Rapid springback prediction method and device for complex components based on CNN-LSTM

Abstract

The present application provides a method and device for predicting the rapid springback of complex components based on CNN-LSTM, including: obtaining three-dimensional model images of complex components with different structural features, and the yield strength and springback of a single structural feature of a complex component under different process conditions, constructing a training data set, and the process conditions include material properties, creep aging forming process, and forming die profile; pre-training a CNN-LSTM model based on the training data set; obtaining three-dimensional model images and process conditions of the complex component to be predicted, and applying the trained CNN-LSTM model to predict the yield strength and springback of the complex component to be predicted based on the three-dimensional model images and process conditions. By introducing structural features and adjusting them to be suitable for the creep aging forming process of complex components, high-precision rapid springback and mechanical property prediction can be achieved, which is convenient for screening out the optimal component size, die target profile, and optimal process.

Description

Technical Field

[0001] The present application belongs to the technical field of thin-walled component forming, and in particular, relates to a method and device for rapid springback prediction of complex components based on CNN-LSTM. Background Art

[0002] Large thin-walled components are important components and key load-bearing structures of aerospace equipment, accounting for more than 40% of the weight of the fuselage / rocket body structure. Creep aging forming technology is often used in the forming of large thin-walled components. Its advantage is that it can simultaneously form and improve the strength of the components, so it is widely used in the manufacturing of large thin-walled components in aerospace.

[0003] Before finalizing the most suitable size of large thin-walled components for creep aging forming, it is necessary to design a three-dimensional model that is constantly iterated. Before the creep aging forming test of large thin-walled components, it is necessary to first perform component-level creep aging forming finite element simulation. For these different three-dimensional models, the determination of key parameters such as the appropriate process system and mold surface requires a large number of tedious finite element calculation iterations, which leads to low efficiency and high cost.

[0004] For example, in patent CN202111291953.4, an anisotropic elastic-plastic constitutive model of aluminum alloy is established, and the anisotropic elastic-plastic constitutive model of aluminum alloy is optimized through simulation test. The elastic-plastic loading of the billet is simulated until it fits the mold surface, the component formation deviation is calculated, and the mold surface is compensated to determine the final mold surface. This patent can only be implemented after the component size is determined. It does not take into account that when the component size is adjusted to adapt to the creep forming process, the subsequent three-dimensional modeling of the component and the finite element simulation of the new three-dimensional model will bring huge time costs. Summary of the invention

[0005] The main purpose of the embodiment of the present invention is to provide a method and device for rapid springback prediction of complex components based on CNN-LSTM. By introducing structural features and adjusting them to be suitable for the creep aging forming process of complex components, high-precision rapid springback and mechanical property prediction can be achieved, which facilitates the screening of the optimal component size, mold target surface and optimal process.

[0006] In the first aspect, a method for rapid springback prediction of complex components based on CNN-LSTM is provided, the method comprising: obtaining three-dimensional model images of complex components with different structural features, and the yield strength and springback of a single structural feature of the complex components under different process conditions, and constructing a training data set, wherein the process conditions include material properties, creep aging forming process, and forming die surface; pre-training a CNN-LSTM model based on the training data set to obtain a trained CNN-LSTM model; obtaining a three-dimensional model image and process conditions of the complex component to be predicted, and applying the trained CNN-LSTM model to predict the yield strength and springback of the complex component to be predicted based on the three-dimensional model image and process conditions.

[0007] In one possible implementation, three-dimensional model images of complex components with different structural features, as well as the yield strength and springback of a single structural feature of the complex components under different process conditions are obtained to construct a training data set, including: obtaining three-dimensional model images of complex components with different structural features; establishing macro- and micro-constitutive equations of materials for creep amount and yield strength under different process conditions; performing creep aging forming finite element simulation at the specimen level and embedding the macro- and micro-constitutive equations of materials to obtain the yield strength and springback of a single structural feature of the complex components under different process conditions; and combining the three-dimensional model images, the yield strength and springback under different process conditions to construct a training data set.

[0008] In another possible implementation, the macro-micro constitutive equations of the creep amount and yield strength under different process conditions are established, including: constructing a creep aging forming experiment with a standard creep specimen; and establishing the macro-micro constitutive equations of the material for predicting the creep amount and yield strength under different process conditions from different times, temperatures and stresses through the experiment:

[0009]

[0010] in, is the creep strain; is the applied creep stress, , , , , , , , , , , is the material constant; is the average radius of the spherical precipitates, , , They represent precipitation strengthening, solid solution strengthening and yield strength changes respectively.

