Modulus-tailored martensitic titanium alloys for additive manufacturing and methods of design thereof

Abstract

The application discloses a modulus control type martensitic titanium alloy suitable for additive manufacturing, and relates to the technical field of additive manufacturing, and discloses the following technical scheme: the Nb content is 22-24%, the Ta content is 0.5-0.9%, the Zr content is 1.5-2.5%, and the balance is Ti; the titanium alloy structure is as follows: the target structure is mainly composed of alpha double prime martensite and alpha prime martensite; the modulus control type martensitic titanium alloy suitable for additive manufacturing realizes the cooperation of 29-35 GPa ultra-low elastic modulus and 30-32% elongation after fracture. The application further discloses a design method of the modulus control type martensitic titanium alloy suitable for additive manufacturing. The modulus control type martensitic titanium alloy suitable for additive manufacturing solves the problems that the elastic modulus of the current high-density additive manufacturing titanium alloy is generally high, the toughness is insufficient, and the modulus difference between the melting state and the additive manufacturing state of the low-modulus titanium alloy is large, so that the alloy can obtain stable ultra-low elastic modulus and good toughness matching in the high-density non-porous printing state.

Description

Technical Field

[0001] This invention belongs to the field of additive manufacturing titanium alloy material technology. Specifically, it relates to a modulus-controlled martensitic titanium alloy suitable for additive manufacturing, and also to a design method for the aforementioned modulus-controlled martensitic titanium alloy suitable for additive manufacturing. Background Technology

[0002] Low-modulus titanium alloys with good toughness have important applications in medical implants, flexible load-bearing structures, lightweight components for aerospace, and complex functional structural parts. These materials need to simultaneously meet requirements for elastic compliance, load-bearing reliability, and damage tolerance. However, there is an inherent contradiction between the low elastic modulus and high toughness of titanium alloys: reducing the elastic modulus usually requires reducing the stability of the β-phase, but reduced β-phase stability can easily lead to decreased strength, insufficient microstructural stability, or reduced damage tolerance. Therefore, existing low-modulus titanium alloys often struggle to achieve a synergistic effect of low modulus and high toughness.

[0003] Additive manufacturing provides an effective way to integrally form complex titanium alloy components. However, the rapid solidification, high temperature gradient and repeated thermal cycling during the laser powder bed melting process can significantly change the phase stability, grain morphology, texture orientation and martensitic transformation behavior of titanium alloys. This often results in problems such as increased modulus and insufficient toughness in the printed state of low-modulus titanium alloys obtained by traditional melting, rolling or heat treatment methods. This indicates that traditional low-modulus components cannot be directly transferred to additive manufacturing.

[0004] Existing high-density additive-manufactured titanium alloys still have relatively high moduli. A Chinese invention patent, filed on January 6, 2020, by South China University of Technology, with publication number CN111118339B, discloses a high-strength, low-modulus medical titanium alloy containing Si and its additive manufacturing method. Although this patent aims for "high strength and low modulus," its technical approach is still based on Ti-Nb-Ta-Zr β-type titanium alloys. Furthermore, the patent proposes an additive-manufactured Ti-30Nb-5Ta-3Zr alloy with an elastic modulus of 64.2 GPa, indicating that current additive-manufactured β-titanium alloys still have a significant gap to close with the modulus of human bone tissue. Secondly, the mainstream approach in existing low-modulus titanium alloy patents is to stabilize the β phase. The aforementioned patent explicitly defines its microstructure as a matrix of columnar and equiaxed β-Ti crystals, with (Ti,Zr)2Si phases discontinuously distributed within the grains and at grain boundaries as the reinforcing phase. In other words, the core of this patent is "β-Ti matrix + silicide reinforced phase", and its low modulus and strength come from β phase stabilization, grain refinement and second phase reinforcement.

[0005] Furthermore, the existing additive and molten states exhibit significant performance differences; even with the same composition, the modulus differs considerably between these two preparation states. For example, a Chinese invention patent filed on September 23, 2020, by South China University of Technology, with publication number CN111118339B, modifies the U-shaped or V-shaped molten pool, controls the heat dissipation direction of the molten pool, and the columnar β-Ti... <001> The growth direction ensures that the implant's specific load-bearing direction satisfies a low-modulus orientation. In this patented embodiment, the elastic modulus of the printed Ti-Nb-Zr-Ta alloy in the stress direction is 42 GPa, which is significantly different from the 62 GPa modulus of the as-cast sample. At the same time, the additively manufactured alloy exhibits obvious anisotropy.

[0006] Therefore, there is an urgent need to develop a special low-modulus, high-toughness titanium alloy composition for microstructure control in additive manufacturing printing, so that it can achieve a match between low modulus and high toughness in the printing state, while the composition is not affected by the preparation state. Summary of the Invention

[0007] The first objective of this invention is to provide a modulus-controlled martensitic titanium alloy suitable for additive manufacturing, which solves the problems of generally high elastic modulus and insufficient toughness of current high-density additive manufacturing titanium alloys, as well as the large modulus difference between the molten state and the additive manufacturing state of traditional low-modulus titanium alloys. This allows the alloy to obtain a stable ultra-low elastic modulus and good toughness matching in a high-density non-porous printing state.

[0008] The second objective of this invention is to provide a design method for the above-mentioned modulus-controlled martensitic titanium alloy suitable for additive manufacturing. This method overcomes the extrapolation problem of machine learning in the process of optimizing alloy structure and helps to discover an unexpected alloy structure.

