Method for optimizing hot rolling process parameters of high-strength titanium / steel composite plate
Through a data-driven approach, data collection, cleaning, feature engineering and neural network prediction models are used to optimize the hot rolling process parameters of titanium steel composite plates, which solves the problem that the multi-parameter coupling effect in traditional processes is difficult to quantify and relies on trial and error, and achieves efficient and accurate process parameter optimization and interface bonding strength improvement.
Patent Information
- Application Number
- CN202510809928.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-06-17
- Publication Date
- 2025-09-23
AI Technical Summary
The existing titanium-steel composite plate rolling process has multi-parameter coupling effects that are difficult to quantify, and process optimization relies on trial and error, resulting in long R&D cycles, high costs, and difficulty in steadily improving interface bonding strength.
Using a data-driven approach, through data collection, cleaning, feature engineering, neural network prediction model and genetic algorithm optimization, a method for optimizing the hot rolling process parameters of high-strength titanium steel composite plates is established, key influencing factors are accurately screened, and scientific quantitative optimization of process parameters is achieved.
The prediction accuracy of the interface bonding strength of titanium-steel composite plates has been greatly improved, the trial rolling cost and cycle have been reduced, the universality and stability of the process plan have been enhanced, and a stable improvement in the interface bonding strength has been achieved.
Smart Images
Figure CN120688359A_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the field of rolling process control, and in particular relates to a method for optimizing hot rolling process parameters of a high-strength titanium / steel composite plate. Background Art
[0002] Titanium-steel composite plates, combining the corrosion resistance of titanium (such as acid and alkali resistance and seawater erosion resistance) with the high strength and low cost of steel, are widely used in key areas such as chemical containers, offshore platforms, nuclear power equipment, and new energy equipment. With the rise of emerging industries such as deep-sea oil and gas development and hydrogen energy storage and transportation, the market has placed higher demands on the composite plates' interfacial bonding strength (≥210MPa), high-temperature rolling stability, and large-scale production efficiency.
[0003] Due to the bottleneck of traditional preparation technology, the current mainstream titanium steel composite plate rolling process has the following defects.
[0004] (1) Multi-parameter coupling effects are difficult to quantify:
[0005] The nonlinear relationship between process parameters (such as rolling temperature, reduction rate, rolling speed) and bonding strength is complex:
[0006] Too high a temperature (>1050°C) intensifies interfacial oxidation, but too low a temperature (<850°C) leads to a titanium / steel plasticity mismatch (titanium deformation resistance >200MPa vs. steel <150MPa).
[0007] The reduction rate and bonding strength show a non-monotonic relationship. Experiments show that the single-pass reduction rate needs to be controlled in the range of 20-40%, but the cumulative mechanism of interface damage caused by multi-pass cumulative deformation is still unclear.
[0008] (2) Process optimization relies on trial and error:
[0009] The existing process parameter setting mainly relies on orthogonal experiments or single-factor methods, which requires a large number of trial rolling samples (usually 50-100 sets of experiments) and cannot analyze the interaction rules of multiple variables (such as temperature-reduction rate-bonding strength), resulting in long R&D cycles and high costs. Summary of the Invention
[0010] In order to solve the above technical problems, the present invention provides a method for optimizing hot rolling process parameters of high-strength titanium / steel composite plates to solve the problems in the prior art. The technical solution adopted by the present invention is:
[0011] A method for optimizing hot rolling process parameters of a high-strength titanium / steel composite plate comprises the following steps:
[0012] Step 1: Data collection and database construction: collect historical experimental data and laboratory measurement data of titanium steel composite plate rolling to construct the original data set;
[0013] Step 2: Data cleaning and standardization, preprocessing the original data set;
[0014] Step 3: Feature engineering and optimization: perform feature analysis on the original dataset. Based on the feature engineering analysis results, remove redundant features with relatively small contributions to generate the final dataset.
[0015] Step 4: Prediction model establishment: establish an artificial neural network prediction model. The network architecture includes input layer, hidden layer, and output layer. Select activation function, loss function and optimizer, and optimize the neural network model through genetic algorithm to obtain the optimal hyperparameters of the prediction model.
