An optimization of the laser powder bed melting process and preparation method for refractory titanium-zirconium alloy

By optimizing the laser powder bed melting process of refractory titanium-zirconium alloy through transfer learning and Gaussian process active learning, the problems of narrow process window and difficult model transfer were solved, achieving efficient and accurate process parameter optimization and performance prediction, and producing high-strength and high-plasticity refractory titanium-zirconium alloy components.

CN123120144APending Publication Date: 2026-11-17SHANGHAI JIAOTONG UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611339295.4
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-09-01
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing technologies in the laser powder bed melting process of refractory titanium-zirconium alloys have a narrow process window. Improper energy input can easily lead to porosity, lack of fusion, evaporation, instability of the molten pool, and cracking. Furthermore, machine learning models have difficulty effectively transferring the process sensitivity of different alloy systems, resulting in high experimental costs and large performance prediction errors.

Method used

By integrating prior knowledge of source alloys through transfer learning and combining active learning of Gaussian processes, the laser powder bed melting process is optimized. A source domain model is constructed using various existing LPBF alloy data, and scale consistency processing and target domain calibration are performed. High-performance or high-uncertainty candidate samples are selected for tensile testing to reduce false data contamination.

Benefits of technology

This study achieved efficient small-sample optimization of laser powder bed melting process parameters for refractory titanium-zirconium alloys, reducing experimental costs, improving the accuracy of performance prediction and process feasibility, and obtaining high-strength and high-ductility refractory titanium-zirconium alloy components.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123120144A_ABST
    Figure CN123120144A_ABST
Patent Text Reader

Abstract

This invention specifically relates to an optimization and preparation method for laser powder bed melting of refractory titanium-zirconium alloys. The optimization method includes: collecting various laser powder bed melting alloy process-performance data to construct a source domain dataset; preprocessing and training / freezing a source domain prediction model to generate source prior features; selecting an initial batch for forming and tensile testing to obtain the measured strength-ductility product; inputting the target process descriptor and source prior features into the target domain calibration model to obtain the transferred prior strength-ductility product; using the difference between the strength-ductility product and the transferred prior strength-ductility product as the residual, and selecting the next set of process parameters from candidate process conditions; preparing samples and determining formability; repeating active selection and verification, and outputting the formable process with the highest strength-ductility product as the optimal parameter after meeting the stopping condition. This invention achieves efficient small-sample optimization of the optimal process parameters for laser powder bed melting of refractory titanium-zirconium alloys by iterative optimization through transfer learning, integrating source alloy prior knowledge, and combining Gaussian process active learning.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the fields of intelligent material design, metal additive manufacturing, and preparation of refractory titanium-zirconium alloys, and more specifically to an optimization method for laser powder bed melting process of refractory titanium-zirconium alloys. Background Technology

[0002] Refractory titanium-zirconium alloys typically contain multiple high-melting-point elements, exhibiting high strength and a wide compositional design space. Laser powder bed melting (LPBF) can achieve near-net-shape forming of complex components, but during the in-situ alloying of elemental powders, powder melting, refractory particle dissolution, molten pool convection mixing, and rapid solidification occur simultaneously, resulting in a narrow process window. Insufficient energy input can easily lead to unfused porosity, unmelted refractory particles, and chemical inhomogeneity at the powder scale; excessive energy input may result in evaporation, molten pool instability, residual stress, and cracking.

[0003] Traditional process optimization typically requires fabrication and testing at numerous combinations of laser power and scanning speed, resulting in high experimental costs. While conventional machine learning can establish process-performance mappings, room temperature tensile data for refractory titanium-zirconium alloys is limited, and the absolute strength and elongation scales of different alloy systems differ significantly. Directly mixing the absolute properties of different alloys for modeling can easily lead the model to learn alloy system differences rather than transferable process sensitivities.

