A CAE-ML-based multi-objective optimization design and performance prediction method for SFRP composite materials
By using the CAE-ML method, combined with computer-aided engineering simulation and machine learning, the problems of complexity and insufficient prediction accuracy in the design of short fiber reinforced composite materials are solved. Multi-objective optimization and efficient performance prediction are achieved, and precise process control guidance is provided, thereby improving design efficiency and quality.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- HUBEI UNIV OF AUTOMOTIVE TECH
- Filing Date
- 2025-12-31
- Publication Date
- 2026-06-09
AI Technical Summary
Existing technologies for the design and optimization of short fiber reinforced composite materials suffer from problems such as complex processes, strong coupling, insufficient performance prediction accuracy, single optimization dimensions, and insufficient exploration of design space. In particular, the characterization of microstructure is insufficient, resulting in low design efficiency and difficulty in meeting multi-dimensional performance requirements.
By adopting a CAE-ML-based approach, an intelligent design system is constructed by integrating computer-aided engineering simulation, machine learning, and multi-objective optimization algorithms. This system achieves tight coupling between process, structure, and performance, establishes a multi-task machine learning proxy model, and performs multi-angle performance prediction and collaborative optimization.
It achieves full-process digital mapping from manufacturing to performance, improving design efficiency and accuracy, enabling simultaneous optimization of multiple performance characteristics, revealing the relationship between microstructure and performance, providing precise process control guidelines, and enhancing the theoretical depth and engineering guidance significance of the design process.
Smart Images
Figure CN122174291A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of computer-aided engineering and intelligent manufacturing technology for composite materials, specifically a multi-objective optimization design and performance prediction method for SFRP composite materials based on CAE-ML. Background Technology
[0002] Short fiber reinforced composites are widely used in the automotive, aerospace, and electronics industries due to their excellent processability, high specific strength, and specific stiffness. Their final properties (such as stiffness, strength, and impact resistance) are strongly dependent on the microstructure (such as fiber orientation distribution, fiber length distribution, and weld seams) determined by the molding process (such as injection pressure, holding time, and melt temperature).
[0003] Currently, the design and optimization in this field mainly face the following technical challenges:
[0004] 1. The process is complex and highly coupled: From process parameters to microstructure and then to macroscopic performance, it is a complex multi-physics coupling process. The traditional "trial and error method" is costly and time-consuming.
[0005] 2. Insufficient accuracy in performance prediction: Existing macroscopic homogenization methods or simple empirical models are insufficient to accurately capture the microstructural non-uniformity caused by complex processes, resulting in large deviations in performance prediction.
[0006] 3. Single optimization dimension: Traditional optimization design often only targets a single performance index or fixed working condition, which cannot meet the comprehensive performance requirements of components under different loads and angles in actual applications (i.e., multi-angle performance).
[0007] 4. Insufficient exploration of the design space: There are numerous process parameters and microstructure variables, which constitute a high-dimensional design space, making it difficult for traditional optimization algorithms to find the best solution efficiently and globally.
[0008] Furthermore, existing technologies for characterizing "microstructure" typically rely on simple fiber orientation tensors, lacking quantitative descriptions of key morphological features such as fiber aggregation, porosity, and uneven local fiber volume fraction. These morphological details have a decisive impact on performance, especially strength and failure behavior.
[0009] Therefore, there is an urgent need for an advanced design method that can efficiently and accurately connect the "process-structure-performance" chain and achieve multi-angle, multi-objective collaborative optimization. Furthermore, establishing a bridging model that can accurately quantify complex microstructures and directly correlate them with macroscopic performance is a key challenge and breakthrough point for achieving high-precision design and optimization. Summary of the Invention
[0010] To address the aforementioned issues, this invention provides a CAE-ML-based multi-objective optimization design and performance prediction method for SFRP composite materials. By integrating CAE simulation, machine learning, and multi-objective optimization algorithms, a closed-loop intelligent design system is constructed to achieve collaborative design and rapid prediction of material formulations, process parameters, and component performance.
[0011] The technical solution of this invention is described below in conjunction with the accompanying drawings:
[0012] This invention provides a multi-objective optimization design and performance prediction method for SFRP composite materials based on CAE-ML, comprising the following steps:
[0013] S1. Modeling is based on a multi-dimensional performance index system and cross-scale parameterization;
[0014] S2. Perform simulation calculations and construct high-dimensional datasets using CAE-based automated simulation.
