Aluminum alloy rotary extrusion solid phase additive manufacturing process design method based on two-step synergistic active learning strategy

Abstract

The application discloses an aluminum alloy rotary extrusion solid-phase additive process design method based on a two-step synergistic active learning strategy, enhances an obtained process-performance experimental data set, trains and selects an optimal machine learning regression model by using the enhanced data set, generates a process parameter design space of a potential high-performance alloy by combining the optimal machine learning regression model and a non-dominated sorting genetic algorithm, recommends a high-value sample in the process parameter design space, and regards process parameters of the high-value sample reaching a preset requirement as aluminum alloy rotary extrusion solid-phase additive process parameters when mechanical property data of an additive body obtained by performing a rotary extrusion solid-phase additive experiment according to the process parameters of the high-value sample reaches the preset requirement, otherwise, adds the high-value sample to the process-performance experimental data set, returns the expanded data set to a process-performance experimental data set enhancement process and iterates until an iteration termination condition is reached.

Description

Technical Field

[0001] This invention belongs to the field of computational materials science and alloy design optimization technology, specifically involving a design method for aluminum alloy rotary extrusion solid-phase additive manufacturing process based on a two-step enhanced active learning strategy. Background Technology

[0002] With the increasing demand for lightweight structures in high-end manufacturing fields such as aerospace and rail transportation, aluminum alloys have become a key choice for critical structural materials due to their low density and good corrosion resistance. However, traditional aluminum alloys often suffer from a strength-to-plasticity inversion problem, making it difficult to meet the mechanical performance requirements of complex structural components for reliable service under high load environments. Rotary extrusion solid-phase additive manufacturing technology, as an emerging method, can avoid defects such as internal porosity, cracks, and segregation caused by traditional melting and solidification preparation techniques. Through intense plastic deformation and dynamic recrystallization, it achieves microstructure refinement and performance improvement, providing an effective technical approach for the preparation of high-strength and high-plasticity aluminum alloy components.

[0003] However, there is a highly nonlinear and strongly coupled relationship between the process parameters of rotary extrusion solid-state additive manufacturing and the microstructure and mechanical properties of aluminum alloys. Traditional methods such as single-factor trial-and-error experiments and orthogonal experiments for process optimization suffer from high costs, long cycles, and low efficiency, making it difficult to meet the practical needs of synergistic optimization of multiple process parameters and multiple performance objectives. In recent years, active learning has gradually become an important method in materials research and development due to its efficient iterative optimization capabilities with small sample sizes. However, it is still limited by the initial sample space, easily gets trapped in local optima, and often focuses on single-objective optimization, making it difficult to handle the trade-off between strength and plasticity. Summary of the Invention

[0004] To address the problems existing in the prior art, the present invention aims to provide a design method for aluminum alloy rotary extrusion solid-phase additive manufacturing process based on a two-step enhanced active learning strategy. The present invention can solve the problems in the prior art where active learning is limited by the initial sample space, easily gets trapped in local optima, and focuses on single-objective optimization, making it difficult to handle the trade-off between strength and plasticity.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: The design method for solid-state additive manufacturing of aluminum alloys based on a two-step efficiency-enhancing active learning strategy includes the following steps: Enhancement of the process-performance experimental dataset: The process-performance experimental dataset of the additive body obtained by the rotary extrusion solid-state additive manufacturing process of aluminum alloy is enhanced to obtain the enhanced dataset; Model selection: The optimal machine learning regression model is trained and selected using the augmented dataset; Process parameter design space generation: Combining the optimal machine learning regression model and the non-dominated sorting genetic algorithm, the process parameter design space of potential high-performance alloys is generated; High-value sample recommendation: Recommend high-value samples within the process parameter design space; Judgment process: When the mechanical property data of the additive obtained by performing a rotary extrusion solid-phase additive manufacturing experiment based on the process parameters of the high-value sample meets the preset requirements, the process parameters of the high-value sample that meets the preset requirements are used as the process parameters for rotary extrusion solid-phase additive manufacturing of aluminum alloy; otherwise, the high-value sample is added to the process-performance experiment dataset to obtain an amplified dataset. Iterative process: The augmented dataset is returned to the process-performance experiment dataset enhancement process and iterated until the mechanical property data of the additive reaches the preset requirements, at which point the iteration stops.

[0006] Preferably, the process-performance experimental dataset includes spindle speed, travel speed, single-layer thickness, wire feed speed, additive path, overlap rate, additive temperature, room temperature tensile strength, and elongation. The process-performance experimental dataset is augmented using a variational autoencoder to obtain an augmented dataset. The specific process includes: learning the feature distribution of the data in the process-performance experimental dataset, generating augmented samples with the same distribution as the data in the process-performance experimental dataset, and the number of samples in the augmented dataset is more than twice the sample size of the process-performance experimental dataset.

