Machine learning based optimization design method for additive manufacturing process parameters

Abstract

The application provides an optimization design method of additive manufacturing process parameters based on machine learning, which can realize rapid and accurate prediction of product forming quality under any process parameters for a wide range of material systems by establishing an additive manufacturing process parameter-material performance gradient boosting regression tree (GBDT) model and double optimizing the model hyperparameters by using random search (RS) and K-fold cross validation (K-CV) algorithms, so as to quickly and accurately determine the best process parameters, and solve the problems of high calculation and test cost and long cycle in the optimization of the process parameter window of laser additive manufacturing.

Description

Technical Field

[0001] This invention relates to the field of metal material preparation technology, and more specifically to a method for optimizing additive manufacturing process parameters based on machine learning. Background Technology

[0002] Additive manufacturing (also known as 3D printing) is a novel digital forming technology that integrates manufacturing technology, information technology, and new materials technology. It is widely used in strategic emerging fields such as aerospace, defense, rail transportation, and biomedicine, and has been listed by McKinsey & Company as one of the twelve disruptive technologies that will determine the future economy. Unlike traditional subtractive or equal-material manufacturing technologies, additive manufacturing is based on the discrete-stacking principle. It uses heat sources such as high-energy laser beams, electric arcs, or electron beams to effectively fuse materials, thereby achieving direct manufacturing driven by three-dimensional data. Therefore, it has low production costs, short cycles, and high freedom in material design, making it particularly suitable for the high-precision and efficient fabrication of complex structural parts. Among these technologies, laser powder bed melting technology, with its advantages of small laser focusing spot, thin powder bed thickness, and rapid melting process, has shown unique advantages and attracted widespread attention in the high-precision fabrication of high-performance complex components.

[0003] Laser powder bed additive manufacturing is a complex nonlinear process involving the continuous interaction of multiple process parameters, including laser power, scanning speed, scanning spacing, and powder layer thickness. Therefore, the selection of these parameters directly impacts product quality. To date, engineering practice still heavily relies on traditional experimental trial-and-error methods and finite element simulation to design and optimize additive manufacturing process parameters within a relatively narrow scope. However, these methods struggle to flexibly and efficiently explore the multi-parameter space of additive manufacturing, and the optimal process parameters vary significantly across different material systems. This results in the process parameter window optimization phase consuming substantial time and resources, hindering rapid product preparation and iterative upgrades, and severely restricting the advancement and widespread application of laser powder bed additive manufacturing technology. Summary of the Invention

[0004] The main objective of this invention is to provide a machine learning-based method for optimizing additive manufacturing process parameters. This method establishes a GBDT model of additive manufacturing process parameters and material properties, and uses random search (RS) and K-Fold Cross Validation (K-CV) algorithms to optimize the model hyperparameters. This enables rapid and accurate prediction of product forming quality under arbitrary process parameters for a wide range of material systems, thereby quickly and accurately determining the optimal process parameters. This solves the problems of high computational and experimental costs and long cycles in optimizing process parameter windows in laser additive manufacturing.

[0005] To achieve the above objectives, this invention provides a machine learning-based method for optimizing additive manufacturing process parameters.

[0006] The machine learning-based method for optimizing additive manufacturing process parameters includes the following steps:

[0007] S1, establish a dataset; wherein the dataset is divided into a training set and a test set; the training set contains multiple sets of training data, and each set of training data includes a first process parameter and its corresponding first material property data;

[0008] S2, Use the training set to train the model and establish an initial GBDT model of additive manufacturing process parameters-material properties;

[0009] S3. The hyperparameters of the initial GBDT model are optimized using Random Search and K-Fold Cross Validation algorithms to obtain an improved GBDT model; where K = 5 and 10, and the hyperparameters include the number of trees (n_estimator), learning rate (learning_rate), maximum tree depth (max_depth), and subsample ratio (subsample).

