Prediction method for apparent density of hard alloy spray-dried mixture
Through the prediction method based on machine learning, using process parameters as characteristics to build an optimized model, the problem of difficult to predict the loose density of the cemented carbide spray-dried mixture is solved, efficient and accurate prediction results are achieved, and the level of material performance evaluation and process optimization is improved.
Patent Information
- Application Number
- CN202411744191.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-30
- Publication Date
- 2025-03-28
AI Technical Summary
The prior art is difficult to accurately predict the loose density of cemented carbide spray-dried mixture, which affects material performance evaluation and process optimization.
Using a machine learning-based prediction method, a model is constructed using a variety of machine learning algorithms by selecting process parameters as features, and the features and models are optimized to predict the loose density of the carbide spray-dried mixture.
It realizes efficient and accurate prediction of the loose density of cemented carbide spray-dried mixture, and improves the accuracy of material performance evaluation and process optimization efficiency.
Smart Images
Figure CN119724440A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the field of prediction technology, in particular to a method for predicting the loose density of a cemented carbide spray-dried mixture. Background Art
[0002] Cemented carbide spray-dried mixture refers to the cemented carbide mixture prepared by spray drying process. The bulk density of cemented carbide spray-dried mixture refers to the mass per unit volume of the mixture powder in the natural stacking state. It reflects the filling degree and porosity between powder particles and is an important indicator for evaluating powder fluidity.
[0003] With the development of artificial intelligence, many researchers have proposed applying machine learning methods to basic disciplines, and they are increasingly being used to scientifically predict material properties. Using different machine learning methods to predict material properties on the large amount of material data sets currently generated can effectively improve the accuracy of material property prediction, save computing costs, and provide guidance for material experiments and applications. Summary of the invention
[0004] The present invention proposes a prediction method for the bulk density of cemented carbide spray-dried mixture based on machine learning and its application. On the basis of the knowledge of physical science, process parameters are selected as machine learning features to efficiently and accurately predict the bulk density of cemented carbide spray-dried mixture, and multiple process parameters can be used as features to solve complex machine learning problems. This method uses process parameters related to the cemented carbide spray drying process as machine learning features, optimizes the features, and constructs a model using multiple machine learning algorithms, and predicts the bulk density of the mixture based on the optimized features and the optimal model.
[0005] In order to achieve the above object, the present invention adopts the following technical solutions:
[0006] A method for predicting the bulk density of a cemented carbide spray-dried mixture comprises the following steps:
[0007] Step 100: Obtain sample data and optimize it through preprocessing;
[0008] Step 200: Obtain a feature subset for modeling based on the preprocessed data, and divide it into a training set and a test set;
[0009] Step 300: Perform principal component analysis and dimensionality reduction on the training set and the test set respectively;
[0010] Step 400: constructing a model using multiple machine learning algorithms, and evaluating and selecting the optimal model;
[0011] Step 500: Optimizing the optimal model selected in step 400;
[0012] Step 600: Predicting the bulk density of the cemented carbide spray-dried mixture based on the feature subset and the optimized optimal model.
[0013] Preferably, the sample data in step 100 includes at least the solid-liquid ratio of the cemented carbide mixture, the particle size and weight ratio of tungsten carbide, the particle size and weight ratio of cobalt powder, the particle size and weight ratio of grain inhibitor, the ball-to-material ratio during ball milling, the amount of forming agent added and the loose density of the mixture obtained under various spray drying process conditions.
[0014] Preferably, the spray drying process conditions are related to spray drying factors, and the spray drying factors include at least spray drying inlet temperature, spray drying outlet temperature, spray pressure, dust collection pressure difference or spray port orifice diameter. Different set values of spray drying factors and a combination of multiple spray drying factors form the diversity of spray drying process conditions.
[0015] Preferably, the preprocessing of the sample data includes at least one or more combinations of data set reading, feature extraction, standardization, feature screening and deletion of missing values. The method for constructing the data set includes the following operations: obtaining sample data, collecting corresponding labels of experimental data, constructing correlation factors between labels, and constructing a data set based on the labels and correlation factors.