[0007] Preferably, the machine learning regression model includes linear regression model, lasso regression model, ridge regression model, support vector regression model, Gaussian process regression model, random forest regression model and extreme gradient regression model; When training and selecting the optimal machine learning regression model using the augmented dataset, the accuracy of the machine learning regression model is evaluated, and the machine learning regression model with the highest accuracy is selected as the optimal machine learning regression model. The accuracy The mathematical expression is as follows:

[0008] In the formula, i To enhance the indexing of samples in the dataset, It is the first in augmented dataset i Measured values ​​of the mechanical properties of each sample. It is the first in the augmented dataset i Predicted values ​​of mechanical properties for each sample It is the average of the measured mechanical properties of all samples in the enhanced dataset.

[0009] Preferably, the optimal machine learning regression model and the second-generation non-dominated sorting genetic algorithm are combined to generate the process parameter design space for potential high-performance alloys. The optimal machine learning regression model is combined with the second-generation non-dominated sorting genetic algorithm by using the prediction error of the optimal machine learning regression model on the optimized process parameters of the second-generation non-dominated sorting genetic algorithm as the fitness function of the second-generation non-dominated sorting genetic algorithm. The mathematical expression of the fitness function is as follows:

[0010] in:

[0011] In the formula, For the reason A decision vector composed of process parameters ; and These are the predicted values ​​of tensile strength and elongation of the selected optimal machine learning regression model, respectively. and The optimal machine learning regression model is in The prediction uncertainty is expressed as the root mean square error. and This represents the uncertainty penalty coefficient. This represents the feasible region for process parameters; This represents the error in an average sense.

[0012] Preferably, the components of the process parameter design space for the generated potential high-performance alloy are the 100 process parameter combinations with the highest fitness function values.

[0013] Preferably, the mechanical property data includes room temperature tensile strength and elongation; In the judgment process: when the room temperature tensile strength of the additive obtained by performing a rotary extrusion solid-phase additive experiment based on the process parameters of the high-value sample is higher than the preset value of room temperature tensile strength and the elongation is higher than the preset value of elongation, then the process parameters of the high-value sample at this time are used as the process parameters of rotary extrusion solid-phase additive for aluminum alloy; otherwise, the high-value sample is added to the process-performance experiment dataset to obtain an amplified dataset.

[0014] Preferably, a weighted summation upper confidence boundary sampling function is used to recommend high-value samples within the process parameter design space.

[0015] Preferably, the weighted summation upper confidence bound sampling function The mathematical expression is as follows:

[0016] In the formula, These are process parameters; These are the weighting coefficients; The machine learning regression model selected during the model screening process as the best predictor of room temperature tensile strength is used for process parameters. Predicted mean room temperature tensile strength of the lower additive body That is the corresponding forecast standard deviation; These are the weighting coefficients; The machine learning regression model selected during the model screening process as the best predictor of room temperature elongation is used for process parameters. Predicted mean room temperature elongation of the additive body. That is the corresponding forecast standard deviation; It is the exploration coefficient.

[0017] Preferably, the recommended high-value samples are those selected from the weighted summation of the upper confidence bound function values ​​under all combinations of process parameters in the process parameter design space, sorted from high to low. The top two samples in each sorting are recommended.

[0018] This invention also provides a design system for aluminum alloy rotary extrusion solid-state additive manufacturing process based on a two-step efficiency-enhancing active learning strategy, used to implement the aluminum alloy rotary extrusion solid-state additive manufacturing process design method based on a two-step efficiency-enhancing active learning strategy as described above, including: Process-performance test dataset enhancement module: used to enhance the process-performance test dataset of the additive body obtained by the aluminum alloy rotary extrusion solid-phase additive process, and obtain the enhanced dataset; Model selection module: used to train and select the optimal machine learning regression model using the augmented dataset; Process parameter design space generation module: used to combine the optimal machine learning regression model and the non-dominated sorting genetic algorithm to generate the process parameter design space of potential high-performance alloys; High-value sample recommendation module: Used to recommend high-value samples within the process parameter design space; The judgment module is used to determine the process parameters of the high-value sample as the process parameters for the rotary extrusion solid-phase additive manufacturing of aluminum alloy when the mechanical property data of the additive obtained by performing the rotary extrusion solid-phase additive manufacturing experiment based on the process parameters of the high-value sample meet the preset requirements; otherwise, the high-value sample is added to the process-performance experiment dataset to obtain an amplified dataset. Iteration module: used to return the augmented dataset to the process-performance experimental dataset enhancement process and perform iterations until the mechanical property data of the additive body reaches the preset requirements, at which point the iteration stops.

