Deep learning model-based hot forming process parameter design method and device

Abstract

The invention relates to a hot forming process parameter design method and device based on a deep learning model, and belongs to the technical field of hot forming. The method comprises the steps that image data, text data and video data in a standard hot forming process and an experimental process are obtained, and the image data, the text data and the video data are fused to obtain a fused image; a multi-mode hot forming process database is obtained; an initial deep learning model is constructed, data in the multi-modal thermal forming process database is used as input parameters to be input into the initial deep learning model for iterative training, and a target deep learning model is obtained; and the product performance requirements serve as input parameters to be input into the target deep learning model, and target hot forming process parameters are obtained. The technical problems of high time cost and low efficiency of process parameter design are solved.

Description

Technical Field

[0001] The present invention relates to the field of thermoforming technology, and in particular to a thermoforming process parameter design method and device based on a deep learning model. Background Art

[0002] The hot forming process has the characteristics of numerous process parameters, continuous process, complex process-performance mapping, and coupling of the form and properties of the formed workpiece. These characteristics greatly increase the time and cost consumption of hot forming process design. Traditional hot forming process design often adopts the trial and error method to carry out process design. By designing different process routes and process parameters to carry out forming experiments, the geometric morphology, microstructure, and macroscopic performance testing and evaluation of the formed workpiece are carried out to obtain the material-process-performance mapping law. Finally, the process and parameters are refined to obtain the final hot forming process design results. This iterative design based on physical experiments consumes a lot of time and money, and imposes a great burden on the design unit.

[0003] As computer computing power increases, simulation technology has been widely used, and more and more designers use computer simulation to assist in hot forming process design. Although the cost consumption has been alleviated, the forming quality of the design cannot be guaranteed due to the large errors between digital simulation and physical experiments. In addition, this kind of computer simulation-assisted design is still a trial-and-error design, which cannot solve the problems of resource consumption and forming quality in the design process.

[0004] Therefore, how to use cutting-edge technology to design a hot forming process that meets quality requirements while improving design efficiency is a technical problem that urgently needs to be solved. Summary of the invention

[0005] In view of this, it is necessary to provide a hot forming process parameter design method and device based on a deep learning model to solve the technical problems of high time cost and low efficiency of process parameter design in the prior art.

[0006] In order to solve the above problems, the present invention provides a method for designing hot forming process parameters based on a deep learning model, comprising: Obtain image data, text data and video data in the standard hot forming process and the experimental process, and fuse the image data, text data and video data to obtain a multimodal hot forming process database; An initial deep learning model is constructed, and the data in the multimodal hot forming process database is input as input parameters into the initial deep learning model for iterative training to obtain a target deep learning model; Product materials and performance requirements are input into the target deep learning model as input parameters to obtain the target hot forming process parameters.

[0007] In a possible implementation, the multimodal thermoforming process database includes thermoforming process data, product material and performance requirement data, and process parameters.

[0008] In a possible implementation, the image data, text data and video data are fused to obtain a multimodal hot forming process database, including: Use a well-trained recurrent neural network model to extract features from text data and obtain text features; Use a well-trained convolutional neural network model to extract features from image data and obtain image features; The video data is extracted using a well-trained long short-term memory neural network model to obtain video features; A preset spatial alignment algorithm and a preset temporal alignment algorithm are used to align and splice text features, graphic features, and video features to obtain fused multimodal thermoforming process data, and a multimodal thermoforming process database is constructed based on the multimodal thermoforming process data.

[0009] In a possible implementation, the initial deep learning model is constructed, and the data in the multimodal hot forming process database is input as input parameters into the initial deep learning model for iterative training to obtain a target deep learning model, including: Slice the data in the multimodal hot forming process database to obtain multiple training data sets; Distribute multiple training sets across different devices and perform distributed training on the models in different devices.

[0010] In a possible implementation, the target deep learning model includes a classifier model and a regression model, and the loss function of the classifier model is expressed by the following formula:

[0011] Among them, For the sample, Indicates the actual label, represents the predicted output, Indicates the total sample size; The loss function of the regression model is expressed by the following formula:

[0012] in, Indicates the actual label, is the predicted label, is a hyperparameter.