[0011] In another possible implementation, the CNN-LSTM model is pre-trained based on the training data set to obtain a trained CNN-LSTM model, including: applying the CNN model to extract the structural features of the complex component based on the three-dimensional model images of different structural features; using the structural features and process conditions as the input of the LSTM model, and applying the LSTM model to output the predicted yield strength and predicted springback of the complex component; calculating the error between the predicted yield strength, predicted springback and the yield strength and springback in the training data set, and adjusting the parameters of the CNN model and the LSTM model according to the error until the training termination condition is met, thereby obtaining the trained CNN-LSTM model.

[0012] In another possible implementation, a CNN model is used to extract structural features of a complex component based on 3D model images of different structural features, including: converting 3D model images of different structural sizes of the complex component into a pixel matrix; using multi-layer convolution in the CNN model to extract priority features and non-priority features in the 3D model image, wherein the convolution kernel is used as , the features of two one-dimensional convolution combinations The extraction formula is:

[0013]

[0014] in, is the index of the output sequence, are the elements of the input sequence, is the convolution kernel weight; all the extracted features are normalized through the normalization layer, the features are compressed and dimensionally reduced through the pooling layer, the adjacent pixels are counted, and all the features are fused to obtain the structural features of complex component image recognition.

[0015] In another possible implementation, the trained CNN-LSTM model is applied to predict the yield strength and springback of the complex component to be predicted according to the three-dimensional model image and process conditions, including: applying the CNN model to extract the structural features of the complex component to be predicted according to the three-dimensional model image; and applying the LSTM model to predict the springback and yield strength of the complex component to be predicted according to the structural features and process conditions.

[0016] In another possible implementation, the LSTM model is applied to predict the springback and yield strength of the complex component to be predicted based on the structural features and process conditions, including: inputting the structural features, material properties, creep aging forming process and forming mold surface into the input gate of the LSTM model; removing part of the structural features, material properties, creep aging forming process and forming mold surface in the input through the forget gate of the LSTM model, inputting new input through the input gate, and after updating the unit state, outputting the yield strength and springback corresponding to the new complex component to be predicted from the output gate.

[0017] In the second aspect, a fast springback prediction device for complex components based on CNN-LSTM is provided, and the device includes: a training data acquisition unit, which is used to obtain three-dimensional model images of complex components with different structural features, and the yield strength and springback of a single structural feature of the complex components under different process conditions, and construct a training data set, and the process conditions include material properties, creep aging forming process and forming mold surface; a model training unit, which is used to pre-train the CNN-LSTM model based on the training data set to obtain a trained CNN-LSTM model; a model prediction unit, which is used to obtain the three-dimensional model image and process conditions of the complex component to be predicted, and apply the trained CNN-LSTM model according to the three-dimensional model image and process conditions to predict the yield strength and springback of the complex component to be predicted.

[0018] In a third aspect, an electronic device is provided, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, a method for rapid springback prediction of complex components based on CNN-LSTM as provided in the first aspect is implemented.

[0019] In a fourth aspect, a non-transitory computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, a fast rebound prediction method for complex components based on CNN-LSTM as provided in the first aspect is implemented. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in describing the embodiments of the present application are briefly introduced below.

[0021] Figure 1 A flowchart of a method for rapid springback prediction of a complex component based on CNN-LSTM provided by an embodiment of the present invention;

[0022] Figure 2 A schematic diagram of a high-rib wall panel according to an embodiment of the present invention;

[0023] Figure 3 The following are examples of high-rib wall panels with different structural dimensions according to the embodiments of the present invention, wherein (a) is an example of a high-rib wall panel with dimensions of 2500 mm×2500 mm×58 mm, and (b) is an example of a high-rib wall panel with dimensions of 2500 mm×2500 mm×60 mm;

[0024] Figure 4 It is a structural diagram of a CNN-LSTM model according to an embodiment of the present invention;

[0025] Figure 5This is an example diagram of the difference between the rebound amount predicted by the CNN-LSTM model and the actual rebound amount in an embodiment of the present invention;

[0026] Figure 6 A structural diagram of a fast springback prediction device for complex components based on CNN-LSTM provided by one embodiment of the present invention;

[0027] Figure 7 A schematic diagram of the physical structure of an electronic device provided by the present invention. DETAILED DESCRIPTION

[0028] The embodiments of the present application are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar modules or modules with the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present application, and cannot be interpreted as limiting the present invention.