[0010] S4, using the test set, evaluate and optimize the prediction performance of the improved GBDT model to obtain the optimal GBDT model; wherein, the evaluation index used is the coefficient of determination R. 2 Mean Absolute Percentage Error (MAPE);

[0011] S5, input multiple sets of preset process parameters into the GBDT optimal model to obtain multiple sets of first predicted material properties corresponding to them;

[0012] S6. Determine the optimal process parameters based on the first prediction data of the multiple sets of material properties.

[0013] Furthermore, the test set contains multiple sets of test data, and each set of test data includes a second process parameter and its corresponding second material performance data.

[0014] Furthermore, the second process parameter includes at least one of the following: laser power, scanning speed, scanning spacing, and powder layer thickness;

[0015] The second material property data includes at least one of density, tensile strength, and elongation.

[0016] Furthermore, the evaluation and optimization design of the GBDT improved model using the test set includes the following steps:

[0017] S4-1, Input the second process parameters into the GBDT improved model to obtain the second predicted data of material properties;

[0018] S4-2, using the second predicted data of material properties and the second material property data, the coefficient of determination R is calculated. 2 and mean absolute percentage error (MAPE);

[0019] S4-3, when the coefficient of determination R 2 <Preset threshold or mean absolute percentage error (MAPE)> After setting the preset threshold, reset the RS algorithm to optimize the hyperparameter range and iteration number, and repeat steps S3 and S4-1 to S4-2 until the coefficient of determination R... 2 If the model hyperparameter optimization design is completed, the result is ≥ the preset threshold and the mean absolute percentage error (MAPE) ≤ the preset threshold.

[0020] Furthermore, the optimization ranges for the number of trees, learning rate, maximum tree depth, and subsampling ratio in the hyperparameters are 1–1000, 0.01–0.5, 1–10, and 0–1, respectively, and the number of iterations is 100–1500.

[0021] Preferably, the R 2 The preset thresholds for MAPE are 0.95% and 5%, respectively.

[0022] Furthermore, the creation of the dataset includes:

[0023] S1-1, Obtain process parameters and material performance data;

[0024] S1-2, classify the process parameters and material performance data according to the material system type to obtain the initial dataset of process parameters and material performance corresponding to different material systems;

[0025] S1-3, Normalize the initial dataset to obtain the dataset.

[0026] Furthermore, the material systems include, but are not limited to, titanium alloys, stainless steel, copper alloys, zinc alloys, and high-entropy alloys.

[0027] Furthermore, the amount of data in the training set accounts for 70-80% of the total amount of data in the dataset, the amount of data in the test set accounts for 20-30% of the total amount of data in the dataset, and the sum of the ratios of the training set and the test set is always 1.

[0028] Furthermore, the first process parameter includes at least one of laser power, scanning speed, scanning spacing, and powder layer thickness;

[0029] The first material property data includes at least one of density, tensile strength, and elongation.

[0030] Furthermore, it also includes the following steps:

[0031] S7. Using the optimal process parameters, the material is prepared by additive manufacturing technology, and the resulting part is subjected to relevant performance tests to obtain experimental data.

[0032] S8, add the optimal process parameters and the experimental data to the dataset as a new set of data in the dataset;

[0033] S9. Repeat steps S1 to S4 to further optimize the GBDT optimal model.

[0034] This invention establishes a GBDT (Gradient Boosting Regression Tree) prediction model for additive manufacturing process parameters and part performance based on a data-driven machine learning method. It then uses random search (RS) and K-Fold Cross Validation (K-CV) algorithms to perform dual optimization of the model's hyperparameters to obtain an improved GBDT model. This enables efficient and accurate prediction of part performance corresponding to unknown process parameters for a wide range of material systems, thereby achieving rapid response fabrication of additive manufacturing parts.

