, a mapping sample dataset between structural parameters and thermal stress performance indicators is constructed and stored in structured data form, providing a data foundation for subsequent surrogate model training.

[0075] In this embodiment, the key geometric parameters of the lattice structure are used as input parameters for the surrogate model. These input parameters include at least the unit cell size parameter, which characterizes the spatial periodicity of the lattice structure, and the wall thickness and shell thickness parameters, which characterize the local stiffness and overall stability of the lattice structure.

[0076] Among them, the unit cell size parameter describes the periodic scale of the lattice structure in three spatial directions, the wall thickness parameter describes the equivalent thickness distribution characteristics of the lattice structure surface, and the shell thickness parameter describes the solid thickness characteristics of the mold shell. These input parameters can be combined within a preset reasonable range to reflect the balance between lightweight design and structural strength of the lattice structure.

[0077] In this embodiment, a thermo-structural coupled finite element simulation is performed on each combination of structural parameters, and thermal stress performance indicators are extracted from the simulation results as model output parameters. The thermal stress performance indicators include at least the maximum equivalent stress value, the minimum equivalent stress value, and the thermal stress difference obtained from the difference between the two, which are used to characterize the overall stress level and its uniformity of distribution of the structure under thermal load.

[0078] The aforementioned thermal stress performance indicators serve as the response outputs corresponding to the structural parameters, and together with the input structural parameters, constitute the sample data required for training the surrogate model.

[0079] In one implementation, by combining and sampling structural parameters within a preset range of structural parameter values, performing multiple thermal-structural coupling simulation calculations, multiple sets of sample data of structural parameters and corresponding thermal stress performance indicators are generated, thereby constructing a dataset for training a surrogate model.

[0080] Each set of sample data includes at least structural parameter fields and thermal stress performance index fields, and is stored in a structured data format, such as in a table or matrix format in a data file, to facilitate subsequent data preprocessing, model training, and parameter optimization calculations.

[0081] It should be noted that the number of samples and the number of data fields can be adjusted according to the model accuracy requirements and computing resources, and do not constitute a limitation on the technical solution of this invention.

[0082] After constructing the simulation dataset, data preprocessing operations are performed on the dataset, including normalization of structural parameter data and thermal stress performance index data, outlier removal, and dataset partitioning, to generate a standardized sample dataset for training and validating the surrogate model. Based on this, the surrogate model is constructed and trained to learn the nonlinear mapping relationship between structural parameters and thermal stress performance indices.

[0083] Furthermore, after the surrogate model is trained, candidate structural parameter combinations are generated based on a preset range of structural parameter values. Each candidate structural parameter combination is then input into the surrogate model to obtain the corresponding predicted thermal stress performance index. By introducing a parameter optimization algorithm, using the predicted thermal stress performance index as the input to the optimization objective function, the candidate structural parameter combinations are evaluated and screened to determine the optimal or near-optimal structural parameter combination.

[0084] In one embodiment, after obtaining the optimal or near-optimal combination of structural parameters, this combination is re-inputted into a thermo-structural coupled finite element model to perform verification simulation calculations, thereby verifying the consistency of the surrogate model's prediction results. Simulation results show that, through the method described in this embodiment, the uniformity of thermal stress distribution in the mold under high-temperature conditions can be effectively improved while ensuring the overall structural stability of the mold, and lightweight design of the mold structure can be achieved. This provides an efficient and reliable technical means for thermal stress prediction and parameter optimization of complex structures.

[0085] In this embodiment, a parameter-thermal stress response prediction model is constructed based on the aforementioned simulation sample data. This prediction model adopts a multi-layer neural network structure, using the unit cell size parameters of the lattice structure, the lattice wall thickness parameters, and the mold shell thickness parameters as inputs, and the thermal stress difference as the output, to achieve rapid prediction of the uniformity of thermal stress distribution under different combinations of structural parameters.

