Multi-dimensional visual analysis method based on titanium alloy components and processing technology
Through systematic data collection and preprocessing, feature selection and multi-dimensional visual display, combined with multiple machine learning algorithms, the data integration difficulties and low prediction accuracy in titanium alloy performance prediction and optimization analysis are solved, and the composition and Intuitive display of processing process relationships and performance improvement.
Patent Information
- Application Number
- CN202411060053.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-02
- Publication Date
- 2024-12-03
- Estimated Expiration
- 2044-08-02
AI Technical Summary
There are problems in the performance prediction and optimization analysis of titanium alloys, such as data integration, insufficient visualization effect and low prediction accuracy. Traditional methods are difficult to intuitively demonstrate the complex relationship between titanium alloy composition and processing technology, and existing tools lack specialized processing modules.
The systematized data collection and preprocessing steps are adopted, combined with Pearson's correlation coefficient selection features, and a variety of machine learning algorithms are used for model training, and the data relationship is visually displayed through multi-dimensional visualization methods such as scatter plots, heat maps and parallel coordinate maps. The model performance is evaluated using the Leave One Cross-Verification method and the composition and processing process combination is optimized through multiple iterations.
It improves the accuracy and efficiency of data analysis, enhances the prediction accuracy and stability of the model, can intuitively display complex data relationships, and provides scientific basis to improve the application performance of titanium alloys.
Smart Images

Figure CN119069038A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the fields of visual analysis and material science, and provides a multidimensional visual analysis method based on titanium alloy composition and processing technology. Background Art
[0002] Titanium alloys are widely used in aerospace, medical devices, and chemical industries due to their superior performance. The performance of titanium alloys is affected by their composition and processing technology, and the relationship between these factors is complex and difficult to display intuitively. Traditional titanium alloy design and optimization mainly rely on experimental trial and error methods, which are time-consuming, labor-intensive, and costly. With the continuous development of science and technology, data analysis and machine learning technologies are gradually applied to the field of materials science to predict material properties and optimize material design.
[0003] At present, some studies have applied data analysis and machine learning to the field of materials science. For example, regression analysis and neural network algorithms are used to predict material properties, and principal component analysis and clustering algorithms are used to classify and optimize material composition and processing technology. The implementation steps of these technical solutions are roughly as follows: data collection, data preprocessing, model training, model evaluation, and optimization analysis. However, the existing technical solutions have the following disadvantages: Difficult data integration: The data sources of titanium alloy composition, processing technology and performance are diverse, which makes integration difficult, and the existing tools lack specialized processing modules; Insufficient visualization effect: The existing technologies mostly use general data visualization tools, which lack specialized designs for titanium alloy composition and processing technology. The performance of titanium alloy is affected by a variety of microstructural factors and chemical composition. There is a highly nonlinear relationship between these factors, which makes it impossible for traditional visualization methods to fully express all the information and potential patterns in the data, and it is difficult to intuitively display complex multidimensional data relationships; Low prediction accuracy: Due to the limited data samples in materials science, the prediction accuracy and generalization ability of existing models are limited, which makes it difficult to meet actual needs. It is necessary to expand the limited data set through data enhancement technology, deeply explore and construct features with high predictive power, combine domain knowledge for feature selection and extraction, and combine the advantages of multiple machine learning models to improve the robustness and prediction accuracy of the model through ensemble learning, thereby improving the training effect and generalization ability of the model. Summary of the invention
[0004] The present invention solves technical problems such as difficulty in data integration, insufficient visualization effect and low prediction accuracy in the prediction and optimization analysis of titanium alloy properties.
[0005] In order to achieve the above purpose, the present invention adopts the following technical means:
[0006] Because the present invention adopts the above technical means, it has the following beneficial effects:
[0007] 1. Systematization of data collection and preprocessing steps: The data collection and preprocessing steps in the present invention are systematically designed to ensure the quality of the data and the accuracy of the analysis. First, the composition, processing technology and performance data of titanium alloys are collected from multiple sources such as literature, experiments and Matminer databases. Then, the collected data is cleaned, including removing noise, processing missing values and outliers to ensure the accuracy of the data. Next, the data set is normalized to the (0, 1) interval to make data of different dimensions comparable. This series of steps ensures that the data set used for modeling is high quality and accurate, laying a solid foundation for subsequent analysis and model training.
