Laser powder bed melting method for preparing high-toughness TC4 titanium alloy based on machine learning

By combining K-means clustering and a hybrid expert model framework with Bayesian optimization and NSGA-II genetic algorithm, the problem of synergistic optimization of high strength and high toughness of TC4 titanium alloy was solved, and the efficient preparation of high-strength and high-toughness TC4 titanium alloy with excellent comprehensive mechanical properties was achieved.

CN121373457APending Publication Date: 2026-01-23XI AN JIAOTONG UNIV

Patent Information

Application Number
CN202511466501.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2025-10-14
Publication Date
2026-01-23

AI Technical Summary

Technical Problem

Existing technologies struggle to achieve synergistic optimization of high strength and high toughness in TC4 titanium alloys across a broad parameter space. Traditional methods are inefficient and costly, while conventional machine learning models lack sufficient prediction accuracy for unknown parameter combinations.

Method used

K-means clustering algorithm was used to classify the performance data of TC4 titanium alloy. A hybrid expert machine learning prediction model was constructed. Combining large step size network search and Bayesian optimization method, the SLM printing parameters of high strength and toughness TC4 alloy were reversed using NSGA-II genetic algorithm. Intelligent design of process parameters and performance prediction were realized through the hybrid expert model framework.

Benefits of technology

It significantly improves the model's generalization ability and prediction accuracy under sparse, multi-source process-performance data, and realizes the efficient preparation of high-strength and high-toughness TC4 titanium alloy with a yield strength exceeding 1050 MPa and an elongation after fracture remaining above 14%, which is superior to conventional SLM forming of TC4 alloy.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN121373457A_ABST
    Figure CN121373457A_ABST
Patent Text Reader

Abstract

The invention provides a laser powder bed melting method for preparing a high-toughness TC4 titanium alloy based on machine learning. The method comprises the steps that performance data of the TC4 titanium alloy prepared through different SLM process parameters are collected; constructing a hybrid expert (MoE) machine learning prediction model framework; classifying the TC4 data by adopting a K-means clustering algorithm, and taking a clustering center as an initial weight of an input layer and a first layer of the gating network; presetting a TC4 performance prediction model through the gating network and the plurality of expert networks after clustering information initialization; performing small-range iteration through large-step network search in combination with a Bayesian optimization method so as to determine an optimal model structure and complete model training; the trained performance prediction model and a genetic algorithm are used for carrying out backstepping to obtain SLM printing parameters of the high-toughness TC4 alloy; and SLM forming is carried out, and the high-toughness TC4 titanium alloy is prepared. According to the method, a calculation framework which is high in precision and suitable for exploring a high-unknown characteristic space is constructed, and far-reaching influences on titanium alloy additive manufacturing and process optimization of other alloy systems are achieved.
Need to check novelty before this filing date? Find Prior Art

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of titanium alloy additive manufacturing, in particular to a laser powder bed fusion method for preparing high-strength and high-toughness TC4 titanium alloy based on machine learning, in particular to a related method for realizing intelligent design of process parameters and performance prediction by combining an expert model and an optimization algorithm. BACKGROUND

[0002] TC4 (Ti-6Al-4V) titanium alloy has become one of the preferred materials for key structural components in the fields of aerospace, national defense and military industry due to its excellent specific strength, corrosion resistance and comprehensive mechanical properties. Laser powder bed fusion (SLM) technology makes it possible to form complex components. However, the combination of SLM process parameters directly affects the mechanical properties of the final part, and high strength and high toughness are usually mutually restrictive. This “strength-toughness inversion” problem is a major bottleneck in the preparation of high-performance TC4 alloy. The traditional “trial and error” experimental method is inefficient and costly, and it is difficult to realize the synergistic optimization of strength and toughness in a wide parameter space.

[0003] In recent years, machine learning methods have provided a new way for SLM process parameter optimization, which can build a process-performance mapping relationship through historical data to accelerate material development. For example, CN119388749A discloses a method, system and method for improving the forming precision of polymer laser powder bed fusion, wherein the method for controlling the forming precision of polymer laser powder bed fusion comprises the following steps: S1: obtaining temperature distribution data in the polymer LPBF process; S2: based on the temperature distribution data and the process parameters used for printing the current layer, a database is constructed; S3: based on machine learning, an accurate control model for the polymer LPBF process is established; S4: the database is used to train the accurate control model for the polymer LPBF process, a mapping relationship between the process parameters, the printing layer thickness and the temperature field distribution of each layer is established, a negative feedback mechanism of the process parameters and the real-time temperature field of the printing layer is obtained, and the accurate control of the forming precision of the polymer laser powder bed fusion is completed.