[0009] The first technical solution adopted in this invention is: This invention is applicable to additively manufactured modulus-controlled martensitic titanium alloys, which, by mass percentage, are composed of the following components: Nb content of 22% to 24%, Ta content of 0.5% to 0.9%, Zr content of 1.5% to 2.5%, and the balance being Ti; The structure of a modulus-controlled martensitic titanium alloy suitable for additive manufacturing is as follows: The target microstructure is mainly composed of α″ martensite and α′ martensite; Modulus-controlled martensitic titanium alloys suitable for additive manufacturing achieve a synergistic effect of ultra-low elastic modulus of 29~35GPa and elongation after fracture of 30~32%.

[0010] The invention is further characterized by: It consists of the following components by mass percentage: 23% Nb, 0.7% Ta, 2% Zr, and the balance Ti.

[0011] Modulus-controlled martensitic titanium alloys suitable for additive manufacturing are compatible with additive manufacturing and smelting processes. Titanium alloys are preferably prepared using additive manufacturing processes. The additive manufacturing process includes: preparing mixed powders of Ti, Nb, Ta, and Zr elements or Ti-Nb-Ta-Zr pre-alloyed powders according to the titanium alloy composition; after sieving, drying, and mixing, forming the alloy using selective laser melting (SLM) under an inert gas atmosphere; wherein the SLM process parameters are: laser power 150–240 W, scanning speed 600–1200 mm / s, scanning spacing 0.06–0.10 mm, and layer thickness 0.02–0.04 mm. The resulting printed titanium alloy has a target microstructure mainly composed of α″ martensite and α′ martensite, with an elastic modulus of 29–35 GPa and an elongation after fracture of 30–32%. Furthermore, the above-mentioned preferred composition is also applicable to the preparation by the smelting process, which includes: feeding the titanium alloy according to the above composition, melting it under vacuum or inert gas protection arc and remelting it 3 to 5 times to obtain an alloy ingot, and then subjecting it to homogenization treatment at 1000℃ for 2 h, water cooling, 80% cold rolling deformation, solution treatment at 900℃ for 10 min and water cooling. The elastic modulus of the resulting smelted-cold-rolled-solution titanium alloy is 33 to 34 GPa, thus achieving a record-breaking ultra-low elastic modulus for the same composition under different preparation conditions.

[0012] The second technical solution adopted in this invention is: a design method for modulus-controlled martensitic titanium alloys suitable for additive manufacturing, the specific method of which is as follows: S1. Using Ti as the matrix element and Nb, Ta, Zr, Mo, Sn and other elements as initial candidate alloying elements, candidate components are screened. Under the constraints of element content range, Bo-Md parameters and martensitic transformation boundary, a candidate component dataset that meets physical rationality is generated. A variational autoencoder is used to pre-train the candidate component dataset. S2. Establish the target stretching curve based on the target and extract the mechanical characteristics from the target curve; S3. Map the target stretching curve features to the composition latent space of the variational autoencoder, and obtain the candidate alloy composition by decoding; predict the elastic modulus and mechanical properties of the candidate alloy, and optimize its composition and additive manufacturing process parameters to obtain the above-mentioned titanium alloy.

[0013] The invention is further characterized by: The specific method for S1 is as follows: S1.1 Establish and screen candidate alloy composition space: With Ti as the balance, Nb, Ta, Zr, Mo and Sn are candidate alloying elements. Set the mass percentage range of each element, and screen metastable candidate compositions close to the martensitic transformation boundary based on Bo-Md phase stability parameters. S1.2. Establishing the latent space of composition using variational autoencoders: A variational autoencoder is used to pre-train the candidate composition dataset after physical constraints to learn a continuous latent space of alloy composition, so that subsequent compositions can be searched in a confined space close to the martensitic transformation boundary, which is suitable for the control of the microstructure of additive manufacturing printing state.

[0014] The specific method for S2 is as follows: S2.1 Establish target stretching curve: Establish target stretching curve based on the selected target; S2.2 Extracting features of the stretching curve: High-dimensional features of the curve are extracted using a convolutional neural network.

[0015] The specific targets for screening in S2.1 are an elastic modulus of 30 GPa, an elongation after fracture of 40%, a yield strength of 600 MPa, and a tensile strength of 900 MPa.

[0016] The specific method for S3 is as follows: S3.1 Establish the mapping relationship between the target tensile curve and the alloy composition: Map the characteristics of the target tensile curve to the composition latent space of the variational autoencoder, and obtain the candidate alloy composition by decoding, so as to realize the reverse design from the target mechanical properties and target microstructure tendency to the alloy composition. S3.2 Optimization using surrogate models and genetic algorithms: Machine learning surrogate models are used to predict the elastic modulus and other properties of candidate alloys, and genetic algorithms are combined to iteratively optimize the alloy composition and additive manufacturing process parameters, ultimately obtaining the preferred composition of Ti-23Nb-0.7Ta-2Zr.