[0015] S3. Establishment and training of a multi-task machine learning agent model that integrates micromorphology;
[0016] S4. Multi-angle performance collaborative optimization based on multi-objective optimization algorithm;
[0017] S5. Optimal design scheme decision and performance prediction.
[0018] Furthermore, the specific method of S1 is as follows:
[0019] S11. Define the design objectives of the component and determine its performance indicators; the performance indicators shall include at least two mechanical properties in different directions and one non-mechanical property.
[0020] S12. Perform three-dimensional parametric modeling of the composite material component and determine the design variables to be optimized, including process parameters and / or material parameters;
[0021] S13. Define a set of micromorphological quantization parameters.
[0022] Furthermore, the specific method of S2 is as follows:
[0023] S21. Sample points are generated in the design variable space using experimental design methods; simulation calculations are performed on each sample point by driving the injection molding simulation and structural mechanics simulation process through automated scripts.
[0024] S22. Extract the microstructure field data and macroscopic performance data corresponding to each sample to construct a high-dimensional "process-structure-performance" dataset.
[0025] Furthermore, the specific method of S3 is as follows:
[0026] S31. Using design variables as input, construct a hierarchical or multi-branch deep learning model; the first stage or an independent branch of the deep learning model is used to extract and predict the microscopic morphological quantization parameters defined in step S1 from the original microscopic structure field data output by CAE simulation; the second stage or other branches of the deep learning model use the design variables and the predicted microscopic morphological quantization parameters as input to predict the final multi-angle macroscopic performance indicators.
[0027] S32. Using the dataset constructed in step S2, perform end-to-end or phased training and validation on the fused deep learning model to obtain a surrogate model that can not only predict performance but also explain the causes of performance.
[0028] S33. Using the trained fusion model, perform global sensitivity analysis to quantify the contribution of each design variable and each micromorphological parameter to the final macroscopic performance indicators; generate a "process-morphology-performance" sensitivity map to identify key control factors.
[0029] Furthermore, the specific method of S4 is as follows:
[0030] S41. Establish a multi-objective optimization mathematical model, with the objective function being to maximize or minimize the multiple performance indicators determined in step S1, and the constraints being the feasible region of the design variables.
[0031] S42. Using the deep learning model trained in step S3 as a fast evaluator for the objective function and constraints, a multi-objective evolutionary algorithm is used to optimize and solve the problem, resulting in a set of Pareto optimal solutions.
[0032] Furthermore, the specific method of S5 is as follows:
[0033] S51. Select one or more optimal design schemes from the Pareto optimal solution set according to actual engineering needs;
[0034] S52. Input the design variables of the selected scheme into the surrogate model trained in step S3 to directly predict the microstructure characterization parameters and multi-angle macroscopic performance, and complete the performance verification.
[0035] The beneficial effects of this invention are as follows:
[0036] 1. Connecting the design and analysis chain: This invention is the first to tightly couple injection molding process, microstructure evolution and multi-angle macroscopic performance through the CAE-ML framework, realizing full-process digital mapping and optimization from the manufacturing end to the performance end.
[0037] 2. Fast prediction speed and high accuracy: By using a pre-trained machine learning proxy model to replace time-consuming CAE simulation, the performance prediction time is shortened from hours to seconds, while ensuring high prediction accuracy, which greatly improves design efficiency.
[0038] 3. Achieve true multi-angle, multi-objective optimization: By constructing a multi-task ML model and introducing a multi-objective optimization algorithm, it is possible to simultaneously optimize the multiple performances of components under different directions and different physical fields, solving the pain point of the single optimization objective of traditional methods, and providing designers with a rich set of optimal solutions after trade-offs.
[0039] 4. Highly exploratory: This method can efficiently explore high-dimensional design spaces, discover optimal process windows and material formulations that are difficult to reach with traditional experience, and help to tap the potential of materials and achieve performance breakthroughs.
[0040] 5. Achieved in-depth quantification of microstructure and interpretability of performance causes: By introducing a set of multi-dimensional micromorphological parameters, the originally "black box" microstructure is transformed into a series of quantifiable engineering indicators. The fusion model can not only predict "what" but also explain "why," revealing how the process affects the final performance by changing specific micromorphologies (such as reducing fiber breakage and improving distribution uniformity), greatly enhancing the theoretical depth and engineering guidance significance of the design process.