[0004] Furthermore, LPBF candidate conditions may include those that result in significant cracking or prevent the formation of effective tensile specimens. Directly assigning these values ​​to zero strength and zero elongation for regression would incorrectly interpret "no physical label" as "performance value is zero"; completely removing them would result in the loss of process feasibility boundaries. Therefore, a process optimization method is needed that can utilize cross-alloy information, continuously calibrate through real-world target experiments, and correctly handle forming failures. Summary of the Invention

[0005] The purpose of this invention is to provide an optimization method for the laser powder bed melting process of refractory titanium-zirconium alloys and a preparation method thereof. By integrating prior knowledge of the source alloy through transfer learning and combining iterative optimization with active learning of Gaussian process, the optimal process parameters for laser powder bed melting of refractory titanium-zirconium alloys can be efficiently optimized with a small sample size.

[0006] To achieve the above objectives, the present invention provides the following technical solution: An optimization method for laser powder bed melting process of refractory titanium-zirconium alloy includes the following steps: S1. Collect various laser powder bed melting alloy process-performance data, construct a source domain dataset, and after performing scale consistency processing on the mechanical property response of each source alloy in the source domain dataset, input it into the trained source domain prediction model to generate source prior features of the target alloy candidate process conditions. S2. Complete the laser powder bed melting and forming of the target alloy sample according to the preset candidate process conditions and determine its formability. Then, form the formable samples into a tensile test candidate pool. An initial batch is selected from the tensile test candidate pool for tensile testing to obtain the measured mechanical properties and strength-ductility product; the target process descriptor corresponding to the candidate sample and the source prior features are jointly input into the target domain calibration model to obtain the transfer prior strength-ductility product. S3. Using the difference between the measured strong plastic product and the prior strong plastic product as the residual, establish a Gaussian process model; Using the established Gaussian process model, the next candidate sample to be tested for tensile testing is selected from the tensile test candidate pool; S4. Perform tensile tests on the selected candidate samples, write the measured mechanical properties and strength-ductility product into the target domain data, and update the target domain data. S5. Based on the updated target domain data, repeat the active selection and tensile test until the preset stop condition is met, and output the process conditions corresponding to the candidate sample with the highest measured strength-ductility product as the optimal process parameters.

[0007] Furthermore, in S1, the process of performing scale consistency processing on the mechanical property response of each source alloy is specifically as follows: The mechanical property response of each source alloy is aligned to the internal alloy scale, and the expression is: ; in y ( i,j ) as source alloy i In the j Mechanical property response under various process conditions and s ( i ( ) represent the average value and standard deviation of the response in the same source alloy, respectively; The process data in the source domain dataset includes source domain descriptors representing laser power P, scanning speed v, channel spacing h, layer thickness t, volumetric energy density, and line energy density; where volumetric energy density is categorized by... log 10 [ P / ( vht Characterization; linear energy density is characterized by P / v.

[0008] Furthermore, in S1, the source domain prediction model is specifically a tree ensemble regression model; In S2, the target domain calibration model is specifically a regularized regression model.

[0009] Furthermore, in S3, the Gaussian process model adopts... Covariance kernel; The process of selecting the next sample to be tested from the tensile test candidate pool is as follows: Using the predicted mean, predicted standard deviation, and the current highest measured formable strength-plasticity product of the Gaussian process model, the expected improvement value of each candidate sample that has not yet undergone tensile testing is calculated by the expected improvement acquisition function, and the candidate sample with the largest expected improvement value is selected as the next sample to be tested.

[0010] Furthermore, in S3, the candidate process conditions include: candidate laser power, candidate scanning speed, fixed channel spacing, fixed layer thickness, and scanning strategy; Among them, candidate laser power P 150-450W, candidate scan speed v 300-1200mm / s, fixed track spacing h The fixed layer thickness is 0.03mm. t The thickness is 0.04 mm, and the scanning strategy is linear fill scanning with the scanning vectors of adjacent layers rotated by 67°.

[0011] Furthermore, in S4, the functional expression for the strong plastic product is: U ( x ) =UTS ( x ) ×Af ( x ); in, UTS ( x ( ) are candidate laser power and scanning speed conditions x The tensile strength below Af ( x () represents elongation after fracture as a percentage.

[0012] Furthermore, in S5, the preset stopping condition is specifically as follows: After obtaining the highest measured strength-ductility product for the first time, at least two confirmatory tensile tests are performed, selected by the desired improved acquisition function. If neither of the confirmatory tensile tests exceeds the highest measured strength-ductility product, the preset stop condition is determined to be met.

