Curve prediction method and device based on hybrid network prediction model

By using a hybrid network prediction model that combines convolutional neural networks and multilayer perceptrons, and inputting microscopic characterization images and experimental parameters, the problem of long prediction time and high cost in the prediction of the mechanical properties of ultra-high strength steel materials in existing technologies is solved, and a wider prediction range and flexible material performance evaluation are achieved.

CN118968506APending Publication Date: 2024-11-15CHONGQING SHUYUANDAO TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410936297.6
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-12
Publication Date
2024-11-15

AI Technical Summary

Technical Problem

Existing computational models for predicting the mechanical properties of ultra-high strength steel under different test conditions suffer from problems such as long processing time, high cost, and limited application scope, especially since they only focus on a single type of data input, such as images of the material's microstructure.

Method used

A hybrid network prediction model, combining convolutional neural networks and multilayer perceptrons, is adopted. By inputting microscopic characterization images and experimental parameters, and through feature extraction, data stitching, and feature refinement, the accurate prediction of material mechanical property curves is achieved.

Benefits of technology

It enables a wider range of predictions and more degrees of freedom, improving the flexibility of material performance evaluation under different conditions and enhancing prediction accuracy and efficiency.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118968506A_ABST
    Figure CN118968506A_ABST
Patent Text Reader

Abstract

The invention discloses a curve prediction method and device based on a hybrid network prediction model, and the method comprises the steps: carrying out the preprocessing of a microcosmic representation image, obtaining a preprocessed microcosmic representation image, inputting the preprocessed microcosmic representation image into a convolutional neural network, and carrying out the feature extraction, thereby obtaining the features of the microcosmic representation image; the test parameters are input into the first multi-layer perceptron for feature extraction to obtain test parameter features, the microscopic representation image features and the test parameter features are subjected to data splicing through feature engineering to obtain spliced features, and finally the spliced features are input into the second multi-layer perceptron for feature refining to obtain a plurality of prediction curve data points; and completing curve prediction. By adopting the hybrid network prediction model, complementary information of images and parameter data can be fully utilized, a wider prediction range and selection of more degrees of freedom are realized, the flexibility of material performance evaluation by a user under different conditions is improved, and accurate prediction of a mechanical property curve of a material under specific conditions is completed.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the technical field of curve prediction, and in particular to a curve prediction method and device based on a hybrid network prediction model. Background Art

[0002] In the field of materials science, especially in the development and application of ultra-high strength steel, it is particularly important to accurately predict the mechanical properties of materials under specific test conditions. Ultra-high strength steel is widely used in industries such as automobiles, aviation, construction and military due to its excellent strength-to-toughness ratio to improve the safety and durability of structures.

[0003] However, the material properties of ultra-high-strength steel depend not only on its chemical composition, but also on process conditions such as heat treatment and cold working during production. These complex variables make it time-consuming and costly to predict its performance in actual applications through traditional experimental methods. For example, in the automotive industry, the use of ultra-high-strength steel can significantly reduce the weight of the vehicle body, improve fuel efficiency and driving performance, while enhancing the safety protection of the vehicle in a collision. In order to design a vehicle body structure that is both light and safe, engineers need to know exactly how the material behaves under different loading conditions.

[0004] In the existing technology, although traditional physical experimental methods can provide accurate data, they are often time-consuming, costly, and difficult to implement in some cases. With the development of computer science, the use of computational models to predict material properties has become an efficient alternative. However, most existing computational models focus on a single type of data input, such as being limited to microstructural images of materials, which limits the scope of application of the model. Summary of the invention

[0005] The purpose of this application is to provide a curve prediction method and device based on a hybrid network prediction model in order to overcome the existing technical defects. The hybrid network prediction model is used to accurately predict the mechanical property curves corresponding to the material under different test conditions based on the input microscopic characterization and test parameters, and to make full use of the complementary information of image and parameter data to achieve a wider prediction range and more freedom of choice, thereby improving the flexibility of users in evaluating material properties under different conditions.

