Method for predicting mechanical properties of additive manufactured nickel-titanium alloy parts based on deep learning

Abstract

The application discloses a method for predicting mechanical properties of a nickel-titanium alloy part based on deep learning, and comprises the following steps: determining initial characteristics of the nickel-titanium alloy required by a mechanical experiment; manufacturing two different nickel-titanium alloy parts, respectively performing tensile and compression tests on the samples, collecting corresponding data, and forming an initial data set by combining all the collected data with printing parameter data; pre-processing the initial data, dividing the initial data into a training set and a test set according to a proportion, using printing parameters of additive manufacturing as input values of a deep learning model, using performance parameters of the mechanical experiment in the data set as output values of the deep learning model, constructing a mechanical property prediction model of the nickel-titanium alloy, training the mechanical property prediction model by using the training set data, and testing the mechanical property prediction model by using the test set data after the training is completed; and the application can predict the mechanical properties of the nickel-titanium alloy with high accuracy and improve the research efficiency of the nickel-titanium alloy.

Description

Technical Field

[0001] This invention relates to the field of material mechanical property prediction technology, and in particular to a method for predicting the mechanical properties of additively manufactured nickel-titanium alloy parts based on deep learning. Background Technology

[0002] Additively manufactured nickel-titanium shape memory alloys are widely used in aerospace, medical devices, and automotive industries due to their excellent strength properties, shape memory effect, and superelasticity. Obtaining their precise mechanical properties can help researchers utilize the materials efficiently.

[0003] Traditional material prediction often uses manual measurement methods, which require a large number of repeated experiments and a significant investment of time and resources, making it difficult to meet the requirements of low cost and high efficiency.

[0004] With the application of deep learning in the field of materials science, a new method based on deep learning to predict the properties of nickel-titanium shape memory alloys can effectively reduce resource waste and accurately and reliably reflect the relationship between the printing parameters and mechanical properties of additively manufactured nickel-titanium alloys. Summary of the Invention

[0005] The purpose of this invention is to address the problems and shortcomings described in the background art by providing a method for predicting the mechanical properties of additively manufactured nickel-titanium alloy parts based on deep learning.

[0006] A method for predicting the mechanical properties of additively manufactured nickel-titanium alloy parts based on deep learning includes the following steps:

[0007] Step 1: Obtain the alloy composition and additive manufacturing method of nickel-titanium alloy from relevant literature on additive manufacturing of nickel-titanium alloys, so as to determine the initial properties of nickel-titanium alloy required for mechanical experiments;

[0008] Step 2: Fabricate two different nickel-titanium alloy parts using the selected additive manufacturing method: a long strip specimen for stretching and a cylindrical specimen for compression.

[0009] Step 3: Perform tensile and compression tests on the strip-shaped and cylindrical specimens respectively, collect the corresponding tensile and compression failure limit data and the displacement data generated when failure occurs, and combine all the collected data with the additive manufacturing printing parameter data to form an initial dataset.

[0010] Step 4: Preprocess the initial data. For some obviously isolated initial samples, delete them directly. For some initial samples with missing data, use the mean of similar data to replace the missing data. Then normalize the data. The preprocessed dataset will be randomly divided into two parts, training set and test set, in a certain proportion.

[0011] Step 5: Use the printing parameters of additive manufacturing as the input values ​​of the deep learning model, and use the performance parameters of the mechanical experiments in the dataset as the output values ​​of the deep learning model to construct a mechanical property prediction model for nickel-titanium alloy. Train the mechanical property prediction model with training set data. After training, test the mechanical property prediction model with test set data.

[0012] Step 6: Use the printing parameters required for additive manufacturing of nickel-titanium alloys as input values ​​for the model to predict the mechanical properties of the target nickel-titanium alloy.

[0013] Furthermore, the nickel-titanium alloy in step 1 comprises 50.8% nickel and 49.2% titanium; the additive manufacturing method for the nickel-titanium alloy is selective laser melting technology.

[0014] Furthermore, in step 2, the tensile specimen of the nickel-titanium alloy has dimensions of 50mm × 2.6mm × 1mm, and the compression specimen of the nickel-titanium alloy is a cylinder with a diameter of 4.496mm and a height of 6.6mm.

[0015] Furthermore, in step 4, the training dataset and the test dataset are normalized using the same method and randomly divided in a ratio of 80%:20%.

