A method for designing high thermal conductivity aluminum silicon alloy based on machine learning model

Abstract

This application discloses a method for designing high thermal conductivity aluminum-silicon alloys based on machine learning models, relating to the field of metallurgical materials technology. The method includes: constructing an original dataset of aluminum-silicon alloys containing alloy composition, process parameters, and thermal conductivity properties; preprocessing and feature filtering the dataset to obtain key features affecting thermal conductivity; training several machine learning models based on these key features and using the coefficient of determination as a preliminary screening indicator; further screening using model performance indicators; constructing a final thermal conductivity prediction model through model fusion technology; finally, predicting the thermal conductivity of the virtual alloy composition, selecting high thermal conductivity alloy formulations based on the prediction results, and conducting experimental verification. This application solves the problems of traditional alloy design relying on trial and error, long development cycles, and insufficient prediction accuracy, achieving efficient and accurate prediction of the thermal conductivity of aluminum-silicon alloys, and providing an effective means for the intelligent design and development of high-performance thermally conductive materials.

Description

Technical Field

[0001] This application relates to the field of metallurgical materials technology, and in particular to a method for designing high thermal conductivity aluminum-silicon alloys based on machine learning models. Background Technology

[0002] Aluminum-silicon alloys are widely used in electronic packaging, automotive radiators, and aerospace thermal management components due to their lightweight, good casting properties, and thermal conductivity. As electronic devices move towards higher power density and miniaturization, higher demands are placed on the thermal conductivity of heat dissipation materials, making the development of aluminum-silicon alloys with higher thermal conductivity an important direction in current materials research.

[0003] The thermal conductivity of aluminum-silicon alloys is influenced by various factors, including silicon phase morphology, the type and content of alloying elements, and microstructure. Traditional alloy design methods rely primarily on trial and error and accumulated experience, involving extensive experimental screening of compositions and processes. This approach suffers from long development cycles, high costs, and low efficiency. Although the silicon phase morphology can be improved by adding modifiers such as strontium and sodium, determining the optimal composition and process parameters still requires repeated experimental verification, making it difficult to systematically explore the broad compositional space.

[0004] In recent years, machine learning technology has shown potential in material property prediction and composition optimization. Existing research has utilized machine learning to predict the mechanical properties and corrosion behavior of aluminum alloys, but predictions of thermal conductivity are still relatively few. Traditional machine learning algorithms are simple and have limited expressive power when dealing with high-dimensional nonlinear relationships; while boosting tree algorithms (such as XGBoost and LightGBM), through ensemble learning and gradient optimization mechanisms, can more effectively capture complex feature interactions, showing superior performance in prediction accuracy and generalization ability. However, existing research is mostly limited to single algorithms or simple feature inputs, lacking systematic feature engineering, model fusion, and interpretability analysis, and often fails to fully incorporate intrinsic physical parameters of the material (such as electronegativity, atomic size, valence electron concentration, etc.), resulting in limited model generalization ability and insufficient prediction reliability. Summary of the Invention

[0005] The purpose of this application is to provide a method for designing high thermal conductivity aluminum-silicon alloys based on machine learning models, which can solve the problems of low R&D efficiency and insufficient model prediction accuracy and reliability in traditional alloy design. To achieve the above objective, this application provides the following solution.

[0006] This application provides a method for designing high thermal conductivity aluminum-silicon alloys based on machine learning models, including the following steps.

[0007] Construct an original dataset of thermal conductivity for aluminum-silicon alloys. The original dataset includes: composition data, process parameters, and thermal conductivity performance data of aluminum-silicon alloys.

[0008] The original dataset is preprocessed to obtain preprocessed data.

[0009] Feature filtering is performed on the preprocessed data to obtain key features that affect thermal conductivity.

[0010] Based on the key features and their corresponding thermal conductivity, several machine learning algorithms are trained to obtain several thermal conductivity prediction models.

[0011] Using a preset determination coefficient as a preliminary evaluation index, the several thermal conductivity prediction models are screened to obtain several preliminary thermal conductivity prediction models.

[0012] The preliminary thermal conductivity prediction models are screened using model performance as a secondary evaluation index to obtain the screened thermal conductivity prediction models; the model performance includes: model complexity, generalization ability, utilization of computing resources and training speed.

[0013] The selected thermal conductivity prediction models are fused to obtain the final thermal conductivity prediction model.