[0029] It will be understood by those skilled in the art that, unless expressly stated, the singular forms "a", "an", "said" and "the" used herein may also include plural forms. It should be further understood that the term "comprising" used in the specification of the present application refers to the presence of the features, integers, steps, operations, modules and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, modules, components and / or groups thereof. It should be understood that when we refer to a module as being "connected" or "coupled" to another module, it may be directly connected or coupled to the other modules, or there may be intermediate modules. In addition, the "connection" or "coupling" used herein may include wireless connection or wireless coupling. The term "and / or" used herein includes all or any modules and all combinations of one or more associated listed items.

[0030] In order to make the objectives, technical solutions and advantages of the present application clearer, the implementation method of the present application will be further described in detail below in conjunction with the accompanying drawings.

[0031] The technical solution of the present application and how to solve the above-mentioned technical problems are described in detail below with specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The embodiments of the present application will be described below in conjunction with the accompanying drawings.

[0032] like Figure 1 The flowchart of a method for rapid springback prediction of a complex component based on CNN-LSTM provided by an embodiment of the present invention is shown, and the method includes:

[0033] Step S11, obtain three-dimensional model images of complex components with different structural features, as well as the yield strength and springback of a single structural feature of the complex component under different process conditions, to construct a training data set, where the process conditions include material properties, creep aging forming process, and forming die surface.

[0034] Step S12, pre-training the CNN-LSTM model based on the training data set to obtain a trained CNN-LSTM model.

[0035] Step S13, obtaining a three-dimensional model image and process conditions of the complex component to be predicted, and applying the trained CNN-LSTM model according to the three-dimensional model image and process conditions to predict the yield strength and springback of the complex component to be predicted.

[0036] The embodiment of the present invention introduces structural features and adjusts it to be suitable for the creep aging forming process of complex components, so as to achieve high-precision rapid springback and mechanical property prediction, and facilitate the screening of the optimal component size, mold target surface and optimal process.

[0037] In step S11, optionally, obtain three-dimensional model images of complex components with different structural features; establish macro- and micro-constitutive equations of materials for creep and yield strength under different process conditions; perform creep aging forming finite element simulation at the specimen level and embed the macro- and micro-constitutive equations of materials to obtain the yield strength and springback of a single structural feature of the complex component under different process conditions; and combine the three-dimensional model images, the yield strength and springback under different process conditions to construct a training data set.

[0038] In the embodiment of the present invention, when constructing the CNN-LSTM model training database, all data are formed by experiments, which is costly and time-consuming, so the experimental data and simulation data are combined. In order to obtain high-accuracy simulation data, it is necessary to first obtain the macro-micro constitutive equations of the material with high prediction accuracy.

[0039] First, a creep aging forming experiment is conducted using a standard creep specimen. Through the experiment, the macro-micro constitutive equations of the material are established to predict the creep amount and yield strength under different process conditions at different times, temperatures and stresses:

[0040]

[0041] in, is the creep strain; is the applied creep stress, , , , , , , , , , , is the material constant; is the average radius of the spherical precipitates, , , They represent precipitation strengthening, solid solution strengthening and yield strength changes respectively.

[0042] The embodiment of the present invention takes a wall panel made of 2219 aluminum alloy as an example. Figure 2 This is a schematic diagram of a high-rib wall panel. Figure 3 The schematic diagram of high-ribbed siding with different structural dimensions is shown as an example. The macro-micro constitutive equation of the material is established to predict the creep amount and yield strength under different process conditions. The macro-micro constitutive equation of the material is embedded in the finite element simulation of creep aging forming at the specimen level. Specifically, the macro-micro constitutive equation of the material is used for secondary development and embedded in the large-scale commercial finite element software MSC.Marc to obtain the yield strength and springback of different material properties, creep aging forming processes and forming die surfaces under the same 2219 siding structural dimensions, and obtain the yield strength and springback of different material properties, creep aging forming processes and forming die surfaces under a single structural dimension of 2219 aluminum alloy siding.

