High-precision injection-molded titanium alloy part and machine learning assisted preparation method thereof

Abstract

The invention belongs to the technical field of powder metallurgy, and particularly relates to a high-precision injection-molded titanium alloy workpiece and a machine learning assisted preparation method thereof.The machine learning assisted preparation method of the high-precision injection-molded titanium alloy workpiece comprises the steps that an original data set is obtained through three-dimensional mold filling numerical simulation, performing data processing on data in the original data set to obtain an expanded data set; based on the expanded data set, multiple different machine learning algorithms are adopted to construct an injection molding process parameter prediction model, and an optimal prediction model is obtained; and optimal process parameters are obtained based on the model, and a target titanium alloy workpiece is prepared. According to the method provided by the invention, three-dimensional mold filling numerical simulation and machine learning are fused, and the process parameter prediction model is constructed and optimized, so that the actual MIM process is guided, the titanium alloy workpiece with high dimensional precision can be obtained, the research and development period is greatly shortened, and the trial-manufacturing cost is reduced.

Description

Technical Field

[0001] This invention belongs to the field of powder metallurgy technology, and specifically relates to a high-precision injection-molded titanium alloy part and its machine learning-aided preparation method. Background Technology

[0002] Ti-6Al-4V alloys, with their high specific strength, excellent corrosion resistance, and biocompatibility, have become crucial materials for lightweight design in fields such as medical devices and aerospace. However, due to their low thermal conductivity and difficult machining, traditional melting and machining processes are limited in efficiency and processing quality when fabricating complex or micro-precision parts, hindering the full realization of the material's intrinsic advantages and restricting its application expansion in emerging fields such as precision manufacturing and consumer electronics. With the increasing demand for high-performance, lightweight, and complex integrated titanium alloy components, efficiently fabricating high-precision complex parts while ensuring material performance has become key to the widespread application of Ti-6Al-4V alloys. Metal injection molding (MIM), combining powder metallurgy with plastic injection molding, enables near-net-shape, mass production of complex-shaped metal parts, offering advantages such as high material utilization and high forming freedom, and is considered an important approach for fabricating micro-sized complex-shaped parts.

[0003] Dimensional accuracy is a key indicator for evaluating the quality of MIM (Metal Injection Molding) parts. Currently, the industry commonly uses mold precision compensation and subsequent mechanical finishing (such as pressing and machining) to improve the dimensional accuracy of injection-molded Ti-6Al-4V alloy parts. However, such methods are essentially "post-processing corrections," which not only increase additional process steps and manufacturing costs but also weaken the near-net-shape advantages of MIM. Furthermore, for geometrically complex or thin-walled parts, the feasibility of subsequent machining is limited, making it difficult to achieve global accuracy control. It should be noted that MIM metal feedstock is a typical "binder-powder" two-phase system, whose rheological behavior is coupled with temperature, shear rate, and powder loading, resulting in a narrow process window for high-quality preform formation. Inappropriate injection molding process parameters can easily lead to severe two-phase separation and uneven filling defects, causing discrete green compact density distribution that is difficult to reverse during subsequent debinding and sintering, thus severely reducing the dimensional accuracy and consistency of the final part. Existing optimization methods rely heavily on trial and error, which is costly and time-consuming. While computational fluid dynamics (CFD) numerical simulations can visualize the filling process, the simulation deviates significantly from reality due to the uncertain coupling between equipment characteristics, complex geometry, and feed rheology.

[0004] Therefore, there is an urgent need to develop a technical method that can break through the limitations of traditional experience and numerical simulation, and achieve rapid prediction and robust optimization of injection molding process parameters, so as to ensure the high-precision preparation of MIM Ti-6Al-4V alloy parts. Summary of the Invention

[0005] The purpose of this invention is to provide a high-precision injection-molded titanium alloy part and its machine learning-assisted preparation method. The method provided by this invention integrates three-dimensional filling numerical simulation and machine learning. It takes the process parameters and dimensional characteristics of different geometric cavities as input and the maximum deviation of powder volume fraction Max_VOF as output. Through data cleaning, correlation screening, feature importance ranking and data expansion, a process parameter prediction model is constructed and optimized, thereby guiding the actual MIM process. This can obtain titanium alloy parts with high dimensional accuracy, significantly shorten the R&D cycle and reduce the trial production cost.

[0006] The first aspect of this invention provides a machine learning-assisted method for preparing high-precision injection-molded titanium alloy parts, comprising the following steps: S1, the original dataset is obtained through three-dimensional filling numerical simulation: the original dataset includes process parameters and size features as input features, and the maximum deviation of powder content distribution as output feature; S2, Data Processing: The data in the original dataset are sequentially cleaned, subjected to Pearson correlation analysis, ranked by feature importance, and augmented to obtain an augmented dataset; S3, Machine Learning Modeling: Based on the expanded dataset, various machine learning algorithms are used to construct a prediction model for injection molding process parameters, and the accuracy of the model prediction is evaluated to obtain the best prediction model; S4, Preparation of target titanium alloy parts: Based on the best prediction model, the optimal process parameters for injection molding are obtained. The titanium alloy parts are prepared by sequentially performing feeding preparation, injection molding, degreasing and sintering treatment using injection molding technology.

[0007] In some embodiments, in step S1, the original dataset is obtained by numerical simulation of the three-dimensional filling process of the injection cavity based on the ANSYS / CFX module and using a two-fluid model. Among them, 5 to 10 part cavities with different sizes and / or shapes are selected; the cavity is divided into unstructured tetrahedral meshes, and the injection gate and venting groove are divided into relatively dense meshes; the process parameters are obtained using a full factorial experimental design scheme.

[0008] In some embodiments, the viscosity parameters in the numerical simulation process are based on actual test data of the titanium alloy feedstock.

[0009] In some embodiments, the powder used for feeding includes at least one of spherical titanium alloy powder, near-spherical titanium alloy powder, and irregularly shaped titanium alloy powder with a particle size of 0~20 μm and not 0.

[0010] In some embodiments, the binder used in the feed is a polyoxymethylene-based binder, and the powder accounts for 52% to 62% of the volume of the feed.