[0008] 2. Feature selection and model training: The present invention improves the prediction accuracy and stability of the model by combining the Pearson correlation coefficient to select features with high correlation with performance indicators, and uses advanced machine learning algorithms for model training. First, the correlation coefficient analysis is performed on the preprocessed data set, and features that are highly correlated with yield strength or tensile strength are selected. Then, a variety of machine learning algorithms, such as random forest, XGBoost, gradient boosting regression, etc., are used to train the model on the feature data. The prediction accuracy and stability of different models are evaluated by the leave-one-out cross-validation method, and the best model is selected based on the evaluation results. This process not only improves the prediction accuracy of the model, but also enhances the generalization ability of the model.
[0009] 3. Multi-dimensional visualization: The present invention uses a variety of visualization methods such as scatter plots, heat maps and parallel coordinates to intuitively display the complex relationship between titanium alloy composition, processing technology and performance, thereby enhancing the user's data analysis experience and efficiency. For example, through the scatter plot after dimensionality reduction by principal component analysis, the user can intuitively see the clustering of different data samples; the heat map can show the correlation between different features; and the parallel coordinates map helps the user analyze data from multiple dimensions. The use of these visualization tools makes complex data relationships intuitive and easy to understand, greatly improving the user's data analysis efficiency.
[0010] 4. Performance prediction and optimization analysis: The present invention can not only perform performance prediction based on the trained model, but also find the best combination of composition and processing technology through multidimensional visualization tools and optimization algorithms, providing a scientific basis for improving the application performance of titanium alloys. Users can use the trained model to predict performance by inputting new titanium alloy composition and processing parameters. Then, new scatter plots and line graphs are generated through multidimensional visualization tools, and the scatter points with similar performance in the same cluster in the scatter plot are compared in combination with the interval to which the performance prediction values belong in the line graph, to find out the features that are positively or negatively correlated with the performance. Based on this information, users can try to increase the components that are positively correlated with the performance or reduce the components that are negatively correlated, form a new combination, and re-perform performance prediction and comparative analysis. Through multiple iterations, users can finally determine the best combination of titanium alloy composition and processing technology to achieve the expected performance indicators. This process not only improves the interactivity and operability of the analysis, but also provides a scientific basis for improving the application performance of titanium alloys. BRIEF DESCRIPTION OF THE DRAWINGS
[0011] Figure 1 It is a flow chart of the present invention;
[0012] Figure 2 This is a diagram showing the interface.
[0013] Among them, a is the control area, b is the data sample cluster scatter plot, c is the custom scatter style, d is the custom scatter card, e is the performance data classification line chart, f is the specific value table of the evaluation index, and g is the evaluation index radar chart. DETAILED DESCRIPTION
[0014] The following is a detailed description of the embodiments of the present invention. Although the present invention will be described and illustrated in conjunction with some specific embodiments, it should be noted that the present invention is not limited to these embodiments. On the contrary, modifications or equivalent substitutions made to the present invention should all be included in the scope of the claims of the present invention.
[0015] In addition, in order to better illustrate the present invention, numerous specific details are given in the following specific embodiments. It will be understood by those skilled in the art that the present invention can also be implemented without these specific details.
[0016] A multi-dimensional visualization analysis tool based on titanium alloy composition and processing technology, comprising the following steps:
[0017] Step 1, data collection and preprocessing: After collecting titanium alloy data from multiple sources, the data is cleaned and normalized to construct a high-quality dataset suitable for modeling.
[0018] Step 2: Model training and evaluation: Extract effective features, use machine learning algorithms to train the model on the feature data, and evaluate and optimize the model.