[0013] This invention also provides a method for preparing refractory titanium-zirconium alloy components, comprising the following steps: (1) Mix Zr, Ti, Nb, Ta and Mo element powders to form a mixed powder with a nominal composition of 39.3% Zr, 19.5% Ti, 17.9% Nb, 16.8% Ta and 6.5% Mo by mass percentage; (2) The optimal process parameters of the mixed powder are determined by using the laser powder bed melting process optimization method for refractory titanium-zirconium alloy; (3) Under the protection of inert gas, the mixed powder is alloyed in situ using the optimal process parameters; wherein, the optimal process parameters include: laser power of 250W, scanning speed of 900mm / s, channel spacing of 0.03mm, layer thickness of 0.04mm and a linear filling strategy with adjacent layer scanning vectors rotated by 67°, to obtain refractory titanium-zirconium alloy components.

[0014] Furthermore, the refractory titanium-zirconium alloy component has a continuous matrix with a BCC structure as its main component, and the continuous matrix has a chemically modulated cellular structure with an average equivalent diameter of 0.30-0.40 μm; The refractory titanium-zirconium alloy component has a relative density of not less than 99.0%, a room temperature yield strength of not less than 800 MPa, a room temperature tensile strength of not less than 1000 MPa, and an elongation after fracture of not less than 5.0%.

[0015] According to specific embodiments provided by the present invention, the present invention has the following technical effects compared to the prior art: This invention addresses the challenges of limited tensile property data for refractory titanium-zirconium alloys and the high cost of conducting tensile tests on all formable samples individually by combining source domain knowledge transfer and active learning. It constructs a source domain model using data from various existing LPBF alloys, performs scale consistency processing and target domain calibration, and transfers the "process-performance" rules of mature alloys to new alloys, allowing performance priors to be obtained with only a few initial tensile tests. A residual Gaussian process model is then established based on the difference between measured values ​​and the transferred priors, and an expected improvement acquisition function is used to select the next sample to be tested. This modeling focuses on the specific biases of the target alloy, prioritizing potentially high-performance or high-uncertainty candidate samples in each round of tensile testing. Simultaneously, samples pre-determined as unformable are only marked with a failure flag and no performance label is generated, avoiding model contamination by false data. Given that candidate samples have already undergone LPBF forming, this invention reduces the number of tensile tests and performance characterization costs required to identify optimal process conditions, providing a feasible approach for process screening of high-strength, high-ductility refractory titanium-zirconium alloy components. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0017] The following description, in conjunction with the accompanying drawings, further illustrates the optimization of the laser powder bed melting process and the preparation method for refractory titanium-zirconium alloy according to the present invention. Figure 1This is a flowchart illustrating the overall process optimization of LPBF based on transfer learning and active learning in this invention. Figure 2 This is a comparison of the tensile property prediction results of the target model without source prior and the transfer learning model under small sample conditions of the target alloy in this invention; where (a) is a schematic diagram of the comparison between the predicted and measured values ​​of yield strength, (b) is a schematic diagram of the comparison between the predicted and measured values ​​of tensile strength, (c) is a schematic diagram of the comparison between the predicted and measured values ​​of elongation after fracture, (d) is a schematic diagram of the learning curve of the yield strength model, (e) is a schematic diagram of the learning curve of the tensile strength model, and (f) is a schematic diagram of the learning curve of the elongation after fracture model. Figure 3 The diagram shows the active tensile test selection path and the number of tensile evaluations required to achieve the target in the candidate Pv space of this invention; where (a) is a schematic diagram of the active learning adaptive point selection process in the laser power-scanning speed parameter space, and (b) is a schematic diagram comparing the number of tensile sample evaluations required to achieve the target performance under different sampling strategies. Figure 4 The tensile response and work hardening diagrams of representative specimens in this invention are shown. Figure 5 This is a multi-scale cellular structure characterization diagram of the sample in this invention. Detailed Implementation

[0018] The specific embodiments of the present invention will be described in further detail below with reference to the accompanying drawings and examples. The following examples are for illustrative purposes only and are not intended to limit the scope of the invention.