[0004] However, in practical applications, the process-performance data from different literatures often have problems such as sparse feature space distribution and significant source difference, which leads to overfitting of conventional machine learning models and poor generalization ability, especially insufficient prediction accuracy for unknown parameter combinations. Although some studies have introduced multi-model strategies to alleviate overfitting by constraining parameter ranges, this strategy often limits the exploration ability of the model, making it difficult to explore the optimal process parameter combination that can achieve high strength and high toughness in a high-unknown feature space. SUMMARY

[0005] Aiming at the problems in the prior art, the application provides a laser powder bed fusion method for preparing high-toughness TC4 titanium alloy based on machine learning.

[0006] The application is realized by the following technical solutions: The laser powder bed fusion method for preparing high-toughness TC4 titanium alloy based on machine learning comprises the following steps: Collect performance data of TC4 titanium alloy prepared by different SLM process parameters from published literature; Classify the performance data of the TC4 titanium alloy by using a K-means clustering algorithm to obtain clustering centers of each "feature category"; Use the clustering centers as initial weights of an input layer and a first layer of a gated network to construct a hybrid expert machine learning prediction model; For the hybrid expert machine learning prediction model, a large-step network search is combined with a Bayesian optimization method for small-range iteration to determine an optimal performance prediction model; Use the optimal performance prediction model and an NSGA-II genetic algorithm to obtain SLM printing parameters of high-toughness TC4 alloy by backstepping; According to the SLM printing parameters, SLM forming is implemented to prepare high-toughness TC4 titanium alloy.

[0007] Preferably, the K-means clustering algorithm is used to classify the performance data of the TC4 titanium alloy to obtain clustering centers of each "feature category", which comprises the following steps: Standardize the feature data, calculate the global mean and standard deviation:

[0008] wherein, is an original feature vector, is a global mean vector of the data, is a standard deviation vector; Then, K-means clustering is performed, and the center is:

[0009] wherein represents the center of the kth cluster, is a set of all samples belonging to the cluster, and is the number of samples in the cluster, is a feature vector of the ith sample in the cluster.

[0010] Preferably, the hybrid expert model framework comprises a gated network, a plurality of expert networks and a load balancing loss; the gated network comprises a hidden layer and an output layer, and the first layer weight is initialized by the clustering center: The expert network is a residual network structure, each expert network comprising an input layer, a plurality of residual blocks and an output layer; The gating network outputs the weights of each expert network and normalizes them through a softmax function; The load balancing loss is used to ensure balanced use of each expert network, and the expert usage rate is defined as , and the ideal uniform distribution is The load balancing loss is the KL divergence:

[0011] wherein is the actual usage rate of the i-th expert, is the ideal usage rate thereof.

[0012] Preferably, the large step network search combines a Bayesian optimization method, comprising: Determine the initial hyperparameter space through grid search; The top 10 best parameter combinations based on the Stage1 results are used as the initial parameter combinations for Bayesian optimization; Use the TPE sampler for Bayesian optimization to maximize the weighted R 2 score of the validation set in five-fold cross-validation to determine the optimal performance prediction model.

[0013] Preferably, the formula for maximizing the weighted R 2 is:

[0014] wherein is the true value of the i-th sample, is the predicted value thereof, is the mean of the true values, and N is the total number of samples.

[0015] Preferably, the optimal performance prediction model is used in conjunction with the NSGA-II genetic algorithm to back-calculate the SLM printing parameters for high-toughness TC4 alloy, comprising: Embed the optimal performance prediction model into the NSGA-II algorithm, and set the laser power, filling speed and filling distance as decision variables; Define a multi-objective optimization problem, and the objective functions are to maximize the elongation after fracture and yield strength:

[0016] With EF ≥ 0 and YS ≥ 0 as constraint conditions, use simulated binary crossover and polynomial mutation operators for iteration; NSGA-II genetic algorithm backstepping is performed to repeatedly update the process parameter space, and a Pareto frontier solution of the SLM printing parameters of the high-strength and high-toughness TC4 alloy is obtained.

[0017] Preferably, SLM forming is performed according to the SLM printing parameters of the high-strength and high-toughness TC4 alloy to prepare the high-strength and high-toughness TC4 titanium alloy, which comprises: According to the optimized process parameters, three groups of parameters are selected from the Pareto frontier solution using purity 99.5% TC4 powder for laser powder bed fusion forming; wherein, in the SLM forming process, high-purity argon gas is used for protection, and the oxygen content is controlled 100 ppm, and the forming chamber temperature is kept at 120 5 ℃.