[0035] The machine learning-based optimization design method for additive manufacturing process parameters in this invention has the following advantages:

[0036] 1) Compared with traditional experimental trial-and-error methods and physical model-based simulation methods, the machine learning-based additive manufacturing process parameter optimization design method proposed in this invention can fully explore the implicit relationship between additive manufacturing process parameters and part performance by analyzing a large amount of existing production and research data. Thus, without fully understanding the internal mechanism of additive manufacturing technology, a process parameter-material performance prediction model can be established, ultimately achieving efficient and low-cost optimization design of additive manufacturing process parameters, significantly reducing production costs and cycle time.

[0037] 2) Compared with the LR model (linear regression model), the GBDT model proposed in this invention has higher prediction accuracy when learning complex multi-dimensional samples and complex nonlinear relationships; compared with the DNN (deep neural network) model, the GBDT prediction model proposed in this invention significantly reduces the model training time while ensuring prediction accuracy, and achieves fast and accurate prediction.

[0038] 3) Compared with traditional shallow machine learning models based on statistical strategies, the dual optimization method for GBDT prediction models proposed in this invention, which combines RandomSearch with K-Fold Cross Validation algorithm, can significantly improve prediction accuracy and eliminate the need for manual parameter tuning, thereby effectively reducing labor costs and shortening the prediction cycle. Attached Figure Description

[0039] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings:

[0040] Figure 1 A flowchart of the optimized design method provided in the embodiments of the present invention. Detailed Implementation

[0041] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0042] This invention develops a new method to unlock the complex implicit relationships between multiple process parameters and part performance during component forming, enabling accurate prediction of product forming quality under arbitrary process parameters for a wide range of material systems. This allows for rapid selection of optimal process parameters, solving the problems of long optimization cycles and high costs associated with laser additive manufacturing technology.

[0043] This invention first obtains process parameters and performance data through material databases, literature, experiments, and simulation calculations to establish a high-quality sample database. On this basis, machine learning methods are used to unlock the complex implicit relationship between the main process parameters and performance of additive manufacturing through data mining, and a process parameter-part performance machine learning model is constructed. Then, the designed process parameters (preset process parameters) are input into the model to obtain their corresponding part performance, thereby quickly and accurately determining the optimal process parameters.

[0044] The machine learning-based additive manufacturing process parameter optimization design method of this invention includes the following steps:

[0045] S1, Create a dataset; the dataset is divided into a training set and a test set.

[0046] In embodiments of the present invention, the amount of data in the training set accounts for 70-80% of the total amount of data in the dataset, the amount of data in the test set accounts for 20-30% of the total amount of data in the dataset, and the sum of the ratios of the training set and the test set is always 1.

[0047] In embodiments of the present invention, the training set contains multiple sets of training data, and each set of training data includes a first process parameter and corresponding first material property data. The first process parameter serves as input data, and the first material property data serves as output data.

[0048] In embodiments of the present invention, the test set contains multiple sets of test data, and each set of test data includes a second process parameter and corresponding second material property data. The second process parameter serves as input data, and the second material property data serves as output data.

[0049] As some embodiments of the present invention, the first process parameter includes at least one of laser power, scanning speed, scanning spacing, and powder layer thickness.

[0050] As some embodiments of the present invention, the first material property data includes at least one of density, tensile strength, and elongation.

[0051] As some embodiments of the present invention, the second process parameter includes at least one of laser power, scanning speed, scanning spacing, and powder layer thickness.

[0052] As some embodiments of the present invention, the second material property data includes at least one of density, tensile strength, and elongation.

[0053] It is worth mentioning that the creation of the dataset includes:

[0054] S1-1: Obtain process parameters and material performance data from materials databases, publicly published high-quality literature, and experimental data to establish a basic dataset.

[0055] S1-2, classify and process the process parameters and material performance data according to the material system type to obtain the initial dataset of process parameters and material performance corresponding to different material systems.

[0056] In embodiments of the present invention, the material system includes, but is not limited to, titanium alloys, stainless steel, copper alloys, zinc alloys, and high-entropy alloys.

[0057] S1-3, normalize the initial dataset to obtain the final dataset.