[0006] The purpose of this application is achieved through the following technical solutions:

[0007] In a first aspect, the present application proposes a curve prediction method based on a hybrid network prediction model, the method is applied to the hybrid network prediction model, the hybrid network prediction model includes a convolutional neural network, a first multi-layer perceptron and a second multi-layer perceptron, the method includes:

[0008] Preprocessing the microscopic characterization image to obtain a preprocessed microscopic characterization image;

[0009] The preprocessed microscopic representation image is input into a convolutional neural network for feature extraction to obtain microscopic representation image features, and the test parameters are input into a first multi-layer perceptron for feature extraction to obtain test parameter features;

[0010] Feature engineering is used to perform data splicing on microscopic characterization image features and test parameter features to obtain splicing features;

[0011] The concatenated features are input into the second multi-layer perceptron for feature refinement to obtain multiple prediction curve data points and complete the curve prediction.

[0012] In a possible implementation, the convolutional neural network includes five convolutional layer structures and global average pooling connected in sequence, and each convolutional layer structure includes a convolution layer and a maximum pooling layer.

[0013] In one possible implementation, a ReLU activation function is set between each convolutional layer and the maximum pooling layer.

[0014] In a possible implementation, the first multilayer perceptron and the second multilayer perceptron both use two fully connected layers, and the two fully connected layers of the first multilayer perceptron are both configured with 256 neurons and are followed by a ReLU activation function;

[0015] The first fully connected layer of the second multilayer perceptron is configured with 256 neurons and a ReLU activation function is set afterwards, and the second fully connected layer is configured with 30 neurons and no activation function is set.

[0016] In a possible implementation, the preprocessing includes one or more of image cutting, image normalization, image zero centering, and image grayscale conversion.

[0017] In a second aspect, the present application proposes a curve prediction device based on a hybrid network prediction model, the device comprising:

[0018] A preprocessing module, used for preprocessing the microscopic representation image to obtain a preprocessed microscopic representation image;

[0019] A feature extraction module is used to input the preprocessed microscopic representation image into a convolutional neural network for feature extraction to obtain microscopic representation image features, and input the test parameters into a first multi-layer perceptron for feature extraction to obtain test parameter features;

[0020] A data splicing module is used to perform data splicing on microscopic characterization image features and test parameter features by using feature engineering to obtain splicing features;

[0021] The feature refining module is used to input the concatenated features into the second multi-layer perceptron for feature refinement to obtain multiple prediction curve data points and complete the curve prediction.

[0022] In a possible implementation, the convolutional neural network includes five convolutional layer structures and global average pooling connected in sequence, and each convolutional layer structure includes a convolution layer and a maximum pooling layer.

[0023] In one possible implementation, a ReLU activation function is set between each convolutional layer and the maximum pooling layer.

[0024] In a possible implementation, the first multilayer perceptron and the second multilayer perceptron both use two fully connected layers, and the two fully connected layers of the first multilayer perceptron are both configured with 256 neurons and are followed by a ReLU activation function;

[0025] The first fully connected layer of the second multilayer perceptron is configured with 256 neurons and a ReLU activation function is set afterwards, and the second fully connected layer is configured with 30 neurons and no activation function is set.

[0026] In a possible implementation, the preprocessing includes one or more of image cutting, image normalization, image zero centering, and image grayscale conversion.

[0027] The above-mentioned main scheme of the present application and its further options can be freely combined to form multiple schemes, all of which are schemes that can be adopted and claimed for protection in the present application; and in the present application, (non-conflicting options) options and other options can also be freely combined. After understanding the scheme of the present application, those skilled in the art can understand that there are multiple combinations based on the prior art and common knowledge, all of which are technical schemes to be protected by the present application, and they are not exhaustively listed here.