[0018] Preferably, the printing parameters are set as follows: the laser power is 450 2 W, the filling speed is 1150 5 mm / s, the filling interval is 150 2 μm, the powder particle size is 37 2 μm, the laser spot diameter is 80 2 μm, the layer thickness is 30 2 μm, the gage area is 12 0.2 mm 2 , and the gage length is 30 0.2 mm; the yield strength of the first TC4 titanium alloy sample obtained after SLM forming in Example 1 is 1074 10 MPa; the ultimate tensile strength is 1220 10 MPa; and the elongation after fracture is 13.7 0.5 %; The printing parameters are set as follows: the laser power is 480 2 W, the filling speed is 1090 5 mm / s, the filling interval is 148 2 μm, the powder particle size is 37 2 μm, the laser spot diameter is 80 2 μm, the layer thickness is 30 2 μm, the gage area is 12 0.2 mm 2 , and the gage length is 30 0.2 mm; the yield strength of the second TC4 titanium alloy sample obtained after SLM forming is 1149 10 MPa; the ultimate tensile strength is 1268 10 MPa; and the elongation after fracture is 10.7 0.5 %; The printing parameters are set as follows: laser power is 455 2 W, filling speed is 1040 5 mm / s, filling interval is 152 2 μm, powder particle size is 37 2 μm, laser spot diameter is 80 2 μm, layer thickness is 30 2 μm, gage length is 30 0.2 mm 2 , gage length is 30 0.2 mm; the yield strength of the No. 3 TC4 titanium alloy sample obtained after SLM shaping is 1062 10 MPa; the ultimate tensile strength is 1286 10 MPa; the elongation after fracture is 10.5 0.5 %; The printing parameters are set as follows: laser power is 440 2 W, filling speed is 1000 5 mm / s, filling interval is 158 2 μm, powder particle size is 37 2 μm, laser spot diameter is 80 2 μm, layer thickness is 30 2 μm, gage length is 30 0.2 mm 2 , gage length is 30 0.2 mm; the yield strength of the No. 3 TC4 titanium alloy sample obtained after SLM shaping is 1133 10 MPa; the ultimate tensile strength is 1270 10 MPa; the elongation after fracture is 10.8 0.5 %; The printing parameters are set as follows: laser power is 475 2 W, filling speed is 1120 5 mm / s, filling interval is 152 2 μm, powder particle size is 37 2 μm, laser spot diameter is 80 2 μm, layer thickness is 30 2 μm, gage length is 30 0.2 mm 2 , gage length is 30 0.2 mm; the yield strength of the No. 3 TC4 titanium alloy sample obtained after SLM shaping is 1136 10 MPa; ultimate tensile strength of 1267 10 MPa; elongation at break of 10.3 0.5 %.

[0019] Compared with the prior art, the present application has the following beneficial effects: The laser powder bed fusion method for preparing high strength and toughness TC4 titanium alloy based on machine learning adopts a mixed expert model structure and a clustering initialization strategy, migrates the domain knowledge of feature classification information to the initial weights of the gating network, can realize "divide and conquer", avoid "hard coding" after direct classification and prediction, realize an integrated model with respective field expertise and cross-field continuity, significantly improve the generalization ability and prediction accuracy of the model under sparse and multi-source process-performance data, provide an extensible and efficient process optimization paradigm suitable for intelligent design of additive manufacturing processes of various metal materials.

[0020] The laser powder bed fusion method for preparing high strength and toughness TC4 titanium alloy based on machine learning adopts a super parameter tuning method combining large step grid search and Bayesian optimization, and can stably realize R 2 2 higher than 0.8 under five-fold cross-validation with only 200 network architecture combinations, efficiently locks the optimal model structure and avoids overfitting and underfitting.

[0021] The TC4 prepared by the traditional SLM process is difficult to realize the combination of strength and toughness, the present application excavates the nonlinear relationship between the characteristics and the target performance in depth through the machine learning method, collects the Pareto frontier solution and recommends the printing parameters that can realize high strength and toughness, successfully breaks through the "strength and toughness inversion" problem and realizes the collaborative optimization of strength and toughness.

[0022] After completing the model training, it is applied to the reverse design of the SLM process parameters, and the NSGA-II algorithm is adopted for multi-objective optimization search.