[0013] In a possible implementation, the step of inputting the data in the multimodal hot forming process database as input parameters into the initial deep learning model for iterative training includes: A preset low-rank adaptive fine-tuning method is used to adjust the parameters of the initial deep learning model.

[0014] In a possible implementation, the product performance requirements are input as input parameters into the target deep learning model to obtain target hot forming process parameters, including: Input product performance requirements as input parameters into the target deep learning model to match the corresponding product material information and product performance information; Determine the target hot forming process parameters based on product material information and product performance information.

[0015] In a second aspect, the present invention further provides a hot forming process parameter design device based on a deep learning model, comprising: A fusion module is used to obtain image data, text data and video data in the standard hot forming process and the experimental process, and fuse the image data, text data and video data to obtain a multi-modal hot forming process database; A training module is used to construct an initial deep learning model, and input the data in the multimodal hot forming process database as input parameters into the initial deep learning model for iterative training to obtain a target deep learning model; The parameter design module is used to input the product performance requirements as input parameters into the target deep learning model to obtain the target hot forming process parameters.

[0016] In a third aspect, the present invention further provides an electronic device, comprising: a processor and a memory; The memory stores a computer-readable program executable by the processor; When the processor executes the computer-readable program, the steps in the hot forming process parameter design method based on the deep learning model as described above are implemented.

[0017] In a fourth aspect, the present invention also provides a computer-readable storage medium, which stores one or more programs, and the one or more programs can be executed by one or more processors to implement the steps in the hot forming process parameter design method based on the deep learning model as described above.

[0018] The beneficial effects of the present invention are as follows: first, image data, text data and video data in the standard hot forming process and the experimental process are obtained, and the image data, text data and video data are fused to obtain a multimodal hot forming process database; thereby, the model can obtain richer context information by integrating multiple data types, which helps to better capture the relationship between process parameters and actual effects. Subsequently, an initial deep learning model is constructed, and the data in the multimodal hot forming process database is input as input parameters into the initial deep learning model for iterative training to obtain a target deep learning model, and finally the product material and performance requirements are input as input parameters into the target deep learning model to obtain the target hot forming process parameters. The present invention uses a deep learning model to automate the design process of hot forming process parameters, thereby reducing manual intervention, shortening the design cycle, improving efficiency, reducing the trial and error process, and reducing R&D costs. At the same time, by using different types of data (images, texts, videos), the model has higher robustness to the diversity and changes of data. Multimodal data can help the model generalize better, thereby reducing the risk of overfitting and improving performance on new data. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 A method flow chart of an embodiment of a method for designing hot forming process parameters based on a deep learning model provided by the present invention; Figure 2 for Figure 1 A method flow chart of an embodiment of step S101; Figure 3 It is a structural schematic diagram of an embodiment of a target deep learning model in the hot forming process parameter design method based on a deep learning model provided by the present invention; Figure 4 It is a schematic diagram of the hot forming process parameter design process in the hot forming process parameter design method based on the deep learning model provided by the present invention; Figure 5 is a schematic diagram of an embodiment of a hot forming process parameter design device based on a deep learning model provided by the present invention; Figure 6 It is a schematic diagram of the operating environment of an electronic device provided by the present invention. DETAILED DESCRIPTION

[0020] The preferred embodiments of the present invention are described in detail below in conjunction with the accompanying drawings, wherein the accompanying drawings constitute a part of this application and are used together with the embodiments of the present invention to illustrate the principles of the present invention, but are not used to limit the scope of the present invention.

[0021] A specific embodiment of the present invention discloses a method for designing hot forming process parameters based on a deep learning model. Figure 1 ,include: S101, acquiring image data, text data and video data in a standard hot forming process and an experimental process, and fusing the image data, text data and video data to obtain a multimodal hot forming process database; S102, constructing an initial deep learning model, inputting the data in the multimodal hot forming process database as input parameters into the initial deep learning model for iterative training, and obtaining a target deep learning model; S103. Input product materials and performance requirements as input parameters into the target deep learning model to obtain target hot forming process parameters.