[0014] Based on the final thermal conductivity prediction model, the thermal conductivity of the virtual alloy composition is predicted, and high thermal conductivity alloy formulations are selected and experimentally verified based on the prediction results.

[0015] Optionally, the data preprocessing includes: missing value imputation, outlier cleaning, and data normalization; the missing value imputation uses a random forest regression method to predict and fill in missing thermal conductivity samples.

[0016] Optionally, the preprocessed data is filtered to obtain key features affecting thermal conductivity. Specifically, this includes: using a random forest algorithm to perform preliminary feature importance ranking and selecting the top-ranked feature groups; the feature groups include: alloying elements and physical characteristics; and removing redundant or irrelevant features from the feature groups to obtain key features affecting thermal conductivity.

[0017] Optionally, the machine learning algorithms include: linear regression, K-nearest neighbors algorithm, decision tree, random forest, gradient boosting decision tree, neural network, LightGBM and XGBoost.

[0018] Optionally, based on the key features and the corresponding thermal conductivity, several machine learning algorithms are trained to obtain several thermal conductivity prediction models, specifically including the following steps.

[0019] The key features are input into each machine learning algorithm to obtain the output of each machine learning algorithm; the output of the machine learning algorithm is the predicted thermal conductivity.

[0020] Based on the thermal conductivity corresponding to each output and the key feature, the parameters of the corresponding machine learning algorithm are optimized to obtain several thermal conductivity prediction models.

[0021] Optionally, the selected thermal conductivity prediction models are fused to obtain the final thermal conductivity prediction model. Specifically, this includes: using a stacking ensemble strategy, taking the selected thermal conductivity prediction models as base learners, using the corresponding prediction results as input features of meta learners, and performing secondary training using a linear regression model to obtain the final thermal conductivity prediction model.

[0022] Optionally, based on the final thermal conductivity prediction model, the thermal conductivity of the virtual alloy composition is predicted, and high thermal conductivity alloy formulations are selected and experimentally verified according to the prediction results, specifically including the following:

[0023] A virtual composition database is constructed; the virtual composition database includes alloying elements and physical characteristics.

[0024] The virtual component database is input into the final thermal conductivity prediction model to obtain the predicted thermal conductivity values ​​of each component.

[0025] Determine whether the predicted thermal conductivity value is greater than the preset thermal conductivity to obtain the determination result.

[0026] If the judgment result is negative, the process ends.

[0027] If the judgment result is yes, then the corresponding alloy is obtained.

[0028] Based on the obtained alloy, alloy samples with the selected alloy composition were prepared by vacuum melting process, and the actual thermal conductivity of the alloy was measured to verify the accuracy of the prediction.

[0029] Optionally, a virtual component database may be constructed, specifically including the following:

[0030] (1) Determine the types and content ranges of the main alloying elements contained in the Al-Si alloy system.

[0031] (2) A number of component combinations that conform to the actual production process are generated by uniform distribution sampling method.

[0032] (3) Randomly match the generated component combinations with physical characteristic parameters to form a virtual component database.

[0033] Optionally, a uniform distribution sampling method is used to generate several component combinations that conform to the actual production process, wherein the component generation adopts the Latin hypercube sampling method.

[0034] Optionally, based on the obtained alloy, an alloy sample of the selected alloy composition is prepared by vacuum melting process, and the actual thermal conductivity of the alloy is measured to verify the accuracy of the prediction. Specifically, this includes: preparing an alloy ingot by non-consumable vacuum arc melting, obtaining a standard test sample by wire cutting, measuring the thermal conductivity using a laser scintillation thermal conductivity meter, and ensuring that the relative error between the predicted value and the experimental value does not exceed 5%.

[0035] According to the specific embodiments provided in this application, this application has the following technical effects.

[0036] This application provides a method for designing high thermal conductivity aluminum-silicon alloys based on machine learning models. By constructing and screening an aluminum-silicon alloy thermal conductivity dataset containing physical characteristic parameters, key features affecting thermal conductivity are obtained. This solves the problem that feature selection in traditional alloy design relies on human experience and cannot systematically identify key influencing factors, providing a clear theoretical basis for alloy design. By training, evaluating, and screening several machine learning algorithms, and then fusing the screened models, a final prediction model is constructed. This solves the problems of insufficient generalization ability and poor prediction stability of single models, achieving high-precision and robust prediction of the thermal conductivity of aluminum-silicon alloys. By constructing a virtual alloy composition library and combining it with the model for rapid prediction, a broad alloy composition space can be systematically and efficiently traversed, quickly screening out candidate compositions with high thermal conductivity potential, greatly improving the efficiency of discovering high-quality alloy formulations. Attached Figure Description

[0037] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0038] Figure 1 This is an application environment diagram of a method for designing high thermal conductivity aluminum-silicon alloys based on machine learning models in one embodiment of this application.