[0019] To better understand the purpose, structure, and function of this invention, the invention will be described in further detail below with reference to the accompanying drawings.

[0020] Example 1 like Figure 1 As shown, this invention provides an optimization method for the laser powder bed melting process of refractory titanium-zirconium alloys, comprising the following steps: S1. Collect various laser powder bed melting alloy process-performance data, construct a source domain dataset, and after performing scale consistency processing on the mechanical property response of each source alloy in the source domain dataset, input it into the trained source domain prediction model to generate source prior features of the target alloy candidate process conditions. In S1, the mechanical property response of each source alloy is subjected to scale consistency processing, specifically as follows: The mechanical property response of each source alloy is aligned to the internal alloy scale, and the expression is: ; in y ( i,j ) as source alloy i In thej Mechanical property response under various process conditions and s ( i ( ) represent the average value and standard deviation of the response in the same source alloy, respectively; The process data in the source domain dataset are characterized using source domain descriptors including laser power P, scanning speed v, pass spacing h, layer thickness t, volumetric energy density, and linear energy density; where volumetric energy density is defined as follows: log 10 [ P / ( vht Characterization; linear energy density is characterized by P / v.

[0021] In S1, the source domain prediction model is specifically a tree ensemble regression model. The output of the source domain prediction model is used only as an additional input feature of the target domain calibration model.

[0022] This embodiment specifically involves: constructing a source domain and performing transfer learning. Data on LPBF alloys with clearly defined relationships between process parameters and tensile properties are collected. The response of each source alloy is standardized within the alloy, enabling the source domain prediction model to learn relative process variations within the same alloy. The source domain model is trained using descriptors such as P, v, h, t, volumetric energy density, and linear energy density. After training, the source domain prediction model is frozen, and prior features of the target candidate conditions are generated.

[0023] S2. Complete the laser powder bed melting and forming of the target alloy sample according to the preset candidate process conditions and determine its formability. Then, form the formable samples into a tensile test candidate pool. An initial batch is selected from the tensile test candidate pool for tensile testing to obtain the measured mechanical properties and strength-ductility product; the target process descriptor corresponding to the candidate sample and the source prior features are jointly input into the target domain calibration model to obtain the transfer prior strength-ductility product. In S2, the target domain calibration model is specifically a regularized regression model; Specifically, this embodiment involves: establishing a target domain calibration model. Under the conditions of fixed target alloy composition, channel spacing, layer thickness, and scanning strategy, 28 sets of samples were LPBF formed according to a preset Pv grid, and their formability was determined. From the formable samples, the initial 10 sets covering the Pv space were selected for room temperature tensile testing, and the target process descriptor and source prior features were jointly input into the target domain model.

[0024] S3. Using the difference between the measured strength-plastic product and the prior strength-plastic product as the residual, establish a Gaussian process model, and use the expected improved acquisition function to select the next candidate sample to be tested for tensile test from the tensile test candidate pool. In S3, the Gaussian process model is adopted. The covariance kernel is used to calculate the expected improvement value of the expected improvement acquisition function using the predicted mean, predicted standard deviation, and the highest measured strength-plasticity product in the current formable conditions of the Gaussian process model.

[0025] In S3, the candidate process conditions for the target alloy include: candidate laser power, candidate scanning speed, fixed channel spacing, fixed layer thickness, and scanning strategy. Among them, candidate laser power P 150-450W, candidate scan speed v 300-1200mm / s, fixed track spacing h The fixed layer thickness is 0.03mm. t The thickness is 0.04 mm, and the scanning strategy is linear fill scanning with the scanning vectors of adjacent layers rotated by 67°.

[0026] In this specific embodiment, the product of tensile strength and elongation after fracture is used as the strength-ductility product. A residual Gaussian process model of the measured strength-ductility product relative to the transfer prior is established. Based on the expected improvement value, the next test sample is selected from the formable samples that have not yet undergone tensile testing.

[0027] S4. Perform tensile tests on the selected candidate samples, write the measured mechanical properties and strength-ductility product into the target domain data, and update the target domain data; among them, samples that are pre-determined to be unformable only retain a failure mark, do not generate mechanical property numerical labels, and are not included in the tensile test candidate pool. In S4, the functional expression for the strong plastic product is: U ( x ) =UTS ( x ) ×Af ( x ); in, UTS ( x ( ) are candidate laser power and scanning speed conditions x The tensile strength below Af ( x () represents elongation after fracture as a percentage.