[0019] Step 3, multi-dimensional visualization: Use the independently designed multi-dimensional visualization tool to display the titanium alloy composition, processing technology and performance data in multiple dimensions to help users intuitively analyze data relationships.
[0020] Step 4: Performance prediction and optimization analysis: Based on the trained model and multi-dimensional visualization tools, the performance of titanium alloys is predicted and optimized to find the best combination of titanium alloy composition and processing technology.
[0021] The above step 1 specifically includes the following steps:
[0022] Step 1.1: Query and collect the composition, processing technology and performance data of titanium alloy from literature, experiments and Matminer database, and organize and construct a data set suitable for modeling.
[0023] Step 1.2: Clean the constructed data set, including removing noise, processing missing values and outliers, to ensure the accuracy of the data;
[0024] Step 1.3: Normalize the data set to the interval (0, 1) to make data of different dimensions comparable.
[0025] The above step 2 specifically includes the following steps:
[0026] Step 2.1, feature selection: for the data set obtained after preprocessing in step 1, use the Pearson correlation coefficient to select features with high correlation with the performance index to obtain a feature data set;
[0027] Step 2.1.1, for the yield strength of titanium alloy, a correlation coefficient analysis is performed to determine 38 characteristic parameters with V alloy composition as the main characteristic as the characteristic set of the yield strength performance prediction model;
[0028] Step 2.1.1.1, calculate the Pearson correlation coefficient between each feature in the data set and the yield strength of titanium alloy;
[0029] Step 2.1.1.2, select the features with the absolute value of the Pearson correlation coefficient greater than 0.5 and determine them as the features that are highly correlated with the yield strength;
[0030] Step 2.1.1.3, selecting 38 characteristic parameters mainly based on V alloy composition from step 2.1.1.2, and constructing a characteristic set of yield strength performance prediction model;
[0031] Step 2.1.2, for the tensile strength of titanium alloy, a correlation coefficient analysis is performed to determine 44 characteristic parameters with Fe alloy composition as the main characteristic as the characteristic set of the tensile strength performance prediction model;
[0032] Step 2.1.2.1, calculate the Pearson correlation coefficient between each feature in the data set and the tensile strength of titanium alloy;
[0033] Step 2.1.2.2, select the features with the absolute value of the Pearson correlation coefficient greater than 0.5 and determine them as the features that are highly correlated with the tensile strength;
[0034] Step 2.1.2.3, selecting 44 characteristic parameters mainly based on Fe alloy components from step 2.1.2.2 to construct a characteristic set of tensile strength performance prediction model;
[0035] Step 2.1.3, integrating the feature sets obtained in step 2.1.1 and step 2.1.2 to form two independent feature data sets, which are used for subsequent yield strength and tensile strength performance prediction model training respectively;
[0036] Step 2.2: Divide the feature data set obtained in step 2.1 into a training set and a test set in a ratio of 9:1;
[0037] Step 2.3: Construct a set of base learners, consisting of the following three parts: a random forest regressor with 50 trees, an XGBoost regressor with 50 iterations, and a gradient boosting regressor with 550 iterations and a maximum depth of 5. The predictions generated by the base learners will be used to construct the meta-feature matrix.
[0038] Step 2.4: Construct a meta-learner, use the linear regression algorithm as the meta-learner, and input the meta-features generated by the base learner in step 2.3;
[0039] Step 2.5: Construct a "Superlearner" model, which is composed of the base learner in step 2.3 and the meta learner in step 2.4;
[0040] Step 2.6: Choose from a variety of machine learning algorithms for model training, including random forest, XGBoost, gradient boosting regression, decision tree, support vector machine, multivariate linear regression, ridge regression with kernel trick, and “Superlearner” model;
[0041] Step 2.7: Evaluate the prediction accuracy and stability of different models by leave-one-out cross-validation method and obtain the R after prediction of each model. 2, MAEf, RMSE and EVS values, compare the performance of each model on the above evaluation indicators, draw radar charts and table views to intuitively display the performance of different models. For the yield strength performance prediction model, the "Superlearner" model with the best performance in all indicators is selected to predict the yield strength performance of the alloy material. For the tensile strength performance prediction model, the support vector machine (SVR) model with the best performance in all indicators is selected to predict the tensile strength performance of the alloy material;
[0042] Step 2.8: Conduct further experimental verification and evaluation on the model determined in step 2.7, optimize the model based on the evaluation results, and adjust the parameters to improve the model performance.