[0039] Figure 2 This is a flowchart illustrating a method for designing high thermal conductivity aluminum-silicon alloys based on a machine learning model, as provided in an embodiment of this application.

[0040] Figure 3 This is a ranking diagram of the importance of features affecting thermal conductivity in a machine learning model, provided as an embodiment of this application.

[0041] Figure 4 An embodiment of this application provides an R method based on eight machine learning models. 2 Compare the bar charts.

[0042] Figure 5 A fitting curve of predicted values ​​and experimental values ​​provided by an integrated model, as an embodiment of this application.

[0043] Figure 6 This is a schematic flowchart illustrating the verification of the accuracy of the thermal conductivity of an alloy, as provided in one embodiment of this application.

[0044] Figure 7 The XRD diffraction pattern of an aluminum-silicon alloy prepared according to an embodiment of this application.

[0045] Figure 8 SEM image of an aluminum-silicon alloy prepared according to an embodiment of this application.

[0046] Figure 9 EDS image of an aluminum-silicon alloy prepared according to an embodiment of this application.

[0047] Figure 10 A surface morphology test image of an aluminum-silicon alloy obtained by preparation, provided in an embodiment of this application.

[0048] Figure 11 The graph shows the thermal conductivity test results of the aluminum-silicon alloy prepared according to an embodiment of this application.

[0049] Figure 12 This application provides a schematic diagram of the overall process for designing high thermal conductivity aluminum-silicon alloys based on machine learning models. Detailed Implementation

[0050] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0051] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0052] This application provides a method for designing high thermal conductivity aluminum-silicon alloys based on machine learning models, which can be applied to, for example... Figure 1In the application environment shown, terminal 102 communicates with server 104 via a network. A data storage system can store the data that server 104 needs to process. The data storage system can be set up independently, integrated into server 104, or placed in the cloud or on another server. Terminal 102 can send the original dataset to server 104. The original dataset includes: composition data, process parameters, and thermal conductivity performance data of aluminum-silicon alloy. After receiving the original dataset, server 104 performs data preprocessing on the original dataset to obtain preprocessed data. The preprocessed data is then filtered to obtain key features affecting thermal conductivity. Based on the key features and the corresponding thermal conductivity, several machine learning algorithms are trained to obtain several thermal conductivity prediction models. The preset coefficient of determination is used as a preliminary evaluation index to filter the several thermal conductivity prediction models to obtain several preliminary thermal conductivity prediction models. Model performance is used as a secondary evaluation index to filter the several preliminary thermal conductivity prediction models to obtain filtered thermal conductivity prediction models. The model performance includes: model complexity, generalization ability, utilization of computing resources, and training speed. The filtered thermal conductivity prediction models are then fused to obtain the final thermal conductivity prediction model. Based on the final thermal conductivity prediction model, the thermal conductivity of the virtual alloy composition is predicted. High thermal conductivity alloy formulations are selected based on the prediction results and experimentally verified. Server 104 can provide feedback of the obtained high thermal conductivity alloy formulations to terminal 102. Furthermore, in some embodiments, a method for designing high thermal conductivity aluminum-silicon alloys based on a machine learning model can also be implemented independently by server 104 or terminal 102. For example, terminal 102 can directly perform corresponding operations on the original dataset, or server 104 can obtain the original dataset from the data storage system and perform corresponding operations on the original dataset.

[0053] The terminal 102 can be, but is not limited to, various desktop computers, laptops, smartphones, tablets, IoT devices, and portable wearable devices. IoT devices can include smart speakers, smart TVs, smart air conditioners, and smart in-vehicle devices. Portable wearable devices can include smartwatches, smart bracelets, and head-mounted devices. The server 104 can be implemented using a standalone server or a server cluster composed of multiple servers, or it can be a cloud server.