[0086] During the training process, the input parameters and output results are normalized, and the model is iteratively trained based on multiple sets of simulation samples. The model with the smaller prediction error is selected as the final prediction model.

[0087] The prediction model has been verified to be able to quickly estimate the thermal stress difference under different parameter combinations without repeating finite element simulations, thus providing a basis for subsequent parameter optimization or structural selection.

[0088] Based on this, the surrogate model is introduced as a performance prediction module into the parameter optimization process to achieve a balance between prediction accuracy and computational efficiency.

[0089] After obtaining the surrogate model, the surrogate model is embedded as a performance prediction module in the parameter optimization process. The predicted thermal stress results are used as the evaluation index to search and update the candidate structural parameter vector, thereby obtaining the target structural parameter vector that meets the requirement of uniform thermal stress distribution.

[0090] In one specific implementation, the parameter optimization calculation can employ a genetic algorithm, which involves selecting, crossing over, and mutating candidate parameter vectors, and calling the surrogate model to generate predicted thermal stress results during each round of updates, thereby gradually approximating the target structural parameter vector that satisfies preset constraints.

[0091] In one specific embodiment, a set of target structural parameter vectors is obtained through the above prediction and optimization process. The corresponding thermal stress difference is significantly reduced compared with that before optimization, indicating that the method can effectively improve the thermal stress distribution while ensuring the lightweight design of the structure.

[0092] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them; although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications can still be made to the specific implementation methods of the present invention or equivalent substitutions can be made to some technical features without departing from the spirit of the technical solutions of the present invention, and all such modifications should be covered within the scope of the technical solutions claimed in the present invention.

Claims

1. A data-driven method for optimizing the thermal stress of TPMS lattice mold structures, characterized in that, Executed by the processor and including the following steps: Step S1: Obtain structural geometric parameter data, material property parameter data, and thermal load boundary condition data of complex structures; perform unified parameterization processing on the structural geometric parameter data, material property parameter data, and thermal load boundary condition data to generate structural parameter vectors. Step S2: Based on the structural parameter vector, call the finite element analysis program to perform thermo-mechanical coupled numerical calculations on the complex structure, obtain thermal stress response data corresponding one-to-one with the structural parameter vector, and construct a sample dataset composed of the structural parameter vector and the thermal stress response data. Step S3: Perform data preprocessing on the sample dataset to generate a standardized sample dataset, and input the standardized sample dataset into the preset proxy model training process; Step S4: Train the surrogate model based on the standardized sample dataset to obtain a surrogate model that can characterize the mapping relationship between structural parameter vectors and thermal stress response data; Step S5: Obtain the set of candidate parameter vectors for the complex structure as the structure to be optimized, and input the set of candidate parameter vectors into the surrogate model to obtain the corresponding predicted thermal stress results; Step S6: Perform parameter optimization calculations based on the predicted thermal stress results, and output the target structural parameter vector that meets the preset constraints.

2. The data-driven TPMS lattice mold structure thermal stress optimization design method according to claim 1, characterized in that, In step S1, the parameterization processing performed on the structural geometric parameter data includes: Numerical encoding is performed on the thickness parameters, aperture parameters, and periodic characteristic parameters of complex structures to generate geometric feature sub-vectors; The elastic modulus, Poisson's ratio, coefficient of thermal expansion and thermal conductivity in the material property parameter data are normalized to generate material property subvectors; The temperature loading path, loading amplitude, and loading time series in the thermal load boundary condition data are discretized to generate thermal load feature sub-vectors. The geometric feature sub-vectors, material property sub-vectors, and thermal load feature sub-vectors are concatenated in a preset order to form a structural parameter vector, which is then used as the unified input data for subsequent finite element analysis and surrogate model training.