[0016] Step 5: Dynamic validation mechanism, select the K-fold cross-validation method, and combine the final dataset of multiple iterations into several different training and test sets;
[0017] Step 6: Model evaluation. The regression evaluation criteria are the errors of specific values, including mean absolute percentage error, mean absolute error, root mean square error, mean square error, and coefficient of determination.
[0018] Step 7: Optimize process validation.
[0019] Furthermore, in step 1, the laboratory measured data temperature, reduction rate, layer thickness ratio, and bonding strength parameters.
[0020] Furthermore, in step 2, preprocessing includes processing missing values, removing data outliers, and normalizing sample data.
[0021] Furthermore, step 1 includes:
[0022] Step 1.1: Collect data from previous literature for the double-layer billet rolling experiment of titanium steel composite plates, and combine it with the data obtained in the laboratory to form an original data set;
[0023] Step 1.2: The collected data include the plate model, plate element composition percentage, plate thickness, rolling temperature, rolling reduction rate, and titanium steel composite plate bonding strength; the plate elements include titanium, steel, manganese, chromium, nickel, vanadium, carbon, nitrogen, oxygen, phosphorus, sulfur, and hydrogen.
[0024] Furthermore, step 2 includes:
[0025] Step 2.1: Delete missing data or replace them with the mean or median;
[0026] Step 2.2: The formula for sample data normalization is as follows:
[0027]
[0028] Where x is the sample value in each feature set, x max is the maximum sample value in each feature set, x min is the maximum sample value in each feature set, and x' is the normalized value.
[0029] Furthermore, step 3 includes:
[0030] Step 3.1: Use the mutual information method to capture parameter correlation, which is expressed as follows:
[0031]
[0032] Where: p(x,y) is the joint probability density function; p(x) and p(y) are marginal probability density functions;
[0033] Step 3.2: Quantify the contribution of each process parameter to the bonding strength using the SHAP value, which is calculated as follows:
[0034]
[0035] Where: N is the set of all features; S is the subset that does not contain feature i; f(S) is the predicted value of the model when only the features in subset S are used;
[0036] is a weight term, which indicates the probability that S precedes i in all possible feature arrangements;
[0037] Step 3.3: The features with smaller contributions are eliminated. The remaining features include the steel elements iron, chromium, nickel, manganese, carbon, and titanium; the titanium elements titanium, carbon, iron, rolling reduction rate, rolling temperature, steel thickness, titanium thickness, and composite plate bonding properties. The remaining features are recombined into the final data set.
[0038] Furthermore, step 4 includes:
[0039] Step 4.1: The number of input layers of the established neural network prediction model is equal to the number of input features, the number of output layers is equal to the number of output features, and the number of hidden layers is optimized based on the genetic algorithm to obtain the optimal number. The hidden layer uses the ReLU function to accelerate convergence, the Adam adaptive learning rate is selected as the optimizer, and the loss function is MAE;
[0040] Step 4.2: Use genetic algorithm to optimize the number of hidden layers and the number of nodes in each layer of the neural network to obtain the best prediction model.
[0041] Furthermore, in step 6: mean absolute percentage error MAPE, mean absolute error MAE, root mean square error RMSE, mean square error MSE and coefficient of determination R 2 The formula is as follows:
[0042]
[0043] Furthermore, step 7 includes: experimentally verifying the trained prediction model, predicting the optimal rolling process parameters by inputting the plate model, and predicting the final bonding strength.