[0054] In one exemplary embodiment, such as Figure 2 As shown, a method for designing high thermal conductivity aluminum-silicon alloys based on a machine learning model is provided. This method is executed by a computer device, specifically a terminal or server, or both. In this embodiment, the method is applied to... Figure 1Taking server 104 as an example, the explanation includes the following steps 201 to 207.

[0055] Step 201: Construct an original dataset of aluminum-silicon alloy thermal conductivity; the original dataset includes: aluminum-silicon alloy composition data, process parameters and thermal conductivity performance data.

[0056] Specifically, in step 201, the dataset is divided into a training set and a test set in a 4:1 ratio. The original dataset comes from alloy compositions reported in various authoritative journals, including 32 physical characteristic parameters such as alloy component information, valence electron concentration, electronegativity, and atomic radius difference, as well as the corresponding thermal conductivity values, totaling 221 data points; the alloy element composition includes (Al, Si, Mn, Fe, Cu, Sr, etc.). The physical characteristic parameters are shown in Table 1.

[0057] Table 1 Physical characteristics and their calculation formulas

[0058] Note: In the above formula, The total number of elements in the alloy is , and the is the in the alloy. Atomic concentration (proportion) of a certain element. Indicates the first The corresponding property parameters of each element For the first atomic radius of an element The average atomic radius of the alloy, Indicates the first Electronegativity of elements For average electronegativity, Indicates the first Type of element and the first Mixture enthalpy between elements It is the ideal gas constant (value 8.314).

[0059] Step 202: Perform data preprocessing on the original dataset to obtain preprocessed data.

[0060] Specifically, in step 202, the input dataset is cleaned by filling in missing values, removing outliers, and normalizing larger values ​​to ensure all data are on the same scale. The cleaned data is then formatted into a fingerprint, creating a data format that the model can recognize.

[0061] Step 203: Filter the preprocessed data to obtain key characteristics affecting thermal conductivity.

[0062] Specifically, in step 203, according to Figure 3The graph showing the importance ranking of features affecting thermal conductivity (only the top twenty most important features are displayed) reveals that the parameter with the greatest weight influencing thermal conductivity is the difference in electronegativity (PEX). In practical applications, the number of folds in cross-validation, the hyperparameter optimization method, and the discrimination criteria can be adjusted according to changes in the dataset.

[0063] Step 204: Based on the key features and the corresponding thermal conductivity, train several machine learning algorithms to obtain several thermal conductivity prediction models.

[0064] Specifically, in step 204, the machine learning algorithms include eight algorithms: linear regression, K-nearest neighbors, decision tree, random forest, gradient boosting decision tree (GBDT), neural network, LightGBM, and XGBoost. Figure 4 The image shows eight machine learning models R selected in this application. 2 Comparing the bar charts, compared to traditional algorithms such as linear regression, K-nearest neighbors, decision trees, and random forests, the XGBoost model performs a second-order Taylor expansion of the cost function, utilizing both first and second derivatives. Furthermore, this model incorporates a regularization term into the cost function to control model complexity; this regularization term includes the number of leaf nodes in the tree, with each leaf node outputting a score L. 2 The regularization term reduces the model's variance from the perspective of bias-variance balance, making the learned model simpler and preventing overfitting. Simultaneously, by scoring each input feature, the model effectively extracts important features, improving the interpretability of the machine learning model. LGBM, on the other hand, is an efficient and scalable decision tree algorithm based on histograms and combining one-sided gradient sampling (GOSS) with mutually exclusive feature binding. Its core purpose is to address the computational inefficiency of the GBDT algorithm framework when handling massive amounts of data, sacrificing minimal computational accuracy to improve efficiency. GOSS reduces significant time and space costs by discarding some unfavorable samples in calculating information gain and utilizing only the remaining data with high gradients.

[0065] Step 205: Use the preset determination coefficient as a preliminary evaluation index to screen several thermal conductivity prediction models to obtain several preliminary thermal conductivity prediction models; use model performance as a secondary evaluation index to screen the several preliminary thermal conductivity prediction models to obtain the screened thermal conductivity prediction models.

[0066] Specifically, in step 205, the machine learning model uses the correlation coefficient R. 2 This serves as a standard for judging prediction accuracy when performing feature engineering. Figure 5The figure shown is a fitting curve between the predicted values ​​and experimental values ​​obtained through the ensemble model in this application. After multiple adjustments, the correlation coefficient R between the XGBoost and LGBM machine learning models on the thermal conductivity test set data was [value missing]. 2 The values ​​are as high as 0.898 and 0.938 respectively, indicating that both models have excellent predictive ability for the thermal conductivity of aluminum-silicon alloys.

