A Machine Learning-Based Coating Performance Prediction Method

Abstract

This invention discloses a machine learning-based method for predicting coating performance, comprising the following steps: 1) acquiring parameters; performing normalization processing to obtain the original dataset; and selecting key features to obtain a key feature dataset; 2) dividing the dataset into four data subsets; 3) training and testing each of the four data subsets using n machine learning algorithms to obtain the optimal physical feature prediction sub-model A, and the optimal performance prediction sub-models B, C, and D; 4) constructing a Stacking ensemble model to predict coating performance. The prediction method of this invention, by selecting the optimal algorithm for modeling and predicting coating performance based on preparation process parameters, possesses high accuracy and generalization ability, achieving precise prediction of coating performance.

Description

Technical Field

[0001] This invention relates to the field of coating technology, and in particular to a method for predicting coating performance based on machine learning. Background Technology

[0002] Coating technology involves applying one or more thin films with special properties to the surface of a product substrate, thereby endowing the product with characteristics such as high hardness, low friction, high temperature resistance, and oxidation resistance, significantly improving product performance and service life. There are many processes for preparing coatings, mainly including physical vapor deposition (PVD), chemical vapor deposition (CVD), and thermal spraying.

[0003] Different preparation process parameters significantly affect the microstructure and properties of coatings, thus influencing their performance in various application scenarios. For a long time, the traditional trial-and-error method, relying on repeated experiments and parameter adjustments, has dominated process parameter optimization. However, this method requires numerous repetitive experiments, consuming vast amounts of raw materials, energy, equipment, and manpower, significantly increasing product development costs. In the context of the booming development of intelligent manufacturing, applying machine learning to process parameter determination, through the establishment of complex mathematical models, can quickly predict coating performance under different parameter combinations. Existing research mainly uses single machine learning algorithms to predict coating performance from preparation processes, lacking universality and reliability, and exhibiting low prediction accuracy. Therefore, it is necessary to develop a machine learning-based method for high-precision prediction of coating performance. Summary of the Invention

[0004] In view of the above-mentioned defects or deficiencies in the existing technology, it is desirable to provide a coating performance prediction method based on machine learning. By selecting the optimal algorithm for modeling, the coating performance is predicted according to the preparation process parameters. It has high accuracy and generalization ability, and achieves accurate prediction of coating performance.

[0005] The present invention provides a coating performance prediction method based on machine learning, comprising the following steps:

[0006] 1) Obtain the preparation process parameters, physical characteristics, and performance-related data of the coating; normalize the above data to obtain the original dataset; use performance as the output feature to screen key features of the preparation process parameters and physical characteristics to obtain the key feature dataset;

[0007] 2) Divide the key feature dataset into four data subsets:

[0008] Using the preparation process parameters as input and the physical characteristics as output, a first data subset is constructed.

[0009] Using the preparation process parameters as input and the performance as output, a second data subset is constructed.

[0010] Using physical characteristics as input and performance as output, a third data subset is constructed.

[0011] Using the preparation process parameters and physical characteristics as inputs and the performance as outputs, a fourth data subset is constructed.

[0012] 3) Divide the four data subsets into training and testing sets according to a preset ratio, and use no less than 4 n machine learning algorithms to train and test them one by one;

[0013] After training and testing on the first subset of data, n sub-models A are obtained. The one with the highest prediction accuracy is selected as the best physical feature prediction sub-model A.

[0014] n sub-models B are obtained through training and testing on the second subset of data; n sub-models C are obtained through training and testing on the third subset of data; and n sub-models D are obtained through training and testing on the fourth subset of data. The optimal combination of these sub-models with high prediction accuracy and different machine learning algorithms is selected to obtain the best-performing prediction sub-models B, C, and D.

[0015] 4) Using the preparation process parameters as input, the predicted physical features are obtained through the optimal physical feature prediction sub-model A; using the preparation process parameters and the predicted physical features as input, and the performance as output, a dataset is constructed, and the training set and test set are divided according to a preset ratio; the optimal performance prediction sub-models B, C and D are used as base models to construct a Stacking ensemble model, and the Stacking ensemble model is used to predict the coating performance.