[0022] In this embodiment, the image data, text data and video data in the standard hot forming process and the experimental process are first obtained, and the image data, text data and video data are fused to obtain a multimodal hot forming process database; thus, the integration of multiple data types enables the model to obtain richer contextual information, which helps to better capture the relationship between process parameters and actual effects. Subsequently, an initial deep learning model is constructed, and the data in the multimodal hot forming process database is input as input parameters into the initial deep learning model for iterative training to obtain a target deep learning model, and finally the product material and performance requirements are input as input parameters into the target deep learning model to obtain the target hot forming process parameters. The present invention uses a deep learning model to automate the design process of hot forming process parameters, thereby reducing manual intervention, shortening the design cycle, improving efficiency, reducing the trial and error process, and reducing R&D costs. At the same time, by using different types of data (images, texts, videos), the model has higher robustness to the diversity and changes of data. Multimodal data can help the model generalize better, thereby reducing the risk of overfitting and improving performance on new data.

[0023] It should be noted that data crawling is used to extract structured and unstructured information from a large number of web pages, mainly using web crawler technology to legally collect authoritative journals, design standards, experimental cases, videos and pictures related to hot forming. Image data includes but is not limited to production line photos, laboratory images, microscope images, etc. of hot forming processes. The content contained in the image data includes equipment status, material surface characteristics, and the appearance of formed products.

[0024] Text data includes but is not limited to process documents, operation manuals, experimental reports, technical specifications, etc. The content contained in the text data includes process steps, parameter settings, operating instructions, experimental results, etc.

[0025] Video data includes but is not limited to real-time monitoring video of the hot forming process, video recording of the experimental process, etc. Video data records the dynamic changes of the entire process, such as the time series of heating, forming, cooling and other processes.

[0026] Furthermore, the multimodal thermoforming process database includes thermoforming process data, product material and performance requirements data, and process parameters. Specifically, process parameters include information such as equipment configuration, settings, and operating status, such as heating temperature, cooling rate, forming pressure, etc. Material type, composition, physical and chemical properties, such as thermal expansion coefficient, fluidity, etc.

[0027] In some embodiments, the image data, text data and video data are fused to obtain a multi-modal hot forming process database, see Figure 2 ,include: S201, using a well-trained recurrent neural network model to extract features from text data to obtain text features; S202, using a well-trained convolutional neural network model to extract features from the image data to obtain image features; S203, extracting features from the video data using a well-trained long short-term memory neural network model to obtain video features; S204, using a preset spatial alignment algorithm and a preset temporal alignment algorithm to align and splice text features, graphic features, and video features to obtain fused multimodal thermoforming process data, and constructing a multimodal thermoforming process database based on the multimodal thermoforming process data.

[0028] In this embodiment, after feature extraction is performed on text data, image data, and video data, they are fused. Early fusion (fusion at the feature extraction stage), mid-term fusion (fusion at the middle layer of the model), or late fusion (fusion at the output layer of the model) can be used. In this embodiment, the feature concatenation method in the early fusion algorithm is used to concatenate feature vectors of different modalities into a long feature vector. This method helps integrate data of different modalities, so that subsequent models can better learn the association and complementarity between data. Compared with late fusion, early fusion can more comprehensively retain the data features of each modality.

[0029] Please note that Figure 3 , the embodiment of the present invention adopts the transformer structure as the architecture of the large model. The transformer includes two parts: encoder and decoder, and uses the self-attention mechanism to train data to generate a prediction model. In the encoder, the formula for calculating the attention score is: ,in, represents the query vector, represents the key vector, represents a value vector, represents the number of dimensions, and the attention score is weighted and summed with the representation of the input sequence to obtain the output of the encoder. is the activation function, The transformer structure does not depend on the order of sequence data and can process all positions in the sequence in parallel, improving the efficiency of training. It can also effectively capture the relationship between distant elements in the sequence and avoid the gradient vanishing and gradient exploding problems.

[0030] In some embodiments, the initial deep learning model is constructed, and the data in the multimodal hot forming process database is input as input parameters into the initial deep learning model for iterative training to obtain a target deep learning model, including: Slice the data in the multimodal hot forming process database to obtain multiple training data sets; Distribute multiple training sets across different devices and perform distributed training on the models in different devices.