[0016] Preferably, in step 200, obtaining the feature subset includes the following steps:
[0017] Step 201: selecting at least the solid-liquid ratio of the cemented carbide mixture, the particle size and weight ratio of tungsten carbide, the particle size and weight ratio of cobalt powder, the particle size and weight ratio of grain inhibitor, the ball-to-material ratio in the ball milling process, the amount of forming agent added and the spray drying process conditions as machine learning features;
[0018] Step 202: Perform Pearson correlation coefficient analysis on the features in the data set to find features that are highly correlated with the fracture toughness of cemented carbide as feature subsets for modeling.
[0019] Preferably, in step 200, 70% of the data in the feature subset is selected as a training set, and 30% of the data is selected as a validation set.
[0020] Preferably, in step 400, the model constructed by the machine learning algorithm includes at least one or more of an XGB model, an RF model, a gradient boosting model and a Bagging model.
[0021] Preferably, in step 400, the calculation results of the model constructed by the machine learning algorithm are evaluated using mean square error, mean absolute error and determination coefficient, and the prediction accuracy of the model constructed by each machine learning algorithm is verified by five-fold cross validation to obtain the optimal model.
[0022] Preferably, in step 500, optimizing the optimal model includes at least one of the following operations: introducing an adaptive learning rate mechanism into the optimal model or introducing a grid search to optimize the parameter space of the optimal model.
[0023] Preferably, the method for predicting the loose density of cemented carbide mixture also includes the following operations: ranking the importance scores of characteristic quantities according to the optimized optimal model; in step 600, designing the process parameters of the cemented carbide mixture based on the characteristic importance scores and the visualized data scatter plot.
[0024] The beneficial effects of the present invention are:
[0025] 1. The present invention provides a method for predicting the loose density of cemented carbide mixtures based on machine learning, and applies the method for predicting the loose density of cemented carbide mixtures. This method uses raw materials and process conditions associated with cemented carbide mixtures as machine learning features, optimizes corresponding features, constructs models with multiple machine learning algorithms, selects the optimal model, and predicts the loose density of the mixture based on the optimized features and the optimal model. At the same time, the importance analysis of feature quantities can be performed based on the optimal model obtained by the method for predicting the loose density of cemented carbide mixtures to determine the factors affecting the loose density, including but not limited to the solid-liquid ratio of the carbide mixture, the particle size and weight ratio of tungsten carbide, the particle size and weight ratio of cobalt powder, the particle size and weight ratio of grain inhibitors, the ball-to-material ratio during ball milling, the amount of forming agent added, and multiple spray drying process conditions. The process parameters of the cemented carbide mixture are designed based on the feature quantity importance score and the visualized data scatter plot.
[0026] 2. This method for predicting the loose density of cemented carbide mixture uses experimental samples from the experimental process and literature as feature sets to establish an efficient prediction model. Combined with the strategy of machine learning prediction performance, it finds the process parameters that have a greater impact on the loose density of cemented carbide mixture. By setting the process parameters, the loose density of the mixture is set, which solves the problem that the loose density of the mixture cannot be quantified. It has the advantages of high prediction accuracy, low cost, and green environmental protection.
[0027] 3. The method for predicting the bulk density of cemented carbide mixtures performs PCA dimensionality reduction on the feature subsets before training and testing. By identifying the principal components in the data, the high-dimensional data set is mapped to a low-dimensional space, thereby reducing the complexity of data processing and the computational cost. BRIEF DESCRIPTION OF THE DRAWINGS
[0028] Figure 1 Schematic diagram of the steps of the method for predicting the bulk density of cemented carbide spray-dried mixture. DETAILED DESCRIPTION
[0029] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.
[0030] refer to Figure 1 A method for predicting the bulk density of a cemented carbide spray-dried mixture comprises the following steps:
[0031] Step 100: Obtain sample data and optimize it through preprocessing;
[0032] Step 200: Obtain a feature subset for modeling based on the preprocessed data, and divide it into a training set and a test set;
[0033] Step 300: Perform principal component analysis and dimensionality reduction on the training set and the test set respectively;
[0034] Step 400: constructing a model using multiple machine learning algorithms, and evaluating and selecting the optimal model;
[0035] Step 500: Optimizing the optimal model selected in step 300;
[0036] Step 600: Predicting the bulk density of the cemented carbide spray-dried mixture based on the feature subset and the optimized optimal model.