[0011] In some embodiments, the adhesive comprises the following components by weight percentage: 70%~85% polyoxymethylene, 2%~10% high-density polyethylene, 2%~10% ethylene-vinyl acetate copolymer, and 5%~10% stearic acid.

[0012] In some embodiments, in step S1, the process parameters include at least one of injection temperature, injection speed, and holding pressure.

[0013] In some embodiments, the injection temperature is 170°C to 190°C; or the injection speed is 40 mm / s to 80 mm / s; or the holding pressure is 60 MPa to 100 MPa.

[0014] In some embodiments, in step S1, the dimensional features include at least one of the following: length, width, height, surface area, volume, specific surface area, maximum wall thickness, minimum wall thickness, maximum aperture, minimum aperture, maximum groove depth, minimum groove depth, maximum curvature, and minimum fillet radius.

[0015] In some embodiments, the dimensional features are automatically obtained by scanning the part model corresponding to the cavity layer by layer using the Open Cascade library in Python.

[0016] In some embodiments, in step S2, the data cleaning is based on the 3σ principle and calculated according to the Z-score method: Where x is the sample value, μ is the mean of the data in the column, and σ is the standard deviation of the data in the column; data with an absolute Z-score of ≥3 are filtered out to remove outliers.

[0017] In some embodiments, the correlation threshold |r| for the Pearson correlation analysis is set to 0.95.

[0018] In some embodiments, a random forest model is used to rank the importance of features and filter out input feature data that are ranked lower in importance among features with a correlation threshold greater than 0.95.

[0019] In some embodiments, the data augmentation is based on standard deviation perturbation, specifically including: keeping the input and output features unchanged and adding or subtracting the standard deviation noise of the data column, so that the augmented data volume is 9 times that of the original data.

[0020] In some embodiments, in step S3, the machine learning modeling is carried out by constructing a model using at least two of gradient boosting regression tree, support vector regression, multilayer perceptron and Gaussian process regression to obtain multiple injection molding process parameter prediction models. During modeling, each prediction model is independently optimized for hyperparameters using grid search combined with five-fold cross-validation. Then, the model prediction accuracy is evaluated using the coefficient of determination, mean absolute error and root mean square error, and the best prediction model is selected based on the coefficient of determination value.

[0021] In some embodiments, in step S4, the optimal process parameters are obtained based on the maximum deviation of powder content distribution as an output feature, using the size characteristics of the target part and multiple sets of process parameters designed by full factorial experiments as input features, and the process parameter corresponding to the minimum value of the maximum deviation of powder content distribution as the optimal process parameter for injection molding of the target part.

[0022] In some embodiments, the minimum value of the maximum deviation of the powder content distribution does not exceed ±0.5% of the powder loading.

[0023] In some embodiments, in step S4, the feeding preparation includes mixing under the protection of high-purity argon gas at a mixing temperature of 170°C to 200°C, a rotation speed of 10 r / min to 30 r / min, and a time of 1 h to 2 h.

[0024] In some embodiments, the degreasing treatment includes oxalic acid catalytic degreasing and thermal degreasing. The oxalic acid catalytic degreasing is carried out under a protective atmosphere of high-purity nitrogen, with a degreasing temperature of 120~145℃, an acid inlet rate of 1.5~2.5 g / min, and a degreasing time of 6h~14h.

[0025] In some embodiments, the thermal degreasing temperature is 450℃~600℃, the heating rate is 1℃ / min~5℃ / min, and the degreasing time is 1 h~5 h.

[0026] In some embodiments, the sintering temperature of the sintering treatment is 1000℃~1200℃, the heating rate is 2℃ / min~10℃ / min, and the sintering time is 4 h~8 h.

[0027] In some embodiments, the thermal degreasing and the sintering treatment are performed under vacuum, wherein the vacuum degree of the thermal degreasing is 1 Pa to 10 Pa, and the vacuum degree of the sintering treatment is 10 Pa. -2 Pa~10 -4 Pa.

[0028] The second aspect of the present invention provides a high-precision injection-molded titanium alloy part, which is prepared by the preparation method described in the first aspect. The titanium alloy part has a density ≥99%, a dimensional deviation within ±0.1 mm, and a surface roughness Ra≤1.6 μm.

[0029] This invention has the following advantages: (1) The machine learning-assisted method proposed in this invention can achieve accurate prediction and rapid screening of injection molding process parameters under limited experimental conditions, which greatly reduces the time cost and material loss caused by repeated process verification. By modeling and optimizing multi-dimensional process features and structural features, this method realizes global search and optimal solution locking of injection molding process window for complex components, which significantly reduces development cost and trial production cycle, and improves preparation efficiency and dimensional consistency.

[0030] (2) This invention introduces a machine learning model to retrain and correct the numerical simulation results, making the prediction results not only physically interpretable but also taking into account the complexity of the process and the diversity of products under actual manufacturing conditions. This method effectively reduces the deviation between the numerical simulation and the actual forming state, improves the stability and reliability of the prediction results, and is suitable for the mass production of complex MIM Ti-6Al-4V alloy parts.

[0031] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, and in order to make the above and other objects, features and advantages of the present invention more apparent and understandable, specific embodiments of the present invention are described below. Attached Figure Description

[0032] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0033] Figure 1 This is a flowchart of the machine learning-assisted high-precision injection molding method for preparing titanium alloy parts in an embodiment of the present invention. Detailed Implementation

[0034] Exemplary embodiments of the invention will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the invention are shown in the drawings, it should be understood that the invention can be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the invention and to fully convey the scope of the invention to those skilled in the art.

[0035] It should be understood that the terminology used herein is for the purpose of describing particular exemplary embodiments only and is not intended to be limiting. Unless the context clearly indicates otherwise, the singular forms “a,” “an,” and “described” as used herein may also include the plural forms. The terms “comprising,” “including,” “containing,” and “having” are inclusive and therefore indicate the presence of the stated features, steps, operations, elements, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, elements, components, and / or combinations thereof. The method steps, processes, and operations described herein are not construed as requiring them to be performed in a particular order described or illustrated unless the order of performance is explicitly indicated. It should also be understood that additional or alternative steps may be used.