[0017] The beneficial effects of this invention are: (1) The modulus-controlled martensitic titanium alloy prepared by the method of the present invention is suitable for additive manufacturing and can fully adapt to the characteristics of rapid solidification and non-equilibrium phase transformation in additive manufacturing, ultimately achieving an ultra-low elastic modulus of 29~35GPa, excellent toughness and high work hardening rate. (2) The martensitic titanium alloy composition designed by the method of the present invention not only maintains an ultra-low elastic modulus of 29~35GPa during the additive manufacturing process, but is also unaffected by the preparation process. Even under the smelting preparation method, it can still maintain a record-breaking ultra-low elastic modulus of 33~34GPa. (3) The modulus-controlled martensitic titanium alloy prepared by the method of the present invention is suitable for additive manufacturing. The stress-induced phase transformation from α″ to α′ in the elastic deformation stage is found and continues to occur throughout the deformation process. This phenomenon can ensure ultra-low modulus, high work hardening and high toughness. Attached Figure Description

[0018] Figure 1This invention provides tensile curves and property comparison diagrams of modulus-controlled martensitic titanium alloys applicable to additive manufacturing. Figure 2 This is a microstructure diagram of the modulus-controlled martensitic titanium alloy before deformation, which is applicable to additive manufacturing according to the present invention. Figure 3 This is a diagram showing the phase transformation behavior of modulus-controlled martensitic titanium alloys in the tensile process of additive manufacturing, which is applicable to the present invention. Figure 4 This invention relates to the tensile curves of modulus-controlled martensitic titanium alloys applicable to additive manufacturing under different preparation processes. Detailed Implementation

[0019] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments.

[0020] Example 1 This invention is applicable to additively manufactured modulus-controlled martensitic titanium alloys, which, by mass percentage, consist of the following components: Nb content of 22%–24%, Ta content of 0.5%–0.9%, Zr content of 1.5%–2.5%, and the balance being Ti; The structure of a modulus-controlled martensitic titanium alloy suitable for additive manufacturing is as follows: The target microstructure is mainly composed of α″ martensite and α′ martensite, and this structure is basically independent of the continuous β phase matrix.

[0021] Modulus-controlled martensitic titanium alloys suitable for additive manufacturing achieve a synergistic effect of low elastic modulus of 29~35 GPa and elongation after fracture of 30~32%.

[0022] The present invention is composed of the following components: Nb content is 23%, Ta content is 0.7%, Zr content is 2%, and the balance is Ti.

[0023] This invention is applicable to additive manufacturing and smelting processes for modulus-controlled martensitic titanium alloys. The additive manufacturing process parameters are as follows: laser power 150-240 W, scanning speed 600-1200 mm / s, scanning spacing 0.06-0.10 mm, layer thickness 0.02-0.04 mm, and the elastic modulus of the resulting printed titanium alloy is 29-35 GPa, with an elongation after fracture of 30-32%. The melting process parameters are as follows: after vacuum or inert gas protected arc melting and 3 to 5 remelting cycles, homogenization treatment at 1000℃ for 2 hours, water cooling, 80% cold rolling deformation, solution treatment at 900℃ for 10 minutes, and water cooling, the elastic modulus of the obtained titanium alloy is 33 to 34 GPa, thus achieving a record-breaking ultra-low elastic modulus for the same composition under different preparation conditions.

[0024] The innovation of the components of this invention lies in: The molybdenum equivalent coefficient of the alloy of this invention is very low, only 8.35, far lower than that of conventionally reported Ti-Nb, Ti-Zr, and Ti-Mo alloys (generally greater than 10). Even within the same Ti-Nb-Ta-Zr system, the Nb, Ta, and Zr content of the alloy of this invention (Ti-23Nb-0.7Ta-2Zr) is very low, far lower than that of other alloys (e.g., Ti-30Nb-5Ta-3Zr in CN111118339B). It is this formulation that makes the microstructure (fully martensitic) and properties (modulus 29-35 GPa) of this alloy significantly different from other reported alloys (β-structure, modulus greater than 42 GPa). The reasons for the selection of the roles and contents of each component in this invention are as follows: Ti: Ti is the matrix element of titanium alloys, used to ensure that the alloy has low density, good corrosion resistance and good biocompatibility, and to provide the basic crystal structure for the formation of α″ martensite and α′ martensite.

[0025] Nb: Nb is a β-stabilizing element and a key element in regulating the phase stability and martensitic transformation tendency of the alloy in this invention. An appropriate amount of Nb can keep the alloy in a region close to the martensitic transformation boundary. An Nb content of 22%–24% can regulate the martensitic transformation without causing excessive β-stabilization of the alloy. If the Nb content is too low, the β-phase stability is insufficient, making it difficult to obtain low-modulus α″ martensite; if the Nb content is too high, it easily leads to excessive β-phase stabilization, causing the alloy to shift towards the traditional β-type low-modulus alloy route, which is not conducive to forming the martensitic dominant structure of this invention.

[0026] The present invention uses a Ta content of 0.5% to 0.9%, which is a low Ta content design to avoid the difficulties in printing melting and composition homogenization caused by high Ta content. Ta is a β-stabilizing element that can synergistically regulate the stability of the β phase with Nb, and has the functions of solid solution strengthening and improving microstructure stability. Ta is mainly used to fine-tune phase stability, rather than to form a continuous β phase matrix. If the Ta content is too low, its synergistic effect with Nb in regulating phase stability is insufficient; if the Ta content is too high, it may increase the difficulty of powder melting and composition homogenization, and lead to excessive β phase stability.

[0027] Zr: Zr is a neutral or near-neutral alloying element that can dissolve in the titanium matrix, regulating lattice distortion and phase transformation behavior, which is beneficial for improving the balance between low modulus and toughness. The Zr content is 1.5%–2.5%. If the Zr content is too low, its effect on regulating lattice distortion, phase transformation behavior, and toughness is limited; if the Zr content is too high, it may change the phase stability range and affect the proportion of martensite.