[0037] Specifically, in step 100, sample documents are obtained, corresponding labels of experimental data are collected, correlation factors between labels are constructed, and a data set is constructed based on the labels and correlation factors. The sample data of the above data set are data related to the preparation of cemented carbide mixtures collected in the experimental process and literature, including the solid-liquid ratio of cemented carbide mixtures, the particle size and weight ratio of tungsten carbide, the particle size and weight ratio of cobalt powder, the particle size and weight ratio of grain inhibitors, the ball-to-material ratio in the ball milling process, the amount of forming agent added, and the bulk density of the mixture under different spray drying process conditions.
[0038] The above preprocessing includes operations based on data set reading, feature extraction, standardization, feature screening, and deletion of sample data with missing values. After reading the data, outliers are deleted, and then standardized. The mean and standard deviation of each feature are calculated. The mean represents the center position of the data, and the standard deviation represents the dispersion of the data. After standardization, the value of each feature will have zero mean and unit variance, which helps to eliminate the dimensional differences between features, making the weights of different features more balanced and better adapted to machine learning algorithms.
[0039] Most of the characteristic quantities of the methods for predicting material properties use components and structural factors, while the embodiments of the present invention select preparation parameters as characteristic quantities. In step 200, the preparation parameters of cemented carbide mixture are used as characteristics; the raw material composition related to cemented carbide mixture, the solid-liquid ratio of cemented carbide mixture, the particle size and weight ratio of tungsten carbide, the particle size and weight ratio of cobalt powder, the particle size and weight ratio of grain inhibitor, the ball-to-material ratio in the ball milling process, the amount of forming agent added and different spray drying process conditions are analyzed and selected. The above-mentioned spray drying process conditions at least include the spray drying inlet temperature, the spray drying outlet temperature, the spray pressure, the dust collection pressure difference and the spray port orifice diameter.
[0040] The complete data set is subjected to Pearson correlation coefficient analysis to find features with strong correlation with cemented carbide mixtures as feature subsets for modeling. The above Pearson correlation coefficient is a statistic used to measure the degree of linear correlation between two continuous variables. It measures the strength and direction of the linear relationship between the two variables. The value range of the Pearson correlation coefficient is between -1 and 1. Among them: when the correlation coefficient is close to 1, it indicates that there is a strong positive correlation, that is, the two variables show a positive linear relationship, when one variable increases, the other variable also increases; when the correlation coefficient is close to -1, it indicates that there is a strong negative correlation, that is, the two variables show a negative linear relationship, when one variable increases, the other variable decreases; when the correlation coefficient is close to 0, it indicates that there is no linear relationship between the two variables, or there is a nonlinear relationship.
[0041] The obtained feature subset samples are randomly divided into training set and test set in a ratio of 7:3, with 70% of the feature subset samples used as training set and 30% of the feature subset samples used as test set.
[0042] In machine learning, we often encounter data that may have high dimensional characteristics. When the dimension of the data to be processed is too high, our analysis and understanding of the data will become more and more complicated. This phenomenon is called the curse of dimensionality. Principal Component Analysis (PCA) is a data analysis technique in machine learning, which is mainly used for dimensionality reduction and generalization of high-dimensional data. Its core goal is to retain the information content of the original data to the greatest extent while reducing the data dimension, which is of great significance for improving the efficiency and accuracy of data processing. Solving the PCA problem is of great significance. First, it allows us to effectively reduce the dimensionality of data, overcome the "curse of dimensionality", and enable applications such as machine learning algorithms to run better. Secondly, PCA can be used for data visualization. By reducing the dimensionality and representing high-dimensional data in a low-dimensional space, we can observe and understand the data more intuitively.