[0019] The present invention has the following beneficial effects: This invention presents a two-step enhanced active learning strategy for designing aluminum alloy rotary extrusion solid-phase additive manufacturing processes. Firstly, through a process-performance experimental dataset enhancement step, the initial experimental dataset is expanded to obtain an enhanced dataset. Existing conventional optimization methods rely solely on a limited amount of original experimental data to train the model. However, the process parameters of aluminum alloy rotary extrusion solid-phase additive manufacturing exhibit a highly nonlinear coupling relationship with tensile strength and elongation. A small number of original samples cannot fully cover the data characteristics, resulting in large prediction biases and a tendency for the trained model to get trapped in local optima. The enhanced dataset in this invention fully preserves the distribution characteristics of the original data. Based on this, subsequent model selection is conducted, reducing the modeling error caused by small samples at the data level and providing a basis for obtaining accurate... The quasi-predictive model provides data support; secondly, model screening is conducted based on the enhanced dataset, training multiple types of machine learning regression models and selecting the optimal machine learning regression model. Unlike directly using a fixed model for prediction, the optimal machine learning regression model can more accurately capture the correlation between multiple process parameters and tensile strength and elongation, reducing the uncertainty in mechanical property prediction and providing a reliable predictive tool for generating the subsequent process parameter design space; thirdly, the optimal machine learning regression model is combined with a non-dominated sorting genetic algorithm to generate the process parameter design space for potential high-performance alloys. Existing active learning methods mostly adopt single-objective optimization, which cannot simultaneously consider tensile strength and elongation, and struggles to resolve the inverse contradiction between the two. Traditional trial-and-error and orthogonal methods... The current method of testing is limited to comparing discrete parameter points. This invention utilizes a non-dominated sorting genetic algorithm to globally search the feasible process domain. Simultaneously, it uses tensile strength and elongation as optimization targets to batch-screen candidate parameter combinations, forming a process parameter design space. This achieves synergistic optimization of both tensile strength and elongation, broadening the search range for high-performance process parameters. High-value samples are then recommended within the process parameter design space. Directly conducting rotational extrusion solid-phase additive manufacturing experiments and mechanical property tests on all candidate parameters would generate numerous invalid experiments, increasing R&D costs and timelines. This invention only screens high-value samples for experiments, prioritizing parameter combinations with better predicted mechanical properties to explore high-performance ranges, while also supplementing data with samples exhibiting high prediction uncertainty. By collecting missing features and balancing the utilization of known high-performance ranges with the exploration of unknown process ranges, unnecessary experimental investment and the R&D cycle are significantly reduced. Finally, a judgment and iteration process is set up. When the mechanical properties of high-value samples obtained by rotational extrusion solid-phase additive manufacturing experiments do not meet the preset requirements, the measured data are merged into the original dataset to obtain an augmented dataset, which is then sent back to the process-performance experiment dataset enhancement link to re-execute each step completely. Each iteration will supplement new measured data, update the augmented dataset and the optimal machine learning regression model, continuously broaden the data coverage and break out of the local optimum range. During the iteration process, the mechanical properties gradually converge to the preset standard until the measured tensile strength and elongation both meet the preset requirements, at which point qualified process parameters can be output.The overall solution forms a two-step efficiency-enhancing closed-loop logic. The pre-process-performance experimental dataset enhancement addresses the fundamental problem of insufficient modeling accuracy with small samples. The post-process parameter design space generation and high-value sample recommendation, combined with iterative iteration, achieve rapid and coordinated optimization of tensile strength and elongation. The entire process requires only a small number of initial experimental samples to initiate optimization, enabling simultaneous improvement of tensile strength and elongation in aluminum alloy rotary extrusion solid-phase additive components with fewer experiments and lower R&D costs. Attached Figure Description

[0020] The accompanying drawings described herein are for illustrative purposes only and are not intended to limit the scope of the invention in any way. Those skilled in the art will be able to derive other drawings from these drawings without any inventive effort.

[0021] Figure 1 This is a flowchart of the method of the present invention.

[0022] Figure 2 This is a scatter plot of the mechanical properties measured experimentally during the iterative process in this embodiment of the invention.

