Multi-component composite material reversely constructed based on machine learning and design method

Abstract

The invention belongs to the technical field of titanium-based composite materials, and particularly relates to a multi-component composite material reversely constructed based on machine learning and a design method. The method comprises the following steps: selecting a reinforcement phase which is expected to realize the high-strength and high-plasticity titanium-based composite material from the existing reinforcements, analyzing the reinforcements, the content and the distribution structure which have great influence on the comprehensive mechanical properties of the titanium-based composite material by utilizing a method of combining machine learning with first principle simulation, and obtaining the high-strength and high-plasticity titanium-based composite material according to the obtained result. And a laser additive manufacturing process and a high-energy ball milling process are utilized to prepare a printed piece. According to the method, the data mining capability of machine learning, the theoretical analysis advantages of a first principle and an advanced material preparation process are integrated, and an efficient, scientific and systematic new way is provided for research and development of the titanium-based composite material.

Description

Technical Field

[0001] This invention belongs to the field of titanium-based composite material technology, and particularly relates to a method for reverse-engineering multi-component composite materials based on machine learning and its design. Background Technology

[0002] Titanium matrix composites (TMCs) possess immense application potential in numerous fields, including aerospace, automotive, and biomedicine, due to their superior specific strength, specific modulus, high-temperature resistance, and corrosion resistance. In aerospace, for example, the application of high-performance TMCs can significantly reduce aircraft structural weight and improve fuel efficiency and flight performance. In biomedicine, their excellent biocompatibility and mechanical properties make them ideal orthopedic implant materials. However, traditional TMC development relies heavily on trial-and-error methods, repeatedly experimenting to explore the relationship between the type, content, distribution structure, and matrix properties of the reinforcing phase. This approach is not only time-consuming, labor-intensive, and costly, but also struggles to overcome the limitations of existing material systems, resulting in low development efficiency and severely hindering further improvements in TMC performance and the construction of novel material systems. Summary of the Invention

[0003] The purpose of this invention is to address the above-mentioned problems by providing a method for reverse engineering multi-component composite materials based on machine learning and designing them.

[0004] To achieve the above objectives, the present invention adopts the following technical solutions:

[0005] A design method for multi-component composite materials based on machine learning is proposed. The method selects reinforcing phases that are expected to achieve high strength and high plasticity in titanium matrix composites from existing reinforcing materials. The method uses machine learning combined with first-principles simulation to analyze the reinforcing materials, their content and distribution structure that have a significant impact on the comprehensive mechanical properties of titanium matrix composites. Based on the results, the printed parts are prepared using laser additive manufacturing process and high-energy ball milling process.

[0006] The above-mentioned design method for multi-component composite materials based on machine learning reverse engineering includes the following steps:

[0007] Step A, establish a dataset: Select some reinforcing phases that are expected to achieve high strength and high plasticity titanium matrix composites, and use the reinforcing phase size, relative molecular mass, element binding energy, reinforcing phase content and thermal expansion coefficient of the above reinforcing phases as input features, and the strength-plasticity product as the output to establish a dataset;

[0008] Step B, establish an asymptotic damage model for the reinforcing phase-titanium matrix composite: using the Random Forest algorithm, machine learning is trained on the dataset from Step A to establish an asymptotic damage model for the reinforcing phase-titanium matrix composite, which can predict the reinforcement that is expected to improve the strength-ductility product of the titanium matrix composite.

[0009] Step C, Simulation Verification: Based on the prediction results in Step B, the selected enhancers are simulated and verified using first-principles simulation methods.

[0010] Step D, Sample Preparation: Based on the simulation verification results of Step C, select one or more reinforcements, mix them with titanium-based powder using a high-energy ball milling process, and then prepare the sample using powder metallurgy laser additive manufacturing process.

[0011] In the above-mentioned design method for reverse-engineering multi-component composite materials based on machine learning, in step A, the reinforcing agents are TiC, TiB, SiC, Ti5Si3, Gr, B4C, TiB2, TiN, and Ti2Cu.

[0012] In the above-mentioned design method for constructing multi-component composite materials by reverse engineering based on machine learning, in step B, the Random Forest algorithm is set to 80 training rounds, a learning rate of 0.02, and a regularization coefficient L2 of 3.