[0043] Among them, the different structural dimensions of the wall panel include: overall height, skin thickness, rib height, rib thickness, and rib position. The process conditions include material properties, creep aging forming process, and forming mold profile. The material properties include yield strength, linear strengthening modulus, and elastic modulus. The creep aging forming process includes aging temperature and aging time. The forming mold profile is preferably a mold profile radius.

[0044] In the embodiment of the present invention, the structure of the CNN-LSTM model is shown in Figure 4 , the three-dimensional model images representing different structural dimensions of the wall panel are input into the CNN layer through the feature input layer, and the structural features are extracted through the CNN layer. The structural features are input into the LSTM model, and the process conditions including material properties, creep aging forming process and forming die surface are input into the LSTM model through other feature input layers for LSTM time series prediction, and the predicted springback and yield strength are output through the output layer.

[0045] In step S12, optionally, the CNN model is first applied to extract the structural features of the complex component based on the three-dimensional model images of different structural features. In the embodiment of the present invention, the three-dimensional model images of different structural features of the wallboard are used to extract the image features using the convolutional neural network (CNN) model, and all the features and dimensions are fused to obtain the structural features of the different component feature image recognition, that is, to obtain the classification information of the different component feature images.

[0046] Optionally, first convert the 3D model images of complex components of different structural sizes into pixel matrices. Since convolutional neural networks (CNN) have excellent predictive capabilities for image recognition, CNN models are used to extract image features. Specifically, multi-layer convolution in the CNN model is used to extract priority features and non-priority features in the 3D model images, where the convolution kernel , the features of two one-dimensional convolution combinations The extracted formula is:

[0047]

[0048] in, is the index of the output sequence, are the elements of the input sequence, is the convolution kernel weight.

[0049] All extracted features are normalized through the normalization layer, features are compressed and dimensionally reduced through the pooling layer, adjacent pixels are counted, and all features are fused to obtain structural features for complex component image recognition. Normalization can reduce the impact of large fluctuations in a few data on data prediction, thereby improving the prediction accuracy of the data volume. Features are compressed and dimensionally reduced through the pooling layer, adjacent pixels are counted, and their spatial size is compressed, thereby reducing the amount of image data and improving subsequent computational efficiency. Finally, all features are fused to obtain structural features for component image recognition. The simplified algorithm of the CNN model is as follows:

[0050] model=Sequential()

[0051] model.add(Conv2D(32,kernel_size=3,activation='relu',input_shape=(28,28,1)))

[0052] model.add(MaxPooling2D(pool_size=(3,3)))

[0053] model.add(Conv2D(24,kernel_size=3,activation='relu'))

[0054] model.add(MaxPooling2D(pool_size=(2,2)))

[0055] model.add(Flatten())

[0056] model.add(Dense(10,activation='softmax'))

[0057] model.compile(optimizer='adam',loss='categorical_crossentropy',metrics=['accuracy'])

[0058] After applying the CNN model to extract the structural features of complex components based on three-dimensional model images with different structural features, the structural features and process conditions are used as the input of the LSTM model, and the LSTM model is applied to output the predicted yield strength and predicted springback of the complex components.

[0059] The errors between the predicted yield strength and predicted springback and the yield strength and springback in the training data set are calculated, and the parameters of the CNN model and LSTM model are adjusted according to the errors until the training termination conditions are met to obtain the trained CNN-LSTM model.

[0060] After obtaining the trained CNN-LSTM model, the trained CNN-LSTM model is applied to predict the yield strength and springback of the complex component to be predicted according to the three-dimensional model image and process conditions.

[0061] Optionally, firstly, a CNN model is applied to extract structural features of the complex component to be predicted based on the 3D model image. The specific process is the same as the process of extracting structural features by the CNN model in the training process of the CNN-LSTM model above, and will not be repeated here.

[0062] Then, the LSTM model is used to predict the springback and yield strength of the complex component to be predicted based on the structural characteristics and process conditions.

[0063] In order to avoid the large differences in the dimensions and magnitudes of various feature values, which may affect the prediction error of the LSTM model, the data needs to be normalized.

[0064]

[0065] in, is the eigenvalue The normalized value of For the i The value of the feature, is the maximum eigenvalue, is the minimum eigenvalue.