[0023] Figure 3 This is the engineering stress-strain curve corresponding to the optimal mechanical properties in the embodiments of the present invention. Detailed Implementation

[0024] The present invention will be further described clearly and in detail below with reference to specific embodiments and the accompanying drawings. Those skilled in the art will be able to implement the present invention based on these descriptions. Furthermore, the embodiments of the present invention described below are generally only some, not all, of the embodiments of the present invention. Therefore, all other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort should fall within the scope of protection of the present invention.

[0025] The theories or mechanisms described and disclosed herein, whether right or wrong, should not in any way limit the scope of the invention, that is, the contents of the invention can be implemented without being limited by any particular theory or mechanism.

[0026] In this document, all features defined by numerical ranges or percentage ranges, such as numerical values, quantities, contents, and concentrations, are for the sake of brevity and convenience only. Accordingly, descriptions of numerical ranges or percentage ranges should be considered as covering and specifically disclosing all possible sub-ranges and individual numerical values ​​(including integers and fractions) within those ranges.

[0027] In this article, unless otherwise specified, “contains,” “includes,” “containing,” “has,” or similar terms cover the meanings of “composed of” and “mainly composed of,” for example, “A contains a” covers the meanings of “A contains a and others” and “A contains only a.”

[0028] For the sake of brevity, not all possible combinations of the technical features in the implementation schemes or embodiments are described herein. Therefore, as long as there is no contradiction in the combination of these technical features, the technical features in the implementation schemes or embodiments can be combined arbitrarily, and all possible combinations should be considered within the scope of this specification.

[0029] This invention relates to a design method for a solid-state additive manufacturing process of aluminum alloy rotary extrusion based on a two-step efficiency-enhancing active learning strategy. The method includes the following steps: Step 1: Construct an initial experimental dataset of process and performance for aluminum alloy rotary extrusion solid-phase additive manufacturing; where process parameters include spindle speed, travel speed, single-layer thickness, wire feed speed, additive path, overlap ratio and additive temperature, and mechanical properties include room temperature tensile strength (UTS) and elongation (EL).

[0030] Step 2: Perform the first step of enhancement: Use a variational autoencoder to enhance the initial experimental dataset and construct an enhanced dataset; the data enhancement method of the variational autoencoder is to learn the feature distribution of the original data and generate enhanced samples with the same distribution as the original data. The number of samples in the enhanced dataset is 10 times the sample size of the original dataset.

[0031] Step 3: Train common machine learning regression models, evaluate their accuracy, and select the optimal model. Common machine learning regression models include linear regression, lasso regression, ridge regression, support vector regression, Gaussian process regression, random forest regression, and extreme gradient regression. The accuracy metric for these models is the coefficient of determination. Its mathematical expression is as follows:

[0032] In the formula, It is the first i Measured values ​​of the mechanical properties of each sample. It is the first i Predicted values ​​of mechanical properties for each sample It is the average of the measured mechanical properties of all samples.

[0033] Step 4: Perform the second efficiency enhancement step: Combine the optimal machine learning regression model and the second-generation non-dominated sorting genetic algorithm to generate the process parameter design space for potential high-performance alloys; wherein, the combination of the optimal model and the second-generation non-dominated sorting genetic algorithm is such that the prediction error of the optimal model on the process parameters optimized by the algorithm is used as the fitness function of the genetic algorithm, and its mathematical expression is as follows:

[0034] in:

[0035] In the formula, For the reason A decision vector composed of process parameters ; and These are the predicted values ​​of tensile strength and elongation of the optimal regression model selected in step three, respectively. Since the algorithm defaults to solving a minimization problem, but we need to maximize performance, we take the negative sign. and The optimal model is in Prediction uncertainty (root mean square error) at this location. and This represents the uncertainty penalty coefficient. This represents the feasible region for process parameters; This represents the error in an average sense. The generated potential process parameter design space consists of the 100 process parameter combinations with the highest fitness function values.

[0036] Step 5: Recommend several high-value samples within the process parameter design space using a weighted summation upper confidence boundary sampling function; the mathematical expression of the weighted summation upper confidence boundary sampling function in Step 5 is as follows:

[0037] In the formula, It is the exploration coefficient; and These are the optimal machine learning models for UTS and EL prediction selected in step three, used to predict process parameters. Predicted mean values ​​of UTS and EL for additive manufacturing. and These are the corresponding forecast standard deviations; All are weighting coefficients; The recommended high-value samples are selected by sorting the weighted summation of the upper confidence bound function values ​​under all process parameter combinations in the design space from high to low, and the top 2 samples in each sorting are recommended.