[0028] The present application discloses a curve prediction method and device based on a hybrid network prediction model, first preprocessing a microscopic characterization image to obtain a preprocessed microscopic characterization image, inputting the preprocessed microscopic characterization image into a convolutional neural network for feature extraction to obtain microscopic characterization image features, and inputting test parameters into a first multi-layer perceptron for feature extraction to obtain test parameter features, using feature engineering to perform data splicing of microscopic characterization image features and test parameter features to obtain splicing features, and finally inputting the splicing features into a second multi-layer perceptron for feature refinement to obtain multiple prediction curve data points, and completing curve prediction. The use of a hybrid network prediction model can make full use of the complementary information of image and parameter data, achieve a wider prediction range and more freedom of choice, improve the flexibility of users in evaluating material properties under different conditions, and complete the accurate prediction of the mechanical property curve of the material under specific conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 A flow chart of a curve prediction method based on a hybrid network prediction model proposed in an embodiment of the present application is shown.

[0030] Figure 2 A schematic diagram of the structure of a convolutional neural network proposed in an embodiment of the present application is shown.

[0031] Figure 3 A schematic diagram of the structure of a two-layer multi-layer perceptron proposed in an embodiment of the present application is shown.

[0032] Figures 4a to 4l A schematic diagram comparing the model prediction proposed in the embodiment of the present application and the actual results is shown. DETAILED DESCRIPTION

[0033] The following describes the embodiments of the present application through specific examples, and those skilled in the art can easily understand other advantages and effects of the present application from the contents disclosed in this specification. The present application can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present application. It should be noted that the following embodiments and features in the embodiments can be combined with each other without conflict.

[0034] Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making any creative work shall fall within the scope of protection of this application.

[0035] In the existing technology, although traditional physical experimental methods can provide accurate data, they are often time-consuming, costly, and difficult to implement in some cases. With the development of computer science, the use of computational models to predict material properties has become an efficient alternative. However, most existing computational models focus on a single type of data input, such as being limited to microstructural images of materials, which limits the scope of application of the model.

[0036] Therefore, in order to solve the above-mentioned technical problems, the embodiment of the present application proposes a curve prediction method and device based on a hybrid network prediction model. The hybrid network prediction model is adopted, which can not only input the material microstructure image, but also simultaneously input the user-defined test condition parameters, so that the model can make full use of the complementary information between the image and parameter data, achieve a wider prediction range, more degrees of freedom in selection, and accurate prediction of the mechanical property curve of the material under specific conditions, thereby improving the flexibility of users in evaluating material properties under different conditions. The curve prediction method and device proposed in the present application are described in detail below.

[0037] Please refer to Figure 1 , Figure 1A flow chart of a curve prediction method based on a hybrid network prediction model proposed in an embodiment of the present application is shown. The method is applied to the hybrid network prediction model, wherein the hybrid network prediction model includes a convolutional neural network, a first multi-layer perceptron, and a second multi-layer perceptron. The method includes:

[0038] Preprocessing the microscopic characterization image to obtain a preprocessed microscopic characterization image;

[0039] The preprocessed microscopic representation image is input into a convolutional neural network for feature extraction to obtain microscopic representation image features, and the test parameters are input into a first multi-layer perceptron for feature extraction to obtain test parameter features;

[0040] Feature engineering is used to perform data splicing on microscopic characterization image features and test parameter features to obtain splicing features;

[0041] The concatenated features are input into the second multi-layer perceptron for feature refinement to obtain multiple prediction curve data points and complete the curve prediction.

[0042] By inputting the input data (microscopic characterization image and test parameters) into the hybrid network prediction model proposed in this application, preprocessing it to obtain the preprocessed input data, then inputting the preprocessed microscopic characterization image into the convolutional neural network, and inputting the preprocessed test parameters into the first multi-layer perceptron for data feature extraction, and then splicing the microscopic characterization image features and the test parameter features to obtain spliced ​​features, wherein the data splicing can be completed using the concatenate function of the Keras library in Python, and finally inputting the spliced ​​features into the second multi-layer perceptron for further feature refinement, and finally outputting multiple predicted curve data points.

[0043] A hybrid network prediction model of convolutional neural network and two-layer perceptron is adopted. In this framework, convolutional neural network is responsible for extracting the features of microstructure images, while perceptron is used to extract the features of test condition parameters.