[0028] Example 2 This embodiment provides a design method for modulus-controlled martensitic titanium alloys suitable for additive manufacturing. The method uses the metastable composition range near the martensitic transformation boundary as the design space, combining target stretching curves, variational autoencoder latent space modeling, and genetic algorithm optimization to reverse-select martensitic-dominant low-modulus titanium alloys suitable for the printed state. The specific method is as follows: S1. Constructing a physically constrained composition latent space: Nb, Ta, Zr, Mo, and Sn are selected as candidate alloying elements, and their content ranges are set. Physically reasonable compositions are screened based on Bo-Md parameters and Ms line distance constraints. A variational autoencoder is used for pre-training to establish a physically constrained continuous composition latent space.

[0029] S2. Establish the mapping from target performance to latent variables: Generate the target tensile curve based on the target elastic modulus, yield strength, tensile strength and plastic deformation capacity, extract the mechanical response characteristics, and map them to the latent space of the variational autoencoder to obtain the target latent variables.

[0030] S3. Joint optimization of composition and additive manufacturing process: Sample the target latent variable and decode it through a variational autoencoder to obtain candidate composition. Combine the machine learning surrogate model to predict the elastic modulus, and then jointly optimize the alloy composition through a genetic algorithm.

[0031] Example 3 This invention relates to a design method for modulus-controlled martensitic titanium alloys manufactured using additive manufacturing. Based on Example 2, the specific method of S1 is as follows: S1.1 Establishing the initial candidate composition space: Taking Ti as the matrix element, Nb, Ta, Zr, Mo and Sn are included in the initial candidate space as preferred candidate alloying elements.

[0032] S1.2 Setting the elemental content range: Candidate components are expressed as mass percentages, with the component vector being X = (Ti, Mo, Nb, Zr, Sn, Ta), where Nb is 15–30 wt.%, Ta is 0–2 wt.%, Mo is 0–9 wt.%, Zr is 0–14 wt.%, Sn is 0–14 wt.%, and Ti is the balance. For each candidate component, the sum of Ti, Mo, Nb, Zr, Sn, and Ta must be 100%.

[0033] S1.3. Perform Bo-Md phase stability screening: Convert candidate components from mass percentage to atomic fraction, and calculate the average Bo value and average Md value by weighting according to atomic fraction. Retain candidate components with Bo≥2.8 and Md≥2.4, placing them in the metastable region close to the β-phase stable region and near the martensitic transformation boundary.

[0034] S1.4. Perform Ms line distance filtering: Calculate the normalized Euclidean distance from the candidate component points to the martensitic transformation boundary Ms line, and retain candidate components with a distance ≤ 0.01. After filtering by element range, Bo-Md parameter, and Ms line distance, approximately 19,025 physically reasonable candidate components are obtained, which are used as the pre-training dataset for the variational autoencoder.

[0035] S1.5. Pre-training the variational autoencoder: The variational autoencoder input is a six-dimensional component vector, with a latent variable dimension of 20. Both the encoder and decoder use multi-layer fully connected networks with 100 hidden layer nodes, LeakyReLU activation function, and a negative slope of 0.2. The encoder outputs the latent variable mean and variance, and the decoder restores the 20-dimensional latent variables to six-dimensional components.

[0036] S1.6 Setting the training parameters for the variational autoencoder: The training loss for the variational autoencoder includes component reconstruction loss, KL divergence loss, element range constraint loss, and component sum constraint loss. Training uses the Adam optimizer with a learning rate of 5×10n. -6 The batch size is 64, and the number of training rounds is 500. The KL weights are gradually increased from 0.01 to 0.1.

[0037] Example 4 This invention relates to a design method for modulus-controlled martensitic titanium alloys manufactured using additive manufacturing. Based on Example 2, the specific method of S2 is as follows: S2.1 Establish the target tensile curve: Use elastic modulus, yield strength, tensile strength, uniform deformation strain, fracture stress, and elongation after fracture as target descriptors. In this embodiment, the elastic modulus is set to 30 GPa, yield strength to 600 MPa, tensile strength to 900 MPa, uniform deformation strain to 15%, fracture stress to 200 MPa, and elongation after fracture to 40%.

[0038] S2.2 Constructing the Curve Morphology: The target tensile curve includes the elastic stage, work hardening stage, and necking stage. The elastic stage is used to limit the low modulus, the post-yield stage is used to limit continuous work hardening, and the necking stage is used to limit the plastic deformation capacity before fracture. The work hardening index is set to 0.25, and the necking curvature factor is set to 3.

[0039] S2.3 Generating curve data: 5000 data points are sampled for each target stretching curve, and Gaussian filtering is used for smoothing with a smoothing parameter of 1.

[0040] S2.4 Extracting Curve Features: The target tensile curve and the experimental tensile curve are converted into curve images and uniformly scaled to 224×224 pixels. A domain adversarial neural network is used to extract curve features, enabling the model to focus on elastic slope, yield transition, work hardening, and pre-fracture response.

[0041] S2.5 Setting the Domain Adversarial Neural Network Structure: The feature extractor of the domain adversarial neural network is a three-layer convolutional neural network with 64, 128, and 256 convolutional channels respectively, a kernel size of 3×3, and the activation function ReLU. The convolutional layers are followed by pooling layers, flattening layers, and fully connected layers, ultimately outputting 128-dimensional curve features.