[0028] In this specific embodiment, the selected candidate samples are subjected to room temperature tensile tests, and the actual performance and strength-ductility product are written into the target domain data. Samples that are pre-determined to be unformable are only recorded as failures and are not included in the tensile test candidate pool, nor are continuous mechanical property labels created.

[0029] S5. Based on the updated target domain data, repeat the active selection and tensile test until the preset stop condition is met, and output the process conditions corresponding to the candidate sample with the highest measured strength-ductility product as the optimal process parameters.

[0030] In step S5, the preset stopping condition is as follows: after obtaining the highest measured strength-ductility volume for the first time, at least two confirmatory tensile tests selected by the desired improved acquisition function are performed. When the confirmatory tensile tests do not exceed the highest measured strength-ductility volume, the preset stopping condition is determined to be met.

[0031] Specifically, in this embodiment, when the continuous confirmatory tensile test does not exceed the current highest measured strength-ductility product, or reaches the preset maximum number of tensile evaluations, the formable process parameters corresponding to the current highest measured strength-ductility product are output.

[0032] The target alloy is Zr39.3Ti19.5Nb17.9Ta16.8Mo6.5 (wt.%), with candidate laser power of 150-450W, scanning speed of 300-1200 mm / s, channel spacing of 0.03 mm, layer thickness of 0.04 mm, and adjacent layer scanning vector rotation of 67°. The preferred output parameters are 250 W and 900 mm / s.

[0033] Compared with the prior art, the present invention also has the following beneficial effects: (1) By aligning the internal scale of the alloy and migrating the frozen source features, the existing LPBF data is used to provide relative process priors, thereby reducing the negative migration caused by the difference in absolute performance scale of different alloys.

[0034] (2) The transfer prior and the actual tensile test of the target alloy are used together for residual modeling, and the next sample to be tested is selected by the prediction uncertainty, reducing the number of tensile tests required to identify high strength-plasticity conditions.

[0035] (3) For non-formable conditions, failure information is retained but false mechanical labels are not generated, thus preserving the process feasibility boundary and avoiding distortion of continuous performance regression.

[0036] (4) The highest measured strength-plasticity volume was obtained under the conditions corresponding to the 13th overall evaluation. After all subsequent tensile tests were completed, it was still the highest value among the 22 groups of formable conditions. Its process parameters were 250W and 900mm / s. The relative density of the resulting component was 99.44%, and it also had a yield strength of 852 MPa, a tensile strength of 1.06 GPa and an elongation after fracture of 5.08%.

[0037] Example 2 The present invention also provides a specific embodiment of preparing the target alloy powder with LPBF using the method in Example 1: 1) Establishment of transfer learning model Source domain data were selected from LPBF records with clearly defined alloy identity, laser power P, scanning speed v, pass spacing h, layer thickness t, and corresponding room temperature tensile properties, requiring multiple processing conditions for the same source alloy. For the j-th condition of source alloy i, yield strength, tensile strength, and elongation after fracture were respectively calculated according to... Standardization is performed. The source domain inputs include P, v, h, t, log10[P / (vht)], and P / v; XGBoost tree ensemble regression models are trained for the three stretch responses respectively, and the model parameters are frozen after training.

[0038] The training process of the XGBoost tree ensemble regression model is as follows: First, target alloy records are excluded, and data lacking laser power P, scanning speed v, trace spacing h, layer thickness t, or corresponding mechanical properties, as well as duplicate records, are removed; only records of the same source alloy with at least two valid process conditions and corresponding performance standard deviations that are non-zero are retained. Subsequently, yield strength, tensile strength, and elongation after fracture are standardized within each source alloy, expressed as P, v, h, t, log... 10 [P / (vht)] and P / v are used as inputs to train XGBoost regression models.

[0039] Each model has 300 decision trees, a maximum tree depth of 3, a learning rate of 0.03, a sample subsampling ratio and a feature subsampling ratio of 0.85, an L2 regularization coefficient of 2.0, a loss function of squared error, and a random seed of 42.