[0013] In the aforementioned design method for multi-component composite materials based on machine learning reverse engineering, step C involves first-principles simulation using the CASTEP module in Materialsstudio, a first-principles quantum mechanics software based on density functional theory. This is used to analyze the DOS curves, charge densities, and differential charge densities of the reinforcements and to determine their interactive reinforcement effects. The calculated parameters are then used to perform convergence tests on the plane wave phase energy and the Brillouin zone K-point to obtain the cutoff energy and Brillouin zone K-point used for calculation.

[0014] In the above-mentioned design method for reverse engineering multi-component composite materials based on machine learning, in step D, the high-energy ball milling process parameters are: ball-to-material ratio 5:1, rotation speed 150-200 r / min, ball milling time 6-8 hours, and argon atmosphere protection.

[0015] In the above-mentioned design method for reverse construction of multi-component composite materials based on machine learning, the powder metallurgy laser additive manufacturing process parameters in step D include: powder spreading speed of 0.04 mm / s to 0.08 mm / s, laser power of 240 W to 300 W, scanning speed of 1800 mm / s to 2000 mm / s, and scanning spacing of 0.10 mm to 0.14 mm.

[0016] Multicomponent composite materials were prepared according to the design method described above.

[0017] Among numerous machine learning algorithms, the Random Forest algorithm has become a commonly used algorithm in materials design due to its advantages such as good resistance to overfitting, ability to handle nonlinear relationships, and effective evaluation of feature importance. By reasonably setting parameters such as the number of training rounds, learning rate, and regularization coefficient, the Random Forest algorithm can learn from a large amount of reinforcing phase data, screening out reinforcing elements that are expected to improve the mechanical properties of the matrix and their optimal content, greatly broadening the selection range of reinforcing phases for titanium-based composites. However, the prediction results of machine learning models require further theoretical verification. First-principles simulations, based on quantum mechanics, can deeply study the structure and properties of materials at the atomic and electronic levels, providing a solid theoretical foundation for materials design. This combination of theoretical simulation and machine learning achieves a leap from macroscopic performance prediction to microscopic mechanism analysis, ensuring the scientific nature and reliability of materials design.

[0018] In terms of material preparation processes, the development of laser additive manufacturing and high-energy ball milling has provided strong support for the preparation of novel titanium-based composite materials. Laser additive manufacturing can achieve near-net-shape forming of materials, precisely controlling the distribution of reinforcing phases and the microstructure of the material; high-energy ball milling can achieve uniform mixing of powders, improving the processing performance of the material. By optimizing laser additive manufacturing process parameters such as powder spreading speed, laser power, scanning speed, and scanning spacing, as well as high-energy ball milling process parameters such as ball-to-powder ratio, rotation speed, and ball milling time, titanium-based composite material samples with excellent comprehensive mechanical properties can be prepared, transforming design concepts into practical materials.

[0019] Compared with existing technologies, the advantages of this invention are:

[0020] The design method for multi-component composite materials based on machine learning integrates the data mining capabilities of machine learning, the theoretical analysis advantages of first-principles methods, and advanced material preparation processes. It provides an efficient, scientific, and systematic new approach for the research and development of titanium-based composite materials, and has significant theoretical and practical value for promoting the performance improvement and widespread application of titanium-based composite materials. Attached Figure Description

[0021] Figure 1 A schematic diagram of TC4 powder used in laser additive manufacturing.

[0022] Figure 2 Charge density plots for Gr / TiB-doped titanium (left) and Gr / TiC-doped titanium (right).

[0023] Figure 3 To predict output for machine learning.