[0043] Technical issues solved in steps 2.1-2.8, results:
[0044] 1. The problem of how to select features when predicting the performance of titanium alloy materials is solved, and the feature set that is highly correlated with different performance indicators is screened out by combining Pearson correlation coefficient analysis.
[0045] 2. It solves the problem of how to comprehensively evaluate the prediction performance of the model. It uses a variety of machine learning algorithms to train the model, which can comprehensively evaluate the performance of each algorithm on different performance indicators, so as to select the optimal algorithm. This multi-algorithm training method can identify patterns and relationships that a single algorithm may ignore, and improve the overall prediction performance and stability of the model.
[0046] 3. Solved the problem of how to objectively evaluate and improve the prediction performance of the model, established the evaluation indicators for the performance prediction model of titanium alloy materials, and used R 2 Four evaluation indicators, MAE, RMSE and EVS, can evaluate the prediction accuracy, stability and deviation of the model from different angles. At the same time, combined with the table view and radar chart, a comprehensive understanding of the model performance can be obtained. These indicators help identify the advantages and disadvantages of the model, guide the selection of the optimal prediction model, and ensure the reliability and effectiveness of the model in practical applications.
[0047] The above step 3 specifically includes the following steps:
[0048] Step 3.1: Construct a scatter plot, use principal component analysis to reduce the dimension of the feature data set, and then use the K-means algorithm to cluster it into three different clusters. Each cluster represents a data set with similar properties, and each point represents a data sample. The similarity between the scattered points is measured by the distance in physical position;
[0049] Step 3.2: Customize the scatter point style based on step 3.1. Each scatter point consists of four parts. The 38 gray bars in the outermost layer represent the features most related to yield strength, and the 44 black bars in the middle layer represent the features most related to tensile strength. The length of the bars in these two parts represent the magnitude of the values. The two arc blocks below the outermost layer represent the yield strength and tensile strength, and the length and thickness represent the relative size. The center circle shows the titanium alloy chemical formula of the current scatter point, and pink, green and blue are used to represent its different categories.
[0050] Step 3.3: Based on Step 3.1 and Step 3.2, click on the scatter point to display the custom scatter point card. Each card contains two parts. One is a bar chart with different colors to represent the characteristics most related to yield strength and tensile strength. The other is to display the chemical formula of the titanium alloy represented by the current card and the specific parameter values of the processing technology that affect the performance. Cards of different colors represent different clusters, which is convenient for comprehensive evaluation and horizontal and vertical comparison.
[0051] Step 3.4: Create a line graph and classify the performance data into three categories. Use a blue line to represent the yield strength and a yellow line to represent the tensile strength. Divide the values into three intervals: [0, 800], [800, 1600], and [1600, 2600], which correspond to the three end nodes 0, 1, and 2 respectively. The end points of the lines correspond to the value ranges to which they belong.
[0052] Step 3.5: Based on steps 3.1, 3.2 and 3.4, the scatter point zoom style is displayed when hovering over the scatter point and two corresponding broken lines are marked in red in the plane coordinate graph;
[0053] Step 3.6: Make a table of specific values of the four evaluation indicators of R2, MAEf, EVS and RMSE for the machine learning model used in step 2.3, for intuitive display and comparison of model effects;
[0054] Step 3.7: Draw a radar chart based on step 3.6. The four vertices of the radar chart represent the four evaluation indicators in step 3.6. The projection axis from the center to the four vertices represents the scale from 0 to 1. To maintain consistency, the vertex MAEf and RMSE are projected with 1 minus the original value. Different colors are used to represent different models. Each model is an irregular quadrilateral. The closer it is to a complete quadrilateral, the better the model effect.