[0040] After training, the tree structure, splitting threshold, and leaf node weights of each source domain model are fixed. During the target domain calibration and active selection process, the source domain models are no longer refitted or updated using target alloy data. Instead, the candidate process conditions of the target alloy are input into the frozen source domain models for prediction, and the source prior features of yield strength, tensile strength, and elongation after fracture are obtained respectively. These source prior features are then used as additional inputs to the target domain regularized regression model.

[0041] Twenty-eight candidate samples of the target alloy have undergone LPBF forming according to a pre-defined Pv grid, and their formability has been determined. Ten initial samples covering the candidate Pv space were selected from the formable samples for tensile testing. The target model uses a basic process descriptor consistent with the source domain, and adds corresponding source prior features. To evaluate the transfer contribution, a benchmark model using the same target tensile test data and the same regression structure, but without source prior features, is set up. Figure 2 As shown, on the same fixed evaluation set, transfer learning increases the determination coefficients of yield strength, tensile strength and elongation after fracture from 0.600 to 0.712, from 0.634 to 0.655 and from -0.131 to 0.273, respectively, while the corresponding mean absolute error decreases.

[0042] 2) Selection of active tensile test and handling of non-formable samples The candidate program for the target alloy includes 28 Pv conditions, of which 22 form complete samples and constitute the tensile test candidate pool. The strength-ductility product is defined as U = UTS × Af. The UTS and Af source prior features output from the frozen source model are input together with the initial target tensile test data into the target domain calibration model to obtain the predicted UTS and Af values ​​at the target scale. These two values ​​are then combined into the transferred prior strength-ductility product. A Gaussian process model is then fitted using the difference between the measured U and the transferred prior U. The acquisition function employs expectation improvement, balancing the predicted mean and uncertainty, to select the next test sample from the formable samples that have not yet undergone tensile testing.

[0043] In this embodiment, candidate conditions are uniformly identified according to the overall evaluation order. After the same initial 10 sets of tensile tests, the acquisition function sequentially determines the test conditions for the 11th to 15th overall evaluations, corresponding to process parameters of 250 W / 600 mm / s, 150 W / 300 mm / s, 250 W / 900 mm / s, 350 W / 900 mm / s, and 450 W / 900 mm / s, respectively. Figure 3 As shown, the measured strength-ductility product obtained in the 13th overall evaluation was the highest value at that time; the measured strength-ductility products obtained in the subsequent 14th and 15th overall evaluations did not exceed this value, and after all tensile tests were completed, the corresponding condition for the 13th overall evaluation was still the highest value among the 22 groups of formable conditions. Therefore, the 15th overall evaluation can be used as a stopping point to confirm support. Compared to completing tensile tests on all 22 groups of formable samples before identifying the optimal value, the number of tensile tests required for the initial identification was reduced from 22 groups to 13 groups, a reduction of 40.9%. For the 6 groups of samples that cracked, did not form complete specimens, or could not obtain effective tensile data, they were only marked as infeasible and were not included in the tensile test candidate pool, nor were zero strength or zero elongation written into the mechanical property regression model.

[0044] 3) Preparation of target alloy powder and LPBF In-situ alloying was performed using commercial elemental powders of Zr, Ti, Nb, Ta, and Mo. A mixed powder of Zr 39.3%, Ti 19.5%, Nb 17.9%, Ta 16.8%, and Mo 6.5% was prepared by mass percentage and mixed under closed conditions to reduce segregation and oxidation. The D90 of Nb and Ta powders is close to 40 μm, which is comparable to a 40 μm layer thickness. Therefore, the process parameters need to simultaneously satisfy the requirements of refractory particle dissolution, melt channel overlap, and melt pool stability.

[0045] The candidate laser power was 150-450 W, the scanning speed was 300-1200 mm / s, the channel spacing was fixed at 30 μm, and the layer thickness was fixed at 40 μm. Linear filling was used, and the scanning vector was rotated 67° layer by layer. Based on transfer learning and active tensile testing, the process parameters corresponding to the 13th overall evaluation were 250 W and 900 mm / s, with a nominal volumetric energy density of 231.5 J / mm². 3 .