[0023] The TC4 titanium alloy obtained by the laser powder bed fusion method for preparing high strength and toughness TC4 titanium alloy based on machine learning has excellent comprehensive mechanical properties, the yield strength is more than 1050 MPa, and the elongation at break is maintained at more than 14%, which is better than the conventional SLM formed TC4 alloy. BRIEF DESCRIPTION OF DRAWINGS

[0024] Figure 1 It is a flowchart of the laser powder bed fusion method for preparing high strength and toughness TC4 titanium alloy based on machine learning; Figure 2A mixed expert model framework schematic diagram in a laser powder bed fusion method for preparing high-toughness TC4 titanium alloy based on machine learning; Figure 3 The engineering stress-strain curve of the TC4 titanium alloy room temperature tensile test of Example 1; Figure 4 The engineering stress-strain curve of the TC4 titanium alloy room temperature tensile test of Example 1; Figure 5 The engineering stress-strain curve of the TC4 titanium alloy room temperature tensile test of Example 2; Figure 6 The engineering stress-strain curve of the TC4 titanium alloy room temperature tensile test of Example 3; Figure 7 The engineering stress-strain curve of the TC4 titanium alloy room temperature tensile test of Example 4; Figure 8 The engineering stress-strain curve of the TC4 titanium alloy room temperature tensile test of Example 5. DETAILED DESCRIPTION

[0025] The present application will be described in more detail by the following specific examples, and other advantages and effects of the present application will be easily understood by those skilled in the art from the disclosure of this specification.

[0026] Exemplary embodiments of the present application will now be described with reference to the accompanying drawings, which are meant to exemplify and do not limit the present application in all aspects. The present application can be implemented in many different forms and is not limited to the embodiments described herein, which are provided to thoroughly and completely disclose the present application and to fully convey the scope of the present application to those skilled in the art. Terms used in the exemplary embodiments shown in the drawings are not limited to the present application. In the drawings, the same elements are designated with the same reference numerals.

[0027] Unless otherwise defined, terms (including technical terms) used herein have meanings that are commonly understood by one of ordinary skill in the art. Also, it is to be understood that the terms defined by commonly used dictionaries are to be interpreted as having a meaning that is consistent with their meaning in the context of the relevant art and not be interpreted in an idealized or overly formal sense unless expressly so defined herein.

[0028] The specific test methods of the following examples are as follows: The yield strength, ultimate tensile strength, and elongation at break of the sample are tested according to the tensile property test of Chinese GB / T 228-2002.

[0029] The present application will be further described in detail below with reference to specific examples, which are explanatory of the present application and not limiting.

[0030] The application discloses a laser powder bed fusion method for preparing high-strength and high-toughness TC4 titanium alloy based on machine learning, and refers to Figure 1 , comprising: S1, collecting performance data of TC4 titanium alloy prepared by different SLM process parameters from published literatures; The SLM process parameters and the performance data of the TC4 titanium alloy are divided into a training set and a verification set, and a K-means clustering algorithm is used to classify the performance data of the TC4 titanium alloy, and the clustering centers of each 'feature category' are obtained.

[0031] Among them, the SLM process parameters refer to the SLM process parameters in the existing research contents, and the performance data of the TC4 titanium alloy prepared by 174 groups of SLM process parameters are collected according to the existing research contents, the process parameters include laser power, filling speed, filling spacing, powder particle size, laser spot diameter, layer thickness, gage length area and gage length, and the performance data include yield strength, ultimate tensile strength and elongation after fracture.

[0032] S2, the K-means clustering algorithm is used to classify the performance data of the TC4 titanium alloy, and the clustering centers of each 'feature category' are obtained.

[0033] The K-means clustering algorithm classifies the TC4 data, comprising: The feature data is standardized, and the global mean and standard deviation are calculated:

[0034] Among them, is the original feature vector, is the global mean vector of the data, is the standard deviation vector.

[0035] Then, K-means clustering is performed, the number of clusters is set to 3, the random seed is set to 42, the initialization number is 10, and the center is:

[0036] Among them, represents the center of the kth cluster, is the sample set belonging to the cluster, is the sample number of the cluster, is the feature vector of the ith sample in the cluster.

[0037] S3, the clustering center is taken as the initial weight of the input layer and the first layer of the gating network, and a hybrid expert machine learning prediction model is constructed; Refer to Figure 2The hybrid expert model framework includes a gating network, multiple expert networks, and a load balancing loss; the gating network learns expert selection weights from clustering information; the expert networks are specially modeled for different process parameter regions; and the load balancing loss is used to ensure effective utilization of each expert network.

[0038] Specifically, the specific architecture of the hybrid expert model is as follows: First, the gating network The task of the gating network is to determine the weight distribution of each expert network according to the input process parameter features. The first layer weights of the gating network are initialized using the cluster centers:

[0039] wherein, is the i-th cluster center, and n is the number of clusters.

[0040] The gating network outputs the weights of each expert network and normalizes them through a softmax function:

[0041] wherein, is the output of the i-th expert network, is the corresponding gating weight, and n is the total number of experts.