[0042] S2.6 Setting Training Parameters for the Domain Adversarial Neural Network: The domain adversarial neural network includes a gradient inversion layer and a domain classifier. The domain classifier consists of a 128-dimensional input layer, a 100-node fully connected layer, a Dropout layer, and a Sigmoid output layer. The model is trained using the Adam optimizer with a learning rate of 0.001, a batch size of 8, and 100 training epochs. The gradient inversion coefficient is gradually increased from 0 to 1.

[0043] Example 5 This invention relates to a design method for modulus-controlled martensitic titanium alloys manufactured using additive manufacturing. Based on Example 2, the specific method of S3 is as follows: S3.1 Predicting the latent variables of the variational autoencoder: Inputting 128-dimensional curve features into the latent space mapping model, the output is a 20-dimensional target latent variable of the variational autoencoder. This mapping model uses a multilayer perceptron with an input dimension of 128, an output dimension of 20, 100 hidden layer nodes, and a Dropout ratio of 0.3.

[0044] S3.2 Setting the training parameters for the mapping model: The mapping model uses the latent variables of the variational autoencoder corresponding to the features of the experimental curve as the supervised target and is trained using mean squared error loss. The optimizer is Adam, the learning rate is 0.001, and the weight decay is 1×10⁻⁶. -5 The batch size is 32, the number of training rounds is 1000, and the gradient clipping threshold is 1.0.

[0045] S3.3. Generating the Initial Genetic Algorithm Population: Centering on the 20-dimensional target latent variable obtained in Example 4, Gaussian sampling is performed in the latent space of the variational autoencoder. The initial population size is 50, the sampling standard deviation is 0.2, and the latent space dimension is 20. Each individual is decoded by the variational autoencoder to obtain a set of candidate alloy compositions.

[0046] S3.4 Establishing a proxy model for elastic modulus: The XGBoost regression model is used to predict the elastic modulus of candidate alloys. The training data includes 46 additive manufacturing titanium alloy samples. The input variables include the contents of Ti, Mo, Nb, Sn, Ta, and Zr, laser power, scanning speed, and volumetric energy density; the output variable is the elastic modulus.

[0047] S3.5. Set the data range for the surrogate model: In the training data, Ti is 44–92.5 wt.%, Mo is 0–18 wt.%, Nb is 0–42 wt.%, Sn is 0–12 wt.%, Ta is 0–50 wt.%, and Zr is 0–31.47 wt.%; the laser power is 100–550 W, the scanning speed is 167–1200 mm / s, the scanning interval is 0.03–0.15 mm, the layer thickness is 0.01–0.10 mm, and the elastic modulus is 41–118 GPa.

[0048] S3.6 Setting XGBoost Model Parameters: XGBoost uses random search and 5-fold cross-validation to determine parameters. The number of trees is 300–700, the learning rate is 0.01–0.08, the maximum tree depth is 4–8, the subsampling ratio is 0.7–0.9, the number of random searches is 20, and the evaluation metric is R0. 2 .

[0049] S3.7 Setting the process parameter search range: Each variational autoencoder decoding component is evaluated in combination with multiple sets of additive manufacturing process parameters. Candidate laser powers are 150 W, 180 W, 210 W, and 240 W; candidate scanning speeds are 600 mm / s, 800 mm / s, 1000 mm / s, and 1200 mm / s; the scanning spacing is fixed at 0.08 mm; and the layer thickness is fixed at 0.03 mm.

[0050] S3.8 Setting Genetic Algorithm Parameters: The genetic algorithm iterates for 30 generations, recording a total of 31 results including the initial generation 0. The best 5 individuals are retained in each generation, with parents selected from the top 50% of individuals in fitness. Crossover is performed using two-point crossover with a crossover probability of 0.8; mutation is performed using Gaussian perturbation with a mutation probability of 0.2 and a mutation standard deviation of 0.2.

[0051] S3.9 Obtaining Optimization Results: During the optimization process, the contents of Mo and Sn gradually decreased and were eventually eliminated, Ta decreased to a low content range, Nb remained at approximately 23 wt.%, and Zr stabilized at around 2 wt.%. Representative results include generation 0 Ti-1Mo-23Nb-1.5Zr-1.6Sn-1.6Ta, generation 10 Ti-23Nb-1.6Zr-0.5Ta, generation 26 Ti-23Nb-1.5Zr-0.7Ta, and finally Ti-23Nb-0.7Ta-2Zr.

[0052] Example 6 Based on Example 1, this example applies to the additive manufacturing process of modulus-controlled martensitic titanium alloys, and the specific process is as follows: This embodiment provides an additive manufacturing process for a Ti-23Nb-0.7Ta-2Zr martensitic titanium alloy. The alloy comprises, by mass percentage: 23 wt.% Nb, 0.7 wt.% Ta, 2 wt.% Zr, with Ti as the balance. The process involves elemental powder preparation, mechanical powder mixing, and in-situ alloying using selective laser melting to obtain a printed martensitic-dominant low-modulus titanium alloy.

[0053] S1. Raw Material Selection and Composition Ratio: Ti, Nb, Ta, and Zr are selected as raw materials, with Ti purity not less than 99.7%, Nb purity not less than 99.9%, Ta purity not less than 99.9%, and Zr purity not less than 99.5%. The final alloy composition is determined according to Ti 74.3 wt.%, Nb 23 wt.%, Ta 0.7 wt.%, and Zr 2 wt.%, with the oxygen content preferably not exceeding 0.20 wt.% and the nitrogen content preferably not exceeding 0.05 wt.%.