[0055] Technical issues solved in steps 3.1-3.7, results:
[0056] 1. Solved the problem of how to reduce the dimension and cluster the feature data set. Use principal component analysis (PCA) to reduce the dimension of the data set, simplify the data structure, and divide the data into three clusters through the K-means algorithm, so that data samples with similar properties are clustered together.
[0057] 2. The problem of how to intuitively display the characteristics related to yield strength and tensile strength is solved. By customizing the scatter point style, the correlation and numerical size of each characteristic and performance indicator are displayed, providing intuitive visual information, while finding the commonalities and differences between different scatter points based on their distribution.
[0058] 3. The problem of how to display the specific information of each data sample in detail and intuitively is solved. Through custom scatter cards, the characteristics related to performance indicators, titanium alloy chemical formula and processing parameters are displayed to facilitate comprehensive evaluation and comparison.
[0059] 4. The problem of how to classify and visualize performance data is solved. By designing a line graph, the yield strength and tensile strength data are classified into three categories and displayed through color and numerical range. At the same time, the line graph will be linked with the scatter graph to enhance the correlation and visualization effect between the data.
[0060] 5. Solved the problem of how to comprehensively evaluate and compare the performance of different models, using tables to intuitively display the performance of various machine learning models in R 2 , MAE, EVS and RMSE indicators, which is convenient for comparing model effects; the performance of each model on the four evaluation indicators is displayed through a radar chart. The closer the model is to a complete quadrilateral, the better the effect, providing a clear visual comparison.
[0061] The above step 4 specifically includes the following steps:
[0062] Step 4.1: Input new titanium alloy composition and processing parameters, use the model trained in step 2 to predict performance, and obtain prediction results;
[0063] Step 4.2: Generate new scatter plots and line graphs through the multidimensional visualization analysis tool built in step 3. Combine the performance prediction value intervals in the line graph, compare the scattered points with similar performance in the same cluster in the scatter plot, find out the features that are positively or negatively correlated with the performance, and thus obtain an optimization plan that can improve the performance.
[0064] Step 4.3: Based on the optimization solution obtained in step 4.2, try to increase the components that are positively correlated with the performance or reduce the components that are negatively correlated to form a new combination;
[0065] Step 4.4: Use the new combination obtained in step 4.3 as the input of step 4.1 to obtain a new prediction result, and compare it with the original prediction result. According to the comparison result, analyze and evaluate the effectiveness of the improvement plan, repeatedly adjust and optimize the titanium alloy composition and processing parameters, and perform multiple iterations to finally determine the best titanium alloy composition and processing technology combination to achieve the expected performance indicators.
[0066] Technical issues solved in steps 4.1-4.4, results:
[0067] 1. The problem of how to predict the performance of titanium alloy materials was solved. The performance prediction results were analyzed through customized visualization tools. With the help of generated scatter plots and line graphs, by comparing the scatter points with similar performance in the same cluster, the characteristics that have a positive or negative impact on the performance were found, and optimization solutions were proposed.
[0068] 2. The problem of how to optimize the performance of titanium alloy materials was solved. According to the positive and negative correlation characteristics of the performance, the titanium alloy composition and processing parameters were adjusted to form a new combination. The new combination was re-input into the model to obtain new prediction results, which were compared with the original prediction results. The effectiveness of the improvement plan was continuously analyzed and evaluated, and adjustments were made repeatedly to achieve the expected performance indicators.
[0069] In summary, the present invention has the following advantages over the prior art:
[0070] 1. Data processing and visualization tools specifically for titanium alloys: The present invention provides a set of data processing and visualization tools specifically for titanium alloys, which can intuitively display complex multidimensional data relationships. Traditional data visualization tools often lack customized designs for specific materials (such as titanium alloys), resulting in poor results in displaying the complex relationship between titanium alloy composition, processing technology and performance. The tools in the present invention, such as scatter plots, heat maps and parallel coordinate graphs, are all designed based on the characteristics of titanium alloys, and can better display the relationship between its composition, processing technology and performance. For example, scatter plots can distinguish different titanium alloy compositions and processing technologies through markers of different colors and shapes, heat maps can display the thermal distribution of correlations between different features, and parallel coordinate graphs can display data features from multiple dimensions at the same time. The use of these tools makes complex data relationships intuitive and easy to understand, greatly improving the user's data analysis efficiency.