[0016] Furthermore, in step 1), the coating preparation process corresponding to the preparation process parameters is one of chemical vapor deposition, physical vapor deposition, laser cladding, or thermal spraying.

[0017] If the coating preparation process is chemical vapor deposition, the preparation process parameters include deposition temperature, deposition time, pressure, type and ratio of reactant gases;

[0018] Physical vapor deposition (PVD) methods include vacuum evaporation, magnetron sputtering, and ion plating. If the coating preparation process is vacuum evaporation, the process parameters include evaporation temperature and holding time. If the coating preparation process is magnetron sputtering, the process parameters include sputtering power, deposition time, sputtering distance, nitrogen flow rate, target current, substrate temperature, and substrate bias voltage. If the coating preparation process is ion plating, the process parameters include substrate bias voltage, gas flow rate, current, deposition temperature, and deposition time.

[0019] If the coating preparation process is laser cladding, the preparation process parameters include laser power, scanning speed, and powder feeding rate.

[0020] Thermal spraying includes plasma spraying, arc spraying, and flame spraying. If the coating preparation process is thermal spraying, the preparation process parameters include spraying distance, spray gun moving speed, gas flow rate, powder / wire feeding rate, current, and voltage.

[0021] Furthermore, in step 1), the physical characteristics include phase content, grain size, residual stress, and coating thickness; the properties include microhardness and adhesion.

[0022] Furthermore, in step 1), the key feature selection adopts Pearson correlation coefficient selection, sets a selection threshold, and selects features whose absolute value of Pearson correlation coefficient is greater than the selection threshold as key features.

[0023] Furthermore, in step 3), during training and testing, principal component analysis is used to reduce the dimensionality of the input features.

[0024] The n machine learning algorithms include random forest, support vector machine, XGBoost, LightGBM, artificial neural network, and decision tree; the prediction accuracy of each machine learning algorithm is analyzed using the coefficient of determination and root mean square error.

[0025] Furthermore, in step 4), during the construction of the Stacking ensemble model, the best performance prediction sub-models B, C, and D are trained using 5-fold cross-validation, and Ridge or Lasso regression is selected as the meta-model; the accuracy of the Stacking ensemble model is evaluated using the coefficient of determination, root mean square error, and mean absolute error.

[0026] Compared with the prior art, the beneficial effects of the present invention are:

[0027] This invention addresses the issue of low accuracy in predicting coating performance using existing single machine learning algorithms. It constructs a Stacking ensemble model that uses fabrication process parameters and predicted physical characteristics as inputs to predict coating performance. The first layer of the Stacking ensemble model is trained using multiple base models of different types, while the second layer utilizes a meta-model to learn the outputs of the base models and generate the final prediction result. This design leverages the strengths of different models, effectively overcoming the limitations of single models in complex scenarios, improving the model's accuracy and generalization ability, and providing guidance for optimizing coating fabrication processes.

[0028] It should be understood that the description in the Summary of the Invention is not intended to limit the key or essential features of the embodiments of the present invention, nor is it intended to restrict the scope of the invention. Other features of the invention will become readily apparent from the following description. Attached Figure Description

[0029] Other features, objects, and advantages of the invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings:

[0030] Figure 1 A flowchart of a coating performance prediction method;

[0031] Figure 2 Flowchart for the Stacking integration model;

[0032] Figure 3 The above figures show a comparison of the predicted and actual performance values ​​of the Stacking ensemble model in the embodiment, as well as the accuracy results. Figure a represents microhardness, and figure b represents bonding force. Detailed Implementation

[0033] The present invention will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and not intended to limit it. Furthermore, it should be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings.

[0034] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0035] Example

[0036] Please refer to Figures 1-3 Taking TiN coating performance prediction as an example, embodiments of the present invention provide a coating performance prediction method based on machine learning, comprising the following steps:

[0037] 1) Establish data for predicting the performance of TiN coatings from relevant literature and experiments;

[0038] The TiN coating is prepared by magnetron sputtering, and the process parameters include sputtering power, deposition time, sputtering distance, nitrogen flow rate, target current, substrate temperature and substrate bias voltage.