[0031] In this embodiment, the entire model copy is copied to each device, and each copy is trained using a different data set. After each device calculates the gradient, the gradient information is synchronized to all devices, the model parameters are updated, and the time required for model training is reduced. In addition, distributed training can be performed with larger data and more computing resources, which may improve the performance and accuracy of the model.

[0032] Furthermore, the large model includes a classifier model and a regression model. The loss function for evaluating the classifier model is the cross-entropy loss function, and the formula is ,in For the sample, Indicates the actual label, represents the predicted output, Indicates the total sample size.

[0033] Preferably, the loss function of the regression model is the Huber Loss function, which is expressed as: ,in Indicates the actual label, is the predicted label, The best value for the hyperparameter can be selected through cross-validation.

[0034] Preferably, the cross entropy loss function and the Huber loss function are used to evaluate the model accuracy. If the evaluation result is qualified, there is no need to perform large model fine-tuning, otherwise the large model needs to be updated.

[0035] Preferably, the large model uses the low-rank adaptive fine-tuning method (Low-Rank-Adaptation) to adjust the model parameters, and the adjustment formula is: , in are the parameters for initializing the pre-trained model, is a Gaussian matrix, is a 0 matrix, is the training data, To generate data.

[0036] In some embodiments, see Figure 4 , the product performance requirements are input as input parameters into the target deep learning model to obtain the target hot forming process parameters, including: Input product performance requirements as input parameters into the target deep learning model to match the corresponding product material information and product performance information; Determine the target hot forming process parameters based on product material information and product performance information.

[0037] In this embodiment, a large model of the hot forming process is generated, and the hot forming process is designed according to the performance requirements of products such as high-strength steel, aluminum alloy, and titanium alloy.

[0038] Based on the above-mentioned hot forming process parameter design method based on deep learning model, the embodiment of the present invention also provides a hot forming process parameter design device based on deep learning model, please refer to Figure 5 ,include: A fusion module 510 is used to obtain image data, text data and video data in the standard hot forming process and the experimental process, and fuse the image data, text data and video data to obtain a multi-modal hot forming process database; A training module 520 is used to construct an initial deep learning model, input the data in the multimodal hot forming process database as input parameters into the initial deep learning model for iterative training, and obtain a target deep learning model; The parameter design module 530 is used to input the product performance requirements as input parameters into the target deep learning model to obtain the target hot forming process parameters.

[0039] like Figure 6 As shown, based on the above-mentioned hot forming process parameter design method based on deep learning model, the present invention also provides an electronic device, which can be a computing electronic device such as a mobile terminal, a desktop computer, a notebook, a palm computer and a server. The electronic device includes a processor 610, a memory 620 and a display 630. Figure 6 Only some components of the electronic device are shown, but it should be understood that it is not required to implement all of the components shown, and more or fewer components may be implemented instead.

[0040] In some embodiments, the memory 620 may be an internal storage unit of the electronic device, such as a hard disk or memory of the electronic device. In other embodiments, the memory 620 may also be an external storage electronic device of the electronic device, such as a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (SecureDigital, SD) card, a flash card (Flash Card), etc. equipped on the electronic device. Further, the memory 620 may also include both an internal storage unit of the electronic device and an external storage electronic device. The memory 620 is used to store application software and various types of data installed in the electronic device, such as program codes for installing the electronic device. The memory 620 may also be used to temporarily store data that has been output or is to be output. In one embodiment, a hot forming process parameter design program 640 is stored on the memory 620, and the hot forming process parameter design program 640 can be executed by the processor 610, thereby realizing the hot forming process parameter design method based on the deep learning model of each embodiment of the present application.

[0041] In some embodiments, the processor 610 can be a central processing unit (CPU), a microprocessor or other data processing chip, which is used to run the program code or process data stored in the memory 620, such as executing a hot forming process parameter design method based on a deep learning model.

[0042] In some embodiments, the display 630 may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, an OLED (Organic Light-Emitting Diode) touch device, etc. The display 630 is used to display information of the electronic device designed in the thermoforming process parameters and to display a visual user interface. The components 610-630 of the electronic device communicate with each other through a system bus.