[0046] This embodiment also provides the process for organization and performance verification: The components obtained under the conditions corresponding to the 13th overall evaluation had a relative density of 99.44%, a room temperature yield strength of 852 MPa, a tensile strength of 1.06 GPa, an elongation after fracture of 5.08%, and an elastic modulus of 99.7 GPa. Low-energy samples exhibited unfused porosity and unmelted particles, while high-energy samples maintained high strength but had lower plasticity; this condition achieved a balance between defect suppression, element mixing, and tensile properties. Figure 4 Representative tensile curves and work hardening behaviors are presented.

[0047] The component obtained under the conditions of the 13th overall evaluation is a BCC structure, which forms a chemically modulated cellular structure with an average equivalent diameter of about 0.35 μm in a continuous matrix. Figure 5 The HAADF-STEM and EDS results show that Nb and Ta have relatively high strength in the specified intracellular regions, while Ti, Zr, and Mo exhibit complementary changes in adjacent cell boundary regions. Dislocation storage, nanotwins, and twin-dislocation interactions can be observed near the cell boundaries after deformation. These results indicate that the selected process not only improves density but also preserves the refined cellular structure that is conducive to dislocation storage and subsequent plastic deformation.

[0048] Example 3 This invention also provides a method for preparing refractory titanium-zirconium alloy components, comprising the following steps: (1) Mix Zr, Ti, Nb, Ta and Mo element powders to form a mixed powder with a nominal composition of 39.3% Zr, 19.5% Ti, 17.9% Nb, 16.8% Ta and 6.5% Mo by mass percentage; (2) Using the laser powder bed melting process optimization method for refractory titanium-zirconium alloy in Example 1, the optimal process parameters of the mixed powder were determined; (3) Under the protection of inert gas, the mixed powder is alloyed in situ using the optimal process parameters; wherein, the optimal process parameters include: laser power of 250W, scanning speed of 900mm / s, channel spacing of 0.03mm, layer thickness of 0.04mm and a linear filling strategy with adjacent layer scanning vectors rotated by 67°, to obtain refractory titanium-zirconium alloy components.

[0049] The refractory titanium-zirconium alloy component has a continuous matrix with a BCC structure as its main component, and the continuous matrix has a chemically modulated cellular structure with an average equivalent diameter of 0.30-0.40 μm. The refractory titanium-zirconium alloy component has a relative density of not less than 99.0%, a room temperature yield strength of not less than 800 MPa, a room temperature tensile strength of not less than 1000 MPa, and an elongation after fracture of not less than 5.0%.

[0050] The above description of the disclosed embodiments enables those skilled in the art to make or use the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the present invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for optimizing the laser powder bed melting process of refractory titanium-zirconium alloy, characterized in that, Includes the following steps: S1. Collect various laser powder bed melting alloy process-performance data, construct a source domain dataset, and after performing scale consistency processing on the mechanical property response of each source alloy in the source domain dataset, input it into the trained source domain prediction model to generate source prior features of the target alloy candidate process conditions. S2. Complete the laser powder bed melting and forming of the target alloy sample according to the preset candidate process conditions and determine its formability. Then, form the formable samples into a tensile test candidate pool. An initial batch is selected from the tensile test candidate pool for tensile testing to obtain the measured mechanical properties and strength-ductility product; the target process descriptor corresponding to the candidate sample and the source prior features are jointly input into the target domain calibration model to obtain the transfer prior strength-ductility product. S3. Using the difference between the measured strong plastic product and the prior strong plastic product as the residual, establish a Gaussian process model; Using the established Gaussian process model, the next candidate sample to be tested for tensile testing is selected from the tensile test candidate pool; S4. Perform tensile tests on the selected candidate samples, write the measured mechanical properties and strength-ductility product into the target domain data, and update the target domain data. S5. Based on the updated target domain data, repeat the active selection and tensile test until the preset stop condition is met, and output the process conditions corresponding to the candidate sample with the highest measured strength-ductility product as the optimal process parameters.