3. The data-driven TPMS lattice mold structure thermal stress optimization design method according to claim 1, characterized in that, In step S2, the finite element analysis program includes: Automatic generation of finite element mesh models based on structural parameter vectors; Assign element material properties to the finite element mesh model based on material property parameter data; Thermal boundary conditions are applied to the finite element mesh model based on thermal load boundary condition data; Perform thermo-mechanical coupling calculations to obtain temperature field data and stress field data corresponding to each element node in the finite element mesh model; The thermal stress response value at at least one pre-selected node or set of nodes in the finite element mesh model is extracted from the stress field data to generate thermal stress response data, and the thermal stress response data is associated and stored with the corresponding structural parameter vector.

4. The data-driven thermal stress optimization design method for TPMS lattice mold structures according to claim 1, characterized in that, In step S3, the data preprocessing operations include: Perform a process to remove outlier samples from the sample dataset; Perform same-scale standardization on the structural parameter vectors and thermal stress response data; The standardized sample dataset is divided into a training subset and a validation subset according to a preset ratio; A training subset is used to train the surrogate model, and a validation subset is used to evaluate the predictive performance of the surrogate model, in order to generate data inputs for updating the surrogate model parameters.

5. The data-driven TPMS lattice mold structure thermal stress optimization design method according to claim 1, characterized in that, In step S4, the surrogate model is a multi-layer feedforward neural network model, which includes an input layer, at least one hidden layer, and an output layer, wherein: The input layer is used to receive the structure parameter vector; Hidden layers are used to perform nonlinear mapping calculations; The output layer is used to output the predicted thermal stress values ​​corresponding to the structural parameter vector; By minimizing the error between the predicted thermal stress output by the surrogate model and the thermal stress response data, the network parameters of the surrogate model are iteratively updated to obtain the trained surrogate model.

6. The data-driven TPMS lattice mold structure thermal stress optimization design method according to claim 1, characterized in that, In step S5, obtaining the candidate parameter vector set includes: Generate an initial set of candidate parameter vectors based on the preset parameter value range; Perform feasibility constraint screening on the initial candidate parameter vector set to generate a feasible candidate parameter vector set; The set of feasible candidate parameter vectors is input into the surrogate model one by one to obtain the corresponding predicted thermal stress results, and the predicted thermal stress results are associated and stored with the corresponding feasible candidate parameter vectors.

7. The data-driven TPMS lattice mold structure thermal stress optimization design method according to claim 1, characterized in that, In step S6, the parameter optimization calculation includes: The predicted thermal stress results are used as evaluation indicators input into the optimization algorithm; Parameter search and update operations are performed on the candidate parameter vector set based on optimization algorithms; During each round of parameter updates, the proxy model is invoked to generate new predicted thermal stress results; When the preset termination condition is met, the target structure parameter vector is output.

8. The data-driven TPMS lattice mold structure thermal stress optimization design method according to claim 7, characterized in that, The optimization algorithm is a genetic algorithm, which includes: An initial population is generated based on the set of candidate parameter vectors; Select, crossover, and mutation operations are performed on the initial population to generate a new generation of parameterized populations; Input the new generation of parameter population into the surrogate model to obtain the corresponding predicted thermal stress results; Fitness assessment of a new generation of parametric populations is performed based on predicted thermal stress results, and the results are used for next generation population updates.

9. The data-driven TPMS lattice mold structure thermal stress optimization design method according to claim 1, characterized in that, The thermal stress response data includes the maximum principal stress value, equivalent stress value, or stress value at preset key nodes. The predicted thermal stress results output by the surrogate model are consistent with the thermal stress response data in terms of data structure and numerical dimensions, so as to be used for unified processing in the parameter optimization calculation process.

10. The data-driven TPMS lattice mold structure thermal stress optimization design method according to claim 1, characterized in that, During the execution of the method, the processor stores the structural parameter vector, thermal stress response data, surrogate model parameters, and target structural parameter vector in memory, and performs subsequent surrogate model calls and parameter optimization calculations based on the data in memory.

Images