[0042] Second, the expert network (Expert Networks) The expert network is a residual network structure, and each expert network includes an input layer, multiple residual blocks, and an output layer. Each expert network adopts a residual network structure and includes multiple residual blocks. The output of the expert network is calculated as follows: Each expert network adopts a residual network structure and includes multiple residual blocks. Let the output of the i-th expert network be:

[0043] Stack all expert outputs along dimension 2 to obtain:

[0044] wherein is the batch size, and n is the feature dimension.

[0045] The layer sequence of each residual block is: .

[0046] ​​Third, output fusion

[0047] Augment it in the first dimension:

[0048] The final output is: .

[0049] The load balancing loss function is used to ensure the balanced use of each expert network, and the expert usage rate is defined as , and the ideal uniform distribution is , and the load balancing loss is the KL divergence:

[0050] Wherein is the actual usage rate of the i-th expert, is the ideal usage rate thereof.

[0051] S4, for the mixed expert machine learning prediction model, through large step network search combined with Bayesian optimization method for small range iteration, determine the optimal performance prediction model, including: Determine the initial hyperparameter space by grid search, and the network structure hyperparameters include the number of nodes in the first layer, the depth of residual blocks, the number of residual blocks, the dropout rate and the number of node attenuations between each residual block; Based on the top 10 best parameter combinations of Stage1 results, intelligently determine the search range of Bayesian optimization; Use TPE sampler for Bayesian optimization to maximize the weighted R 2 Score of the validation set in five-fold cross-validation to determine the optimal performance prediction model, R 2 The formula is:

[0052] Wherein, the is the true value of the i-th sample, is the predicted value thereof, is the mean of the true value, and N is the total number of samples.

[0053] In the embodiments of the application, the MoE architecture initialized by clustering effectively alleviates the feature space sparsity and data imbalance problem, and the R 2 score of the prediction model is 0.9.

[0054] S5, use the optimal performance prediction model and NSGA-II genetic algorithm to obtain the SLM printing parameters of high strength and toughness TC4 alloy, specifically, including: The optimal performance prediction model is embedded in the NSGA-II algorithm, and the laser power, filling speed and filling spacing are set as decision variables, such as the laser power range 120-500 W, the filling speed range 1000-2000 mm / s, and the filling spacing range 30-200 μm; A multi-objective optimization problem is defined, and the objective function is to maximize the elongation at break and yield strength:

[0055] The objective function is to maximize the elongation at break and yield strength, and the constraint condition is added to ensure that EF 0 and YS 0.

[0056] The algorithm parameters are set as follows: population size 2000, maximum iteration number 1000, crossover probability 0.9, crossover distribution index 20, and mutation distribution index 70.

[0057] In addition, the fixed process parameters are as follows: powder particle size 37 2 μm, laser spot diameter 80 2 μm, layer thickness 30 2 μm.

[0058] The NSGA-II genetic algorithm is performed to repeatedly update the process parameter space, and the Pareto front solution of the SLM printing parameters of the high-toughness TC4 alloy is obtained.

[0059] S6, SLM forming is performed according to the SLM printing parameters of the high-toughness TC4 alloy to prepare a high-toughness TC4 titanium alloy, comprising: According to the SLM printing parameters of the high-toughness TC4 alloy, 3 groups of parameters are selected from the Pareto front solution using TC4 powder with purity 99.5% to perform laser powder bed fusion forming; wherein, in the SLM forming process, high-purity argon gas is used for protection, the oxygen content is controlled 100 ppm, and the forming chamber temperature is maintained at 120 5℃.

[0060] The SLM forming uses an iSLM-160 device equipped with a 500 W fiber laser. The scanning strategy uses zoned scanning, and the rotation angle between adjacent layers is 67°.

[0061] In addition, the prepared high-toughness TC4 titanium alloy is tested for mechanical properties using an Instron 5565 universal testing machine on standard tensile specimens, the tensile direction is parallel to the sample building direction, and the tensile strain rate is 5×10 -4 s -1 The elongation at break and yield strength are obtained by stress-strain curve calculation.

[0062] Example 1 The printing parameters were set as follows: laser power was 450 2 W, filling speed was 1150 5 mm / s, filling interval was 150 2 μm, powder particle size was 37 2 μm, laser spot diameter was 80 2 μm, layer thickness was 30 2 μm, gage area was 12 0.2 mm 2 , gage length was 30 0.2 mm.