[0058] S2, use the training set to train the model and establish the initial GBDT model of additive manufacturing process parameters-material properties.

[0059] It is worth mentioning that the present invention uses a training set for model training, with the first process parameter in the training set as input data and the first material property data as output data. The number of trees (n_estimator) can be set to 100, the learning rate (learning_rate) to 0.02, the maximum tree depth (max_depth) to 3, and the subsample ratio (subsample) to 1.

[0060] Of course, the settings of the above four hyperparameters can be adjusted as needed, without specific limitations.

[0061] S3. The hyperparameters of the initial GBDT model are optimized using both Random Search and K-Fold Cross Validation algorithms to obtain an improved GBDT model. K = 5 and 10. The Random Search algorithm optimizes the hyperparameters n_estimator, learning_rate, max_depth, and subsample ranges as 1 to 1000, 0.01 to 0.5, 1 to 10, and 0 to 1, respectively. The number of iterations n_iter is 100 to 1500.

[0062] It should be noted that the value of K can be 5 or 10, and of course, it can be designed according to actual needs without specific requirements.

[0063] S4. The GBDT improved model is evaluated and optimized using the test set to improve its prediction performance and generalization ability, thus obtaining the best GBDT model.

[0064] The model evaluation and optimization design specifically includes the following steps:

[0065] S4-1, Input the second process parameters into the GBDT improved model to obtain the second predicted data of material properties;

[0066] S4-2, using the second predicted data of material properties and the second material property data, the coefficient of determination R is calculated. 2 and mean absolute percentage error (MAPE);

[0067] S4-3, Determine the coefficient of determination R. 2 Whether it is greater than or equal to a preset threshold or whether the mean absolute percentage error (MAPE) is less than or equal to a preset threshold;

[0068] When the coefficient of determination R 2 If the model hyperparameter optimization design is completed when the mean absolute percentage error (MAPE) is ≥ the preset threshold and ≤ the preset threshold, the GBDT improved model is the optimal GBDT model.

[0069] When the coefficient of determination R2 <Preset threshold or mean absolute percentage error (MAPE)> After setting the preset threshold, reset the RS algorithm optimization parameters (n_estimator, learning_rate, max_depth, subsample) range and iteration count (n_iter), and repeat optimization design steps S3 and S4-1 to S4-2 until the coefficient of determination R... 2 If the mean absolute percentage error (MAPE) is greater than or equal to the preset threshold and less than or equal to the preset threshold, the optimization design is completed, and the best GBDT model is obtained.

[0070] In an embodiment of the present invention, R 2 The preset thresholds for MAPE can be 0.95% and 5%, respectively.

[0071] In embodiments of the present invention, the model evaluation index used includes the coefficient of determination R. 2 And mean absolute percentage error (MAPE), R 2 Its value is in the range of 0 to 1, and the closer it is to 1, the higher the model accuracy. The smaller the MAPE value, the higher the model accuracy.

[0072] S5 inputs multiple sets of preset process parameters into the GBDT optimal model to obtain multiple sets of first predicted data of material properties corresponding to them.

[0073] In embodiments of the present invention, multiple sets of process parameters can be selected from a pre-set range of process parameters and input into the GBDT optimal model to obtain corresponding multiple sets of first predicted data on material properties.

[0074] S6 determines the optimal process parameters based on multiple sets of first-prediction data of material properties.

[0075] In embodiments of the present invention, the optimal process parameters can be screened and determined by comparing and analyzing the performance requirements of the material to be obtained with the first predicted data of the material performance. This enables rapid and accurate prediction of product forming quality under any process parameters for a wide range of material systems, solving the problems of long optimization cycles and high costs in laser additive manufacturing technology.

[0076] In order to achieve faster and more accurate design of additive manufacturing process parameters, this invention optimizes the model based on the above-mentioned machine learning-based additive manufacturing process parameter design method to further improve the model accuracy.