[0066] Since the LSTM algorithm has a significant prediction effect on time series data. The structural features obtained by the CNN model and other features of the same structure: material properties (yield strength, linear strengthening modulus, elastic modulus), creep aging forming process (aging temperature, aging time) and forming die surface (die surface radius) are used as LSTM algorithm inputs and input into the input gate of the LSTM model. The forget gate of the LSTM model removes some of the structural features, material properties, creep aging forming process and forming die surface in the input, and inputs new inputs through the input gate. After the unit state is updated, the yield strength and springback of the corresponding new complex component to be predicted are output from the output gate. It should be noted that the forget gate removes unimportant structural features, material properties, creep aging forming process and forming die surface, where unimportant means that they have little effect on yield strength and springback.

[0067] Taking the siding made of 2219 aluminum alloy as an example, the input gate formula of the LSTM model is:

[0068]

[0069] Indicates that the input gate is at time step The output, is the sigmoid activation function, is the output state of the previous time step, is the input of the current time step, and are the weight and bias of the input gate respectively.

[0070] Then, the forget gate of the LSTM algorithm is used to remove unimportant siding structural features and other features from these inputs, and new inputs are input through the input gate. After the unit state is updated, the output gate outputs (siding yield strength and rebound) corresponding to the new output. The output gate formula of the LSTM model is:

[0071]

[0072] is the output gate value at the current moment, is the input at the current moment, and are the weight matrices from the input gate and the hidden layer to the output gate, is the bias term of the output gate.

[0073] Use Keras Tuner to optimize the LSTM algorithm based on mean squared error (MSE) hyperparameters, and the reference loss function is:

[0074]

[0075] in, is the calculated value of LSTM surface springback, is the surface springback of the actual component, p Indicates the number of surface points taken. The simplified algorithm of LSTM model is as follows:

[0076] Import kerastuner as kt

[0077] def build_model(hp):

[0078] model=Sequential()

[0079] model.add(LSTM(units=hp.Int('units_1',min_value=50,max_value=100,step=50),return_sequences=True,input_shape=(time_step,1)))

[0080] …

[0081] model.add(Dropout(hp.Float('dropout_2',min_value=0.1,max_value=0.5,step=0.1)))

[0082] model.add(Dense(units=1))

[0083] model.compile(optimizer='adam',loss='mean_squared_error')

[0084] return model

[0085] tuner = kt.RandomSearch(

[0086] build_model,

[0087] objective = 'val_loss',

[0088] max_trials=5,

[0089] executions_per_trial=3,

[0090] directory='my_dir',

[0091] project_name='stock_price_prediction')

[0092] tuner.search(X_train,y_train,epochs=50,validation_data=(X_test,y_test))

[0093] best_model=tuner.get_best_models(num_models=1)[0]

[0094] best_model.summary()

[0095] In the embodiment of the present invention, the structural features of the component three-dimensional model are extracted using a convolutional neural network (CNN). After extracting features from the entire component partition, discrete point modeling parameters are used to build a finite element simulation model of the process and mold profile in combination with discrete feature points of the component, and a large amount of training data is provided through a reliable finite element simulation model. Finally, based on the long short-term memory network (LSTM) algorithm, high-precision rapid springback and mechanical properties of creep aging forming of components of different sizes are predicted to screen out the optimal component size, mold target profile and optimal process.

[0096] The embodiment of the present invention carries out finite element simulation of 810 groups of different wall panels with different structural sizes, different aging times and different forming molds under aging temperatures of 160°C, 170°C and 180°C. As shown in Table 1, 81,000 wall panel model sample points are extracted from the finite element simulation results to train the CNN-LSTM algorithm, increase the diversity and complexity of the training data, and predict the rebound and yield strength of different grid-type wall panel components.

[0097]

[0098] Figure 5 The wall panel verification part with experimental conditions of aging temperature of 175℃, aging time of 2.2h, mold radius of 1350mm, wall panel height of 63mm, and annular rib thickness of (14mm, 16mm, 18mm) is used as the prediction object, and the difference diagram of the springback amount predicted by the CNN-LSTM model and the surface springback amount of the actual component is used. The value in the figure is the difference between the predicted springback amount output by the CNN-LSTM model and the actual springback amount. It can be seen from the figure that the difference between the two is very small. The CNN-LSTM model algorithm applied in the embodiment of the present invention can realize high-precision rapid springback and mechanical property prediction, which is convenient for screening out the optimal component size, mold target surface and optimal process.