[0063] Referring to Figure 3 , 4 , the yield strength of the first TC4 titanium alloy sample obtained after SLM forming in Example 1 was 1074 10 MPa; the ultimate tensile strength was 1220 10 MPa; the elongation after fracture was 13.7 0.5 %.

[0064] Example 2 The printing parameters were set as follows: laser power was 480 2 W, filling speed was 1090 5 mm / s, filling interval was 148 2 μm, powder particle size was 37 2 μm, laser spot diameter was 80 2 μm, layer thickness was 30 2 μm, gage area was 12 0.2 mm 2 , gage length was 30 0.2 mm.

[0065] Referring to Figure 3 , 5 , the yield strength of the second TC4 titanium alloy sample obtained after SLM forming was 1149 10 MPa; the ultimate tensile strength was 1268 10 MPa; the elongation after fracture was 10.7 0.5 %.

[0066] Example 3 The printing parameters were set as follows: laser power was 455 2 W, filling speed was 1040 5 mm / s, filling interval was 152 2 μm, powder particle size is 37 2 μm, laser spot diameter is 80 2 μm, layer thickness 30 2μm, gauge length area is 12 0.2 mm 2 The gauge length is 30. 0.2 mm.

[0067] Reference Figure 3 , 6 The yield strength of the No. 3 TC4 titanium alloy sample obtained after SLM forming is 1062. 10MPa; ultimate tensile strength is 1286. 10 MPa; Elongation after fracture 10.5 0.5%.

[0068] Example 4 The printing parameters are set as follows: laser power 440. 2 W, filling speed 1000 5 mm / s, with a fill spacing of 158 2 μm, powder particle size is 37 2 μm, laser spot diameter is 80 2 μm, layer thickness 30 2μm, gauge length area is 12 0.2 mm 2 The gauge length is 30. 0.2 mm.

[0069] Reference Figure 3 , 7 The yield strength of the No. 3 TC4 titanium alloy sample obtained after SLM forming was 1133. 10 MPa; ultimate tensile strength is 1270. 10 MPa; Elongation after fracture 10.8 0.5%.

[0070] Example 5 The printing parameters are set as follows: laser power 475. 2 W, filling speed 1120 5 mm / s, with a fill spacing of 152 2 μm, powder particle size is 37 2 μm, laser spot diameter is 80 2 μm, layer thickness 30 2μm, gauge length area is 12 0.2 mm 2 The gauge length is 30. 0.2 mm.

[0071] Referring to Figure 3 , 8 The yield strength of the No. 3 TC4 titanium alloy sample obtained after SLM shaping is 1136 10 MPa; the ultimate tensile strength is 1267 10 MPa; and the elongation after fracture is 10.3 0.5 %.

[0072] The laser powder bed fusion method for preparing high strength and toughness TC4 titanium alloy based on machine learning adopts a mixed expert (MoE) machine learning model framework, integrates K-means clustering initialization and multi-objective optimization, trains a preset TC4 performance prediction model through the mixed expert machine learning model framework, the training set and the clustering initialization information, and obtains the SLM printing parameters of the high strength and toughness TC4 alloy through the trained performance prediction model and the NSGA-II genetic algorithm. The high strength and toughness TC4 titanium alloy is designed and prepared, and the elongation after fracture, yield strength and ultimate tensile strength are all significantly improved, and can be applied to application fields such as aerospace structural parts.

[0073] Referring to Figure 6 Compared with the products in other literatures, the TC4 titanium alloy prepared in the application realizes excellent plastic matching while maintaining high strength, and the yield strength and elongation after fracture are improved, and the comprehensive strength and toughness is better than most of the SLM formed TC4 alloys reported in the literatures, and the precise design and preparation of high strength and toughness TC4 titanium alloy in the high unknown process parameter space is realized (as shown in Figure 5 ​The method for designing and preparing high strength and toughness TC4 titanium alloy provides new insights for traditional SLM process parameter design, builds a high-precision computing framework suitable for exploring high-unknown characteristic space, and develops a batch of printing parameters that break through the inherent cognition of the field. Many studies believe that too high laser power and filling speed will reduce the forming quality of alloy samples (Thijs L, Verhaeghe F, Craeghs T, et al. A study of the microstructural evolution during selective laser melting of Ti-6Al-4V[J]. Acta materialia, 2010, 58(9): 3303-3312.) (Kasperovich G, Hausmann J. Improvement of fatigue resistance and ductility of TiAl6V4 processed by selective laser melting[J]. Journal of Materials Processing Technology, 2015, 220: 202-214.) (Pal S, Gubeljak N, Hudak R, et al. Tensile properties of selective laser melting products affected by building orientation and energy density[J]. Materials Science and Engineering: A, 2019, 743: 637-647.), thereby adversely affecting the performance of the sample. In the printing parameters designed by the application, high filling speed and filling spacing can prevent high energy density caused by high laser power, and at the same time, the nonlinear influence of each process parameter on the performance is accurately captured from the perspective of the machine learning model, thereby accurately positioning the best process parameters using reverse design, solving the long-standing problem of strength and toughness synergy in the field, and having a profound impact on titanium alloy additive manufacturing and process optimization of other alloy systems.