[0044] In a possible implementation, the preprocessing includes one or more of image cutting, image normalization, image zero centering, and image grayscale conversion.

[0045] First, the input image data needs to be preprocessed, including conventional operations such as image cutting, image normalization, image zero centering, and image grayscale. Other image processing operations may also be included, which are not listed one by one in the embodiments of the present application. In addition, in addition to data standardization processing of the input test condition data (test parameters), data preprocessing operations will also be performed, such as eliminating duplicate data and data that does not meet the conditions. This is to ensure that the input microscopic characterization image and test parameters are correct and will not have any other impact on the subsequent image processing and curve generation, resulting in inaccurate data results.

[0046] In a possible implementation, the convolutional neural network includes five convolutional layer structures and global average pooling connected in sequence, and each convolutional layer structure includes a convolution layer and a maximum pooling layer.

[0047] Please refer to Figure 2 , Figure 2 A schematic diagram of the structure of a convolutional neural network proposed in an embodiment of the present application is shown, wherein the convolutional neural network comprises five convolutional layer structures connected in sequence and global average pooling, wherein each convolutional layer structure comprises a convolutional layer and a maximum pooling layer, wherein the maximum pooling layer can enhance the spatial hierarchy of the model and reduce the computational complexity,

[0048] In the convolutional neural network, the number of neurons starts from 64 and increases layer by layer to 512 neurons in the 4th layer. The reason why the number of neurons increases gradually in layers 1-4 is to capture more complex and abstract features. Then in the 5th layer, the number of neurons drops to 216.

[0049] In the early stages of convolutional neural networks, a smaller number of neurons is sufficient to capture simple features, such as edges and corners. As the depth of the network increases, the features processed by each layer gradually change from simple to complex, such as from simple shapes to parts of entire objects, and the number of neurons must be increased to allow the network to have more parameters to learn these complex features.

[0050] The fifth layer shows a decreasing trend relative to the fourth layer in order to reduce the risk of overfitting, improve the generalization ability of the model, enhance computational efficiency, and optimize feature compression and refinement, which is not only conducive to reducing the prediction error of the model on unknown data, but also makes the deployment of the model in resource-constrained environments more feasible. The convolution kernel size of the first convolution layer in the convolutional neural network is slightly larger and is designed to be 5x5 in size, which can quickly reduce the image matrix size and extract shallow features of the image. The convolution kernel size and step size of the second to fifth convolution layers are smaller and the image matrix size is lower and slower, which is used to extract high-order features of the image. After each convolution layer, a maximum pooling layer is added to further process the feature map, simplifying the complexity of the convolutional neural network through downsampling and enhancing the robustness of the features.

[0051] Global Average Pooling (GAP) is the last layer of the convolutional neural network. It is placed after the last pooling layer. By calculating the average value of each channel of the deep feature map, it not only significantly reduces the number of model parameters, but also effectively prevents overfitting and maintains sensitivity to the global features of the image. Through the processing of this layer, a fixed-size feature vector is formed, which effectively represents the features extracted from the microscopic tissue image and ensures the average response of the feature map globally.

[0052] In one possible implementation, a ReLU activation function is set between each convolutional layer and the maximum pooling layer.

[0053] The ReLU activation function is applied after each convolutional layer, introducing necessary nonlinear characteristics to enhance the expressiveness of the model.

[0054] In a possible implementation, the first multilayer perceptron and the second multilayer perceptron both use two fully connected layers, and the two fully connected layers of the first multilayer perceptron are both configured with 256 neurons and are followed by a ReLU activation function;

[0055] The first fully connected layer of the second multilayer perceptron is configured with 256 neurons and a ReLU activation function is set afterwards, and the second fully connected layer is configured with 30 neurons and no activation function is set.