[0036] In the description of the embodiments of this invention, technical terms such as "first" and "second" are used only to distinguish different objects and should not be construed as indicating or implying relative importance or implicitly specifying the number, specific order, or primary and secondary relationship of the indicated technical features. In the description of the embodiments of this invention, "multiple" means two or more, unless otherwise explicitly defined.

[0037] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the invention. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0038] In the description of the embodiments of this invention, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Additionally, the character " / " in this document generally indicates that the preceding and following related objects have an "or" relationship.

[0039] In the description of the embodiments of the present invention, the term "multiple" refers to two or more (including two), similarly, "multiple groups" refers to two or more (including two groups), and "multiple pieces" refers to two or more (including two pieces).

[0040] In the description of the embodiments of the present invention, unless otherwise explicitly specified and limited, the technical terms such as "installation," "connection," "joining," and "fixing" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components. Those skilled in the art can understand the specific meaning of the above terms in the embodiments of the present invention according to the specific circumstances.

[0041] This invention addresses the challenges of narrow high-precision process windows and difficult dimensional accuracy control in complex MIM titanium alloy parts. It proposes an injection molding process parameter optimization design method that integrates three-dimensional filling numerical simulation and machine learning. An injection molding process parameter prediction model under the influence of model feature parameters is established, enabling rapid design of high-quality green blank preparation processes for complex shapes. This improves the high-precision window control capability of the parts and shortens the preparation cycle, thus helping to broaden the application fields of Ti-6Al-4V alloy.

[0042] See Figure 1 As shown, the machine learning-assisted high-precision injection molding method for preparing titanium alloy parts provided by the present invention is carried out according to the following steps.

[0043] (1) Obtain feed rheological data (viscosity parameters).

[0044] In this embodiment of the invention, titanium alloy powder and binder are mixed to form a uniform feedstock, with the feedstock powder loading being 52%~62%. Then, a capillary rheometer is used to test the feedstock at test temperatures of 170℃, 180℃, and 190℃, and for 500 seconds. -1 1000 s -1 2000 s -1 5000 s -1 8000 s -1 Viscosity data under shear rate conditions were used as feed rheological data for numerical simulation of the filling process.

[0045] In this embodiment of the invention, the powder used for feeding includes at least one of spherical Ti-6Al-4V powder, near-spherical Ti-6Al-4V powder, and irregular Ti-6Al-4V powder with a particle size of 0~20 μm and not 0.

[0046] In this embodiment of the invention, the binder used in the feed is a polyoxymethylene-based binder, and the volume ratio (powder loading) of the powder in the feed is 52% to 62%. Exemplarily, the volume ratio of the powder in the feed can be one of 52%, 53%, 54%, 55%, 56%, 57%, 58%, 59%, 60%, 61%, 62%, or any value that satisfies the above range.

[0047] In this embodiment of the invention, the adhesive comprises the following components by mass percentage: 70%~85% polyoxymethylene, 2%~10% high-density polyethylene, 2%~10% ethylene-vinyl acetate copolymer, and 5%~10% stearic acid.

[0048] The mass percentage of polyoxymethylene provided by this invention can be one of 70%, 71%, 72%, 73%, 74%, 75%, 76%, 77%, 78%, 79%, 80%, 81%, 82%, 83%, 84%, or 85%, or any value satisfying the above range. The mass percentage of high-density polyethylene provided by this invention can be one of 2%, 3%, 4%, 5%, 6%, 7%, 8%, 9%, or 10%, or any value satisfying the above range. The mass percentage of ethylene-vinyl acetate copolymer provided by this invention can be one of 2%, 3%, 4%, 5%, 6%, 7%, 8%, 9%, or 10%, or any value satisfying the above range. The mass percentage of stearic acid provided by this invention can be one of 5%, 6%, 7%, 8%, 9%, or 10%, or any value satisfying the above range.

[0049] (2) Obtain the dimensional characteristics of different cavities.

[0050] In this embodiment of the invention, the Open Cascade library (OCC) based on the Python language scans the part model corresponding to the cavity layer by layer, automatically capturing and obtaining 14 dimensional features, including length (L), width (W), height (H), surface area (S), volume (V), specific surface area (S / V), maximum wall thickness (Max_WT), minimum wall thickness (Min_WT), maximum aperture (Max_HD), minimum aperture (Min_HD), maximum groove depth (Max_SD), minimum groove depth (Min_SD), maximum curvature (Max_C), and minimum fillet radius (Min_FR).

[0051] (3) Establish the original dataset for machine learning.

[0052] In this embodiment of the invention, the filling process of different cavities is numerically simulated using the ANSYS / CFX module based on a two-fluid model with full factorial design of injection parameters. The process parameters and dimensional features are used as input features, and the maximum deviation of powder content distribution (Max_VOF) is used as the output feature to obtain the raw data for machine learning.

[0053] In this embodiment of the invention, 5 to 10 different component cavities with different sizes and / or shapes are selected.

[0054] In this embodiment of the invention, the cavity is divided into unstructured tetrahedral meshes, and the injection gate and venting groove are divided into relatively dense meshes.

[0055] (4) Data processing.

[0056] In this embodiment of the invention, redundant data is filtered out based on the 3σ principle, Pearson correlation (PCC) analysis, and random forest feature importance ranking, and the remaining data is augmented based on the standard deviation perturbation method to obtain an augmented dataset.

[0057] Specifically, the purpose of data cleaning is to remove the disturbances caused by factors such as convergence error and extreme operating conditions; the purpose of PCC analysis and feature importance ranking is to remove redundant features, reduce information overlap, and improve modeling efficiency and prediction accuracy; the purpose of data augmentation is to enhance the robustness and generalization ability of the model and improve the prediction accuracy of the machine learning model.

[0058] In this embodiment of the invention, the data cleaning is based on the 3σ principle and calculated according to the Z-score method, as shown in formula (1).