[0074] The laser powder bed fusion method for preparing high strength and toughness TC4 titanium alloy based on machine learning can alleviate the overfitting problem caused by sparse characteristic space and different sources of data, thereby realizing high-precision exploration of high-unknown printing parameter combinations.

[0075] The method for designing and preparing high-toughness TC4 titanium alloy provides new insights for traditional SLM process parameter design, builds a high-precision calculation framework suitable for exploring high-unknown characteristic space, and has a profound influence on process optimization of titanium alloy additive manufacturing and other alloy systems.

[0076] The above only describes the preferred embodiments of the present application, and does not use any limitation on the technical solutions of the present application. Those skilled in the art should understand that without departing from the spirit and principles of the present application, the technical solutions can also be modified and replaced in several simple ways, and these modifications and replacements also belong to the protection scope covered by the claims.

Claims

1. A laser powder bed melting method for preparing high-strength and high-toughness TC4 titanium alloy based on machine learning, characterized in that, include: Performance data of TC4 titanium alloys prepared by different SLM process parameters were collected from publicly published literature; The K-means clustering algorithm was used to classify the performance data of TC4 titanium alloy and obtain the cluster centers of each "feature category". A hybrid expert machine learning prediction model is constructed by using cluster centers as the initial weights of the input layer and the first layer of the gated network. For hybrid expert machine learning prediction models, the optimal performance prediction model is determined by small-scale iterations using a large-step network search combined with Bayesian optimization. The SLM printing parameters of high-strength and high-toughness TC4 alloy were obtained by back-calculating using the optimal performance prediction model and the NSGA-II genetic algorithm. SLM forming was performed based on the SLM printing parameters of the high-strength and high-toughness TC4 alloy to prepare the high-strength and high-toughness TC4 titanium alloy.

2. The laser powder bed melting method for preparing high-strength and high-toughness TC4 titanium alloy based on machine learning as described in claim 1, characterized in that, The K-means clustering algorithm was used to classify the performance data of TC4 titanium alloy, obtaining the cluster centers for each "feature category", including: Standardize the feature data and calculate the global mean and standard deviation: in, The original feature vector, The global mean vector of the data. The standard deviation vector; Then, K-means clustering is performed, with the centers being: in Indicates the center of the k-th cluster. For the set of all samples belonging to this cluster, | | represents the number of samples in this cluster. Let be the feature vector of the i-th sample within the cluster.

3. The laser powder bed melting method for preparing high-strength and high-toughness TC4 titanium alloy based on machine learning as described in claim 2, characterized in that, The hybrid expert model framework includes a gated network, multiple expert networks, and a load balancing loss; the gated network comprises a hidden layer and an output layer, with the first layer weights initialized using cluster centers. The expert network is a residual network structure, and each expert network contains an input layer, multiple residual blocks and an output layer. The gated network outputs the weights of each expert network and normalizes them using the softmax function. The load balancing loss is used to ensure balanced utilization across expert networks, and the expert utilization rate is defined as follows: The ideal uniform distribution is The load balancing loss is then the KL divergence: in Let i be the actual usage rate of the i-th expert. To achieve its ideal utilization rate.

4. The laser powder bed melting method for preparing high-strength and high-toughness TC4 titanium alloy based on machine learning as described in claim 1, characterized in that, For hybrid expert machine learning prediction models, a small-scale iteration using five-fold cross-validation with a large-step network search combined with Bayesian optimization is employed to determine the optimal performance prediction model, including: The initial hyperparameter space is determined by grid search; The top 10 optimal parameter combinations based on the Stage 1 results are used as the initial parameter combinations for Bayesian optimization. Bayesian optimization is performed using the TPE sampler to maximize the weighted R-value of the validation set in the five-fold cross-validation. 2 The score determines the optimal performance prediction model.

5. The laser powder bed melting method for preparing high-strength and high-toughness TC4 titanium alloy based on machine learning as described in claim 4, characterized in that, Maximize weighted R 2 The formula is: in, For the true value of the i-th sample, For its predicted value, is the mean of the true values, and N is the total number of samples.