[0041] 6. Provides precise process control guidelines: Through the newly added cross-scale correlation sensitivity analysis, it can accurately identify which process parameter should be adjusted most to improve a specific performance, and how it mainly works by changing which microstructural feature. This provides a clear and direct roadmap for process optimization, avoiding blind adjustments. Attached Figure Description
[0042] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.
[0043] Figure 1 This is a schematic diagram of the overall process of the present invention;
[0044] Figure 2 A schematic diagram of the structure of the multi-task machine learning model being constructed;
[0045] Figure 3 A schematic diagram of a hierarchical machine learning model structure that incorporates micromorphology. Detailed Implementation
[0046] The present invention will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and not intended to limit it. Furthermore, it should be noted that, for ease of description, the accompanying drawings show only the parts relevant to the present invention, and not all of the structures.
[0047] Example 1
[0048] See Figure 1 This embodiment provides a multi-objective optimization design and performance prediction method for SFRP composite materials based on CAE-ML, including the following steps:
[0049] S1. Modeling is based on a multi-dimensional performance index system and cross-scale parameterization. The specific method is as follows:
[0050] S11. Define the design objectives of the component and determine the key performance indicators; the performance indicators shall include at least two mechanical properties in different directions and one non-mechanical property.
[0051] S12. Perform three-dimensional parametric modeling of the composite material component and determine the design variables to be optimized, including process parameters and / or material parameters;
[0052] S13. Define a set of micromorphological quantification parameters. These parameters include, but are not limited to:
[0053] Global and local fiber orientation tensors and their degree of anisotropy.
[0054] Statistical characteristics of fiber length distribution (such as number average length and volume average length).
[0055] Indicators of uniformity in local fiber volume fraction distribution (such as standard deviation).
[0056] Identification of fiber aggregates and statistical parameters (such as the number of aggregates and average size).
[0057] Porosity and its distribution.
[0058] S2. Automated simulation based on CAE is used for simulation calculations and the construction of high-dimensional datasets. The specific methods are as follows:
[0059] S21. Sample points are generated in the design variable space using experimental design methods; the injection molding simulation and structural mechanics simulation processes are driven by automated scripts to perform simulation calculations for each sample point; the injection molding simulation uses Moldex3D or Moldflow software, and the structural mechanics simulation uses Abaqus or ANSYS software, and the simulation process and data transfer are automated through Python or MATLAB scripts; S22. Microstructural field data (such as fiber orientation tensor field) and macroscopic performance data (such as isotropic / anisotropic elastic modulus, tensile strength, heat distortion temperature, etc.) corresponding to each sample are extracted to construct a high-dimensional "process-structure-performance" dataset.
[0060] S3. The establishment and training of a multi-task machine learning agent model integrating micromorphology, the specific method is as follows:
[0061] S31. Using design variables as input, construct a hierarchical or multi-branch deep learning model; the first stage or an independent branch of the deep learning model is used to extract and predict the microscopic morphological quantification parameters defined in step S1 from the raw microscopic structural field data (such as fiber orientation and concentration field) output by CAE simulation; the second stage or other branches of the deep learning model use the design variables and the predicted microscopic morphological quantification parameters as input to predict the final multi-angle macroscopic performance indicators; wherein, the branch used to extract microscopic morphological parameters can use convolutional neural networks or graph convolutional networks to better process the field data;
[0062] S32. Using the dataset constructed in step S2, perform end-to-end or phased training and validation on the fused deep learning model to obtain a surrogate model that can not only predict performance but also explain the causes of performance.
[0063] S33. Using the trained fusion model, perform global sensitivity analysis (such as the Sobol index-based method) to quantify the contribution of each design variable (process parameter) and each micromorphological parameter to the final macroscopic performance indicators; and generate a "process-morphology-performance" sensitivity map based on the contribution to identify key control factors.
[0064] S4. Multi-angle performance collaborative optimization based on multi-objective optimization algorithm, the specific method is as follows:
[0065] S41. Establish a multi-objective optimization mathematical model, with the objective function being to maximize or minimize the multiple performance indicators determined in step S1, and the constraints being the feasible region of the design variables.