[0043] Since there are many feature input values for cemented carbide spray-dried mixture, in order to improve the accuracy of the model, in step 300, PCA dimension reduction is performed on the feature subset of cemented carbide spray-dried mixture. Using the PCA algorithm to reduce the dimension of the training set can improve the model training speed, reduce the consumption of computing resources, and try to keep the main information of the data. However, in the process of dimensionality reduction, it is necessary to select a suitable k value (i.e., the number of principal components) to ensure that the data after dimensionality reduction can retain enough information for model training. The test set is subjected to dimensionality reduction to ensure that the test data and the training data are compared and evaluated in the same low-dimensional space. When the test set is subjected to dimensionality reduction, the test set needs to be subjected to dimensionality reduction using the same PCA model (i.e., the same feature vector) as the training set. It should be noted here that the fit_transform method cannot be used to reduce the dimension of the test set, but the transform method should be used. Because the features have been determined when the training set is reduced in dimension, the features extracted from the test set should be consistent with the training set.
[0044] In step 400, in this embodiment, four machine learning algorithms are used to build models, and the four machine learning algorithms include XGB model (eXtreme Gradient Boosting), RF model (Random Forest), gradient boosting model (Gradient Boosting) and Bagging. After the four machine learning algorithms build the models respectively, the mean square error (MSE), mean absolute error (MAE) and determination coefficient (R2) are used to evaluate the results of the four models. The above-mentioned mean square error (MSE) is used to evaluate the average difference between the predicted value and the actual value. The smaller its value is, the smaller the difference between the prediction and the actual value of the model is, and the better the model performance is. The mean absolute error (MAE) is used to evaluate the average difference between the predicted value and the actual value. It can better handle outliers. The smaller its value is, the smaller the difference between the prediction and the actual value of the model is, and the better the model performance is. RR is one of the commonly used performance metrics in regression problems. Its value range is from 0 to 1. The closer it is to 1, the better the model explains the variability of the dependent variable and the better the model performance. When RR is equal to 1, it means that the model can perfectly explain the variability of the dependent variable and the predicted value is completely consistent with the actual value.
[0045] According to the comparison of the evaluation indicators, it is found that the XGB model has the best effect, so the XGB model is optimized in step 500. In this embodiment, the optimization of the XGB model includes the following steps:
[0046] Step 501: Change the adaptive learning rate. The XGB model uses a fixed learning rate to update model parameters. The introduction of an adaptive learning rate mechanism can dynamically adjust the learning rate according to the performance of the model during training to better adapt to different data and model states.
[0047] Step 502: In addition to the learning rate, the parameter space of the XGB model also includes the maximum tree depth, the number of trees, the column sampling ratio, the row sampling ratio and the Gamma parameter. This embodiment introduces grid search in the XGB model to optimize these parameter spaces. Grid search is a method of finding the best parameter combination by exhaustively searching the parameter space. Optimizing the parameter space of the XGB model through grid search can significantly improve the performance and generalization ability of the model.
[0048] In step 600, the accuracy of the prediction is verified by five-fold cross validation. The data set is divided into five subsets of equal size, four of which are used as training data and the remaining subset is used as validation data. Then, the XGB model is evaluated on the validation set, and the above five subsets are used as separate validation sets and need to be evaluated separately. Therefore, the XGB model needs to be evaluated on five subsets separately, and a different subset is selected as the validation set each time until all subsets are evaluated once. Five-fold cross validation reduces the dependence on specific data segmentation and provides more reliable model evaluation results by using the XGB model in combination with different validation sets. Finally, the feature importance score is ranked according to the optimized XGB model to obtain the important factors affecting the target performance.
[0049] The prediction method of the bulk density of cemented carbide spray-dried mixture in this embodiment is a prediction method based on process parameters as features, and the bulk density sample data of cemented carbide mixture is modeled to establish an efficient prediction model of the bulk density of cemented carbide mixture. The prediction model obtained in this embodiment is a prediction model of fracture toughness, and its determination coefficient is 0.91, and the prediction accuracy rate reaches 92%. The consistency between the model prediction value and the actual value of this prediction model is high, and it has high reliability and credibility in practical applications.
[0050] serial number Actual value of bulk density (g / cm3) Predicted bulk density (g / cm3) 1 2.85 2.91 2 2.82 2.9 3 2.81 2.85 4 2.94 2.96 5 3.12 3.05 6 3.21 3.17 7 2.98 2.90 8 2.91 2.96 9 2.94 2.87 10 2.95 2.85 11 2.96 2.99 12 3.02 3.12
[0051] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.