[0059] (1) In the formula, x is the sample value, μ is the mean of the data in the column, and σ is the standard deviation of the data in the column; data with an absolute Z-score of ≥3 are screened out to remove outliers.

[0060] In this embodiment of the invention, the correlation threshold |r| for PCC analysis is set to 0.95, and a random forest model is used to rank the importance of features, filtering out input feature data with lower importance rankings among features with a correlation threshold or coefficient greater than 0.95.

[0061] In this embodiment of the invention, data augmentation is performed based on standard deviation perturbation. While preserving the original data distribution characteristics, a small amount of noise is added to each sample to generate multiple approximate samples. This alleviates the overfitting problem caused by the scarcity of samples and maintains the consistency of the statistical distribution of the augmented data, ensuring that the patterns learned by the model have physical interpretability and predictive stability.

[0062] In this embodiment of the invention, data augmentation specifically includes: performing unchanged operation on the input and output features and adding or subtracting the standard deviation noise of the data in the column. After augmentation, the data volume is 9 times that of the original data, as shown in equations (2) to (5).

[0063] (2) (3) (4) (5) In the formula, and These are sample data representing the input and output features, respectively. and These are the average value of the current column of the input feature and the average value of the output feature, respectively. Input the standard deviation of the current column of the feature; is the standard deviation of the output feature; n is the number of data points, which is the amount of data remaining after cleaning, PCC analysis, and importance ranking and filtering of the original data. For expanded input feature data; To expand the output feature data, NF is the noise coefficient, which is preferably set to 0.2 in this embodiment of the invention to enhance the robustness and generalization of the model while maintaining the consistency of statistical distribution.

[0064] (5) Construct the best prediction model.

[0065] In this embodiment of the invention, the expanded dataset is divided into a training set and a test set in an 8:2 or 7:3 ratio. Using the training set as the data foundation, two or more of the following methods are employed to construct models: Gradient Boosting Regression Tree (GBDT), Support Vector Regression (SVR), Multilayer Perceptron (MLP), and Gaussian Process Regression (GPR). This results in multiple injection molding process parameter prediction models. Furthermore, during modeling, each injection molding process parameter prediction model undergoes hyperparameter optimization using grid search combined with five-fold cross-validation. Then, using the test set as the data foundation, the coefficient of determination (R²) is used... 2 The accuracy of model prediction is evaluated by the mean absolute error (MAE) and root mean square error (RMSE), and the calculation formulas are shown in equations (6) to (8).

[0066] (6) (7) (8) In the formula, Based on predicted values; These are experimental values; This represents the average of the experimental values.

[0067] Specifically, R 2 The closer the value is to 1, the higher the goodness of fit, while the smaller the MAE and RMSE values, the lower the prediction error.

[0068] In this embodiment of the invention, R 2 As a selection criterion, R is selected from all models. 2 The prediction model with the higher score is considered the best prediction model.

[0069] (6) Injection molding of target titanium alloy parts.

[0070] In this embodiment of the invention, the dimensional features of the target part are captured using the OCC library. The 27 sets of injection temperature (Injection_T), injection speed (Injection_V), and holding pressure (Holding_P) and dimensional feature combinations of the full factorial experimental design are used as input features. The optimal prediction model obtained in the above steps is used to predict the injection molding process parameters. The injection molding process parameters corresponding to the minimum Max_VOF value are used as the optimal injection molding process parameters to prepare titanium alloy green blanks.

[0071] The preparation of the feed is the same as in step (1), and the powder used for the feed includes at least one of spherical titanium alloy powder, near-spherical titanium alloy powder, and irregularly shaped titanium alloy powder with a particle size in the range of 0~20 μm and not equal to 0.

[0072] The binder used in the feed is a polyoxymethylene-based binder, and the volume ratio of powder in the feed (powder loading) is 52%~62%.

[0073] The adhesive comprises the following components by weight percentage: 70%~85% polyoxymethylene, 2%~10% high-density polyethylene, 2%~10% ethylene-vinyl acetate copolymer, and 5%~10% stearic acid.

[0074] The feed preparation involves mixing titanium alloy powder with a binder under the protection of high-purity argon gas at a temperature of 170°C to 200°C, a rotation speed of 10 r / min to 30 r / min, and a time of 1 h to 2 h.

[0075] The mixing temperature provided by this invention can be one of 170°C, 175°C, 180°C, 185°C, 190°C, 195°C, and 200°C, or any value satisfying the above range. The mixing speed can be one of 10 r / min, 15 r / min, 20 r / min, 25 r / min, and 30 r / min, or any value satisfying the above range. The mixing time can be one of 1 h, 1.2 h, 1.5 h, 1.8 h, and 2 h, or any value satisfying the above range.

[0076] Next, the titanium alloy green billet was placed in a degreasing furnace for catalytic degreasing under a high-purity nitrogen protective atmosphere. The degreasing temperature was 120℃~145℃, the acid inlet rate was 1.5 g / min~2.5 g / min, and the degreasing time was 6 h~14 h. Afterward, the catalytically degreased billet was placed in a vacuum degreasing and sintering furnace for hot degreasing and sintering. The vacuum degree during the hot degreasing stage was 1 Pa~10 Pa, the hot degreasing temperature was 450℃~600℃, the heating rate was 1℃ / min~5℃ / min, and the degreasing time was 1 h~5 h. The vacuum degree during the sintering stage was 10 Pa. -2 Pa~10 -4 The sintering temperature is 1000℃~1200℃, the heating rate is 2℃ / min~10℃ / min, the sintering time is 4 h~8 h, and the titanium alloy parts are obtained after furnace cooling.

[0077] The catalytic degreasing temperature provided by this invention can be one of 120℃, 125℃, 130℃, 135℃, 140℃, and 145℃, or any value satisfying the above range. The acid ingress rate for catalytic degreasing can be one of 1.5 g / min, 1.6 g / min, 1.7 g / min, 1.8 g / min, 1.9 g / min, 2.0 g / min, 2.1 g / min, 2.2 g / min, 2.3 g / min, 2.4 g / min, and 2.5 g / min, or any value satisfying the above range. The catalytic degreasing time can be one of 6 h, 7 h, 8 h, 9 h, 10 h, 11 h, 12 h, 13 h, and 14 h, or any value satisfying the above range.