[0077] The machine learning-based additive manufacturing process parameter optimization design method of this invention includes the following steps:

[0078] S1. Establish a dataset; the dataset is divided into a training set and a test set; the training set contains multiple sets of training data, and each set of training data includes the first process parameter and the corresponding first material property data.

[0079] S2, use the training set to train the model and establish the initial GBDT model of additive manufacturing process parameters-material properties.

[0080] S3 uses Random Search and K-Fold Cross Validation algorithms to optimize the hyperparameters of the initial GBDT model, resulting in an improved GBDT model; where K = 5 and 10, and the hyperparameters include the number of trees (n_estimator), learning rate (learning_rate), maximum tree depth (max_depth), and subsample ratio (subsample).

[0081] S4. The GBDT improved model is evaluated and optimized using the test set to obtain the best GBDT model.

[0082] S5 inputs multiple preset process parameters into the GBDT optimal model to obtain the first predicted data of multiple material properties corresponding to them.

[0083] S6 determines the optimal process parameters based on multiple material property prediction data.

[0084] S7 utilizes optimal process parameters for additive manufacturing and performs relevant performance tests on the resulting material parts to obtain experimental performance data.

[0085] It is worth mentioning that additive manufacturing using optimal process parameters specifically includes the following steps:

[0086] Use SOLIDWORKS software to create a 3D model of the required material part and export it.

[0087] The CAD model is imported into the rapid prototyping software Materialise Magics. The 3D model is sliced ​​according to the optimal process parameters, and the resulting 2D data information is input into the metal printing equipment for part preparation.

[0088] After printing is complete and the temperature has cooled to room temperature, remove the part to obtain the desired component.

[0089] S8 adds the optimal process parameters and experimental data to the dataset as a set of data in the dataset; where the optimal process parameters are used as input data and the experimental data are used as optimal output data.

[0090] S9. Repeat steps S1 to S4 to further optimize the GBDT optimal model.

[0091] It is worth mentioning that the coefficient of determination (R²) of the prediction model can be calculated. 2 Mean absolute percentage error (MAPE), typically R 2 A value ≥0.95 and MAPE ≤5% indicate that the model accuracy meets the requirements.

[0092] The following will provide a detailed description of the machine learning-based additive manufacturing process parameter optimization design method of the present invention through specific embodiments.

[0093] Example 1:

[0094] Taking the rapid and accurate prediction of density in the properties of titanium alloys formed by selective laser melting as an example, this paper illustrates the optimization design of the corresponding process parameters. The specific steps are as follows:

[0095] 1) Establish the initial dataset: Extract the density data of selective laser melting technology process parameters from public literature, experimental and production data to establish the initial dataset.

[0096] 2) Normalization: The process parameters and density data in the initial dataset are normalized (0 to 1) to obtain the process parameter-density dataset. The process parameter-density dataset is divided into a training set and a test set. The data in the training set accounts for 70% of the total data in the dataset, and the data in the test set accounts for 30% of the total data in the dataset.

[0097] 3) Establish the initial GBDT model for selective laser melting process parameters-density: Use the first process parameter in the training set as input data and density data (corresponding to the first material property data) as output data. Set n_estimator to 100, learning_rate to 0.02, max_depth to 5 and subsample to 1 to train the initial GBDT model.

[0098] 4) The hyperparameters of the initial GBDT model were optimized using both Random Search and K-Fold Cross Validation algorithms to obtain an improved GBDT model, where K = 10, and the optimization intervals for hyperparameters n_estimator, learning_rate, max_depth, and subsample were 1–1000, 0.01–0.5, 1–10, and 0–1, respectively, with 500 iterations n_iter. Simultaneously, the coefficient of determination R0 of the models before and after the improvement could be compared. 2 Among them, R before improvement 2 =0.91, Improved R2 =0.99.