[0071] 2. Efficient model training and evaluation methods: The present invention adopts efficient model training and evaluation methods, which improve the prediction accuracy and stability of the model. In the model training stage, the present invention uses a variety of advanced machine learning algorithms, such as random forest, XGBoost, gradient boosting regression, etc., which can effectively learn complex patterns from data. In the model evaluation stage, the present invention adopts the leave-one-out cross-validation method, which evaluates the prediction accuracy and stability of the model by taking each sample as a validation set in turn and the remaining samples as training sets. This evaluation method can comprehensively evaluate the performance of the model and avoid the problems of overfitting and underfitting. Through this efficient model training and evaluation method, the present invention can obtain a model with high prediction accuracy and stability.
[0072] 3. Comprehensive data processing flow: The present invention adopts a comprehensive data processing flow, which ensures the quality of the data and the accuracy of the analysis. First, in the data collection stage, the present invention collected a large amount of titanium alloy data from multiple sources, including composition, processing technology and performance data. Then, in the data preprocessing stage, the present invention cleaned and normalized the collected data to ensure the quality of the data. Next, in the feature selection stage, the present invention selected features with a high correlation with performance indicators through the Pearson correlation coefficient. Finally, in the model training and evaluation stage, the present invention used a variety of machine learning algorithms for model training, and evaluated the prediction accuracy and stability of the model through the leave-one-out cross-validation method. This comprehensive data processing flow ensures the quality of the data and the accuracy of the analysis.
[0073] 4. Performance prediction and optimization analysis: The present invention can not only perform performance prediction based on the trained model, but also find the best combination of composition and processing technology through multidimensional visualization tools and optimization algorithms. In the performance prediction stage, the user can use the trained model to predict performance by inputting new titanium alloy composition and processing parameters. Then, in the optimization analysis stage, the user can generate new scatter plots and line graphs through multidimensional visualization tools, combine the interval to which the performance prediction values in the line graph belong, compare the scatter points with similar performance in the same cluster in the scatter plot, and find out the features that are positively or negatively correlated with the performance. Based on this information, the user can try to increase the components that are positively correlated with the performance or reduce the negatively correlated components to form a new combination, and re-perform performance prediction and comparative analysis. Through multiple iterations, the user can finally determine the best combination of titanium alloy composition and processing technology to achieve the expected performance indicators. This process not only improves the interactivity and operability of the analysis, but also provides a scientific basis for improving the application performance of titanium alloys.
Claims
1. A multidimensional visualization analysis method based on titanium alloy composition and processing technology, characterized in that: The following steps are involved: Step 1, data collection and preprocessing: After collecting titanium alloy data from multiple sources, the data is cleaned and normalized to construct a high-quality dataset suitable for modeling. Step 2: Model training and evaluation: Extract effective features, use machine learning algorithms to train the model on the feature data, and evaluate and optimize the model. Step 3, multi-dimensional visualization: Use the independently designed multi-dimensional visualization tool to display the titanium alloy composition, processing technology and performance data in multiple dimensions to help users intuitively analyze data relationships. Step 4: Performance prediction and optimization analysis: Based on the trained model and multi-dimensional visualization tools, the performance of titanium alloys is predicted and optimized to find the best combination of titanium alloy composition and processing technology.
2. The multi-dimensional visualization analysis method based on titanium alloy composition and processing technology according to claim 1 is characterized in that: Step 1 specifically includes the following steps: Step 1.1: Query and collect the composition, processing technology and performance data of titanium alloy from literature, experiments and Matminer database, and organize and construct a data set suitable for modeling. Step 1.2: Clean the constructed data set, including removing noise, processing missing values and outliers, to ensure the accuracy of the data; Step 1.3: Normalize the data set to the interval (0, 1) to make data of different dimensions comparable.