[0044] The present invention has the following beneficial effects: the present invention accurately screens key influencing factors, greatly improves the model's prediction accuracy for the interface bonding strength of titanium-steel composite plates, and provides a scientific quantitative basis for the optimization of complex nonlinear process parameters; the neural network hyperparameter optimization based on genetic algorithms enables the model to adapt to the rolling characteristics of different material combinations, significantly enhancing the universality and stability of the process scheme; this method shifts process research and development from reliance on manual trial and error to data-driven intelligent optimization, which not only greatly reduces the trial rolling cost and cycle, but also achieves a stable improvement in interface bonding strength. BRIEF DESCRIPTION OF THE DRAWINGS
[0045] Figure 1 It is a flow chart of the present invention;
[0046] Figure 2 Collect information for data;
[0047] Figure 3 The result of mutual information feature selection;
[0048] Figure 4 Calculation results of SHAP values for each feature;
[0049] Figure 5 A line chart comparing the prediction results of the prediction model with the actual values;
[0050] Figure 6 The diagonal error graph is the prediction result of the prediction model. DETAILED DESCRIPTION
[0051] The following is a combination of the embodiments of the present invention Figures 1-6 , the technical solutions in the embodiments of the present invention are described clearly and completely. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Unless otherwise specified, the technical means used in the embodiments are conventional means well known to those skilled in the art.
[0052] To address the problem that the current hot-rolled titanium-steel composite plate process optimization relies on empirical trial and error and is difficult to accurately match the combination of multiple parameters such as layer thickness ratio / temperature / reduction rate, this application provides a method for optimizing the hot-rolling process parameters of high-strength titanium-steel composite plates based on machine learning, comprising the following steps:
[0053] Step 1: Data Collection and Database Construction. Collect historical experimental data and laboratory measurement data on titanium-steel composite plate rolling from historical literature to construct a multi-source, heterogeneous process parameter-performance database, i.e., the original data set, ensuring that the data covers different rolling conditions and material combinations.
[0054] Step 2: Data cleaning and standardization: Preprocess the original data set to remove invalid data due to sensor anomalies and missing records.
[0055] Step 3: Feature Engineering and Optimization: Perform feature analysis on the original dataset preprocessed in Step 2. Based on the feature engineering analysis results, remove redundant features with relatively small contributions to generate the final dataset.
[0056] Step 4: Prediction model establishment. Build an artificial neural network prediction model. The network architecture includes an input layer, a hidden layer, and an output layer. Select appropriate activation functions, loss functions, and optimizers. Optimize the neural network model using a genetic algorithm to obtain the optimal hyperparameters of the prediction model.
[0057] Step 5: Dynamic Validation Mechanism. A K-fold cross-validation method is selected, where the final dataset from multiple iterations is combined into several different training and test sets. This allows the data to participate in error calculations as both training and test sets, avoiding arbitrary dataset partitioning that can lead to incorrect evaluation of model errors.
[0058] Step 6: Model evaluation. For regression problems, the regression evaluation criteria are errors of specific values, usually including mean absolute percentage error (MAPE), mean absolute error (MAE), root mean square error (RMSE), mean square error (MSE), and coefficient of determination (R). 2 .
[0059] Step 7: Optimize process validation.
[0060] Specifically: In step 1, the laboratory measured data of the titanium steel composite plate rolling include temperature, reduction rate, layer thickness ratio, and bonding strength parameters.
[0061] In step 2, the preprocessing includes processing missing values, eliminating data outliers, and normalizing sample data.
[0062] In step 3, the feature analysis method includes using the mutual information method to capture parameter correlation; and using the SHAP value to quantify the contribution of each parameter to the binding strength.
[0063] In step 4, the hyperparameters of the neural network include the number of hidden layers and the number of nodes in each layer.
[0064] In step 5, within the K-fold cross-validation framework, each fold is trained independently by the ANN model to ensure that the diversity of data distribution is fully learned and to avoid single-time partitioning bias.
[0065] The following steps are specifically described in conjunction with the accompanying drawings in actual application:
[0066] Step 1 includes:
[0067] Step 1.1: Collect data from previous literature for the double-layer billet rolling experiment of titanium steel composite plates, and combine it with the data obtained in the laboratory to form an original data set.
[0068] Step 1.2: The collected data includes the titanium-steel composite plate model, plate element composition percentage, plate thickness, rolling temperature, rolling reduction ratio, and titanium-steel composite plate bonding strength. Plate elements include titanium, steel, manganese, chromium, nickel, vanadium, carbon, nitrogen, oxygen, phosphorus, sulfur, and hydrogen. This data is used as sample data features to form the initial sample data set.