2. The method for optimizing the laser powder bed melting process of refractory titanium-zirconium alloy according to claim 1, characterized in that, In S1, the process of performing scale consistency processing on the mechanical property response of each source alloy is as follows: The mechanical property response of each source alloy is aligned to the internal alloy scale, and the expression is: ; in y ( i,j ) as source alloy i In the j Mechanical property response under various process conditions and s ( i ( ) represent the average value and standard deviation of the response in the same source alloy, respectively; The process data in the source domain dataset includes source domain descriptor representations of laser power P, scanning speed v, channel spacing h, layer thickness t, volumetric energy density, and line energy density. Where volume energy density is according to log 10 [ P / ( vht Characterization; linear energy density is characterized by P / v.

3. The method for optimizing the laser powder bed melting process of refractory titanium-zirconium alloy according to claim 1, characterized in that, In S1, the source domain prediction model is specifically a tree ensemble regression model. In S2, the target domain calibration model is specifically a regularized regression model.

4. The method for optimizing the laser powder bed melting process of refractory titanium-zirconium alloy according to claim 1, characterized in that, In S3, the Gaussian process model is adopted. Covariance kernel; The process of selecting the next sample to be tested from the tensile test candidate pool is as follows: Using the predicted mean, predicted standard deviation, and the current highest measured formable strength-plasticity product of the Gaussian process model, the expected improvement value of each candidate sample that has not yet undergone tensile testing is calculated by the expected improvement acquisition function, and the candidate sample with the largest expected improvement value is selected as the next sample to be tested.

5. The method for optimizing the laser powder bed melting process of refractory titanium-zirconium alloy according to claim 1, characterized in that, In S3, the candidate process conditions include: candidate laser power, candidate scanning speed, fixed channel spacing, fixed layer thickness, and scanning strategy. Among them, candidate laser power P 150-450W, candidate scan speed v 300-1200mm / s, fixed track spacing h The fixed layer thickness is 0.03mm. t The thickness is 0.04 mm, and the scanning strategy is linear fill scanning with the scanning vectors of adjacent layers rotated by 67°.

6. The method for optimizing the laser powder bed melting process of refractory titanium-zirconium alloy according to claim 1, characterized in that, In S4, the functional expression for the strong plastic product is: U ( x ) =UTS ( x ) ×Af ( x ); in, UTS ( x ( ) are candidate laser power and scanning speed conditions x The tensile strength below Af ( x () represents elongation after fracture as a percentage.

7. The method for optimizing the laser powder bed melting process of refractory titanium-zirconium alloy according to claim 1, characterized in that, In S5, the preset stopping condition is specifically as follows: After obtaining the highest measured strength-ductility product for the first time, at least two confirmatory tensile tests are performed, selected by the desired improved acquisition function. If neither of the confirmatory tensile tests exceeds the highest measured strength-ductility product, the preset stop condition is determined to be met.

8. A method for preparing a refractory titanium-zirconium alloy component, characterized in that, Includes the following steps: (1) Mix Zr, Ti, Nb, Ta and Mo element powders to form a mixed powder with a nominal composition of 39.3% Zr, 19.5% Ti, 17.9% Nb, 16.8% Ta and 6.5% Mo by mass percentage; (2) Using the laser powder bed melting process optimization method for refractory titanium-zirconium alloy according to any one of claims 1-7, determine the optimal process parameters of the mixed powder; (3) Under the protection of inert gas, the mixed powder is alloyed in situ using the optimal process parameters; wherein, the optimal process parameters include: laser power of 250W, scanning speed of 900mm / s, channel spacing of 0.03mm, layer thickness of 0.04mm and a linear filling strategy with adjacent layer scanning vectors rotated by 67°, to obtain refractory titanium-zirconium alloy components.

9. The method for preparing refractory titanium-zirconium alloy components according to claim 8, characterized in that, The refractory titanium-zirconium alloy component has a continuous matrix with a BCC structure as its main component, and the continuous matrix has a chemically modulated cellular structure with an average equivalent diameter of 0.30-0.40 μm. The refractory titanium-zirconium alloy component has a relative density of not less than 99.0%, a room temperature yield strength of not less than 800 MPa, a room temperature tensile strength of not less than 1000 MPa, and an elongation after fracture of not less than 5.0%.