[0016] Furthermore, the printing parameters used in additive manufacturing in step 5 are laser power P, scanning speed V, scanning spacing H, and layer thickness T; the mechanical property parameters of the nickel-titanium alloy include tensile strength σ. t δ, tensile deformation limit t σ compressive strength c δ, compressive deformation limit c .

[0017] Furthermore, the deep learning model in step 6 is a long short-term memory network model.

[0018] Furthermore, in step 4, the data normalization process uses the Min-Max method, and its calculation formula is as follows:

[0019]

[0020] Where X is the original data in the sample dataset, and X' is the sample data after normalization. Max X is the maximum value in the original sample dataset. Min It is the minimum value in the original sample dataset.

[0021] The beneficial effects of this invention are as follows:

[0022] This invention uses long short-term memory networks to predict the mechanical properties of nickel-titanium alloy samples, which can replace traditional manual testing methods, effectively improving the efficiency of analyzing the mechanical properties of nickel-titanium alloys, reducing time and labor costs, improving economic benefits, and facilitating the use of nickel-titanium alloys.

[0023] This invention is based on an established neural network model, which can predict the properties of nickel-titanium alloys by adjusting different inputs, making it convenient for researchers to use.

[0024] The present invention has a simple operation method and is highly practical. This method can be extended to the prediction of mechanical properties of other alloys produced by additive manufacturing. Attached Figure Description

[0025] Figure 1 This is a flowchart illustrating the present invention;

[0026] Figure 2 This is a graph showing the mechanical property data of nickel-titanium alloy collected in the embodiments of this invention;

[0027] Figure 3 The Long Short-Time Memory (LSTM) network in the embodiments of this invention affects the tensile strength σ of nickel-titanium alloy. t A schematic diagram of the training results;

[0028] Figure 4 The embodiment of this invention uses a Long Short-Time Memory (LSTM) network to measure the tensile deformation limit δ of nickel-titanium alloy. t A schematic diagram of the training results;

[0029] Figure 5 The Long Short-Time Memory (LSTM) network in the embodiments of this invention affects the compressive strength σ of nickel-titanium alloy. c A schematic diagram of the training results;

[0030] Figure 6 The Long Short-Time Memory (LSTM) network in the embodiments of this invention is used to measure the compressive deformation limit δ of nickel-titanium alloy. c A diagram illustrating the training results. Detailed Implementation

[0031] To enable those skilled in the art to better understand the purpose, technical solution, and advantages of this invention, the invention will be described in detail below with reference to specific embodiments. Based on the embodiments of this invention, those skilled in the art can make several improvements without departing from the concept of this invention. These improvements all fall within the protection scope of this invention.

[0032] Referring to the attached figure, a method for predicting the mechanical properties of additively manufactured nickel-titanium alloy parts based on deep learning includes the following steps:

[0033] Step 1: Obtain the alloy composition and additive manufacturing method of nickel-titanium alloy from relevant literature on additive manufacturing of nickel-titanium alloys, so as to determine the initial properties of nickel-titanium alloy required for mechanical experiments;

[0034] Step 2: Fabricate two different nickel-titanium alloy parts using the selected additive manufacturing method: a long strip specimen for stretching and a cylindrical specimen for compression.

[0035] Step 3: Perform tensile and compression tests on the strip-shaped and cylindrical specimens respectively, collect the corresponding tensile and compression failure limit data and the displacement data generated when failure occurs, and combine all the collected data with the additive manufacturing printing parameter data to form an initial dataset.

[0036] Step 4: Preprocess the initial data. For some obviously isolated initial samples, delete them directly. For some initial samples with missing data, use the mean of similar data to replace the missing data. Then normalize the data. The preprocessed dataset will be randomly divided into two parts, training set and test set, in a certain proportion.

[0037] Step 5: Use the printing parameters of additive manufacturing as the input values ​​of the deep learning model, and use the performance parameters of the mechanical experiments in the dataset as the output values ​​of the deep learning model to construct a mechanical property prediction model for nickel-titanium alloy. Train the mechanical property prediction model with training set data, and test the mechanical property prediction model with test set data after training.

[0038] Step 6: Use the printing parameters required for additive manufacturing of nickel-titanium alloys as input values ​​for the model to predict the mechanical properties of the target nickel-titanium alloy.