[0038] Step 6: Perform a rotary extrusion solid-phase additive manufacturing experiment based on the process parameters of several high-value samples, and test the mechanical properties of the resulting additives. Step 7: Feed back the mechanical property data samples obtained from the experiment to the initial experimental dataset, and repeat steps 2 to 6 until the iteration termination condition is met, and then stop the iteration; wherein, the iteration termination condition is that the experimentally measured UTS and EL are both higher than the preset value.

[0039] This invention integrates variational autoencoder for small-sample data augmentation with machine learning regression modeling to construct a two-step efficiency-enhancing active learning strategy: First, an augmented dataset is generated using a variational autoencoder to train a high-precision surrogate model; second, a second-generation non-dominated sorting genetic algorithm is combined with bi-objective optimization and weighted summation upper confidence bound active learning sampling to generate a design space for process parameters of potential high-performance alloys and accurately recommend high-value experimental samples. This strategy overcomes the limitations of traditional rotary extrusion solid-phase additive manufacturing process design, which relies on numerous trial-and-error experiments and suffers from insufficient model accuracy under small sample sizes. It provides a reliable theoretical tool for the synergistic optimization of strength and plasticity in rotary extrusion solid-phase additive manufacturing of aluminum alloys, significantly improving R&D efficiency.

[0040] Example Reference Figure 1 The aluminum alloy rotary extrusion solid-phase additive manufacturing process design method based on a two-step efficiency-enhancing active learning strategy in this embodiment includes the following steps: Step 1: Construct the initial experimental dataset for the process-performance of aluminum alloy rotary extrusion solid-phase additive manufacturing; Step 2: Perform the first step of enhancement: Use a variational autoencoder to enhance the initial experimental dataset and construct an enhanced dataset; Step 3: Train common machine learning regression models, evaluate the accuracy of machine learning regression models, and select the best machine learning regression model among them; Step 4: Perform the second step of efficiency enhancement: combine the optimal machine learning regression model and the second-generation non-dominated sorting genetic algorithm to generate the process parameter design space for potential high-performance alloys; Step 5: Use the weighted summation upper confidence boundary sampling function to recommend several high-value samples within the process parameter design space; Step 6: Perform a rotary extrusion solid-phase additive manufacturing experiment based on the process parameters of several high-value samples, and test the mechanical properties of the resulting additives. Step 7: Feed back the mechanical property data samples obtained from the experiment to the initial experimental dataset, and repeat steps 2 to 6 until the iteration termination condition is met, then stop the iteration.

[0041] In this embodiment, in step one, for the 2A10 aluminum alloy rotary extrusion solid-state additive manufacturing process, a total of 15 initial experimental datasets were collected, including 7 process parameters and 2 mechanical property parameters. The process parameters cover spindle speed (rpm), feed speed (mm / min), single-layer thickness (mm), wire feed speed (m / min), additive path, overlap rate (%), and additive temperature (°C). The mechanical properties are room temperature tensile strength (UTS) (MPa) and elasticity (EL) (%). The process parameters and mechanical property parameters of this embodiment are shown in Table 1. Table 1

[0042] In this embodiment, in step two, a variational autoencoder model is used to learn the feature distribution of the original data for the initial experimental dataset to obtain an augmented dataset. The sample size of the augmented dataset is 10 times that of the initial experimental dataset, that is, it contains 150 augmented data samples.

[0043] In this embodiment, step three employs seven different types of machine learning regression models: linear regression, lasso regression, ridge regression, support vector regression, Gaussian process regression, random forest regression, and extreme gradient regression. Model performance is assessed using the coefficient of determination. The evaluation is conducted using the following calculation formula:

[0044] In the formula It is the first The true value of the mechanical properties of each sample It is the first Predicted values ​​of mechanical properties for each sample It is the average of the true mechanical properties of all samples.

[0045] Before modeling, all continuous features were normalized to eliminate the influence of different parameters' dimensions. For scenarios with extremely small samples, leave-one-out cross-validation was used to obtain the average prediction accuracy, and the results are shown in Table 2.

[0046] Table 2

[0047] The results show that the model established by limiting gradient regression has the highest prediction accuracy and the best tensile strength prediction. =0.96, elongation prediction =0.89, therefore, limiting gradient regression is chosen as the optimal prediction model in this embodiment.

[0048] In this embodiment, in step four, the population size of the second-generation non-dominated sorting genetic algorithm is set to 100, the number of iterations is set to 1000, the crossover probability is 0.9, and the mutation probability is 0.1. Within the range of process parameters for aluminum alloy rotary extrusion solid-phase additive manufacturing, the algorithm iteratively searches through selection, crossover, and mutation operations to finally select 100 sets of candidate process parameters with the best prediction performance, thereby constructing the design space.