[0043] Those skilled in the art will appreciate that all or part of the processes of the above-mentioned embodiments can be implemented by instructing related hardware through a computer program, and the program can be stored in a computer-readable storage medium, wherein the computer-readable storage medium is a disk, an optical disk, a read-only storage memory, or a random access memory, etc.

[0044] 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 changes or substitutions that can be easily conceived by any technician familiar with the technical field within the technical scope disclosed by the present invention should be covered within the protection scope of the present invention.

Claims

1. A method for designing hot forming process parameters based on a deep learning model, characterized in that: include: Obtain image data, text data and video data in the standard hot forming process and the experimental process, and fuse the image data, text data and video data to obtain a multimodal hot forming process database; An initial deep learning model is constructed, and the data in the multimodal hot forming process database is input as input parameters into the initial deep learning model for iterative training to obtain a target deep learning model; The product performance requirements are input as input parameters into the target deep learning model to obtain the target hot forming process parameters.

2. The method for designing hot forming process parameters based on a deep learning model according to claim 1, characterized in that: The multimodal thermoforming process database includes thermoforming process data, product material and performance requirement data, and process parameters.

3. The method for designing hot forming process parameters based on a deep learning model according to claim 2, characterized in that: The image data, text data and video data are fused to obtain a multi-modal hot forming process database, including: Use a well-trained recurrent neural network model to extract features from text data and obtain text features; Use a well-trained convolutional neural network model to extract features from image data and obtain image features; The video data is extracted using a well-trained long short-term memory neural network model to obtain video features; A preset spatial alignment algorithm and a preset temporal alignment algorithm are used to align and splice text features, graphic features, and video features to obtain fused multimodal thermoforming process data, and a multimodal thermoforming process database is constructed based on the multimodal thermoforming process data.

4. The method for designing hot forming process parameters based on a deep learning model according to claim 1, characterized in that: The initial deep learning model is constructed, and the data in the multimodal hot forming process database is input as input parameters into the initial deep learning model for iterative training to obtain a target deep learning model, including: Slice the data in the multimodal hot forming process database to obtain multiple training data sets; Distribute multiple training sets across different devices and perform distributed training on the models in different devices.

5. The method for designing hot forming process parameters based on deep learning model according to claim 1, characterized in that: The target deep learning model includes a classifier model and a regression model. The loss function of the classifier model is expressed by the following formula: Among them, For the sample, Indicates the actual label, represents the predicted output, Indicates the total sample size; The loss function of the regression model is expressed by the following formula: in, Indicates the actual label, is the predicted label, is a hyperparameter.

6. The method for designing hot forming process parameters based on a deep learning model according to claim 1, characterized in that: The method of inputting the data in the multimodal hot forming process database as input parameters into the initial deep learning model for iterative training includes: A preset low-rank adaptive fine-tuning method is used to adjust the parameters of the initial deep learning model.

7. The method for designing hot forming process parameters based on deep learning model according to claim 1, characterized in that: The product performance requirements are input as input parameters into the target deep learning model to obtain target hot forming process parameters, including: Input product performance requirements as input parameters into the target deep learning model to match the corresponding product material information and product performance information; Determine the target hot forming process parameters based on product material information and product performance information.

8. A hot forming process parameter design device based on a deep learning model, characterized in that: include: A fusion module is used to obtain image data, text data and video data in the standard hot forming process and the experimental process, and fuse the image data, text data and video data to obtain a multi-modal hot forming process database; A training module is used to construct an initial deep learning model, and input the data in the multimodal hot forming process database as input parameters into the initial deep learning model for iterative training to obtain a target deep learning model; The parameter design module is used to input the product performance requirements as input parameters into the target deep learning model to obtain the target hot forming process parameters.

9. An electronic device, characterized in that: include: Processor and memory; The memory stores a computer-readable program executable by the processor; When the processor executes the computer-readable program, the steps in the hot forming process parameter design method based on the deep learning model as described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores one or more programs, and the one or more programs can be executed by one or more processors to implement the steps in the hot forming process parameter design method based on the deep learning model as described in any one of claims 1-7.

Images