[0039] Furthermore, the nickel-titanium alloy in step 1 comprises 50.8% nickel and 49.2% titanium; the additive manufacturing method for the nickel-titanium alloy is selective laser melting technology.

[0040] Furthermore, in step 2, the tensile specimen of the nickel-titanium alloy has dimensions of 50mm × 2.6mm × 1mm, and the compression specimen of the nickel-titanium alloy is a cylinder with a diameter of 4.496mm and a height of 6.6mm.

[0041] Furthermore, in step 4, the training dataset and the test dataset are normalized using the same method and randomly divided in a ratio of 80%:20%.

[0042] Furthermore, the printing parameters used in additive manufacturing in step 5 are laser power P, scanning speed V, scanning spacing H, and layer thickness T; the mechanical property parameters of the nickel-titanium alloy include tensile strength σ. tδ, tensile deformation limit t σ compressive strength c δ, the limit of compressive deformation c .

[0043] Furthermore, the deep learning model in step 6 is a long short-term memory network model.

[0044] Furthermore, in step 4, the data normalization process uses the Min-Max method, and its calculation formula is as follows:

[0045]

[0046] Where X is the original data in the sample dataset, and X' is the sample data after normalization. Max X is the maximum value in the original sample dataset. Min It is the minimum value in the original sample dataset.

[0047] Example 1:

[0048] Step 1: Based on relevant literature on nickel-titanium alloys, the alloy composition of the nickel-titanium alloy required for the experiment is determined to be 50.8%-51.8% nickel and the remainder titanium. The additive manufacturing method adopts selective laser melting technology. The parameters affecting the mechanical properties of the nickel-titanium alloy include laser power P, scanning speed V, scanning spacing H, and layer thickness T. The experimental specimens manufactured are strip tensile specimens with dimensions of 50mm×2.6mm×1mm and cylindrical compression specimens with dimensions of 4.496mm in diameter and 6.6mm in height. To avoid affecting the final prediction results, at least 50 sets of experimental specimens of each are required.

[0049] When collecting samples, experimental specimens whose shape does not conform to the original dimensions will not be used to avoid interfering with the dataset. The mechanical properties of nickel-titanium alloys include tensile strength σ. t δ, tensile deformation limit t σ compressive strength c δ, the limit of compressive deformation c The following table shows the additive manufacturing printing parameters and mechanical properties of some nickel-titanium alloys collected:

[0050] 1 100 50 0.03 0.08 42807 2.431 1087 2.963 2 200 75 0.03 0.08 43257 2.439 1093 2.956 3 300 100 0.03 0.08 43872 2.452 1110 2.991 4 400 125 0.03 0.08 44892 2.464 1124 3.013 5 500 150 0.03 0.08 46089 2.472 1142 3.028 6 600 175 0.03 0.08 47069 2.543 1181 3.062 7 700 200 0.03 0.08 47814 2.581 1202 3.082 8 800 225 0.03 0.08 48774 2.643 1219 3.097 9 … … … … … … … …

[0051] Step 2: Preprocess the initial data. For some obviously isolated initial samples, delete them directly. For some initial samples with missing data, use the mean of similar data to replace the missing data points. Then normalize the data. The preprocessed dataset will be randomly divided into two parts, training set and test set, in a ratio of 80%:20%.

[0052] Data normalization uses the Min-Max method, and its calculation formula is as follows:

[0053]

[0054] Where X is the original data in the sample dataset, and X' is the sample data after normalization. Max X is the maximum value in the original sample dataset. Min It is the minimum value in the original sample dataset;

[0055] This embodiment is not limited to a specific normalization calculation formula; other calculation formulas are also applicable.

[0056] Step 3: Using the four additive manufacturing printing parameters in the dataset as input values ​​for the deep learning model, and the performance parameters from the mechanical experiments in the dataset as output values, a mechanical property prediction model for nickel-titanium alloy is constructed. The model is trained using the training dataset and then tested using the test dataset. The correlation coefficient R0 is then used to determine the model's performance. 2 The accuracy of mechanical property prediction models is analyzed using parameters such as root mean square error (RMSE) and mean absolute error (MAE).