[0067] Step 206: The selected thermal conductivity prediction models are fused to obtain the final thermal conductivity prediction model.

[0068] Specifically, in step 206, the two learners are further combined into a stronger learner using a stacking algorithm, thus constructing an ensemble model (EN1). The purpose is to leverage the different advantages of the two models and ensure that the prediction results for thermal conductivity are more controllable.

[0069] Step 207: Based on the final thermal conductivity prediction model, predict the thermal conductivity of the virtual alloy composition, select high thermal conductivity alloy formulas based on the prediction results, and conduct experimental verification.

[0070] By implementing steps 201 to 207 above, a smart design method for thermal conductivity aluminum-silicon alloys, based on data-driven and machine learning modeling, is constructed without relying on a large number of traditional trial-and-error experiments. This method effectively solves the challenge of high-precision prediction under complex nonlinear composition-performance relationships by integrating material physical characteristics with ensemble learning algorithms. Furthermore, by constructing a virtual alloy composition library and using a trained model for efficient screening, it achieves a systematic exploration of a broad composition space, enabling rapid identification of alloy formulations with high thermal conductivity potential. This significantly shortens the material development cycle, reduces trial-and-error costs, and provides a data-driven system solution for the rapid design and verifiable development of high-performance aluminum-silicon alloys.

[0071] In another exemplary embodiment of this application, in order to accurately predict the thermal conductivity of the virtual alloy composition based on the final thermal conductivity prediction model, a high thermal conductivity alloy formula is selected based on the prediction results and experimentally verified. Figure 6 As shown, step 207 above is replaced by steps 301 to 306.

[0072] Step 301: Construct a virtual composition database; the virtual composition database includes alloying elements and physical characteristics.

[0073] Step 302: Input the virtual component database into the final thermal conductivity prediction model to obtain the predicted thermal conductivity values ​​of each component; Step 303: Determine whether the predicted thermal conductivity value is greater than the preset thermal conductivity, and obtain the determination result.

[0074] Step 304: If the judgment result is negative, then the process ends.

[0075] Step 305: If the judgment result is yes, then the corresponding alloy is obtained.

[0076] Step 306: Based on the obtained alloy, prepare an alloy sample with the selected alloy composition by vacuum melting process, and measure the actual thermal conductivity of the alloy to verify the accuracy of the prediction.

[0077] In practical applications, the following should be included.

[0078] (1) Construct a virtual composition database. Using the established prediction model, predict 10,000 different aluminum-silicon alloy compositions. Select one composition with a thermal conductivity higher than 190 W / ( Furthermore, an aluminum-silicon alloy with a relatively reasonable composition was smelted to verify the accuracy of the prediction model.

[0079] (2) To detect the composition and structure of the prepared alloy discs, X-ray diffraction (XRD) tests were performed on them. Figure 7 The XRD diffraction pattern of the aluminum-silicon alloy prepared in this application is shown. As can be seen from the figure, Fm-3m (225) diffraction peaks of Al appear at 2Θ≈38°, 2Θ≈45°, 2Θ≈65°, and 2Θ≈78°; Fd-3m (227) diffraction peaks of Si appear at 2Θ≈28°, 2Θ≈47°, and 2Θ≈56°; and Fm-3m (225) diffraction peaks of Cu appear at 2Θ≈44°. The diffraction peak intensities of Al and Si are relatively consistent, while the diffraction peak of Cu is weaker, thus the composition and structure are consistent with expectations.

[0080] (3) The alloy samples were subjected to scanning electron microscopy (SEM) and energy-dispersive X-ray spectroscopy (EDS) to determine the elemental composition and phase composition of the alloy samples. The test results are as follows: Figure 8 , Figure 9 , Figure 10 As shown in the figure, the alloy surface morphology exhibits a general microstructure including blocky, granular, fibrous, and branched intermetallic phases. EDS results indicate that the alloy was well-melted, with uniform elemental distribution that closely matches the raw material proportions.

[0081] (4) The thermal conductivity of the sample was tested using a German Netzsch LFA467 / LF427 thermal conductivity meter at a temperature of 25℃. Figure 11 The graph shows the thermal conductivity test results of the aluminum-silicon alloy prepared in this embodiment. As can be seen from the graph, the thermal conductivity of the aluminum-silicon alloy is 190.03 W / (m²). The results are largely consistent with the predictions, demonstrating the feasibility of applying machine learning to materials design.