3. The multi-dimensional visualization analysis method based on titanium alloy composition and processing technology according to claim 1 is characterized in that: Step 2 specifically includes the following steps: Step 2.1, feature selection: for the data set obtained after preprocessing in step 1, use the Pearson correlation coefficient to select features with high correlation with the performance index to obtain a feature data set; Step 2.2: Divide the feature data set obtained in step 2.1 into a training set and a test set in a ratio of 9:1; Step 2.3: Construct a set of base learners, consisting of the following three parts: a random forest regressor with 50 trees, an XGBoost regressor with 50 iterations, and a gradient boosting regressor with 550 iterations and a maximum depth of 5. The predictions generated by the base learners will be used to construct the meta-feature matrix. Step 2.4: Construct a meta-learner, use the linear regression algorithm as the meta-learner, and input the meta-features generated by the base learner in step 2.3; Step 2.5: Construct a "Superlearner" model, which is composed of the base learner in step 2.3 and the meta learner in step 2.4; Step 2.6: Choose a variety of machine learning algorithms for model training, including random forest, XGBoost, gradient boosting regression, decision tree, support vector machine, multivariate linear regression, ridge regression based on kernel technique, and "Superlearner" model; Step 2.7: Evaluate the prediction accuracy and stability of different models through leave-one-out cross-validation method and obtain the R after prediction of each model. 2 , MAEf, RMSE and EVS values, compare the performance of each model on the above evaluation indicators, draw radar charts and table views to intuitively display the performance of different models. For the yield strength performance prediction model, the "Superlearner" model with the best performance in all indicators is selected to predict the yield strength performance of the alloy material. For the tensile strength performance prediction model, the support vector machine model with the best performance in all indicators is selected to predict the tensile strength performance of the alloy material; Step 2.8: Conduct further experimental verification and evaluation on the model determined in step 2.7, optimize the model based on the evaluation results, and adjust the parameters to improve the model performance.
4. The multi-dimensional visualization analysis method based on titanium alloy composition and processing technology according to claim 3 is characterized in that: Step 2.1 specifically includes the following steps: Step 2.1.1, for the yield strength of titanium alloy, a correlation coefficient analysis is performed to determine 38 characteristic parameters with V alloy composition as the main characteristic as the characteristic set of the yield strength performance prediction model; Step 2.1.1.1, calculate the Pearson correlation coefficient between each feature in the data set and the yield strength of titanium alloy; Step 2.1.1.2, select the features with the absolute value of the Pearson correlation coefficient greater than 0.5 and determine them as the features that are highly correlated with the yield strength; Step 2.1.1.3, selecting 38 characteristic parameters mainly based on V alloy composition from step 2.1.1.2, and constructing a characteristic set of yield strength performance prediction model; Step 2.1.2, for the tensile strength of titanium alloy, a correlation coefficient analysis is performed to determine 44 characteristic parameters with Fe alloy composition as the main characteristic as the characteristic set of the tensile strength performance prediction model; Step 2.1.2.1, calculate the Pearson correlation coefficient between each feature in the data set and the tensile strength of titanium alloy; Step 2.1.2.2, select the features with the absolute value of the Pearson correlation coefficient greater than 0.5 and determine them as the features that are highly correlated with the tensile strength; Step 2.1.2.3, selecting 44 characteristic parameters mainly based on Fe alloy components from step 2.1.2.2 to construct a characteristic set of tensile strength performance prediction model; Step 2.1.3: Integrate the feature sets obtained in step 2.1.1 and step 2.1.2 to form two independent feature data sets, which are used for subsequent yield strength and tensile strength performance prediction model training respectively.