[0056] Please refer to Figure 3 , Figure 3 The schematic diagram of the structure of the two-layer multilayer perceptron proposed in the embodiment of the present application is shown. The left side is the first multilayer perceptron, and the right side is the second multilayer perceptron. The first multilayer perceptron extracts the features of the test condition parameters, including two fully connected layers, each fully connected layer is configured with 256 neurons. This structural design enables the multilayer perceptron to process and learn complex input data, and effectively extract features and perform nonlinear transformations through two layers of dense neural network connections.

[0057] In addition, the activation function ReLU is used after each fully connected layer to introduce necessary nonlinearity to help the model capture complex patterns and relationships in the data. After data feature engineering and fusing the extracted image features with the experimental parameter features, a second multi-layer perceptron is needed to output the regression prediction results.

[0058] The second multi-layer perceptron includes two fully connected layers. The first fully connected layer is a dense layer with 256 neurons, which can further process and integrate the previously extracted features. The activation function ReLU is used after the first fully connected layer. The second fully connected layer is a dense layer with 30 neurons, which is designed to output 30 independent numerical results. These results specifically represent 30 equally spaced data points of the mechanical property curve. With equally spaced data, the predicted mechanical property curve can be drawn.

[0059] In order to better illustrate the curve prediction method proposed in the embodiment of the present application, it is verified next.

[0060] The data set of the model is set up and randomly divided into a training set and a test set with a ratio of 9:1. After training, the test set data is input into the hybrid network model for prediction. The model is able to generate multiple prediction curve data points for each set of data and draw the corresponding prediction curve based on them. In order to evaluate the prediction accuracy of the model in detail, the prediction curve and the actual curve obtained by the experiment are plotted on the same chart at the same time for intuitive comparison. Figures 4a to 4l A schematic diagram comparing the model prediction proposed in the embodiment of the present application and the actual results is shown.

[0061] After verification testing, the average prediction accuracy of the model is as high as 85.248%, which is very outstanding in the field of materials science. Because predicting the mechanical properties of materials involves multiple complex variables and profound nonlinear relationships, these factors greatly increase the difficulty of prediction. Therefore, such an accuracy rate not only proves the superior performance of this hybrid network model, but also reflects its outstanding ability in analyzing and processing complex data.

[0062] In summary, the present application discloses a curve prediction method and device based on a hybrid network prediction model, first preprocessing the microscopic characterization image to obtain a preprocessed microscopic characterization image, inputting the preprocessed microscopic characterization image into a convolutional neural network for feature extraction to obtain microscopic characterization image features, and inputting the test parameters into a first multi-layer perceptron for feature extraction to obtain test parameter features, using feature engineering to perform data splicing of microscopic characterization image features and test parameter features to obtain splicing features, and finally inputting the splicing features into a second multi-layer perceptron for feature refinement to obtain multiple prediction curve data points, and completing curve prediction. The use of a hybrid network prediction model can make full use of the complementary information of image and parameter data, achieve a wider prediction range and more freedom of choice, improve the flexibility of users in evaluating material properties under different conditions, and complete the accurate prediction of the mechanical property curve of the material under specific conditions.

[0063] A possible implementation of a curve prediction device based on a hybrid network prediction model is given below, which is used to execute the various execution steps and corresponding technical effects of the curve prediction method shown in the above embodiment and possible implementation. The device includes:

[0064] A preprocessing module, used for preprocessing the microscopic representation image to obtain a preprocessed microscopic representation image;

[0065] A feature extraction module is used to input the preprocessed microscopic representation image into a convolutional neural network for feature extraction to obtain microscopic representation image features, and input the test parameters into a first multi-layer perceptron for feature extraction to obtain test parameter features;

[0066] A data splicing module is used to perform data splicing on microscopic characterization image features and test parameter features by using feature engineering to obtain splicing features;

[0067] The feature refining module is used to input the concatenated features into the second multi-layer perceptron for feature refinement to obtain multiple prediction curve data points and complete the curve prediction.

[0068] In a possible implementation, the convolutional neural network includes five convolutional layer structures and global average pooling connected in sequence, and each convolutional layer structure includes a convolution layer and a maximum pooling layer.