[0082] As another feasible implementation, the following are included.

[0083] (1) Composition parameters, process conditions, and thermal conductivity data of Al-Si alloys were systematically collected from authoritative literature and databases to establish an initial dataset containing 215 samples. The dataset was divided into training and testing sets at a ratio of 4:1. The random forest regression algorithm was used to fill in the missing thermal conductivity data and remove outlier samples, finally obtaining 147 sets of valid data that could be used for modeling. The key physical characteristic parameters affecting thermal conductivity were calculated based on the composition ratio, as shown in Table 1.

[0084] (2) The preprocessed dataset was normalized, and the initial 32 features were selected by combining the SHAP global importance bar chart and the random forest importance permutation. The results showed that the difference in electronegativity, atomic size difference, Sr content, Si content and mixing enthalpy had a significant impact on thermal conductivity.

[0085] (3) Model training was performed on eight machine learning algorithms (linear regression, K-nearest neighbors, decision tree, random forest, gradient boosting tree, neural network, LightGBM, and XGBoost). The hyperparameter optimization strategies were as follows: traditional algorithms used grid search for parameter tuning; deep learning and boosting tree algorithms used Bayesian optimization methods.

[0086] (4) This application evaluates the prediction accuracy of each basic model and determines the accuracy based on its corresponding R. 2 In terms of size, LightGBM and XGBoost performed best in thermal conductivity prediction, with a test set determination coefficient R0. 2 They reached 0.938 and 0.898 respectively.

[0087] (5) A stacking ensemble strategy was adopted, using LightGBM and XGBoost as base learners, combined with a ridge regression meta-learner to construct the final prediction model. The ensemble model achieved R on the test set. 2 The mean square error reached 0.941, with a root mean square error of 8.7 W / (m·K), which is significantly better than the performance of the single model.

[0088] (6) 10,000 virtual alloy compositions were generated based on the Dirichlet distribution, and a virtual database was constructed in conjunction with process parameters. Thermal conductivity was predicted using a trained ensemble model, and the predicted thermal conductivity value of each virtual composition was output. Candidate compositions with predicted values ​​higher than 180 W / (m·K) were selected. The preferred composition range was: Si: 6.5-8.5%, Sr: 0.4-5%, Cu: 2.5-3.5%, Mn: 0.1-0.3%, Fe: 0.1-0.3%, with the balance being aluminum.

[0089] Typical components (Al-7.3Si-9.07Mn-2.6Fe-3.51Cu-5.67Sr, wt%) were selected for experimental verification.

[0090] Taking the target composition Al-7.3Si-9.07Mn-2.6Fe-3.51Cu-5.67Sr (mass percentage) as an example, the following raw materials are used for formulation: Al-50Cu, Al-50Si, Al-20Sr, Al-50Fe, Al-30Mn master alloy, and high-purity aluminum ingot (99.99%). The mass of each raw material is calculated according to the proportions, with a total mass of 200g. Specifically, this includes: Al-50Si: 29.20g, Al-20Sr: 56.70g, Al-50Fe: 10.40g, Al-30Mn: 60.47g, Al-50Cu: 14.04g, and pure aluminum: 29.19g.

[0091] After accurately weighing all raw materials using an electronic balance, they were mixed thoroughly. A vacuum arc melting system was used. First, a mechanical pump was turned on to evacuate the atmosphere for approximately 15 minutes. Then, high-purity argon gas was introduced, and evacuation continued for another 10 minutes. This cycle of gas purification was repeated three times. Subsequently, a molecular pump was started to further purify the atmosphere, and melting was carried out under argon protection, controlling the melting temperature at 1480℃. After the alloy material was completely melted, it was allowed to cool statically. The ingot was then flipped, and the melting process was repeated four times on each surface, ultimately yielding an alloy button ingot weighing approximately 200g. The refined alloy was cut into alloy discs with a diameter of 12.7mm and a thickness of 1.5-2mm using wire cutting. Before characterization and tensile testing, the samples needed to be coarsely ground, finely ground, and mechanically polished with sandpaper.