[0054] S2. Elemental Powder Preparation: Ti powder, Nb powder, Ta powder, and Zr powder were prepared separately. Ti powder and Zr powder were prepared using hydrogenation-dehydrogenation, gas atomization, or plasma spheroidization. Nb powder and Ta powder were prepared using plasma rotating electrode atomization, gas atomization, or plasma spheroidization. The resulting powders were spherical or near-spherical, with a particle size controlled between 15 and 53 μm to meet the requirements for selective laser melting and powder spreading.

[0055] S3. Powder Sieving and Drying: The Ti powder, Nb powder, Ta powder, and Zr powder obtained in S2 were sieved through a 15–53 μm sieve to remove fine powder, coarse powder, and agglomerated particles. The sieved powders were placed in a vacuum drying oven and dried at 60–100℃ for 4–12 h. After drying, the powders were cooled under argon protection and sealed for storage to prevent oxygen and moisture absorption.

[0056] S4. Powder Weighing and Mechanical Mixing: Weigh Ti powder, Nb powder, Ta powder, and Zr powder according to the target composition, so that the overall composition of the mixed powder is Ti-23Nb-0.7Ta-2Zr. The weighing error is preferably controlled within ±0.05 wt.%, and the weighing process is preferably carried out in an argon-protected environment.

[0057] S5. Graded Mechanical Mixing: Mechanical mixing is carried out using a three-dimensional motion powder mixer, a V-type powder mixer, or a drum-type powder mixer. The mixing atmosphere is argon, and the filling rate of the mixing tank is 30%–60%. Preferably, Ta powder or Ta-containing pre-alloyed powder is first pre-mixed with a portion of Ti powder at a mass ratio of 1:(5–20) for 30–60 min to form Ta pre-dispersed powder; then Nb powder, Zr powder, or Nb and Zr-containing pre-alloyed powder are added and mixed for 30–60 min; finally, the remaining Ti powder is added for total mixing. The total mixing time is 2 h, and the mixing speed is 50–120 r / min.

[0058] In another embodiment, a pre-alloyed powder route is employed. Sponge titanium, niobium particles, tantalum particles, and zirconium particles are weighed according to the Ti-23Nb-0.7Ta-2Zr composition. These are pressed into electrodes and then subjected to vacuum arc melting, remelted 3–5 times to obtain a uniformly composed Ti-Nb-Ta-Zr alloy ingot. The ingot is machined into rods suitable for powder preparation, and the surface oxide scale is removed. Subsequently, pre-alloyed powder is prepared using an electrode induction melting gas atomization method. The atomizing gas is high-purity argon, the atomization pressure is 3.5–4.5 MPa, the melting power is 20–30 kW, and the feed rate is 35–45 mm / min. After atomization, the powder is subjected to airflow classification and sieving to obtain spherical pre-alloyed powder with a particle size of 15–53 μm.

[0059] S6. Re-sieving and Pre-printing Inspection of Mixed Powder: The mixed powder is sieved again through a 15–53 μm sieve and vacuum dried at 80 ℃ for 6 h. Before printing, the particle size distribution, flowability, and gaseous impurity content of the mixed powder are inspected. Preferably, the Hall flow rate is not greater than 35 s / 50 g, the loose bulk density is not less than 45% of the theoretical density, the oxygen content is not higher than 0.20 wt.%, the nitrogen content is not higher than 0.05 wt.%, and the hydrogen content is not higher than 0.015 wt.%.

[0060] S7. Model Construction and Slicing: Establish a 3D model of the component or sample to be formed, import it into the slicing software for placement, support design, and layered slicing. The slice thickness is set to 0.03 mm, and the scanning interval is set to 0.08 mm. The scanning strategy adopts interlayer rotation scanning, with the scanning direction of adjacent layers rotated by 67° to improve the uniformity of heat input and the uniformity of component diffusion during multilayer remelting.

[0061] S8. Substrate Preparation and Atmosphere Control: Ti-6Al-4V alloy is selected as the substrate. Before printing, the substrate surface is polished, cleaned with anhydrous ethanol, and dried. The substrate is preheated to 100–200 °C, preferably 160 °C. Before forming, the forming cavity is evacuated and filled with high-purity argon gas. During the printing process, the oxygen content is controlled below 1000 ppm, preferably below 500 ppm.

[0062] S9. Selective Laser Melting: The mixed powder obtained in step S7 is shaped using a selective laser melting (SLM) device. The laser spot diameter is no greater than 40 μm; the process parameters are: laser power 150–240 W, scanning speed 600–1200 mm / s, scanning interval 0.06–0.10 mm, and layer thickness 0.02–0.04 mm. The preferred process parameters are: laser power 180 W, scanning speed 1000 mm / s, scanning interval 0.08 mm, layer thickness 0.03 mm, and interlayer rotation angle 67°.

[0063] S10. Cooling and Removal: After printing, the formed part is cooled with the forming platform in an argon protective atmosphere. After the forming cavity temperature drops below 80 ℃, it is removed and the substrate is separated and the residual powder on the surface is cleaned.

[0064] Example 7 Based on Example 1, the specific process flow for the melting and preparation of modulus-controlled martensitic titanium alloys applicable to additive manufacturing in this example is as follows: In this embodiment, the titanium alloy is composed of the following components by mass percentage: Nb: 23%, Ta: 0.7%, Zr: 2%, with the balance being Ti, and the sum of the mass percentages of each component is 100%.