[0049] The optimal model and the second-generation non-dominated sorting genetic algorithm are combined in such a way that the prediction error of the optimal model on the optimized process parameters is used as the fitness function of the genetic algorithm, and its mathematical expression is as follows:

[0050] in:

[0051] In the formula, For the reason A decision vector composed of process parameters ; and These are the predicted values ​​of tensile strength and elongation of the optimal regression model selected in step three, respectively. Since the algorithm defaults to solving a minimization problem, but we need to maximize performance, we take the negative sign. and The optimal model is in Prediction uncertainty (root mean square error) at this location. and The uncertainty penalty coefficients are 0.3 and 0.35, respectively. This represents the feasible region for process parameters; This represents the error in an average sense.

[0052] In this embodiment, in step five, the weighted summation upper confidence bound function score is calculated for all candidate process parameters within the design space. The mathematical expression of the weighted summation upper confidence bound formula is as follows:

[0053] In the formula, This is the exploration coefficient, with a value of 2.5; Candidate point The output predicted mean; Candidate point The standard deviation of the output prediction; This represents the weights for strength and plasticity, both with a value of 0.5.

[0054] In this embodiment, the two samples with the highest scores are selected as high-value samples each time.

[0055] In this embodiment, in step six, based on the process parameters corresponding to the two selected high-value samples, as shown in Table 3, a rotary extrusion solid-phase additive manufacturing experiment is carried out to prepare 2A10 aluminum alloy additive samples.

[0056] Table 3

[0057] In accordance with GB / T 228.1-2021 "Metallic materials - Tensile testing - Part 1: Test at room temperature", the additive test specimens were processed into standard tensile specimens and subjected to room temperature tensile testing using an INSTRON 5969 universal testing machine. The measured tensile strength and elongation of each specimen were obtained, and the results are shown in Table 4.

[0058] Table 4

[0059] In this embodiment, in step seven, the two sets of new data samples obtained in this experiment are fed back to the initial experimental dataset, and steps two through six are repeated for iterative optimization. The preset iterative indicators are that the UTS is at least 350 MPa and the EL is at least 35%.

[0060] Performance changes during the iterative process, such as Figure 2 As shown. After three rounds of iteration, the highest mechanical property combination was obtained, and the iteration was stopped. Using the process parameters corresponding to this sample, a rotary extrusion solid-state additive manufacturing experiment was carried out to prepare 2A10 aluminum alloy additive specimens. According to GB / T 228.1-2021 "Metallic materials - Tensile testing - Part 1: Room temperature test method", the additive specimens were processed into standard tensile specimens, and room temperature tensile tests were performed using an INSTRON 5969 universal testing machine. The test results are shown below. Figure 3 As shown, the measured tensile strength of the sample was 359.98 MPa and the elongation was 38%. It can be seen that this method can optimize the mechanical properties of aluminum alloys produced by rotary extrusion solid-phase additive manufacturing.

[0061] The results of the above embodiments show that the two-step enhanced active learning strategy proposed in this invention can efficiently generate the optimal parameter combination for aluminum alloy rotary extrusion solid-phase additive manufacturing process with fewer experimental samples while ensuring prediction accuracy. This significantly shortens the R&D cycle, reduces experimental costs, and has good reliability and engineering applicability.

[0062] Furthermore, embodiments of the present invention also provide a design system for aluminum alloy rotary extrusion solid-state additive manufacturing process based on a two-step efficiency-enhancing active learning strategy, used to implement the aluminum alloy rotary extrusion solid-state additive manufacturing process design method based on a two-step efficiency-enhancing active learning strategy as described above. The system includes: Process-performance test dataset enhancement module: used to enhance the process-performance test dataset of the additive body obtained by the aluminum alloy rotary extrusion solid-phase additive process, and obtain the enhanced dataset; Model selection module: used to train and select the optimal machine learning regression model using the augmented dataset; Process parameter design space generation module: used to combine the optimal machine learning regression model and the non-dominated sorting genetic algorithm to generate the process parameter design space of potential high-performance alloys; High-value sample recommendation module: Used to recommend high-value samples within the process parameter design space; The judgment module is used to determine the process parameters of the high-value sample as the process parameters for the rotary extrusion solid-phase additive manufacturing of aluminum alloy when the mechanical property data of the additive obtained by performing the rotary extrusion solid-phase additive manufacturing experiment based on the process parameters of the high-value sample meet the preset requirements; otherwise, the high-value sample is added to the process-performance experiment dataset to obtain an amplified dataset. Iteration module: used to return the augmented dataset to the process-performance experimental dataset enhancement process and perform iterations until the mechanical property data of the additive body reaches the preset requirements, at which point the iteration stops.