[0092] The testing process included the following: The Al-Si alloy prepared in this embodiment was characterized by X-ray diffraction (XRD), with a scanning range of 20°-90° and a scanning speed of 7° / min. Finally, the obtained diffraction patterns were used to characterize the phases and calculate the lattice constants using Jade 9.0 software, and the results are as follows. Figure 7 As shown in the figure. XRD analysis reveals that the main phase is an Al matrix, supplemented by a eutectic Si phase.

[0093] To determine the elemental and phase composition of the sample, scanning electron microscopy (SEM) and energy-dispersive X-ray spectroscopy (EDS) were performed. The test results are as follows: Figure 8 , Figure 9 , Figure 10 As shown, the alloy surface morphology exhibits a general microstructure including blocky, granular, fibrous, and branched intermetallic phases. EDS results indicate that the alloy was well-melted, with uniform elemental distribution that closely matches the raw material proportions.

[0094] The thermal conductivity of the alloy disc described in S53 was measured using a Netzsch LFA467 / LF427 thermal conductivity meter at a temperature of 25°C. The average thermal conductivity of this example was found to be 190.03 W / (m²). The relative error between the prediction value and the machine learning model's prediction value of 195.51 W / (m·K) was only 2.8%, which verified the accuracy of the prediction model.

[0095] To verify that the Al-Si alloy obtained in this embodiment has good thermal conductivity, a comparison of its thermal conductivity with existing alloys is shown in Table 2. A380, 6061, and A360 are all commonly used Al-Si alloys in industry. Under the conditions of the smelting process, the Al-Si alloy obtained in this embodiment can maintain good thermal conductivity, indicating that the composition ratio of this alloy is superior to that of existing Al-Si alloys.

[0096] Table 2 Comparison of thermal conductivity between Al-Si alloys and existing alloys

[0097] like Figure 12 As shown, this application also provides an application scenario in which the above-mentioned method for designing high thermal conductivity aluminum-silicon alloys based on machine learning models is applied. Specifically, this includes: creating a raw dataset related to the thermal conductivity performance of the aluminum-silicon alloy; cleaning the dataset and "fingerprinting" the cleaned data into a unified format that the model can recognize; evaluating the prediction accuracy of each basic model and determining its corresponding R-value. 2 The size was determined, and ultimately, both XGBoost and LGBM algorithms were used to construct the prediction model. This model predicted 10,000 different aluminum-silicon alloy compositions, and one composition with a thermal conductivity higher than 190 W / (m²·K) was selected. A relatively reasonable aluminum-silicon alloy with a suitable composition is smelted; the smelted master alloy is wire-cut into alloy discs with a diameter of 12.7 mm and a thickness of 1.5-2 mm. Thermal conductivity and related characterization tests are performed on the alloy discs to verify the accuracy of the prediction model. Specifically, the method for designing high thermal conductivity aluminum-silicon alloys based on machine learning models provided in this embodiment can be applied to the research and development and production process of aluminum-silicon alloy materials. This process typically includes material design, process development, performance testing, and product application. The alloy composition is determined from the design stage, undergoing iterative calculation simulation and experimental verification to finally determine the optimal formula and proceed to the production and application stages. The thermal conductivity prediction and design method provided in this application, specifically in the research and development process of aluminum-silicon alloys, can quickly screen alloy compositions with high thermal conductivity potential based on the synergistic approach of machine learning prediction and traditional experimental verification provided in this application, thus providing precise composition and performance guidance for new alloy development.

[0098] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0099] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for designing high thermal conductivity aluminum-silicon alloys based on machine learning models, characterized in that, The method for designing high thermal conductivity aluminum-silicon alloys based on machine learning models includes: Construct an original dataset of thermal conductivity for aluminum-silicon alloys; the original dataset includes: composition data, process parameters, and thermal conductivity performance data of aluminum-silicon alloys; The original dataset is preprocessed to obtain preprocessed data; The preprocessed data is filtered to obtain the key characteristics affecting thermal conductivity; Based on the key features and corresponding thermal conductivity, several machine learning algorithms are trained to obtain several thermal conductivity prediction models. The predetermined coefficient of determination is used as a preliminary evaluation index to screen the several thermal conductivity prediction models to obtain several preliminary thermal conductivity prediction models; the model performance is used as a secondary evaluation index to screen the several preliminary thermal conductivity prediction models to obtain the screened thermal conductivity prediction models; the model performance includes: model complexity, generalization ability, utilization of computing resources, and training speed. The selected thermal conductivity prediction models are fused to obtain the final thermal conductivity prediction model; Based on the final thermal conductivity prediction model, the thermal conductivity of the virtual alloy composition is predicted, and high thermal conductivity alloy formulations are selected and experimentally verified based on the prediction results.