[0024] Figure 4 Samples prepared by printing titanium-based composite materials. Detailed Implementation

[0025] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0026] Example 1

[0027] A design method for multi-component composite materials based on machine learning is proposed. The specific steps are as follows: Data on the reinforcing phase size, relative molecular mass, elemental binding energy, reinforcing content, and thermal expansion coefficient of common reinforcing agents such as TiC, TiB, SiC, Ti5Si3, Gr, B4C, TiB2, TiN, and Ti2Cu are collected as input features. The corresponding strength-ductility product measured experimentally is used as the output, thus establishing a dataset. The Random Forest algorithm is used, with 80 training epochs, a learning rate of 0.02, and a regularization coefficient L2 of 3, to train the dataset, obtaining an asymptotic damage model for the reinforcing phase-titanium matrix composite material. Based on the predictions of this model, TiC, TiB, and Gr are identified as three reinforcing agents that are expected to improve the strength-ductility product of titanium matrix composite materials. Subsequently, the CASTEP block in Materials Studio, a first-principles quantum mechanics software based on density functional theory, was used to simulate the selected TiC, TiB, and Gr reinforcements. The plane wave cutoff energies were measured at 200 eV, 250 eV, and 300 eV, respectively, and the Brillouin zone K-points were measured at 2×2×2, 3×3×3, and 4×4×4, respectively. Convergence tests confirmed that the cutoff energy was 250 eV, and the calculation results were stable and reliable at a K-point of 3×3×3. Analysis of the DOS curves, charge density, and differential charge density revealed that TiC and the titanium matrix had good energy level matching and moderate charge transfer. Gr exhibited a unique charge distribution at the interface, effectively transferring stress, and TiB also showed good reinforcing interaction effects. Next, high-energy ball milling was used to mix TiC, TiB2, Gr, and titanium-based powders, with TiC accounting for 2.0 wt%, TiB2 for 1.0 wt%, and Gr for 1.0 wt%. The average particle size of Gr powder was 5-10 μm, TiB2 was 200-300 nm, and TiC was 200-300 nm. A ball-to-powder ratio of 5:1, a rotation speed of 180 r / min, and a ball milling time of 7 hours were used under argon atmosphere protection. Samples were then prepared using a powder metallurgy laser additive manufacturing process with a powder spreading speed of 0.06 mm / s, a laser power of 270 W, a scanning speed of 1900 mm / s, and a scanning spacing of 0.12 mm. Testing showed that the sample achieved a strength-ductility product of 27 GPa%, 31 GPa%, and 34 GPa%, respectively, representing a significant improvement compared to traditional titanium-based materials without reinforcement (strength-ductility product of approximately 25 GPa%).

[0028] Example 2

[0029] A dataset was constructed by collecting the reinforcing phase size, relative molecular mass, elemental binding energy, reinforcing phase content, and thermal expansion coefficient of reinforcing phases such as TiC, TiB, SiC, Ti5Si3, Gr, B4C, TiB2, TiN, and Ti2Cu as input features and the corresponding strength-ductility product as output. The Random Forest algorithm was then used for training in 80 epochs with a learning rate of 0.02 and a regularization coefficient L2 of 3 to retrain the dataset, resulting in a new asymptotic damage model for the reinforcing phase-titanium matrix composite. This model predicted that TIC+TIB and Gr+TIB composite reinforcing phases have potential in improving the strength-ductility product of titanium matrix composites. Subsequently, the CASTEP block in Materials Studio, a first-principles quantum mechanics software based on density functional theory, was used to simulate TiC+TiB and Gr+TiB. Convergence tests confirmed a plane wave cutoff energy of 280 eV and a Brillouin zone K-point of 4×4×4. By analyzing its DOS curve, charge density, differential charge density, and other physical quantities, it was found that Ti5Si3 can form an effective coherent interface with the matrix, Gr exhibits good cooperative deformation capability at the interface, and the electronic structure of TiB2 enables it to effectively disperse stress under load. Subsequently, high-energy ball milling was used to mix TiC+TiB and Gr+TiB with titanium-based powders, with 0.5 wt% each of TiC and TiB2 composite reinforcements added; and 0.5 wt% each of Gr and TiB2 added to the TiC and Gr composite reinforcements. The average particle size of Gr was 5-10 μm, TiC was 200-300 nm, and TiB2 was 200-300 nm. The ball-to-powder ratio was 5:1, the rotation speed was 180 r / min, the ball milling time was 7 hours, and an argon atmosphere was used for protection. Then, a powder metallurgy laser additive manufacturing process was used, with the laser power adjusted to 290W and the scanning speed to 1850 mm / s, while keeping other parameters unchanged (powder spreading speed 0.04 mm / s~0.08 mm / s, scanning spacing 0.10 mm~0.14 mm). Testing showed that the sample achieved a strength-ductility product of 29 GPa% and 37 GPa%, respectively. Compared with the previously prepared titanium-based composite material (strength-ductility product approximately 25 GPa%), the performance was significantly improved, and even improved compared to the sample in Example 1. This is presumably due to the different synergistic strengthening effects of different reinforcement combinations.