[0063] In addition, the system also includes an initial experimental dataset construction module for constructing an initial experimental dataset of process-performance for aluminum alloy rotary extrusion solid-phase additive manufacturing.

[0064] In summary, the aluminum alloy rotary extrusion solid-state additive manufacturing process design method based on a two-step enhanced active learning strategy provided by this invention addresses the problem of insufficient training data for extremely small initial experimental samples by first using variational autoencoder data augmentation. Then, it utilizes machine learning regression modeling and a second-generation non-dominated sorting genetic algorithm to establish a high-precision process-performance prediction model, providing a design space for process parameters of potential high-performance alloys. This effectively overcomes the challenges of insufficient accuracy and vast exploration space in machine learning models under small sample scenarios. Finally, it learns accurate recommendations for sampling functions based on weighted summation upper confidence bounds and combines this with feedback from rotary extrusion solid-state additive manufacturing experiments. This efficiently overcomes the bottlenecks of traditional process design, which relies on numerous trial-and-error experiments, resulting in long development cycles and high costs. It also solves the problem of difficulty in balancing strength and plasticity during co-optimization. This invention provides a new method for the co-optimization of strength and plasticity in rotary extrusion solid-state additive manufacturing of aluminum alloys and offers a new strategy for aluminum alloy process design. Through data augmentation and design space optimization, this invention can accelerate process parameter design based on small sample datasets, providing reliable guidance and technical support for improving the mechanical properties of rotary extrusion solid-state additive manufactured aluminum alloys.

[0065] Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort should fall within the scope of protection of the present invention.

Claims

1. A design method for solid-state additive manufacturing of aluminum alloys based on a two-step efficiency-enhancing active learning strategy, characterized in that, The process includes the following: Enhancement of the process-performance experimental dataset: The process-performance experimental dataset of the additive body obtained by the rotary extrusion solid-state additive manufacturing process of aluminum alloy is enhanced to obtain the enhanced dataset; Model selection: The optimal machine learning regression model is trained and selected using the augmented dataset; Process parameter design space generation: Combining the optimal machine learning regression model and the non-dominated sorting genetic algorithm, the process parameter design space of potential high-performance alloys is generated; High-value sample recommendation: Recommend high-value samples within the process parameter design space; Judgment process: When the mechanical property data of the additive obtained by performing a rotary extrusion solid-phase additive manufacturing experiment based on the process parameters of the high-value sample meet the preset requirements, the process parameters of the high-value sample that meet the preset requirements are used as the process parameters for rotary extrusion solid-phase additive manufacturing of aluminum alloy. Otherwise, the high-value samples are added to the process-performance experiment dataset to obtain an expanded dataset; Iterative process: The augmented dataset is returned to the process-performance experiment dataset enhancement process and iterated until the mechanical property data of the additive reaches the preset requirements, at which point the iteration stops.

2. The method for designing a solid-state additive manufacturing process for aluminum alloy rotary extrusion based on a two-step efficiency-enhancing active learning strategy according to claim 1, characterized in that, The process-performance experimental dataset includes spindle speed, travel speed, single-layer thickness, wire feed speed, additive path, overlap rate, additive temperature, room temperature tensile strength, and elongation. The process-performance experimental dataset is augmented using a variational autoencoder to obtain an augmented dataset. The specific process includes: learning the feature distribution of the data in the process-performance experimental dataset, generating augmented samples with the same distribution as the data in the process-performance experimental dataset, and the number of samples in the augmented dataset is more than twice the sample size of the process-performance experimental dataset.

3. The method for designing a solid-state additive manufacturing process for aluminum alloy rotary extrusion based on a two-step efficiency-enhancing active learning strategy according to claim 1, characterized in that, The machine learning regression models include linear regression, lasso regression, ridge regression, support vector regression, Gaussian process regression, random forest regression, and extreme gradient regression. When training and selecting the optimal machine learning regression model using the augmented dataset, the accuracy of the machine learning regression model is evaluated, and the machine learning regression model with the highest accuracy is selected as the optimal machine learning regression model. The accuracy The mathematical expression is as follows: In the formula, i To enhance the indexing of samples in the dataset, It is the first in augmented dataset i Measured values ​​of the mechanical properties of each sample. It is the first in augmented dataset i Predicted values ​​of mechanical properties for each sample It is the average of the measured mechanical properties of all samples in the enhanced dataset.