[0069] In one possible implementation, a ReLU activation function is set between each convolutional layer and the maximum pooling layer.

[0070] In a possible implementation, the first multilayer perceptron and the second multilayer perceptron both use two fully connected layers, and the two fully connected layers of the first multilayer perceptron are both configured with 256 neurons and are followed by a ReLU activation function;

[0071] The first fully connected layer of the second multilayer perceptron is configured with 256 neurons and a ReLU activation function is set afterwards, and the second fully connected layer is configured with 30 neurons and no activation function is set.

[0072] In a possible implementation, the preprocessing includes one or more of image cutting, image normalization, image zero centering, and image grayscale conversion.

[0073] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present application should be included in the protection scope of the present application.

Claims

1. A curve prediction method based on a hybrid network prediction model, characterized in that: The method is applied to a hybrid network prediction model, the hybrid network prediction model includes a convolutional neural network, a first multi-layer perceptron, and a second multi-layer perceptron, and the method includes: Preprocessing the microscopic characterization image to obtain a preprocessed microscopic characterization image; The preprocessed microscopic representation image is input into a convolutional neural network for feature extraction to obtain microscopic representation image features, and the test parameters are input into a first multi-layer perceptron for feature extraction to obtain test parameter features; Feature engineering is used to perform data splicing on microscopic characterization image features and test parameter features to obtain splicing features; The concatenated features are input into the second multi-layer perceptron for feature refinement to obtain multiple prediction curve data points and complete the curve prediction.

2. The curve prediction method according to claim 1, characterized in that: The convolutional neural network consists of five convolutional layer structures and global average pooling connected in sequence. Each convolutional layer structure includes a convolution layer and a maximum pooling layer.

3. The curve prediction method according to claim 2, characterized in that: A ReLU activation function is set between each convolutional layer and the maximum pooling layer.

4. The curve prediction method according to claim 1, characterized in that: The first multilayer perceptron and the second multilayer perceptron both use two fully connected layers, and the two fully connected layers of the first multilayer perceptron are both configured with 256 neurons and are followed by a ReLU activation function; The first fully connected layer of the second multilayer perceptron is configured with 256 neurons and a ReLU activation function is set afterwards, and the second fully connected layer is configured with 30 neurons and no activation function is set.

5. The curve prediction method according to claim 1, characterized in that: The preprocessing includes one or more of image cutting, image normalization, image zero centering, and image grayscale conversion.

6. A curve prediction device based on a hybrid network prediction model, characterized in that: The device includes: A preprocessing module, used for preprocessing the microscopic representation image to obtain a preprocessed microscopic representation image; A feature extraction module is used to input the preprocessed microscopic representation image into a convolutional neural network for feature extraction to obtain microscopic representation image features, and input the test parameters into a first multi-layer perceptron for feature extraction to obtain test parameter features; A data splicing module is used to perform data splicing on microscopic characterization image features and test parameter features by using feature engineering to obtain splicing features; The feature refining module is used to input the concatenated features into the second multi-layer perceptron for feature refinement to obtain multiple prediction curve data points and complete the curve prediction.

7. The curve prediction device according to claim 6, characterized in that: The convolutional neural network consists of five convolutional layer structures and global average pooling connected in sequence. Each convolutional layer structure includes a convolution layer and a maximum pooling layer.

8. The curve prediction device according to claim 7, characterized in that: A ReLU activation function is set between each convolutional layer and the maximum pooling layer.

9. The curve prediction device according to claim 6, characterized in that: The first multilayer perceptron and the second multilayer perceptron both use two fully connected layers, and the two fully connected layers of the first multilayer perceptron are both configured with 256 neurons and are followed by a ReLU activation function; The first fully connected layer of the second multilayer perceptron is configured with 256 neurons and a ReLU activation function is set afterwards, and the second fully connected layer is configured with 30 neurons and no activation function is set.

10. The curve prediction device according to claim 6, characterized in that: The preprocessing includes one or more of image cutting, image normalization, image zero centering, and image grayscale conversion.