[0069] Step 2 includes:
[0070] Step 2.1: Since the experimental conditions in the collected data are not uniform, some feature data may be missing. The missing data should be deleted or replaced by the mean or median.
[0071] Step 2.2: The formula for sample data normalization is as follows:
[0072]
[0073] Where x is the sample value in each feature set, x max is the maximum sample value in each feature set, x min is the maximum sample value in each feature set, and x' is the normalized value.
[0074] Step 3 includes:
[0075] Step 3.1: Use the mutual information method to capture parameter correlation, which is expressed as follows:
[0076]
[0077] Where: p(x,y) is the joint probability density function; p(x) and p(y) are the marginal probability density functions.
[0078] Step 3.2: Quantify the contribution of each process parameter to the bonding strength using the SHAP value, which is calculated as follows:
[0079]
[0080] Where: N is the set of all features; S is the subset that does not contain feature i; f(S) is the predicted value of the model when only the features in subset S are used;
[0081] is a weight term, which represents the probability that S precedes i in all possible feature arrangements.
[0082] The results are as attached Figure 4 shown.
[0083] Step 3.3: Based on the above steps, the features with smaller contributions are eliminated. The remaining features include the steel elements iron, chromium, nickel, manganese, carbon, and titanium; the titanium elements titanium, carbon, iron, rolling reduction rate, rolling temperature, steel thickness, titanium thickness, and composite plate bonding properties. These features are recombined into the final data set.
[0084] Step 4 includes:
[0085] Step 4.1: The number of input layers in the established neural network prediction model is equal to the number of input features, the number of output layers is equal to the number of output features, and the number of hidden layers is optimized using a genetic algorithm to obtain the optimal number. The hidden layers use the ReLU function to accelerate convergence and alleviate the vanishing gradient problem. The optimizer uses the Adam adaptive learning rate and the MAE loss function.
[0086] Step 4.2: Use genetic algorithm to optimize the number of hidden layers and the number of nodes in each layer of the neural network to obtain the best prediction model.
[0087] Step 5: Based on the final data set processed by steps 1 to 3, randomly shuffle it and select a K value of 5 to divide the training set into 5 mutually exclusive subsets with equal number of samples in each subset. Perform 5 rounds of iterations, select a different subset as the test set in each round, and the remaining 4 as the training set. Train the prediction model established in step 4 on the training set. The evaluation indicators are shown in step 6. Record the evaluation indicators for each round, and finally take the average of the K results as the estimate of the model generalization performance.
[0088] Step 6: Evaluation indicators of the established prediction model, including mean absolute percentage error (MAPE), mean absolute error (MAE), root mean square error (RMSE), mean square error (MSE), and coefficient of determination (R) 2 The detailed formula is as follows:
[0089]
[0090] Step 7: Optimization process verification. The trained prediction model is experimentally verified. By inputting the plate model, the optimal rolling process parameters, including rolling temperature, reduction rate, plate layer thickness ratio, and the final bond strength can be predicted.
[0091] The embodiments described above are merely descriptions of preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Without departing from the spirit of the present invention, various deformations, modifications, and substitutions made to the technical solutions of the present invention by ordinary technicians in this field should fall within the scope of protection determined by the claims of the present invention.
Claims
1. A method for optimizing hot rolling process parameters of high-strength titanium / steel composite plates, characterized in that: The following steps are involved: Step 1: Data collection and database construction: collect historical experimental data and laboratory measurement data of titanium steel composite plate rolling to construct the original data set; Step 2: Data cleaning and standardization, preprocessing the original data set; Step 3: Feature engineering and optimization: perform feature analysis on the original dataset. Based on the feature engineering analysis results, remove redundant features with relatively small contributions to generate the final dataset. Step 4: Prediction model establishment: establish an artificial neural network prediction model. The network architecture includes input layer, hidden layer, and output layer. Select activation function, loss function and optimizer, and optimize the neural network model through genetic algorithm to obtain the optimal hyperparameters of the prediction model. Step 5: Dynamic validation mechanism, select the K-fold cross-validation method, and combine the final dataset of multiple iterations into several different training and test sets; Step 6: Model evaluation. The regression evaluation criteria are the errors of specific values, including mean absolute percentage error, mean absolute error, root mean square error, mean square error, and coefficient of determination. Step 7: Optimize process validation.