[0099] 5) Test the accuracy of the GBDT improved model using a test set and optimize it; specifically, use the second process parameter in each set of test data as input data to obtain the second predicted density data (corresponding to the second predicted material property data), and then use this second predicted density data and the density data (corresponding to the second material property data) in each set of test data to calculate the coefficient of determination R. 2 =0.98, MAPE = 1.6% (satisfies R) 2 If the value is ≥0.95 and MAPE ≤5%, the improved GBDT model is considered the optimal GBDT model.

[0100] 6) The powder layer thickness was set to a fixed value of 0.03 μm. The process parameters were designed by combining laser power, scanning speed and scanning spacing as three factors and determining four levels for each factor. Multiple sets of preset process parameters were obtained and used as input data into the GBDT optimal model to obtain multiple sets of density first prediction data (corresponding to the first prediction data / predicted value of material properties) corresponding to the multiple sets of preset process parameters, as shown in Table 1.

[0101] 7) The optimal process parameters were determined by screening multiple sets of first-prediction density data in Table 1, with the screening criterion being a density ≥ 98%.

[0102] 8) Using the above-mentioned optimal process parameters, selective laser melting technology is used to form the part. The specific steps are as follows: A three-dimensional model is drawn and exported using SOLIDWORKS software according to the required part shape; the CAD model is imported into the rapid prototyping auxiliary software Materialise Magics, and layered slicing is performed according to the optimal process parameters in step 7). The obtained two-dimensional data information is input into the metal printing equipment and printing begins; the part is obtained after printing; the part is ultrasonically cleaned in alcohol for 20 minutes, and then dried in a vacuum drying oven for 30 minutes at a drying temperature of 60℃; the density of the dried part is tested using the Archimedes displacement method (corresponding experimental data / values), and the results are shown in Table 1.

[0103] 9) Compare the percentage error between the predicted values ​​and the experimental values, and add the experimental data to the established dataset as a new set of data. Repeat steps 2) to 7) to continuously improve the model's prediction accuracy and generalization ability through iterative feedback.

[0104] Table 1. Process parameter design and corresponding first predicted density data and experimental data.

[0105]

[0106]

[0107] As shown in Table 1, the percentage error between the experimental and predicted density values ​​of the formed titanium parts is within the range of 0.81% to 1.18%, indicating that the GBDT model for selected area laser melting process parameters of titanium alloys established in this invention has high accuracy (generally, an error within 5% indicates that the model accuracy meets the requirements), and can achieve rapid and accurate prediction of the density of titanium alloys obtained by selected area laser melting. Specifically, the optimal process parameters in Example 1 are selected within the range of laser power 125–225 W, scanning speed 300–1200 mm / s, and scanning spacing 0.09–0.18 mm.

[0108] Example 2:

[0109] Taking the rapid and accurate prediction of tensile strength and elongation in the properties of pure titanium parts formed by selective laser melting as an example, this paper illustrates the optimization design of the corresponding process parameters. The specific steps are as follows:

[0110] 1) Establish the initial dataset: Extract the process parameters of selective laser melting technology - performance data of pure titanium parts (tensile strength, elongation) from public literature, experimental and production data to establish the initial dataset.

[0111] 2) Normalization: The process parameters and performance data in the initial dataset are normalized (0 to 1) to obtain the process parameter-pure titanium part performance dataset. The process parameter-pure titanium part performance dataset is divided into a training set and a test set. The data in the training set accounts for 80% of the total data in the dataset, and the data in the test set accounts for 20% of the total data in the dataset.

[0112] 3) Establish the initial GBDT model for selective laser melting process parameters and pure titanium part performance: The first process parameter in the training set is used as the input data and the pure titanium part performance data (corresponding to the first material performance data) is used as the output data. The initial GBDT model is trained by setting n_estimator to 50, learning_rate to 0.02, max_depth to 8 and subsample to 1.