[0065] The preparation method of the titanium alloy in this embodiment includes the following steps: S1. Alloy Batching and Melting: High-purity Ti, Nb, Ta, and Zr raw materials are weighed according to the target composition of Ti-23Nb-0.7Ta-2Zr. After surface cleaning, the weighed raw materials are placed in a vacuum arc melting furnace and melted under an inert gas protective atmosphere to obtain an alloy ingot with the composition Ti-23Nb-0.7Ta-2Zr. To improve the uniformity of the alloy composition, the alloy ingot is turned over and remelted during the melting process, preferably 3 to 5 times.

[0066] S2. Homogenization treatment: The alloy ingot obtained in S1 is placed in a vacuum or inert gas protective atmosphere for homogenization heat treatment at a temperature of 1000℃ for 2 hours. After the holding time, the alloy ingot is quickly removed and water-cooled to suppress the precipitation of equilibrium phases during slow cooling, thus obtaining a homogenized alloy billet.

[0067] S3. Cold Rolling Deformation: After removing the surface oxide layer from the alloy billet obtained in S2, cold rolling is performed with a total deformation of 80%. During cold rolling, multi-pass rolling can be used to gradually thin the billet, avoiding excessive deformation in a single pass that could lead to cracking. The cold-rolled billet yields a cold-deformed Ti-23Nb-0.7Ta-2Zr alloy sheet.

[0068] S4. Solution Treatment: The cold-rolled sheet obtained in S3 is placed in a vacuum or inert gas protective atmosphere for solution treatment at a temperature of 900℃ for 10 minutes. After the heat treatment, it is water-cooled to obtain a solution-treated Ti-23Nb-0.7Ta-2Zr alloy sheet.

[0069] S5. Sample Processing and Performance Testing: The solution-treated sheet obtained in S4 was machined into tensile specimens. The specimen surface was subjected to progressive grinding and polishing to remove machining marks and surface defects. Subsequently, room temperature tensile performance tests were conducted on the specimens to obtain the tensile curves of the melt-cold-rolled-solution-treated Ti-23Nb-0.7Ta-2Zr alloy.

[0070] Testing revealed that the elastic modulus of the Ti-23Nb-0.7Ta-2Zr alloy obtained in this embodiment is 33-34 GPa. This result indicates that, with unchanged composition, the alloy of this invention can maintain a low elastic modulus of 33-34 GPa after conventional melting, homogenization, cold rolling, and solution treatment, which is essentially consistent with the 29-35 GPa elastic modulus of the additively manufactured state of this invention. This demonstrates that the performance of the alloy of this invention is not limited to a single preparation process; unexpectedly low modulus can be obtained under different preparation processes.

[0071] The core innovation of this invention lies in: firstly, proposing an additive manufacturing method for printing a low-modulus, high-toughness titanium alloy structure dominated by martensite, and further proposing a machine learning-aided composition design method and corresponding alloy composition for obtaining this structure.

[0072] This invention proposes a low-modulus, high-toughness structure dominated by martensite in a printed state. Unlike the continuous β-phase matrix in traditional low-modulus β-titanium alloys, this structure consists of α″-martensite and α′-martensite microstructures formed at room temperature in a printed state. This structure is largely independent of the continuous β-phase matrix, yet achieves a balance between a low elastic modulus of 29-35 GPa and high toughness, which is the main structural innovation of this invention.

[0073] To achieve the aforementioned structure, this invention proposes a machine learning-assisted composition design method for the target printed microstructure. Instead of simply predicting the low elastic modulus, this method aims to "form a martensite-dominated low-modulus, high-toughness microstructure." It combines Bo-Md phase stability screening, variational autoencoder composition latent space, domain adversarial neural network stretching curve feature extraction, and genetic algorithm optimization to screen critically stable compositions suitable for rapid solidification in additive manufacturing from the commonly used element space of traditional low-modulus β-titanium alloys.

[0074] Compared to existing technologies, the key breakthrough of this invention is not a "reduction in modulus," but rather the direct reduction of the elastic modulus of a high-density, non-porous, additively manufactured solid titanium alloy to 29-35 GPa. This value is significantly lower than the 42 GPa or higher levels of existing additively manufactured low-modulus titanium alloys, and also significantly lower than the common range of 50-80 GPa for most β-type low-modulus titanium alloys. More importantly, this invention maintains a 30-32% elongation after fracture at the low modulus of 29-35 GPa, indicating that it does not sacrifice plasticity or rely on porous structure to achieve a low modulus, but rather achieves a simultaneous breakthrough in low modulus and high toughness in a dense solid material. Figure 1 As shown.

[0075] This invention does not continue to stabilize the β phase, nor does it rely on a continuous β phase matrix or β phase. <001> Instead of using texture or β-phase second-phase strengthening to reduce modulus, the alloy is intentionally designed within a compositional window with low β stability and a high susceptibility to non-equilibrium martensite transformation. This allows the alloy to form a refined martensite microstructure dominated by α″ and α′ martensite after rapid solidification during laser powder bed melting. Figure 2 As shown.

[0076] Therefore, the true innovation of this invention lies in transforming the traditionally considered "unstable" martensitic structure into a functional microstructure with low modulus and high toughness. α″ martensite provides a low-modulus response, while α′ martensite and refined interfaces provide deformation coordination. Further stress-induced phase transformation from α″ to α′ occurs during tensile testing. This represents a completely different technical approach from existing β-type low-modulus titanium alloys. Figure 3 As shown.