[0078] The vacuum degree for thermal degreasing provided by this invention can be one of 1 Pa, 2 Pa, 3 Pa, 4 Pa, 5 Pa, 6 Pa, 7 Pa, 8 Pa, 9 Pa, 10 Pa, or any value satisfying the above range. The thermal degreasing temperature can be one of 450℃, 500℃, 550℃, 600℃, or any value satisfying the above range. The heating rate can be one of 1℃ / min, 2℃ / min, 3℃ / min, 4℃ / min, 5℃ / min, or any value satisfying the above range. The degreasing time can be one of 1 h, 2 h, 3 h, 4 h, 5 h, or any value satisfying the above range.

[0079] The vacuum degree of the sintering process provided by this invention can be 10. -2 Pa, 10 -3 Pa, 10 -4 The sintering temperature can be one of the values ​​in Pa or any value within the range described above. The sintering temperature can be one of the values ​​in 1000℃, 1100℃, and 1200℃ or any value within the range described above. The heating rate can be one of the values ​​in 2℃ / min, 3℃ / min, 4℃ / min, 5℃ / min, 6℃ / min, 7℃ / min, 8℃ / min, 9℃ / min, and 10℃ / min or any value within the range described above. The sintering time can be one of the values ​​in 4 h, 5 h, 6 h, 7 h, and 8 h or any value within the range described above.

[0080] It is worth mentioning that titanium alloys include, but are not limited to, Ti-6Al-4V alloys.

[0081] Unless otherwise defined, the technical terms used in the following embodiments have the same meaning as commonly understood by those skilled in the art. Unless otherwise specified, the experimental reagents used in the following embodiments are all conventional biochemical reagents; the raw materials, instruments, and equipment used in the following embodiments can all be obtained commercially or through existing methods; unless otherwise specified, the amounts of experimental reagents used are the amounts used in conventional experimental operations; unless otherwise specified, the experimental methods are conventional methods. It should be further noted that the following description is merely exemplary and not a specific limitation of the present invention.

[0082] Example 1 Spherical Ti-6Al-4V alloy powder with a particle size of 0-20 μm (not zero) was used as raw material, with a loading of 62%. Then, 80% polyoxymethylene, 5% high-density polyethylene, 5% ethylene-vinyl acetate copolymer, and 5% stearic acid were weighed out as binders by mass percentage. Metal feedstock was prepared under high-purity argon protection at a mixing temperature of 185℃, a rotation speed of 20 r / min, and a mixing time of 2 h. The prepared feedstock was then tested using a capillary rheometer at temperatures of 170℃, 180℃, and 190℃ for 500 s. -1 1000 s -1 2000 s -1 5000 s -1 8000 s -1 Viscosity data under shear rate conditions were used as feed rheological data for numerical simulation of the filling process.

[0083] The OCC library based on Python scans the part models corresponding to five cavities layer by layer, automatically capturing and obtaining 14 dimensional features, including L, W, H, S, V, S / V, Max_WT, Min_WT, Max_HD, Min_HD, Max_SD, Min_SD, Max_C, and Min_FR. Numerical simulation of the filling process with full factorial design of injection parameters for different cavities is performed based on a two-fluid model. Process parameters and dimensional features are used as input features, and the maximum deviation of powder content distribution (Max_VOF) is used as the output feature. 135 sets of raw machine learning data are obtained to establish the original dataset.

[0084] Data with an absolute Z-score ≥ 3 was removed based on the 3σ principle. After cleaning, 6 data points were eliminated, leaving 129 data points. Then, redundant features (W, S, Max_SD, Min_SD, and Max_WT) were removed through PCC analysis and random forest feature importance ranking. Finally, the standard deviation perturbation method was used to augment the 129 data points by keeping the input and output features unchanged, adding or subtracting the standard deviation noise of the column, resulting in 1161 augmented data points, which formed the augmented dataset.

[0085] The expanded dataset was divided into training and test sets in an 8:2 ratio. Using the training set as the foundation, three algorithms—SVR, MLP, and GPR—were used to build models. After hyperparameter optimization, three process parameter prediction models were obtained. Using the test set as the foundation, R... 2 The model prediction accuracy was evaluated using MAE and RMSE, and the calculation results are shown in Table 1. RSE was prioritized over MAE. 2 As a screening criterion, R is selected as the preferred option. 2 The GPR prediction model with the highest score is selected as the best prediction model.

[0086] Using a "dog bone" shaped tensile specimen as a typical target part for verification and preparation, the dimensional characteristics of the target part were first extracted using the OCC library. Using 27 combinations of Injection_T, Injection_V, and Holding_P from a full-factor experimental design, along with the dimensional characteristic combinations, as input features, the optimal prediction model was used to predict the injection molding process parameters. The injection molding process parameters corresponding to the minimum Max_VOF value were selected as the optimal injection molding process parameters for preparing the Ti-6Al-4V alloy green billet. The optimal injection molding process parameters included: Injection_T of 180℃, Injection_V of 60 mm / s, Holding_P of 60 MPa, and Max_VOF of 0.24%. Then, the green billet was placed in a degreasing furnace for catalytic degreasing under a high-purity nitrogen protective atmosphere. The degreasing temperature was 135℃, the acid inlet rate was 2.5 g / min, and the degreasing time was 8 h. After catalytic degreasing, the degreased billet was placed in a vacuum degreasing and sintering furnace for hot degreasing and sintering. The vacuum level during the hot degreasing stage was 8... Pa, the vacuum degree during the sintering stage is 10. -3 Pa; the hot degreasing temperature was 500℃, the heating rate was 5℃ / min, and the degreasing time was 4 h; the vacuum sintering temperature was 1050℃, the heating rate was 6℃ / min, and the sintering time was 4 h. After furnace cooling, Ti-6Al-4V alloy parts were obtained.

[0087] Table 1

[0088] The results show that the Ti-6Al-4V alloy parts prepared in Example 1 have a density of 99.6%, a dimensional deviation within ±0.08 mm, high dimensional accuracy, and a surface roughness Ra of only 1.01 μm.