[0113] 4) The hyperparameters of the initial GBDT model were optimized using Random Search and K-Fold Cross Validation algorithms to obtain an improved GBDT model, where K=5, and the optimization intervals for hyperparameters n_estimator, learning_rate, max_depth, and subsample are 1–1000, 0.01–0.5, 1–10, and 0–1, respectively. The number of iterations n_iter is 200. Simultaneously, the coefficient of determination R0 of the models before and after the improvement can be compared. 2Among them, R before improvement 2 =0.92, Improved R 2 =0.96.

[0114] 5) Test the accuracy of the GBDT improved model using the test set and optimize it; specifically, use the second process parameter in each set of training data as input data to obtain the second predicted data of the pure titanium part performance, and then use the second predicted data of the pure titanium part performance and the output data (corresponding to the second material performance data) in each set of test data to calculate the coefficient of determination R. 2 =0.94≤0.95, MAPE=5.7%; then reset the number of iterations n_iter to 500, rerun the program, and calculate the coefficient of determination R. 2 =0.97≥0.95, MAPE=2.4%, the best GBDT model is obtained.

[0115] 6) Perform process parameter combination design to obtain multiple sets of preset process parameters and input them as input data into the GBDT optimal model to obtain the first predicted data of tensile strength and elongation corresponding to the multiple sets of preset process parameters (corresponding to the first predicted data / predicted value of material properties), as shown in Table 2.

[0116] 7) The optimal process parameters were determined by screening the first predicted data of tensile strength and elongation in Table 2. The screening criteria were tensile strength ≥ 895 MPa and elongation ≥ 10%.

[0117] 8) Using the above-mentioned optimal process parameters, selective laser melting technology is used to form the part. The specific steps are as follows: Draw a three-dimensional model according to the required part shape using SOLIDWORKS software and export it; import the CAD model into the rapid prototyping auxiliary software Materialise Magics, perform layer slicing according to the optimal process parameters in step 7), input the obtained two-dimensional data information into the metal printing equipment and start printing; after printing, the part is obtained; ultrasonically clean the obtained part in alcohol for 20 minutes, and then dry it in a vacuum drying oven for 30 minutes at a drying temperature of 60℃; tensile properties are tested using a tensile testing machine, and the results are shown in Table 2.

[0118] 9) Compare the percentage error between the predicted values ​​and the experimental values, and add the experimental data to the established dataset as a new set of data. Repeat steps 2) to 7) to continuously improve the model's prediction accuracy and generalization ability through iterative feedback.

[0119] Table 2. Process parameter design and corresponding first predicted data and experimental data of tensile strength and elongation.

[0120]

[0121] Table 2 shows the prediction results of the GBDT model for the selected area laser melting process parameters and pure titanium material properties established in this invention. It can be seen that the error ranges for tensile strength and elongation are between 0.98% and 1.38% and 1.46% and 2.11%, respectively, both less than 5%. Therefore, the trained model has high accuracy and can accurately predict the tensile mechanical properties of pure titanium parts produced by selected area laser melting, thereby efficiently determining the optimal process parameters. Specifically, the optimal process parameters in Example 2 are selected within the range of laser power 175–225 W, scanning speed 300–900 mm / s, and scanning spacing 0.10–0.18 mm.

[0122] This invention constructs an additive manufacturing process parameter-material property prediction model based on a data-driven machine learning method, which can significantly reduce the time and cost of process parameter optimization.

[0123] It should be noted that the term "comprising" and any variations thereof in the specification and claims of this invention are intended to cover non-exclusive inclusion, for example, including a series of components that are not necessarily limited to those explicitly listed, but may include other components that are not explicitly listed or that are inherent to the component.

[0124] The descriptions using terms such as "first," "second," etc., in this invention are for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined with "first" or "second" may explicitly or implicitly include at least one of those features.

[0125] Furthermore, the technical solutions of the various embodiments can be combined with each other, but only if they are feasible for those skilled in the art. If the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such combination of technical solutions does not exist and is not within the scope of protection claimed by this invention.