[0039] The physical characteristics of the coating include phase content, grain size, residual stress, and coating thickness;

[0040] Coating properties include microhardness and adhesion;

[0041] Specifically, X-ray diffraction analysis was used to analyze the phase content, grain size, and residual stress of the coating; scanning electron microscopy was used to analyze the coating thickness.

[0042] All data is normalized and mapped to the interval [0, 1] to obtain the original feature dataset. The calculation formula is as follows:

[0043] ;

[0044] Where, x norm This is the normalized value, where x is the original data value. max and x min These are the maximum and minimum values, respectively.

[0045] Using performance as the output feature, the Pearson correlation coefficient was used to screen the preparation process parameters and physical characteristics. The screening threshold was set to 0.8. Finally, sputtering power, deposition time, nitrogen flow rate, substrate temperature, substrate bias voltage, phase content, grain size, residual stress, and coating thickness were selected as key features to obtain a key feature dataset.

[0046] 2) Divide the key feature dataset into 4 data subsets:

[0047] Using the preparation process parameters as input and the physical characteristics as output, a first data subset is constructed.

[0048] Using the preparation process parameters as input and the performance as output, a second data subset is constructed.

[0049] Using physical characteristics as input and performance as output, a third data subset is constructed.

[0050] Using the preparation process parameters and physical characteristics as inputs and the performance as outputs, a fourth data subset is constructed.

[0051] 3) Divide the four data subsets into training and test sets in an 80%:20% ratio. Train the model on the training set and evaluate it on the test set. To avoid increased computational complexity and model overfitting, principal component analysis (PCA) is used to reduce the dimensionality of the input features.

[0052] The machine learning algorithms used include random forest, support vector machine, XGBoost, LightGBM, artificial neural network, and decision tree;

[0053] Using the coefficient of determination (R²) 2 The accuracy of sub-models constructed by different machine learning algorithms was evaluated using the root mean square error (RMSE) to obtain the best algorithm. The evaluation results are shown in Table 1. It was found that the sub-models with the best prediction performance after training and testing for the first, second, third, and fourth data subsets were random forest, artificial neural network, random forest, and XGBoost, respectively.

[0054] Table 1: Accuracy of 4 Machine Learning Algorithms

[0055]

[0056] 4) Construct a stacking ensemble model to predict coating performance:

[0057] Using the fabrication process parameters as input, the predicted physical features are obtained through the optimal physical feature prediction sub-model A-random forest; using the fabrication process parameters and the predicted physical features as input, and the performance as output, a dataset is constructed, and the training set and test set are divided in a ratio of 80%:20%.

[0058] Three machine learning models—B—Artificial Neural Network, C—Random Forest, and D—XGBoost—are used as base models for prediction. Each model is trained using 5-fold cross-validation to obtain the prediction results for the training set. Each trained base model is then used to make predictions on the test set.

[0059] To address the overfitting problem of the base model, regularized Ridge regression was chosen as the meta-model. The complete prediction results of the training set were input into the meta-model for training and prediction, and the prediction results of the base model on the test set were used as input features. The trained meta-model was then used to obtain the final prediction value.

[0060] Using the coefficient of determination (R²) 2 The accuracy of the TiN coating performance prediction model was evaluated using root mean square error (RMSE) and mean absolute error (MAE); the comparison results between the predicted and measured values ​​of the coating performance are as follows: Figure 3 As shown in the comparison results, the coating performance prediction method of this application has high accuracy, ensuring accurate prediction of coating performance.