2. The method for optimizing hot rolling process parameters of a high-strength titanium / steel composite plate according to claim 1, characterized in that: In step 1, the laboratory measured data temperature, reduction rate, layer thickness ratio, and bonding strength parameters.
3. The method for optimizing hot rolling process parameters of a high-strength titanium / steel composite plate according to claim 1, characterized in that: In step 2, preprocessing includes processing missing values, removing data outliers, and normalizing sample data.
4. The method for optimizing hot rolling process parameters of a high-strength titanium / steel composite plate according to claim 1, characterized in that: Step 1 includes: Step 1.1: Collect data from previous literature for the double-layer billet rolling experiment of titanium steel composite plates, and combine it with the data obtained in the laboratory to form an original data set; Step 1.2: The collected data include the plate model, plate element composition percentage, plate thickness, rolling temperature, rolling reduction rate, and titanium steel composite plate bonding strength; the plate elements include titanium, steel, manganese, chromium, nickel, vanadium, carbon, nitrogen, oxygen, phosphorus, sulfur, and hydrogen.
5. The method for optimizing hot rolling process parameters of a high-strength titanium / steel composite plate according to claim 1, characterized in that: Step 2 includes: Step 2.1: Delete missing data or replace them with the mean or median; Step 2.2: The formula for sample data normalization is as follows: Where x is the sample value in each feature set, x max is the maximum sample value in each feature set, x min is the maximum sample value in each feature set, and x' is the normalized value.
6. The method for optimizing hot rolling process parameters of a high-strength titanium / steel composite plate according to claim 1, characterized in that: Step 3 includes: Step 3.1: Use the mutual information method to capture parameter correlation, which is expressed as follows: Where: p(x,y) is the joint probability density function; p(x) and p(y) are marginal probability density functions; Step 3.2: Quantify the contribution of each process parameter to the bonding strength using the SHAP value, which is calculated as follows: Where: N is the set of all features; S is the subset that does not contain feature i; f(S) is the predicted value of the model when only the features in subset S are used; is a weight term, which indicates the probability that S precedes i in all possible feature arrangements; Step 3.3: The features with smaller contributions are eliminated. The remaining features include the steel elements iron, chromium, nickel, manganese, carbon, and titanium; the titanium elements titanium, carbon, iron, rolling reduction rate, rolling temperature, steel thickness, titanium thickness, and composite plate bonding properties. The remaining features are recombined into the final data set.
7. The method for optimizing hot rolling process parameters of a high-strength titanium / steel composite plate according to claim 1, characterized in that: Step 4 includes: Step 4.1: The number of input layers of the established neural network prediction model is equal to the number of input features, the number of output layers is equal to the number of output features, and the number of hidden layers is optimized based on the genetic algorithm to obtain the optimal number. The hidden layer uses the ReLU function to accelerate convergence, the Adam adaptive learning rate is selected as the optimizer, and the loss function is MAE; Step 4.2: Use genetic algorithm to optimize the number of hidden layers and the number of nodes in each layer of the neural network to obtain the best prediction model.
8. The method for optimizing hot rolling process parameters of a high-strength titanium / steel composite plate according to claim 1, characterized in that: In step 6: mean absolute percentage error (MAPE), mean absolute error (MAE), root mean square error (RMSE), mean square error (MSE), and coefficient of determination (R) 2 The formula is as follows:
9. The method for optimizing hot rolling process parameters of a high-strength titanium / steel composite plate according to claim 1, characterized in that: Step 7 includes: experimentally verifying the trained prediction model, predicting the optimal rolling process parameters by inputting the plate model, and predicting the final bonding strength.