[0126] 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 optimizing additive manufacturing process parameters based on machine learning, characterized in that, Includes the following steps: S1, establish a dataset; wherein the dataset is divided into a training set and a test set; the training set contains multiple sets of training data, and each set of training data includes a first process parameter and its corresponding first material property data; S2, Use the training set to train the model and establish an initial GBDT model of additive manufacturing process parameters-material properties; S3, the hyperparameters of the initial GBDT model are optimized using Random Search and K-Fold Cross Validation algorithms to obtain an improved GBDT model; where K=5 and 10, and the hyperparameters include the number of trees, learning rate, maximum tree depth, and subsampling ratio; S4, using the test set, evaluate and optimize the prediction performance of the improved GBDT model to obtain the optimal GBDT model; wherein, the evaluation index used is the coefficient of determination R. 2 Mean Absolute Percentage Error (MAPE); S5, input multiple sets of preset process parameters into the GBDT optimal model to obtain multiple sets of first predicted material properties corresponding to them; S6. Determine the optimal process parameters based on the first prediction data of the multiple sets of material properties.

2. The optimization design method as described in claim 1, characterized in that, The test set contains multiple sets of test data, and each set of test data includes a second process parameter and its corresponding second material performance data.

3. The optimization design method as described in claim 2, characterized in that, The second process parameter includes at least one of the following: laser power, scanning speed, scanning spacing, and powder layer thickness; The second material property data includes at least one of density, tensile strength, and elongation.

4. The optimization design method as described in claim 2, characterized in that, The evaluation and optimization design of the GBDT improved model using the test set includes the following steps: S4-1, Input the second process parameters into the GBDT improved model to obtain the second predicted data of material properties; S4-2, using the second predicted data of material properties and the second material property data, the coefficient of determination R is calculated. 2 and mean absolute percentage error (MAPE); S4-3, when the coefficient of determination R 2 <Preset threshold or mean absolute percentage error (MAPE)> After setting the preset threshold, reset the RS algorithm to optimize the hyperparameter range and iteration number, and repeat steps S3 and S4-1~S4-2 until the coefficient of determination R... 2 If the model hyperparameter optimization design is completed, the result is ≥ the preset threshold and the mean absolute percentage error (MAPE) ≤ the preset threshold.

5. The optimization design method as described in claim 4, characterized in that, The optimization ranges for the number of trees, learning rate, maximum tree depth, and subsampling ratio in the hyperparameters are 1~1000, 0.01~0.5, 1~10, and 0~1, respectively, and the number of iterations is 100~1500.

6. The optimization design method as described in claim 4, characterized in that, The R 2 The preset thresholds for MAPE are 0.95 and 5%, respectively.

7. The optimization design method as described in claim 1, characterized in that, The creation of the dataset includes: S1-1, Obtain process parameters and material performance data; S1-2, classify the process parameters and material performance data according to the material system type to obtain the initial dataset of process parameters and material performance corresponding to different material systems; S1-3, Normalize the initial dataset to obtain the dataset.

8. The optimization design method as described in claim 7, characterized in that, The material systems include, but are not limited to, titanium alloys, stainless steel, copper alloys, zinc alloys, and high-entropy alloys.

9. The optimization design method as described in claim 1, characterized in that, The training set contains 70-80% of the total data in the dataset, and the test set contains 20-30% of the total data in the dataset. The sum of the ratios of the training set and the test set is always 1.

10. The optimization design method as described in claim 1, characterized in that, The first process parameter includes at least one of laser power, scanning speed, scanning spacing, and powder layer thickness; The first material property data includes at least one of density, tensile strength, and elongation.

11. The optimization design method as described in claim 1, characterized in that, It also includes the following steps: S7. Using the optimal process parameters, the material is prepared by additive manufacturing technology, and the resulting part is subjected to relevant performance tests to obtain experimental data. S8, add the optimal process parameters and the experimental data to the dataset as a new set of data in the dataset; S9. Repeat steps S1 to S4 to further optimize the GBDT optimal model.

Images