Method for designing key manufacturing parameters of aluminum oxide three-period minimal curved surface-steel interpenetrating structure in additive manufacturing based on machine learning

Abstract

The invention discloses a method for designing key manufacturing parameters of an aluminum oxide additive manufacturing three-period minimal curved surface-steel interpenetrating structure based on machine learning, and relates to the field of composite material additive manufacturing and structure optimization design. Comprising the following steps: establishing three-dimensional models with different printing densities and curved surface thicknesses; preparing a test sample under corresponding parameters; obtaining mechanical property training data by adopting finite element simulation; performing a mechanical experiment on the experimental sample to obtain verification data; constructing and training a KNN model, and training the model by using the mechanical property training data; and printing density and curved surface thickness parameters are designed by adopting a gradient descent algorithm. According to the method, through combination of machine learning and an optimization algorithm, the problem that key parameters of the aluminum oxide three-period minimal curved surface-steel interpenetrating structure are difficult to determine in the additive manufacturing process is effectively solved.

Description

Technical Field

[0001] This invention belongs to the field of additive manufacturing parameter and structural optimization design technology of interpenetrating composite materials, and particularly relates to a key manufacturing parameter design method for additive manufacturing alumina three-period minimal surface-steel interpenetrating structure based on machine learning. Background Technology

[0002] Tri-periodic minimal surface structures are widely used in aerospace, biomedicine, and other fields due to their high specific strength and high energy absorption capacity. Combining such structures with metal matrices to form interpenetrating phase composite materials is of great significance for the research of high-performance composite materials.

[0003] The mechanical properties of the interpenetrating material formed by the additively manufactured alumina ceramic three-period minimal surface structure and steel depend on the printing and shape parameters of the three-period minimal surface, namely the printing density and the final surface thickness during the additive manufacturing process. However, the relationship between printing density, surface thickness, and the final macroscopic mechanical properties exhibits a highly complex nonlinear relationship. Traditional design methods heavily rely on engineers' experience and repeated trial-and-error experiments, resulting in long development cycles, high costs, and difficulty in obtaining globally optimal solutions.

[0004] Machine learning has been explored for predicting material properties, offering new research avenues for materials science. However, its application to the reverse design of additive manufacturing parameters for complex interpenetrating structures such as alumina three-period minimal surfaces-steel presents a lack of reverse design pathways. Existing research largely focuses on "forward prediction," predicting performance given parameters, while neglecting the more pressing engineering problem of "reverse design"—finding optimal parameters given target performance—lacking a reverse design technique that combines prediction models with optimization algorithms. Therefore, this paper proposes a machine learning-based method for designing key manufacturing parameters of alumina three-period minimal surfaces-steel interpenetrating structures. The KNN algorithm combined with gradient descent is an innovative aspect of this invention. Summary of the Invention

[0005] To address the challenges in determining the additive manufacturing printing density and the thickness of the interpenetrating phase minimal surface in interpenetrating composite materials, as mentioned in this invention, this invention proposes a machine learning-based method for designing key manufacturing parameters of an additively manufactured alumina three-period minimal surface-steel interpenetrating structure. This method combines finite element simulation data with physical experimental data to construct an efficient machine learning prediction model and utilizes optimization algorithms to achieve reverse design of manufacturing parameters, implemented according to the following steps:

[0006] Step 1: Construct a parametric model library for alumina three-period minimal surface-steel interpenetrating structure: Establish three-dimensional models of alumina three-period minimal surface-steel interpenetrating structure with different printing densities and different surface thicknesses;

[0007] Step 2: Obtaining the mechanical property dataset: Based on the three-dimensional model in Step 1, prepare the corresponding alumina three-period minimal surface-steel interpenetrating structure experimental specimen; use finite element simulation software to simulate the mechanical properties of the three-dimensional model to obtain mechanical property training data; and conduct mechanical experiments on the experimental specimen to obtain mechanical property verification data; use the verification data to verify the training data and continuously adjust the finite element model to finally form a high-quality mechanical property training dataset.

[0008] Step 3: Construct and train the mechanical performance prediction model: Establish a KNN model and train it using the training dataset formed in Step 2. Before training, normalize the mechanical performance data and scale it to the [0,1] interval. The calculation formula is as follows:

[0009]

[0010] Where X is the original data, X min and X max X represents the minimum and maximum values ​​of this feature data, respectively. n The data is normalized; the KNN model uses Euclidean distance as the distance metric, and its calculation formula is as follows:

[0011]

[0012] Where x and y represent two sample points, and n is the number of features, x i and y i The sample points are defined by their respective feature values. Then, the optimal number of nearest neighbors K is determined using cross-validation, and the model is tested using the validation dataset.