[0089] Example 2 Near-spherical Ti-6Al-4V alloy powder with a particle size of 0-20 μm (not zero) was used as raw material, with a loading of 56%. Then, 78% polyoxymethylene, 8% high-density polyethylene, 8% ethylene-vinyl acetate copolymer, and 6% stearic acid were weighed out as binders by mass percentage and prepared as a metal feedstock under high-purity argon protection. The mixing temperature was 190℃, the rotation speed was 30 r / min, and the mixing time was 1.5 h. The prepared feedstock was then tested using a capillary rheometer at temperatures of 170℃, 180℃, and 190℃, and for 500 s. -1 1000 s -1 2000 s -1 5000 s -1 8000 s -1Viscosity data under shear rate conditions were used as feed rheological data for numerical simulation of the filling process.

[0090] The OCC library based on Python scans the part models corresponding to eight cavities layer by layer, automatically capturing and obtaining 14 dimensional features, including L, W, H, S, V, S / V, Max_WT, Min_WT, Max_HD, Min_HD, Max_SD, Min_SD, Max_C, and Min_FR. Numerical simulation of the filling process with full factorial design of injection parameters for different cavities is performed based on a two-fluid model. Process parameters and dimensional feature parameters are used as input features, and the maximum deviation of powder content distribution (Max_VOF) is used as the output feature, obtaining 216 sets of raw machine learning data to form the original dataset.

[0091] Data with an absolute Z-score ≥ 3 was removed based on the 3σ principle. After cleaning, 11 data points were eliminated, leaving 205 data points. Then, redundant features W, S, Max_SD, and Max_WT were removed through PCC analysis and random forest feature importance ranking. Finally, the standard deviation perturbation method was used to augment the 205 data points by keeping the input and output features unchanged, adding or subtracting the standard deviation noise of the data column, resulting in 1845 augmented data points, forming the augmented dataset.

[0092] The expanded dataset was divided into training and testing sets in a 7:3 ratio. Using the training set as the foundation, three algorithms—GBDT, SVR, and GPR—were used to build models. After hyperparameter optimization, three process parameter prediction models were obtained. Using the testing set as the foundation, R... 2 The model prediction accuracy was evaluated using MAE and RMSE, with R being the preferred metric. 2 As the screening criterion, the calculation results are shown in Table 2, and R is selected as the preferred option. 2 The GBDT prediction model with the highest score is selected as the best prediction model.

[0093] Using a "dog bone" shaped tensile specimen as a typical target part for verification, the dimensional characteristics of the target part were first extracted using the OCC library. Using 27 combinations of Injection_T, Injection_V, and Holding_P from a full factorial experimental design, along with dimensional feature combinations, as input features, the optimal prediction model was used to predict injection molding process parameters. The injection molding process parameters corresponding to the minimum Max_VOF value were selected as the optimal injection molding process parameters for preparing the Ti-6Al-4V alloy green billet. The optimal injection molding process parameters included: Injection_T of 190℃, Injection_V of 60 mm / s, Holding_P of 80 MPa, and Max_VOF of 0.43%. Then, the green billet was placed in a degreasing furnace for catalytic degreasing treatment under a high-purity nitrogen protective atmosphere. The degreasing temperature was 135℃, the acid inlet rate was 2.2 g / min, and the degreasing time was 12 minutes. h; After catalytic degreasing, the catalytically degreased blank is placed in a vacuum degreasing sintering furnace for hot degreasing and sintering. The vacuum degree of the hot degreasing stage is 10 Pa, the hot degreasing temperature is 550℃, the heating rate is 2℃ / min, and the degreasing time is 6 h; the vacuum degree of the sintering stage is 10 Pa. -3 The sintering temperature was 1100℃, the heating rate was 3℃ / min, the sintering time was 7 h, and the Ti-6Al-4V alloy parts were obtained after furnace cooling.

[0094] Table 2

[0095] The results show that the Ti-6Al-4V alloy part prepared in Example 2 has a density of 99.2%, a dimensional deviation within ±0.09 mm, and a surface roughness Ra of 1.31 μm.

[0096] Example 3 Irregular Ti-6Al-4V alloy powder with a particle size of 0-20 μm (not zero) was used as raw material, with a loading of 54%. Then, 75% polyoxymethylene, 10% high-density polyethylene, 8% ethylene-vinyl acetate copolymer, and 7% stearic acid were weighed out as binders by mass percentage. Metal feedstock was prepared under high-purity argon protection at a mixing temperature of 190℃, a mixing speed of 20 r / min, and a mixing time of 2 h. The prepared feedstock was then tested using a capillary rheometer at temperatures of 170℃, 180℃, and 190℃ for 500 s. -1 1000 s -1 2000 s -1 5000 s -1 8000 s -1 Viscosity data under shear rate conditions were used as feed rheological data for numerical simulation of the filling process.

[0097] Based on the Python language OCC library, the part models corresponding to five cavities are scanned layer by layer, automatically capturing and obtaining 14 dimensional features, including L, W, H, S, V, S / V, Max_WT, Min_WT, Max_HD, Min_HD, Max_SD, Min_SD, Max_C, and Min_FR. Numerical simulation of the filling process of different cavities is carried out based on a two-fluid model with full factorial design of injection parameters. The process parameters and dimensional features are used as input features, and the maximum deviation of powder content distribution (Max_VOF) is used as the output feature. 135 sets of raw machine learning data are obtained to establish the original dataset.

[0098] Data with an absolute Z-score ≥ 3 was removed based on the 3σ principle. After cleaning, 7 data points were eliminated, leaving 128 data points. Then, redundant features W, S, Max_SD, and Max_WT were eliminated through PCC analysis and random forest feature importance ranking. Finally, the standard deviation perturbation method was used to augment the input and output features of the 128 data points by keeping them unchanged, adding or subtracting the standard deviation noise of the column, resulting in 1152 augmented data points, which were then used to build the augmented dataset.