Claims
1. A method for predicting the bulk density of cemented carbide spray-dried mixture, characterized in that: The following steps are involved: Step 100: Obtain sample data and optimize it through preprocessing; Step 200: Obtain a feature subset for modeling based on the preprocessed data, and divide it into a training set and a test set; Step 300: Perform principal component analysis and dimensionality reduction on the training set and the test set respectively; Step 400: constructing a model using multiple machine learning algorithms, and evaluating and selecting the optimal model; Step 500: Optimizing the optimal model selected in step 300; Step 600: Predicting the bulk density of the cemented carbide spray-dried mixture based on the feature subset and the optimized optimal model.
2. The method for predicting the bulk density of cemented carbide spray-dried mixture according to claim 1, characterized in that: The sample data in step 100 at least include the solid-liquid ratio of the cemented carbide mixture, the particle size and weight ratio of tungsten carbide, the particle size and weight ratio of cobalt powder, the particle size and weight ratio of grain inhibitor, the ball-to-material ratio during ball milling, the amount of forming agent added and the loose density of the mixture obtained under various spray drying process conditions.
3. The method for predicting the bulk density of cemented carbide spray-dried mixture according to claim 2, characterized in that: The spray drying process conditions are related to spray drying factors, which include at least spray drying inlet temperature, spray drying outlet temperature, spray pressure, dust collection pressure difference or spray port orifice diameter. Different set values of spray drying factors and a combination of multiple spray drying factors form the diversity of spray drying process conditions.
4. The method for predicting the bulk density of cemented carbide spray-dried mixture according to claim 2 or 3, characterized in that: The preprocessing of the sample data at least includes one or more combinations of data set reading, feature extraction, standardization, feature screening and deletion of missing values; The method for constructing the data set includes the following operations: acquiring sample data, collecting corresponding labels of experimental data, constructing correlation factors between labels, and constructing the data set based on the labels and the correlation factors.
5. The method for predicting the bulk density of cemented carbide spray-dried mixture according to claim 4, characterized in that: In step 200, the acquisition of the feature subset includes the following steps: Step 201: selecting at least the solid-liquid ratio of the cemented carbide mixture, the particle size and weight ratio of tungsten carbide, the particle size and weight ratio of cobalt powder, the particle size and weight ratio of grain inhibitor, the ball-to-material ratio in the ball milling process, the amount of forming agent added and the spray drying process conditions as machine learning features; Step 202: Perform Pearson correlation coefficient analysis on the features in the data set to find features that are highly correlated with the fracture toughness of cemented carbide as feature subsets for modeling.
6. The method for predicting the bulk density of cemented carbide spray-dried mixture according to claim 5, characterized in that: In step 200, 70% of the data in the feature subset is selected as a training set, and 30% of the data is selected as a validation set.
7. The method for predicting the bulk density of cemented carbide spray-dried mixture according to claim 5, characterized in that: In step 400, the model constructed by the machine learning algorithm includes at least one or more of an XGB model, a RF model, a gradient boosting model and a Bagging model.
8. The method for predicting the bulk density of cemented carbide spray-dried mixture according to claim 7, characterized in that: In step 400, the calculation results of the model constructed by the machine learning algorithm are evaluated using mean square error, mean absolute error and determination coefficient, and the prediction accuracy of the model constructed by each machine learning algorithm is verified through five-fold cross validation to obtain the optimal model.
9. The method for predicting the bulk density of cemented carbide spray-dried mixture according to claim 8, characterized in that: In step 500, optimizing the optimal model includes at least one of the following operations: introducing an adaptive learning rate matrix into the optimal model or introducing a grid search to optimize the parameter space of the optimal model.
10. The method for predicting the bulk density of cemented carbide spray-dried mixture according to claim 9, characterized in that: The following operations are also included: ranking the importance scores of feature quantities according to the optimized optimal model; In step 600, sintering process parameters of the cemented carbide mixture are designed based on the feature quantity importance score and the visualized data scatter plot.