4. The method for designing a solid-state additive manufacturing process for aluminum alloy rotary extrusion based on a two-step efficiency-enhancing active learning strategy according to claim 1, characterized in that, By combining the optimal machine learning regression model and the second-generation non-dominated sorting genetic algorithm, a design space for process parameters of potential high-performance alloys is generated. The optimal machine learning regression model is combined with the second-generation non-dominated sorting genetic algorithm by using the prediction error of the optimal machine learning regression model on the optimized process parameters of the second-generation non-dominated sorting genetic algorithm as the fitness function of the second-generation non-dominated sorting genetic algorithm. The mathematical expression of the fitness function is as follows: in: In the formula, For the reason A decision vector composed of process parameters ; and These are the predicted values ​​of tensile strength and elongation of the selected optimal machine learning regression model, respectively. and The optimal machine learning regression model is in The prediction uncertainty is expressed as the root mean square error. and This represents the uncertainty penalty coefficient. This represents the feasible region for process parameters; This represents the error in an average sense.

5. The method for designing a solid-state additive manufacturing process for aluminum alloy rotary extrusion based on a two-step efficiency-enhancing active learning strategy according to claim 4, characterized in that, The components of the process parameter design space for the generated potential high-performance alloys are the 100 process parameter combinations with the highest fitness function values.

6. The method for designing a solid-state additive manufacturing process for aluminum alloy rotary extrusion based on a two-step efficiency-enhancing active learning strategy according to claim 1, characterized in that, The mechanical property data include room temperature tensile strength and elongation. In the judgment process: when the room temperature tensile strength of the additive obtained by performing a rotary extrusion solid-phase additive experiment based on the process parameters of the high-value sample is higher than the preset value of room temperature tensile strength and the elongation is higher than the preset value of elongation, then the process parameters of the high-value sample at this time are used as the process parameters of rotary extrusion solid-phase additive for aluminum alloy; otherwise, the high-value sample is added to the process-performance experiment dataset to obtain an amplified dataset.

7. The method for designing a solid-state additive manufacturing process for aluminum alloy rotary extrusion based on a two-step efficiency-enhancing active learning strategy according to claim 6, characterized in that, High-value samples are recommended within the process parameter design space by using a weighted summation upper confidence boundary sampling function.

8. The method for designing a solid-state additive manufacturing process for aluminum alloy rotary extrusion based on a two-step efficiency-enhancing active learning strategy according to claim 7, characterized in that, The weighted summation upper confidence boundary sampling function The mathematical expression is as follows: In the formula, These are process parameters; These are the weighting coefficients; The machine learning regression model selected during the model screening process as the best predictor of room temperature tensile strength is used for process parameters. Predicted mean room temperature tensile strength of the lower additive body That is the corresponding standard deviation of the forecast; These are the weighting coefficients; The machine learning regression model selected during the model screening process as the best predictor of room temperature elongation is used for process parameters. Predicted mean room temperature elongation of the additive body. That is the corresponding standard deviation of the forecast; It is the exploration coefficient.

9. The method for designing a solid-state additive manufacturing process for aluminum alloy rotary extrusion based on a two-step efficiency-enhancing active learning strategy according to claim 8, characterized in that, The recommended high-value samples are those selected from the weighted summation of the upper confidence bound function values ​​under all combinations of process parameters in the process parameter design space, sorted from high to low. The top two samples in each sorting are recommended.

10. A design system for a solid-state additive manufacturing process of aluminum alloy rotary extrusion based on a two-step efficiency-enhancing active learning strategy, characterized in that, The method for designing a solid-state additive manufacturing process for aluminum alloy rotary extrusion based on a two-step enhanced active learning strategy as described in any one of claims 1-9 includes: Process-performance test dataset enhancement module: used to enhance the process-performance test dataset of the additive body obtained by the aluminum alloy rotary extrusion solid-phase additive process, and obtain the enhanced dataset; Model selection module: used to train and select the optimal machine learning regression model using the augmented dataset; Process parameter design space generation module: used to combine the optimal machine learning regression model and the non-dominated sorting genetic algorithm to generate the process parameter design space of potential high-performance alloys; High-value sample recommendation module: Used to recommend high-value samples within the process parameter design space; The judgment module is used to determine the process parameters of the high-value sample as the process parameters for the rotary extrusion solid-phase additive manufacturing of aluminum alloy when the mechanical property data of the additive obtained by performing the rotary extrusion solid-phase additive manufacturing experiment based on the process parameters of the high-value sample meet the preset requirements; otherwise, the high-value sample is added to the process-performance experiment dataset to obtain an amplified dataset. Iteration module: used to return the augmented dataset to the process-performance experimental dataset enhancement process and perform iterations until the mechanical property data of the additive body reaches the preset requirements, at which point the iteration stops.

Images