[0077] The present invention preferably uses a Ti-23Nb-0.7Ta-2Zr alloy with an elastic modulus of 29-35 GPa in the laser powder bed molten printing state; simultaneously, as Figure 4 As shown, the tensile modulus of this alloy in the molten state is 33-34 GPa. The modulus is almost identical under two completely different preparation methods, indicating that the low modulus of the alloy composition of this invention can be extended to different preparation processes.

Claims

1. A modulus-controlled martensitic titanium alloy suitable for additive manufacturing, characterized in that, By mass percentage, it consists of the following components: Nb content of 22%–24%, Ta content of 0.5%–0.9%, Zr content of 1.5%–2.5%, and the balance being Ti; The structure of the modulus-controlled martensitic titanium alloy suitable for additive manufacturing is as follows: The target microstructure is mainly composed of α″ martensite and α′ martensite; The modulus-controlled martensitic titanium alloy suitable for additive manufacturing achieves a synergistic effect of an ultra-low elastic modulus of 29~35GPa and an elongation after fracture of 30~32%.

2. The modulus-controlled martensitic titanium alloy suitable for additive manufacturing according to claim 1, characterized in that, It consists of the following components by mass percentage: 23% Nb, 0.7% Ta, 2% Zr, and the balance Ti.

3. The modulus-controlled martensitic titanium alloy suitable for additive manufacturing according to claim 1 or 2, characterized in that, The modulus-controlled martensitic titanium alloy suitable for additive manufacturing is adapted to additive manufacturing and smelting processes. The additive manufacturing process parameters are as follows: laser power 150-240 W, scanning speed 600-1200 mm / s, scanning spacing 0.06-0.10 mm, layer thickness 0.02-0.04 mm, and the resulting printed titanium alloy has an elastic modulus of 29-35 GPa and an elongation after fracture of 30-32%. The smelting process parameters are as follows: after vacuum or inert gas protected arc melting and 3 to 5 remelting cycles, homogenization treatment at 1000℃ for 2 hours, water cooling, 80% cold rolling deformation, solution treatment at 900℃ for 10 minutes, and water cooling, the elastic modulus of the obtained titanium alloy is 33 to 34 GPa, thus achieving a record-breaking ultra-low elastic modulus for the same composition under different preparation conditions.

4. A design method for modulus-controlled martensitic titanium alloys suitable for additive manufacturing, characterized in that, Specifically as follows: S1. Using Ti as the matrix element and Nb, Ta, Zr, Mo, Sn and other elements as initial candidate alloying elements, candidate components are screened. Under the constraints of element content range, Bo-Md parameters and martensitic transformation boundary, a candidate component dataset that meets physical rationality is generated. A variational autoencoder is used to pre-train the candidate component dataset. S2. Establish the target stretching curve based on the target and extract the mechanical characteristics from the target curve; S3. Map the target stretching curve features to the composition latent space of the variational autoencoder, and obtain the candidate alloy composition by decoding; predict the elastic modulus and mechanical properties of the candidate alloy, and optimize its composition and additive manufacturing process parameters to obtain the titanium alloy as described in claim 1 or 2.

5. The design method for modulus-controlled martensitic titanium alloys suitable for additive manufacturing according to claim 4, characterized in that, The specific method of S1 is as follows: S1.1 Establish and screen candidate alloy composition space: With Ti as the balance, Nb, Ta, Zr, Mo and Sn are candidate alloying elements. Set the mass percentage range of each element, and screen metastable candidate compositions close to the martensitic transformation boundary based on Bo-Md phase stability parameters. S1.

2. Establishing the latent space of composition using variational autoencoders: A variational autoencoder is used to pre-train the candidate composition dataset after physical constraints to learn a continuous latent space of alloy composition, so that subsequent compositions can be searched in a confined space close to the martensitic transformation boundary, which is suitable for the control of the microstructure of additive manufacturing printing state.

6. The design method for modulus-controlled martensitic titanium alloys suitable for additive manufacturing according to claim 4, characterized in that, The specific method of S2 is as follows: S2.1 Establish target stretching curve: Establish target stretching curve based on the selected target; S2.2 Extracting features of the stretching curve: High-dimensional features of the curve are extracted using a convolutional neural network.

7. The design method for modulus-controlled martensitic titanium alloys suitable for additive manufacturing according to claim 6, characterized in that, The specific targets for screening in S2.1 are an elastic modulus of 30 GPa, an elongation after fracture of 40%, a yield strength of 600 MPa, and a tensile strength of 900 MPa.

8. The design method for modulus-controlled martensitic titanium alloys suitable for additive manufacturing according to claim 4, characterized in that, The specific method of S3 is as follows: S3.1 Establish the mapping relationship between the target tensile curve and the alloy composition: Map the characteristics of the target tensile curve to the composition latent space of the variational autoencoder, and obtain the candidate alloy composition by decoding, so as to realize the reverse design from the target mechanical properties and target microstructure tendency to the alloy composition. S3.2 Optimization using surrogate models and genetic algorithms: Machine learning surrogate models are used to predict the elastic modulus and other properties of candidate alloys, and genetic algorithms are combined to iteratively optimize the alloy composition and additive manufacturing process parameters, ultimately obtaining the preferred composition of Ti-23Nb-0.7Ta-2Zr.

Images