6. The laser powder bed melting method for preparing high-strength and high-toughness TC4 titanium alloy based on machine learning as described in claim 1, characterized in that, The SLM printing parameters for high-strength and high-toughness TC4 alloy were obtained by back-calculating using the optimal performance prediction model and the NSGA-II genetic algorithm, including: The optimal performance prediction model is embedded into the NSGA-II algorithm, and laser power, filling speed, and filling spacing are set as decision variables. Define a multi-objective optimization problem, with the objective function being to maximize elongation after fracture and yield strength: With EF 0 and YS 0 is the constraint condition, and iterative processing is performed using simulated binary crossover and polynomial mutation operators; The NSGA-II genetic algorithm was used to back-engineer and repeatedly update the process parameter space to obtain the Pareto front solution as the SLM printing parameter for the high-strength and high-toughness TC4 alloy.

7. The laser powder bed melting method for preparing high-strength and high-toughness TC4 titanium alloy based on machine learning as described in claim 1, characterized in that, Based on the SLM printing parameters of the high-strength and high-toughness TC4 alloy, SLM forming was performed to prepare the high-strength and high-toughness TC4 titanium alloy, including: Based on the optimized process parameters, three sets of parameters were selected from the Pareto front solution, using purity as the primary parameter. 99.5% TC4 powder is used for laser powder bed fusion forming; during the SLM forming process, high-purity argon gas is used for protection and oxygen content is controlled. 100ppm, molding chamber temperature maintained at 120 5℃.

8. The laser powder bed melting method for preparing high-strength and high-toughness TC4 titanium alloy based on machine learning as described in claim 7, characterized in that, The printing parameters are set as follows: laser power 450. 2 W, filling speed 1150 5 mm / s, with a fill spacing of 150. 2μm, powder particle size is 37 2 μm, laser spot diameter is 80 2 μm, layer thickness 30 2 μm, gauge length area is 12 0.2 mm 2 The gauge length is 30. 0.2 mm; The yield strength of the No. 1 TC4 titanium alloy sample obtained after SLM forming in Example 1 was 1074 mm. 10 MPa; ultimate tensile strength is 1220 MPa 10 MPa; Elongation after fracture 13.7 0.5%; The printing parameters are set as follows: laser power 480. 2 W, filling speed 1090 5 mm / s, with a fill spacing of 148 2μm, powder particle size is 37 2 μm, laser spot diameter is 80 2 μm, layer thickness 30 2 μm, gauge length area is 12 0.2 mm 2 The gauge length is 30. 0.2 mm; the yield strength of the No. 2 TC4 titanium alloy sample obtained after SLM forming is 1149. 10 MPa; ultimate tensile strength is 1268 MPa 10 MPa; Elongation after fracture 10.7 0.5%; The printing parameters are set as follows: laser power 455. 2 W, filling speed 1040 5 mm / s, with a fill spacing of 152 2 μm, powder particle size is 37 2 μm, laser spot diameter is 80 2 μm, layer thickness 30 2μm, gauge length area is 12 0.2 mm 2 The gauge length is 30. 0.2 mm; the yield strength of the No. 3 TC4 titanium alloy sample obtained after SLM forming is 1062 oz. 10 MPa; ultimate tensile strength is 1286 MPa 10 MPa; Elongation after fracture 10.5 0.5%; The printing parameters are set as follows: laser power 440. 2 W, filling speed 1000 5 mm / s, with a fill spacing of 158 2 μm, powder particle size is 37 2 μm, laser spot diameter is 80 2 μm, layer thickness 30 2μm, gauge length area is 12 0.2 mm 2 The gauge length is 30. 0.2 mm; the yield strength of the No. 3 TC4 titanium alloy sample obtained after SLM forming is 1133. 10 MPa; ultimate tensile strength is 1270 MPa 10 MPa; Elongation after fracture 10.8 0.5%; The printing parameters are set as follows: laser power 475. 2 W, filling speed 1120 5 mm / s, with a fill spacing of 152 2 μm, powder particle size is 37 2 μm, laser spot diameter is 80 2 μm, layer thickness 30 2μm, gauge length area is 12 0.2 mm 2 The gauge length is 30. 0.2 mm; the yield strength of the No. 3 TC4 titanium alloy sample obtained after SLM forming is 1136. 10 MPa; ultimate tensile strength is 1267 MPa 10 MPa; elongation after fracture 10.3 0.5%.

Citation Information

Patent Citations

  • Regulation and control method and regulation and control system for melt forming precision of polymer laser powder bed and method for improving forming precision

    CN119388749A

Cited By

  • A technology for preparing ultra-high strength titanium alloy by recovering residual titanium based on data driving

    CN122462535A