5. The multi-dimensional visualization analysis method based on titanium alloy composition and processing technology according to claim 1 is characterized in that: Step 3 specifically includes the following steps: Step 3.1: Construct a scatter plot, use principal component analysis to reduce the dimension of the feature data set, and then use the K-means algorithm to cluster it into three different clusters. Each cluster represents a data set with similar properties, and each point represents a data sample. The similarity between the scattered points is measured by the distance in physical position; Step 3.2: Customize the scatter point style based on step 3.
1. Each scatter point consists of four parts. The 38 gray bars in the outermost layer represent the features most related to yield strength, and the 44 black bars in the middle layer represent the features most related to tensile strength. The length of the bars in these two parts represent the magnitude of the values. The two arc blocks below the outermost layer represent the yield strength and tensile strength, and the length and thickness represent the relative size. The center circle shows the titanium alloy chemical formula of the current scatter point, and pink, green and blue are used to represent its different categories. Step 3.3: Based on Step 3.1 and Step 3.2, click on the scatter point to display the custom scatter point card. Each card contains two parts. One is a bar chart with different colors to represent the characteristics most related to yield strength and tensile strength. The other is to display the chemical formula of the titanium alloy represented by the current card and the specific parameter values of the processing technology that affect the performance. Cards of different colors represent different clusters, which is convenient for comprehensive evaluation and horizontal and vertical comparison. Step 3.4: Create a line graph and classify the performance data into three categories. Use a blue line to represent the yield strength and a yellow line to represent the tensile strength. Divide the values into three intervals: [0, 800], [800, 1600], and [1600, 2600], which correspond to the three end nodes 0, 1, and 2 respectively. The end points of the lines correspond to the value ranges to which they belong. Step 3.5: Based on steps 3.1, 3.2 and 3.4, the scatter point zoom style is displayed when hovering over the scatter point and two corresponding broken lines are marked in red in the plane coordinate graph; Step 3.6: Make a table of specific values of the four evaluation indicators of R2, MAEf, EVS and RMSE for the machine learning model used in step 2.3, for intuitive display and comparison of model effects; Step 3.7: Draw a radar chart based on step 3.
6. The four vertices of the radar chart represent the four evaluation indicators in step 3.
6. The projection axis from the center to the four vertices represents the scale from 0 to 1. To maintain consistency, the vertex MAEf and RMSE are projected with 1 minus the original value. Different colors are used to represent different models. Each model is an irregular quadrilateral. The closer it is to a complete quadrilateral, the better the model effect.
6. The multi-dimensional visualization analysis method based on titanium alloy composition and processing technology according to claim 1 is characterized in that: Step 4 specifically includes the following steps: Step 4.1: Input new titanium alloy composition and processing parameters, use the model trained in step 2 to predict performance, and obtain prediction results; Step 4.2: Generate new scatter plots and line graphs through the multidimensional visualization analysis tool built in step 3. Combine the performance prediction value intervals in the line graph, compare the scattered points with similar performance in the same cluster in the scatter plot, find out the features that are positively or negatively correlated with the performance, and thus obtain an optimization plan that can improve the performance. Step 4.3: Based on the optimization solution obtained in step 4.2, try to increase the components that are positively correlated with the performance or reduce the components that are negatively correlated to form a new combination; Step 4.4: Use the new combination obtained in step 4.3 as the input of step 4.1 to obtain a new prediction result, and compare it with the original prediction result. According to the comparison result, analyze and evaluate the effectiveness of the improvement plan, repeatedly adjust and optimize the titanium alloy composition and processing parameters, and perform multiple iterations to finally determine the best titanium alloy composition and processing technology combination to achieve the expected performance indicators.
Citation Information
Patent Citations
Multi-feature thin plate part quality prediction and process parameter optimization based on machine learning
CN113176761A
Visual analysis method and system based on industrial big data model
CN114510519A
Magnesium alloy mechanical property prediction method, device and equipment and storage medium
CN115527636A
Titanium alloy mechanical property prediction method and device based on machine learning
CN116230136A
Method for screening high-hardness and high-entropy alloy through particle swarm optimization algorithm assisted machine learning
CN116312890A