2. The method for designing high thermal conductivity aluminum-silicon alloys based on machine learning models according to claim 1, characterized in that, The data preprocessing includes: missing value imputation, outlier cleaning, and data normalization; wherein the missing value imputation uses the random forest regression method to predict and fill in the missing thermal conductivity samples.

3. The method for designing high thermal conductivity aluminum-silicon alloys based on machine learning models according to claim 1, characterized in that, The preprocessed data is filtered to obtain key characteristics affecting thermal conductivity, specifically including: A random forest algorithm is used to perform preliminary feature importance ranking, and the feature group with the highest ranking is selected; the feature group includes: alloying elements and physical features; Redundant or irrelevant features are removed from the feature set to obtain the key features that affect thermal conductivity.

4. The method for designing high thermal conductivity aluminum-silicon alloys based on machine learning models according to claim 1, characterized in that, The machine learning algorithms include: linear regression, K-nearest neighbors algorithm, decision tree, random forest, gradient boosting decision tree, neural network, LightGBM and XGBoost.

5. The method for designing high thermal conductivity aluminum-silicon alloys based on machine learning models according to claim 1, characterized in that, Based on the key features and corresponding thermal conductivity, several machine learning algorithms are trained to obtain several thermal conductivity prediction models, specifically including: The key features are input into each machine learning algorithm to obtain the output of each machine learning algorithm; the output of the machine learning algorithm is the predicted thermal conductivity. Based on the thermal conductivity corresponding to each output and the key feature, the parameters of the corresponding machine learning algorithm are optimized to obtain several thermal conductivity prediction models.

6. The method for designing high thermal conductivity aluminum-silicon alloys based on machine learning models according to claim 1, characterized in that, The selected thermal conductivity prediction models are fused to obtain the final thermal conductivity prediction model, which specifically includes: A stacking ensemble strategy is adopted, using the selected thermal conductivity prediction model as the base learner and the corresponding prediction results as the input features of the meta learner. A linear regression model is used for secondary training to obtain the final thermal conductivity prediction model.

7. The method for designing high thermal conductivity aluminum-silicon alloys based on machine learning models according to claim 1, characterized in that, Based on the final thermal conductivity prediction model, the thermal conductivity of the virtual alloy composition is predicted. High thermal conductivity alloy formulations are selected based on the prediction results and experimentally verified. Specifically, this includes: Construct a virtual composition database; the virtual composition database includes alloying elements and physical characteristics; The virtual component database is input into the final thermal conductivity prediction model to obtain the predicted thermal conductivity values ​​of each component. Determine whether the predicted thermal conductivity value is greater than the preset thermal conductivity, and obtain the determination result; If the judgment result is negative, the process ends. If the judgment result is yes, then the corresponding alloy is obtained; Based on the obtained alloy, alloy samples of the selected alloy composition are prepared by vacuum melting process, and the actual thermal conductivity of the alloy is measured to verify the accuracy of the prediction.

8. The method for designing high thermal conductivity aluminum-silicon alloys based on machine learning models according to claim 7, characterized in that, Constructing a virtual component database specifically includes: (1) Determine the types and content ranges of the main alloying elements contained in the Al-Si alloy system; (2) A number of component combinations that conform to the actual production process are generated by uniform distribution sampling method; (3) Randomly match the generated component combinations with physical characteristic parameters to form a virtual component database.

9. The method for designing high thermal conductivity aluminum-silicon alloys based on machine learning models according to claim 8, characterized in that, Several component combinations that conform to the actual production process are generated using a uniform distribution sampling method, wherein the component generation adopts the Latin hypercube sampling method.

10. The method for designing high thermal conductivity aluminum-silicon alloys based on machine learning models according to claim 7, characterized in that, Based on the obtained alloy, alloy samples with the selected alloy composition are prepared by vacuum melting process, and the actual thermal conductivity of the alloy is measured to verify the accuracy of the prediction, specifically including: Alloy ingots were prepared by non-consumable vacuum arc melting, and standard test samples were obtained by wire cutting. Thermal conductivity was measured using a laser flare thermal conductivity meter, and the relative error between the predicted and experimental values ​​did not exceed 5%.

Images