[0066] S42. Using the deep learning model trained in step S3 as a fast evaluator for the objective function and constraints, a multi-objective evolutionary algorithm is employed for optimization to obtain a set of Pareto optimal solutions. Optional constraints for the optimization model may include micromorphological parameters, such as "fiber orientation distribution uniformity index > a certain threshold"; the multi-objective evolutionary algorithm is NSGA-II, NSGA-III, or MOEA / D algorithm.
[0067] S5. Optimal design scheme decision and performance prediction, the specific methods are as follows:
[0068] S51. Based on actual engineering requirements, select one or more optimal design schemes from the Pareto optimal solution set; if an optimal solution appears under the given conditions, that is sufficient.
[0069] S52. Input the design variables of the selected scheme into the surrogate model trained in step S3 to directly predict the microstructure characterization parameters and multi-angle macroscopic performance, and complete the performance verification.
[0070] Example 2
[0071] This embodiment uses a short glass fiber reinforced polylactic acid (PLA+SBF40 wt%) front hatch inner panel as an example to conduct multi-objective optimization design and performance prediction, which meets the national pedestrian protection requirements. The specific method is as follows:
[0072] S1. Modeling is based on a multi-dimensional performance index system and cross-scale parameterization;
[0073] Performance Indicators: Table 1 shows the optimized design results of the composite material front hatch cover's performance indicators, including mass, stiffness, and main low-order modes, as well as the pedestrian protection performance index HIC. 15 .
[0074] Table 1 Optimization Design Results
[0075] Name [y1 mass / kg] [ y2 torsional deformation / mm ] [y3 impact deformation / mm] [ y4 free mode / Hz ] HIC 15 ]] Indicator value 17.85 ≤1.0 ≤1.0 ≥25.0 ≤1000 After optimization 5.55 0.1933 0.1191 40.87 All meet
[0076] Design variable: Melt temperature (T) m Injection speed (Vinj), holding pressure (Ppack), and holding time (tpack).
[0077] Parametric modeling: Create a 3D model of the front hatch cover in CAD software.
[0078] Quantitative parameters of micromorphology: Definitions: (a) Average fiber orientation factor (between 0 and 1, where 1 indicates complete orientation); (b) Fiber length retention rate (the ratio of final average length to initial average length).
[0079] S2. Perform simulation calculations and construct high-dimensional datasets using CAE-based automated simulation.
[0080] Latin hypercube sampling was used to generate 200 sample points in the four-dimensional design variable space.
[0081] A Python script was written to automatically call the Moldflow API to perform injection molding simulation and obtain the fiber orientation tensor field. Then, this orientation field was used as a material property to call an Abaqus Python script for structural and thermal analysis, calculating E1, IS2, and HDT respectively.
[0082] Finally, a dataset containing 200 samples was constructed, each containing 4 input variables and multiple output responses (microstructure parameters, performance parameters (Table 1)).
[0083] During data extraction, the microscopic morphological parameters defined above are calculated from the CAE results using image processing algorithms and added to the dataset.
[0084] S3. Establishment and training of a multi-task machine learning agent model that integrates micromorphology;
[0085] Construct a hierarchical model:
[0086] The first layer (microstructure encoder): a small CNN, which takes as input a fiber orientation field slice from CAE simulation output and outputs five predicted morphological parameters (a), (b), (c), (d), and (e).
[0087] The second layer (performance predictor): an MLP whose input is [process parameters, morphological parameters predicted by the first layer], and whose output is performance parameters (Table 1).
[0088] First, pre-train the first layer, then fine-tune the entire model end-to-end.
[0089] The Sobol sensitivity analysis method was used to analyze the contribution of four process parameters and three morphological parameters to the performance parameters (Table 1). The results may show that injection speed mainly dominates stiffness by affecting the average fiber orientation factor; while holding pressure has a significant impact on fiber length retention.
[0090] S4. Multi-angle performance collaborative optimization based on multi-objective optimization algorithm;
[0091] The optimization model can be set as: Maximize [y1(X), y2(X), y3(X), y4(X), HIC15(X)], while adding the constraint that the fiber length retention rate is > 0.8 to ensure toughness; a multi-objective evolutionary algorithm is used to optimize and solve the problem, and a set of Pareto optimal solutions is obtained, as shown in Table 2.