[0030] The specific embodiments described herein are merely illustrative examples of the spirit of the invention. Those skilled in the art to which this invention pertains may make various modifications or additions to the described specific embodiments or use similar methods to substitute them, without departing from the spirit of the invention.

Claims

1. A design method for multi-component composite materials based on machine learning reverse engineering, characterized in that, We selected reinforcing phases from existing reinforcing materials that are expected to achieve high-strength and high-plasticity titanium matrix composites. Using machine learning combined with first-principles simulation, we analyzed the reinforcing materials, their content, and distribution structure that have a significant impact on the comprehensive mechanical properties of titanium matrix composites. Based on the results, we prepared printed parts using laser additive manufacturing and high-energy ball milling processes.

2. A design method for multi-component composite materials based on machine learning reverse engineering, characterized in that, Includes the following steps: Step A, establish a dataset: Select some reinforcing phases that are expected to achieve high strength and high plasticity titanium matrix composites, and use the reinforcing phase size, relative molecular mass, element binding energy, reinforcing phase content and thermal expansion coefficient of the above reinforcing phases as input features, and the strength-plasticity product as the output to establish a dataset; Step B, establish an asymptotic damage model for the reinforcing phase-titanium matrix composite: using the Random Forest algorithm, machine learning is trained on the dataset from Step A to establish an asymptotic damage model for the reinforcing phase-titanium matrix composite, which can predict the reinforcement that is expected to improve the strength-ductility product of the titanium matrix composite. Step C, Simulation Verification: Based on the prediction results in Step B, the selected enhancers are simulated and verified using first-principles simulation methods. Step D, Sample Preparation: Based on the simulation verification results of Step C, select one or more reinforcements, mix them with titanium-based powder using a high-energy ball milling process, and then prepare the sample using powder metallurgy laser additive manufacturing process.

3. The design method for multi-component composite materials based on machine learning reverse engineering as described in claim 2, characterized in that, In step A, the reinforcing agents are TiC, TiB, SiC, Ti5Si3, Gr, B4C, TiB2, TiN, and Ti2Cu.

4. The design method for multi-component composite materials based on machine learning reverse engineering as described in claim 2, characterized in that, In step B, the Random Forest algorithm is set to 80 training rounds, a learning rate of 0.02, and a regularization coefficient L2 of 3.

5. The design method for multi-component composite materials based on machine learning reverse engineering as described in claim 2, characterized in that, In step C, the first-principles simulation method uses the CASTEP module in Materials Studio, a first-principles quantum mechanics software based on density functional theory, to perform calculations. This is used to analyze the DOS curves, charge density, and differential charge density of the screened reinforcements and to determine their interactive reinforcement effects. The convergence test of the plane wave phase energy and Brillouin zone K-point is then performed using the above calculation parameters to obtain the cutoff energy and Brillouin zone K-point used for calculation.

6. The design method for multi-component composite materials based on machine learning reverse engineering as described in claim 2, characterized in that, In step D, the high-energy ball milling process parameters are: ball-to-material ratio 5:1, rotation speed 150-200 r / min, ball milling time 6-8 hours, and argon atmosphere protection.

7. The design method for multi-component composite materials based on machine learning reverse engineering as described in claim 2, characterized in that, In step D, the powder metallurgy laser additive manufacturing process parameters include: powder spreading speed of 0.04 mm / s to 0.08 mm / s, laser power of 240 W to 300 W, scanning speed of 1800 mm / s to 2000 mm / s, and scanning spacing of 0.10 mm to 0.14 mm.

8. The multi-component composite material prepared by the design method according to any one of claims 1-7.

Images