[0057] The deep learning model is a Long Short-Term Memory (LSTM) network, and its training parameters are shown in the table below:

[0058] Input features 4 Output features 4 Hidden layer unit 600 Initial learning rate 0.01 Training rounds 800 BatchSize 30

[0059] Step 4: Based on the trained nickel-titanium alloy mechanical property prediction model, the printing parameters required for additive manufacturing of nickel-titanium alloy are used as the input values ​​of the model to predict the mechanical properties of the target nickel-titanium alloy. In this embodiment, the nickel-titanium alloy used is Ni50.8Ti49.2 (at%).

[0060] In this invention, the correlation coefficient R 2 The root mean square error (RMSE) and mean absolute error (MAE) are used as evaluation indicators to assess the magnitude of the error between the predicted and actual values. The relevant parameters for multiple sets of data are shown in the table below:

[0061]

[0062] Comparison revealed that the predicted compressive strength σ c The root mean square error is relatively large, but it still has some reference value in practical applications. The overall prediction accuracy of the nickel-titanium alloy mechanical property prediction model can reach 98%.

[0063] The specific embodiments of the present invention have been described in detail above. It should be noted that the present invention is not limited to the specific embodiments described above. Those skilled in the art should understand that modifications and substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the present invention, which is defined by the appended claims and their equivalents.

Claims

1. A method for predicting the mechanical properties of additively manufactured nickel-titanium alloy parts based on deep learning, characterized in that: Includes the following steps: Step 1: Obtain the alloy composition and additive manufacturing method of nickel-titanium alloy from relevant literature on additive manufacturing of nickel-titanium alloys, so as to determine the initial properties of nickel-titanium alloy required for mechanical experiments; Step 2: Fabricate two different nickel-titanium alloy parts using the selected additive manufacturing method: a long strip specimen for stretching and a cylindrical specimen for compression. Step 3: Perform tensile and compression tests on the strip-shaped and cylindrical specimens respectively, collect the corresponding tensile and compression failure limit data and the displacement data generated when failure occurs, and combine all the collected data with the additive manufacturing printing parameter data to form an initial dataset. Step 4: Preprocess the initial data. For some obviously isolated initial samples, delete them directly. For some initial samples with missing data, use the mean of similar data to replace the missing data. Then normalize the data. The preprocessed dataset will be randomly divided into two parts, training set and test set, in a certain proportion. Step 5: Using the additive manufacturing printing parameters as input values ​​to the deep learning model and the performance parameters from the centralized mechanical experiments in the dataset as output values, construct a mechanical property prediction model for nickel-titanium alloys. Train the mechanical property prediction model using the training set data, and then test the model using the test set data. The printing parameters used in additive manufacturing are laser power P, scanning speed V, scanning spacing H, and layer thickness T. The mechanical property parameters of nickel-titanium alloys include tensile strength. Tensile deformation limit Compressive strength Compression deformation limit ; Step 6: Use the printing parameters required for additive manufacturing of nickel-titanium alloys as input values ​​for the model to predict the mechanical properties of the target nickel-titanium alloy; The deep learning model is a long short-term memory network model.

2. The method for predicting the mechanical properties of additively manufactured nickel-titanium alloy parts based on deep learning according to claim 1, characterized in that: The nickel-titanium alloy in step 1 comprises 50.8% nickel and 49.2% titanium; the additive manufacturing method for the nickel-titanium alloy is selective laser melting technology.

3. The method for predicting the mechanical properties of additively manufactured nickel-titanium alloy parts based on deep learning according to claim 1, characterized in that: In step 2, the tensile specimen of the nickel-titanium alloy has dimensions of 50mm × 2.6mm × 1mm, and the compression specimen of the nickel-titanium alloy is a cylinder with a diameter of 4.496mm and a height of 6.6mm.

4. The method for predicting the mechanical properties of additively manufactured nickel-titanium alloy parts based on deep learning according to claim 1, characterized in that: In step 4, the training dataset and the test dataset are normalized using the same method and randomly divided in a ratio of 80%:20%.

5. The method for predicting the mechanical properties of additively manufactured nickel-titanium alloy parts based on deep learning according to claim 1, characterized in that: In step 4, the data is normalized using the Min-Max method, and the calculation formula is as follows: Where X is the original data in the sample dataset, and X' is the sample data after normalization. It is the maximum value in the original sample dataset. It is the minimum value in the original sample dataset.

Images