[0061] In the description of this specification, the terms "one embodiment," "some embodiments," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0062] The above are merely preferred embodiments of this application and are not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A method for predicting coating performance based on machine learning, characterized in that, Includes the following steps: 1) Obtain the preparation process parameters, physical characteristics, and performance-related data of the coating; normalize the above data to obtain the original dataset; use performance as the output feature to screen key features of the preparation process parameters and physical characteristics to obtain the key feature dataset; 2) Divide the key feature dataset into four data subsets: Using the preparation process parameters as input and the physical characteristics as output, a first data subset is constructed. Using the preparation process parameters as input and the performance as output, a second data subset is constructed. Using physical characteristics as input and performance as output, a third data subset is constructed. Using the preparation process parameters and physical characteristics as inputs and the performance as outputs, a fourth data subset is constructed. 3) Divide the four data subsets into training and testing sets according to a preset ratio, and use no less than 4 n machine learning algorithms to train and test them one by one; After training and testing on the first subset of data, n sub-models A are obtained. The one with the highest prediction accuracy is selected as the best physical feature prediction sub-model A. n sub-models B are obtained through training and testing on the second subset of data; n sub-models C are obtained through training and testing on the third subset of data; and n sub-models D are obtained through training and testing on the fourth subset of data. The optimal combination of these sub-models with high prediction accuracy and different machine learning algorithms is selected to obtain the best-performing prediction sub-models B, C, and D. 4) Using the preparation process parameters as input, the predicted values ​​of physical features are obtained through the optimal physical feature prediction sub-model A; A dataset is constructed using the predicted values ​​of the preparation process parameters and physical characteristics as inputs and the performance as the output, and the training set and test set are divided according to a preset ratio. The optimal performance prediction sub-models B, C, and D are used as base models to construct a Stacking ensemble model, which is then used to predict coating performance.

2. The coating performance prediction method based on machine learning according to claim 1, characterized in that, In step 1), the coating preparation process corresponding to the preparation process parameters is one of chemical vapor deposition, physical vapor deposition, laser cladding, or thermal spraying. If the coating preparation process is chemical vapor deposition, the preparation process parameters include deposition temperature, deposition time, pressure, type and ratio of reactant gases; Physical vapor deposition methods include vacuum evaporation, magnetron sputtering, and ion plating; if the coating preparation process is vacuum evaporation, the preparation process parameters include evaporation temperature and holding time; If the coating preparation process is magnetron sputtering, the preparation process parameters include sputtering power, deposition time, sputtering distance, nitrogen flow rate, target current, substrate temperature and substrate bias voltage. If the coating preparation process is ion plating, the preparation process parameters include substrate bias voltage, gas flow rate, current, deposition temperature and deposition time. If the coating preparation process is laser cladding, the preparation process parameters include laser power, scanning speed, and powder feeding rate. Thermal spraying includes plasma spraying, arc spraying, and flame spraying; If the coating preparation process is thermal spraying, the preparation process parameters include spraying distance, spray gun moving speed, gas flow rate, powder / wire feeding rate, current, and voltage.

3. The coating performance prediction method based on machine learning according to any one of claims 1 or 2, characterized in that, In step 1), the physical characteristics include phase content, grain size, residual stress, and coating thickness; the properties include microhardness and adhesion.

4. The coating performance prediction method based on machine learning according to claim 1, characterized in that, In step 1), the key feature selection adopts the Pearson correlation coefficient selection. A selection threshold is set, and features with an absolute value of Pearson correlation coefficient greater than the selection threshold are selected as key features.

5. The coating performance prediction method based on machine learning according to claim 1, characterized in that, In step 3), during training and testing, principal component analysis is used to reduce the dimensionality of the input features. The n machine learning algorithms include random forest, support vector machine, XGBoost, LightGBM, artificial neural network, and decision tree; the prediction accuracy of each machine learning algorithm is analyzed using the coefficient of determination and root mean square error.

6. The coating performance prediction method based on machine learning according to claim 1, characterized in that, In step 4), during the construction of the Stacking ensemble model, the best performance prediction sub-models B, C and D are trained using 5-fold cross-validation, and Ridge or Lasso regression is selected as the meta-model. The accuracy of the Stacking ensemble model is evaluated using the coefficient of determination, root mean square error, and mean absolute error.

Images