[0013] Step 4: Key Manufacturing Parameter Optimization Design: Taking the preset target mechanical properties as the optimization objective, construct the following optimization problem:

[0014] L(ω)=|P T -P P (ω)| (3)

[0015] Where ω=(ρ,t) represents the combination of manufacturing parameters, including printing density ρ and surface thickness t, P T P is the preset target mechanical property value. P (ω) represents the mechanical performance value predicted by the KNN model trained in step three based on the parameter ω. Using the gradient descent algorithm, based on the KNN model trained and verified in step three, the printing density and surface thickness are iteratively optimized until convergence, and the optimal combination of manufacturing parameters is output.

[0016] Furthermore, in step one, the three-period minimal surface structure types include three types: Gyroid, Diamond, and Primitive.

[0017] Furthermore, in step two, the alumina three-period minimal surface structure in the sample preparation is prepared by additive manufacturing. During the additive manufacturing process, samples with different printing densities and corresponding different surface thicknesses are prepared for each type of three-period minimal surface. Then, the alumina preform is printed by a DLP photopolymer ceramic 3D printer and densified by atmospheric pressure sintering.

[0018] Furthermore, in step two, the mechanical property data includes yield strength, tensile strength, and elastic modulus.

[0019] Furthermore, in step two, the finite element simulation is a nonlinear quasi-static compression simulation, and the mechanical experiment is a corresponding quasi-static compression experiment to ensure data consistency.

[0020] Furthermore, in step three, the distance metric of the KNN model is Euclidean distance.

[0021] Furthermore, in step three, the number of nearest neighbors K in the KNN model is determined using the ten-fold cross-validation method.

[0022] Furthermore, in step four, the gradient descent algorithm employs the Adam optimizer.

[0023] Explanation of the rationality of the technical solution

[0024] The rationality and feasibility of the technical solution of this invention are mainly reflected in the following aspects:

[0025] First, the scientific rigor of the dataset lies in its simultaneous consideration of the advantages of both simulation and experiment. Specifically, a representative and universally applicable training dataset is generated using high-quality finite element simulations; simultaneously, physical experiments are used as the verification standard to validate the simulation model, ensuring the data's authenticity. Both work together to construct a high-quality dataset, providing crucial support for training high-precision prediction models.

[0026] Second, although the KNN model is essentially a nonparametric model, it establishes a definite "parameter-performance" mapping relationship within the input space. Based on this, this invention treats the trained KNN model as a surrogate model and embeds it into the gradient descent optimization framework. During the iteration process, the optimizer calls this surrogate model to obtain the performance prediction value of the current parameter point to calculate the objective function, and then determines the search direction based on the changing trend of the function value. Modern optimizers such as Adam do not need to rely on the analytical gradient of the model; they can determine the search direction solely based on the objective function value. This strategy cleverly circumvents the nondifferentiability barrier of the model by combining the KNN surrogate model with a derivativeless optimization algorithm, ultimately achieving inverse optimization design from performance objectives to manufacturing parameters. Attached Figure Description

[0027] Figure 1 This is a schematic diagram of the KNN model training of the present invention.

[0028] Figure 2 This is a flowchart of the Adam optimizer optimization process of the present invention.

[0029] Figure 3 This is the overall flowchart of the present invention. Detailed Implementation

[0030] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. These embodiments will help those skilled in the art to further understand the present invention, but do not limit the invention in any way.

[0031] Example 1:

[0032] This embodiment takes the Gyroid-type alumina three-period minimal surface-steel interpenetrating structure as an example to illustrate the implementation process of the present invention in detail.

[0033] Step one: Using 3D modeling software, generate multiple sets of 3D models of the Gyroid structure, each containing different combinations of printing densities and surface thickness parameters. All models maintain a uniform cubic configuration to ensure geometric consistency.

[0034] Step 2: Obtain the mechanical properties dataset.

[0035] (1) Finite element simulation: Import the three-dimensional model generated in step one into the finite element analysis software. In the software, the alumina phase is set as an elastic-brittle material and the steel phase is set as an elastic-plastic material, and the interface relationship between the two phases is defined. Uniaxial quasi-static compression simulation is performed to calculate the stress-strain curves of each model, and mechanical property indicators such as yield strength and tensile strength are extracted from them as training data.