[0092] Table 2 Fitting Accuracy Verification
[0093] Number [y1 - mass / kg] [ y2 - torsional deformation / mm ] [y3 - impact deformation / mm] [y4 - mode / Hz] [R 2 ]] 0.994 0.952 0.988 0.94 RMSE 0.0288 0.0812 0.0218 0.0682
[0094] S5. Optimal design scheme decision and performance prediction;
[0095] Designers select the optimal solution from the Pareto set based on the specific product emphasis (e.g., prioritizing stiffness and heat resistance).
[0096] By inputting the corresponding process parameters into the trained ML model, the fiber orientation state and performance parameters of the component under this scheme can be predicted instantly (Table 1), confirming that it meets all design requirements. No further time-consuming CAE simulation verification is needed.
[0097] In summary, this invention achieves rapid and accurate mapping and optimization from process to performance, solving the problems of low efficiency and single optimization target of traditional methods, and significantly improving the design and development efficiency of composite material products.
[0098] 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 multi-objective optimization design and performance prediction method for SFRP composite materials based on CAE-ML, characterized in that, Includes the following steps: S1. Modeling is based on a multi-dimensional performance index system and cross-scale parameterization; S2. Perform simulation calculations and construct high-dimensional datasets using CAE-based automated simulation. S3. Establishment and training of a multi-task machine learning agent model that integrates micromorphology; S4. Multi-angle performance collaborative optimization based on multi-objective optimization algorithm; S5. Optimal design scheme decision and performance prediction.
2. The method for multi-objective optimization design and performance prediction of SFRP composite materials based on CAE-ML according to claim 1, characterized in that, The specific method of S1 is as follows: S11. Define the design objectives of the component and determine its performance indicators; the performance indicators shall include at least two mechanical properties in different directions and one non-mechanical property. S12. Perform three-dimensional parametric modeling of the composite material component and determine the design variables to be optimized, including process parameters and / or material parameters; S13. Define a set of micromorphological quantization parameters.
3. The method for multi-objective optimization design and performance prediction of SFRP composite materials based on CAE-ML according to claim 1, characterized in that, The specific method of S2 is as follows: S21. Sample points are generated in the design variable space using experimental design methods; simulation calculations are performed on each sample point by driving the injection molding simulation and structural mechanics simulation process through automated scripts. S22. Extract the microstructure field data and macroscopic performance data corresponding to each sample to construct a high-dimensional "process-structure-performance" dataset.
4. The method for multi-objective optimization design and performance prediction of SFRP composite materials based on CAE-ML according to claim 2, characterized in that, The specific method of S3 is as follows: S31. Using design variables as input, construct a hierarchical or multi-branch deep learning model; the first stage or an independent branch of the deep learning model is used to extract and predict the micromorphological quantization parameters defined in step S1 from the original microstructure field data output by CAE simulation. The second stage or other branch of the deep learning model uses the design variables and the predicted micromorphological quantification parameters as inputs to predict the final multi-angle macroscopic performance indicators. S32. Using the dataset constructed in step S2, perform end-to-end or phased training and validation on the fused deep learning model to obtain a surrogate model that can not only predict performance but also explain the causes of performance. S33. Using the trained fusion model, perform global sensitivity analysis to quantify the contribution of each design variable and each micromorphological parameter to the final macroscopic performance indicators. Generate a sensitivity map of "process-morphology-performance" to identify key control factors.
5. The method for multi-objective optimization design and performance prediction of SFRP composite materials based on CAE-ML according to claim 1, characterized in that, The specific method of S4 is as follows: S41. Establish a multi-objective optimization mathematical model, with the objective function being to maximize or minimize the multiple performance indicators determined in step S1, and the constraints being the feasible region of the design variables. S42. Using the deep learning model trained in step S3 as a fast evaluator for the objective function and constraints, a multi-objective evolutionary algorithm is used to optimize and solve the problem, resulting in a set of Pareto optimal solutions.
6. The method for multi-objective optimization design and performance prediction of SFRP composite materials based on CAE-ML according to claim 5, characterized in that, The specific method of S5 is as follows: S51. Select one or more optimal design schemes from the Pareto optimal solution set according to actual engineering needs; S52. Input the design variables of the selected scheme into the surrogate model trained in step S3 to directly predict the microstructure characterization parameters and multi-angle macroscopic performance, and complete the performance verification.