[0099] The expanded dataset was divided into training and test sets in an 8:2 ratio. Using the training set as the foundation, four algorithms—GBDT, SVR, MLP, and GPR—were employed for model construction. After hyperparameter optimization, four process parameter prediction models were obtained. Using the test set as the foundation, R... 2 The prediction accuracy was evaluated using MAE and RMSE, with R being the preferred metric. 2 As the screening criterion, the calculation results are shown in Table 3, and R is selected as the preferred option. 2 The GBDT prediction model with the highest score is selected as the best prediction model.

[0100] Using a "dog bone" shaped tensile specimen as a typical target part for verification and preparation, the characteristic dimensions of the target part were first extracted using the OCC library. Using 27 combinations of Injection_T, Injection_V, and Holding_P from a full-factor experimental design, along with dimensional feature combinations, as input features, the optimal prediction model was used to predict injection molding process parameters. The injection molding process parameters corresponding to the minimum Max_VOF value were selected as the optimal injection molding process parameters for preparing Ti-6Al-4V alloy green blanks. The optimal injection molding process parameters included: Injection_T of 190℃, Injection_V of 80 mm / s, Holding_P of 100 MPa, and Max_VOF of 0.47%. Then, the green blank was placed in a degreasing furnace for catalytic degreasing under a high-purity nitrogen protective atmosphere. The degreasing temperature was 130℃, the acid inlet rate was 1.5 g / min, and the degreasing time was 14 h. After catalytic degreasing, the degreased green blank was placed in a vacuum degreasing and sintering furnace for hot degreasing and sintering. The vacuum degree during the hot degreasing stage was 8... Pa, the hot degreasing temperature is 580℃, the heating rate is 2℃ / min, and the degreasing time is 8 h; the vacuum degree during the sintering stage is 10. -4 The sintering temperature was 1080℃, the heating rate was 4℃ / min, the sintering time was 8 h, and the Ti-6Al-4V alloy parts were obtained after furnace cooling.

[0101] Table 3

[0102] The results show that the Ti-6Al-4V alloy part prepared in Example 3 has a density of 99.1%, a dimensional deviation within ±0.1 mm, and a surface roughness Ra of 1.42 μm.

[0103] Example 4 A mixture of two types of Ti-6Al-4V alloy powders with a particle size of 0–20 μm (not zero) and near-spherical and spherical particles was used as raw material in a 1:1 mass ratio with a loading of 58%. Then, 82% polyoxymethylene, 6% high-density polyethylene, 4% ethylene-vinyl acetate copolymer, and 8% stearic acid were weighed out as binders by mass percentage. Metal feedstock was prepared under high-purity argon protection at a mixing temperature of 185℃, a mixing speed of 30 r / min, and a mixing time of 1.5 h. The prepared feedstock was then tested using a capillary rheometer at temperatures of 170℃, 180℃, and 190℃ for 500 s. -1 1000 s -1 2000 s -1 5000 s -1 8000 s -1Viscosity data under shear rate conditions were used as feed rheological data for numerical simulation of the filling process.

[0104] The OCC library based on Python scans the part models corresponding to eight cavities layer by layer, automatically capturing and obtaining 14 dimensional features, including L, W, H, S, V, S / V, Max_WT, Min_WT, Max_HD, Min_HD, Max_SD, Min_SD, Max_C, and Min_FR. Numerical simulations of the filling process for different cavities are performed using a two-fluid model with full factorial design of injection parameters. Process parameters and dimensional features are used as input features, and the maximum deviation of powder content distribution (Max_VOF) is used as the output feature, resulting in 216 sets of raw machine learning data to establish the original dataset.

[0105] Data with an absolute Z-score ≥ 3 was removed based on the 3σ principle. After cleaning, 15 data points were eliminated, leaving 201 data points. Then, redundant features W, S, Max_SD, Min_SD, and Max_WT were removed through PCC analysis and random forest feature importance ranking. Finally, the standard deviation perturbation method was used to augment the input and output features of the 201 data points by keeping them unchanged, adding or subtracting the standard deviation noise of the column, resulting in 1809 augmented data points, which formed the augmented dataset.

[0106] The expanded dataset was divided into training and testing sets in an 8:2 ratio. Using the training set as the data foundation, both SVR and GPR algorithms were employed for model construction. After hyperparameter optimization, two process parameter prediction models were obtained. Using the testing set as the data foundation, R... 2 The model prediction accuracy was evaluated using MAE and RMSE, with R being the preferred metric. 2 As the screening criterion, the calculation results are shown in Table 4, and R is selected as the preferred option. 2 The GPR prediction model with the highest score is selected as the best prediction model.

[0107] Using a "dog bone" shaped tensile specimen as a typical target part for verification and preparation, the dimensional characteristics of the target part were first extracted using the OCC library. Using 27 combinations of Injection_T, Injection_V, and Holding_P from a full-factor experimental design, along with dimensional feature combinations, as input features, the optimal prediction model was used to predict injection molding process parameters. The injection molding process parameters corresponding to the minimum Max_VOF value were selected as the optimal injection molding process parameters for preparing Ti-6Al-4V alloy green blanks. The optimal injection molding process parameters included: Injection_T of 180℃, Injection_V of 60 mm / s, Holding_P of 80 MPa, and Max_VOF of 0.36%. Then, the green blank was placed in a degreasing furnace for catalytic degreasing under a high-purity nitrogen protective atmosphere. The degreasing temperature was 135℃, the acid inlet rate was 1.8 g / min, and the degreasing time was 10 h. After catalytic degreasing, the degreased green blank was placed in a vacuum degreasing and sintering furnace for hot degreasing and sintering. The vacuum degree during the hot degreasing stage was 10... Pa, the hot degreasing temperature is 530℃, the heating rate is 3℃ / min, and the degreasing time is 5 h; the vacuum degree during the sintering stage is 10. -2 The sintering temperature was 1060℃, the heating rate was 5℃ / min, the sintering time was 5 h, and the Ti-6Al-4V alloy parts were obtained after furnace cooling.

[0108] Table 4

[0109] The results show that the Ti-6Al-4V alloy part prepared in Example 4 has a density of 99.3%, a dimensional deviation within ±0.08 mm, and a surface roughness Ra of 1.24 μm.