[0036] (2) Physical Experiments: Several representative sets of parameters were selected from the parameter combinations, and corresponding alumina-steel interpenetrating structure experimental specimens were prepared using additive manufacturing technology and corresponding post-processing techniques. Quasi-static compression tests were conducted on a mechanical testing machine to obtain the corresponding mechanical property verification data.

[0037] (3) Data preprocessing: The training data obtained from the simulation is verified using the validation data obtained from the experiment. The validated training dataset is normalized and scaled to the [0,1] interval to prepare for model training.

[0038] Step 3: Build and train the KNN prediction model.

[0039] A KNN regression model was constructed using printing density and surface thickness as input features and normalized mechanical properties as the output target. Cross-validation was employed to optimize the nearest neighbor count K on the training set to determine the optimal K value that minimizes the model's prediction error. After training, the model's performance was validated using a physical experiment dataset. The results show that the model's predictions are in good agreement with the actual values, demonstrating that the established KNN model has reliable prediction accuracy.

[0040] Step 4: Optimize the design of key manufacturing parameters.

[0041] A specific target mechanical property is set as the optimization objective. This performance index is used as the optimization objective function, with a trained KNN model as the core surrogate model, and printing density and surface thickness as optimization variables. The Adam optimizer is used for iterative optimization, with an appropriate initial learning rate. After multiple iterations, the optimization process converges, outputting the optimal combination of manufacturing parameters. Comparative verification shows that the structural performance under these optimized parameters meets the performance target.

[0042] In summary, this invention effectively solves the problem of determining the additive manufacturing parameters for alumina three-period minimal curved surface-steel interpenetrating structures, and provides a new technical path for the intelligent design of high-performance interpenetrating composite materials.

[0043] The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for designing key manufacturing parameters of additive manufacturing alumina three-period minimal surface-steel interpenetrating structure based on machine learning, characterized in that, The key manufacturing parameter design method for the additive manufacturing alumina three-period minimal surface-steel interpenetrating structure based on machine learning includes the following steps: Step 1: Construct a parametric model library for alumina three-period minimal surface-steel interpenetrating structure: Establish three-dimensional models of alumina three-period minimal surface-steel interpenetrating structure with different printing densities and different surface thicknesses; Step 2: Prepare experimental specimens and obtain mechanical property data: Based on the three-dimensional model in Step 1, prepare corresponding alumina three-period minimal surface-steel interpenetrating structure experimental specimens; use finite element simulation software to simulate the mechanical properties of the three-dimensional model and obtain mechanical property training data; and conduct mechanical experiments on the experimental specimens to obtain mechanical property verification data. Step 3: Construct and train the mechanical performance prediction model: Build a K-nearest neighbor (KNN) model, train the KNN model using the mechanical performance training data obtained in Step 2, and verify the trained model using the mechanical performance verification data; Step 4: Key Manufacturing Parameter Optimization Design: With the preset mechanical properties as the target, the gradient descent algorithm is used to iteratively optimize the printing density and surface thickness based on the K-nearest neighbor (KNN) model trained and verified in Step 3, and output the optimal combination of manufacturing parameters.

2. The method according to claim 1, characterized in that, In step one, the three-period minimal surface structure types include three types: Gyroid, Diamond, and Primitive.

3. The method according to claim 1, characterized in that, In step two, the alumina three-period minimal surface structure in the sample preparation is prepared by additive manufacturing. During the additive manufacturing process, samples with different printing densities and corresponding surface thicknesses are prepared for each type of three-period minimal surface. Then, the alumina preform is printed by a DLP photopolymer ceramic 3D printer and densified by atmospheric pressure sintering.

4. The method according to claim 1, characterized in that, In step two, the mechanical property data includes yield strength, tensile strength, and elastic modulus.

5. The method according to claim 1, characterized in that, In step two, the finite element simulation is a nonlinear quasi-static compression simulation.

6. The method according to claim 1, characterized in that, In step three, the distance metric used in the KNN model is Euclidean distance.

7. The method according to claim 1, characterized in that, In step three, before training the KNN model, the mechanical performance training data is normalized.

8. The method according to claim 1, characterized in that, In step four, the gradient descent algorithm is implemented using the Adam optimizer.

Images