[0110] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for preparing high-precision injection-molded titanium alloy parts with machine learning assistance, characterized in that, Includes the following steps: S1, the original dataset is obtained through three-dimensional filling numerical simulation: the original dataset includes process parameters and size features as input features, and the maximum deviation of powder content distribution as output feature; S2, Data Processing: The data in the original dataset are sequentially cleaned, subjected to Pearson correlation analysis, ranked by feature importance, and augmented to obtain an augmented dataset; S3, Machine Learning Modeling: Based on the expanded dataset, various machine learning algorithms are used to construct a prediction model for injection molding process parameters, and the accuracy of the model prediction is evaluated to obtain the best prediction model; S4, Preparation of target titanium alloy parts: Based on the best prediction model, the optimal process parameters for injection molding are obtained. The titanium alloy parts are prepared by sequentially performing feeding preparation, injection molding, degreasing and sintering treatment using injection molding technology.

2. The preparation method according to claim 1, characterized in that, In step S1, the original dataset was obtained by numerical simulation of the three-dimensional filling process of the injection cavity using the ANSYS / CFX module and a two-fluid model. Among them, 5 to 10 types of component cavities with different sizes and / or shapes are selected; The cavity is divided into unstructured tetrahedral meshes, while the injection gate and venting groove are divided into relatively dense meshes. The process parameters were obtained using a full factorial experimental design.

3. The preparation method according to claim 2, characterized in that, The viscosity parameters in the numerical simulation process are based on actual test data of titanium alloy feed. Preferably, the powder used in the feeding includes at least one of spherical titanium alloy powder, near-spherical titanium alloy powder, and irregularly shaped titanium alloy powder with a particle size of 0~20 μm and not 0. Preferably, the binder used in the feed is a polyoxymethylene-based binder, and the powder accounts for 52% to 62% of the volume of the feed; Preferably, the adhesive comprises the following components by weight percentage: 70%~85% polyoxymethylene, 2%~10% high-density polyethylene, 2%~10% ethylene-vinyl acetate copolymer, and 5%~10% stearic acid.

4. The preparation method according to claim 1, characterized in that, In step S1, the process parameters include at least one of injection temperature, injection speed, and holding pressure; Preferably, the injection temperature is 170℃~190℃; or, the injection speed is 40 mm / s~80 mm / s; or, the holding pressure is 60 MPa~100 MPa.

5. The preparation method according to claim 1, characterized in that, In step S1, the dimensional features include at least one of the following: length, width, height, surface area, volume, specific surface area, maximum wall thickness, minimum wall thickness, maximum aperture, minimum aperture, maximum groove depth, minimum groove depth, maximum curvature, and minimum fillet radius. Preferably, the dimensional features are automatically obtained by scanning the part model corresponding to the cavity layer by layer using the Open Cascade library in Python.

6. The preparation method according to claim 1, characterized in that, In step S2, the data cleaning is based on the 3σ principle and calculated according to the Z-score method: ； Where x is the sample value, μ is the mean of the data in the column, and σ is the standard deviation of the data in the column; data with an absolute Z-score of ≥3 are filtered out to remove outliers; Preferably, the correlation threshold |r| for the Pearson correlation analysis is set to 0.95; Preferably, a random forest model is used to rank the importance of features, and input feature data with lower importance rankings among features with a relevance threshold greater than 0.95 are filtered out. Preferably, the data augmentation is based on the standard deviation perturbation method, specifically including: keeping the input and output features unchanged and adding or subtracting the standard deviation noise of the data in the column, so that the amount of data after augmentation is 9 times that of the original data.

7. The preparation method according to claim 1, characterized in that, In step S3, the machine learning modeling is carried out by constructing at least two of the following: gradient boosting regression tree, support vector regression, multilayer perceptron and Gaussian process regression, to obtain multiple injection molding process parameter prediction models. During modeling, each prediction model is independently optimized for hyperparameters using grid search combined with five-fold cross-validation. Then, the model prediction accuracy is evaluated using the coefficient of determination, mean absolute error and root mean square error, and the best prediction model is selected based on the coefficient of determination value.

8. The preparation method according to claim 1, characterized in that, In step S4, the optimal process parameters are obtained based on the maximum deviation of powder content distribution as an output feature. The size characteristics of the target part and multiple sets of process parameters designed by full factorial experiment are used as input features. The process parameters corresponding to the minimum value of the maximum deviation of powder content distribution are used as the optimal process parameters for injection molding of the target part. Preferably, the minimum value of the maximum deviation of the powder content distribution does not exceed ±0.5% of the powder loading.

9. The preparation method according to claim 1, characterized in that, In step S4, the feed preparation includes mixing under the protection of high-purity argon gas, with a mixing temperature of 170°C to 200°C, a rotation speed of 10 r / min to 30 r / min, and a time of 1 h to 2 h; Preferably, the degreasing treatment includes oxalic acid catalytic degreasing and thermal degreasing. The oxalic acid catalytic degreasing is carried out under a protective atmosphere of high-purity nitrogen, with a degreasing temperature of 120~145℃, an acid inlet rate of 1.5~2.5 g / min, and a degreasing time of 6 h~14 h. Preferably, the thermal degreasing temperature is 450℃~600℃, the heating rate is 1℃ / min~5℃ / min, and the degreasing time is 1h~5h; Preferably, the sintering temperature of the sintering treatment is 1000℃~1200℃, the heating rate is 2℃ / min~10℃ / min, and the sintering time is 4 h~8 h; Preferably, the thermal degreasing and sintering processes are performed under vacuum conditions, wherein the vacuum degree of the thermal degreasing is 1 Pa to 10 Pa, and the vacuum degree of the sintering process is 10 Pa. -2 Pa~10 -4 Pa.

10. A high-precision injection-molded titanium alloy part, characterized in that, The titanium alloy part prepared by any one of claims 1 to 9 has a density ≥99%, a dimensional deviation within ±0.1 mm, and a surface roughness Ra ≤1.6 μm.

Images