[0099] In the past, creep aging forming processes have considered the influence of a single wall panel structural feature on the final springback and yield strength after creep aging forming. The embodiments of the present invention introduce wall panel structural features to adjust the wall panel structural features to be suitable for the creep aging forming process; the wall panel structural features are displayed as pictures through different CAD three-dimensional models, and the CNN algorithm is used to identify its features, and the different features of many feature points are converted from discrete display to centralized overall display, and the wall panel structural features are input as a whole into the subsequent model to improve the prediction accuracy; in the past, the wall panel structural features and other features could only be simulated and solved one by one through finite element analysis. When solving long time series, the calculation amount is greater and the time is longer. As the number of features increases, the efficiency is lower, but the LSTM algorithm can be used to achieve efficient, fast, and long-term prediction of wall panel structural features and other features.

[0100] The embodiment of the present invention constructs a training data set by acquiring three-dimensional model images of complex components with different structural features, and the yield strength and springback of a single structural feature of the complex component under different process conditions, wherein the process conditions include material properties, creep aging forming process and forming die surface; pre-training a CNN-LSTM model based on the training data set to obtain the trained CNN-LSTM model; acquiring a three-dimensional model image and process conditions of the complex component to be predicted, and applying the trained CNN-LSTM model according to the three-dimensional model image and the process conditions to predict the yield strength and springback of the complex component to be predicted, and by introducing structural features and adjusting to be suitable for the creep aging forming process of the complex component, high-precision rapid springback and mechanical property prediction can be achieved, which facilitates the screening of the optimal component size, the target die surface and the optimal process.

[0101] like Figure 6 The figure shows a structural diagram of a fast springback prediction device for complex components based on CNN-LSTM provided by one embodiment of the present invention, wherein the device comprises:

[0102] The training data acquisition unit 601 is used to acquire three-dimensional model images of complex components with different structural features, and the yield strength and springback of a single structural feature of the complex component under different process conditions, to construct a training data set, wherein the process conditions include material properties, creep aging forming process and forming die profile;

[0103] A model training unit 602 is used to pre-train the CNN-LSTM model based on the training data set to obtain the trained CNN-LSTM model;

[0104] The model prediction unit 603 is used to obtain the three-dimensional model image and process conditions of the complex component to be predicted, and apply the trained CNN-LSTM model according to the three-dimensional model image and the process conditions to predict the yield strength and springback of the complex component to be predicted.

[0105] The device of the above embodiment is applied to the corresponding method in the above embodiment, and has the beneficial effects of the corresponding method embodiment, which will not be described in detail here.

[0106] Figure 7 An example of a physical structure diagram of an electronic device is shown in FIG. Figure 7 As shown, the electronic device may include: a processor (processor) 701, a communication interface (Communications Interface) 702, a memory (memory) 703 and a communication bus 704, wherein the processor, the communication interface, and the memory complete mutual communication through the communication bus. The processor can call the logic instructions in the memory to execute a fast springback prediction method for complex components based on CNN-LSTM, the method comprising: obtaining a three-dimensional model image of a complex component with different structural features, and the yield strength and springback of a single structural feature of the complex component under different process conditions, constructing a training data set, the process conditions including material properties, creep aging forming process and forming die profile; pre-training the CNN-LSTM model based on the training data set to obtain the trained CNN-LSTM model; obtaining a three-dimensional model image and process conditions of the complex component to be predicted, and applying the trained CNN-LSTM model according to the three-dimensional model image and the process conditions to predict the yield strength and springback of the complex component to be predicted.

[0107] In addition, the logic instructions in the above-mentioned memory can be implemented in the form of software functional units and can be stored in a computer-readable storage medium when sold or used as an independent product. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc., which can store program code.

[0108] On the other hand, an embodiment of the present invention also provides a computer program product, which includes a computer program stored on a non-transitory computer-readable storage medium, and the computer program includes program instructions. When the program instructions are executed by a computer, the computer can execute a method for rapid springback prediction of a complex component based on CNN-LSTM provided by the above-mentioned method embodiments, the method including: obtaining three-dimensional model images of complex components with different structural features, and the yield strength and springback of a single structural feature of the complex component under different process conditions, and constructing a training data set, wherein the process conditions include material properties, creep aging forming process and forming mold surface; pre-training the CNN-LSTM model based on the training data set to obtain the trained CNN-LSTM model; obtaining the three-dimensional model image and process conditions of the complex component to be predicted, and applying the trained CNN-LSTM model according to the three-dimensional model image and the process conditions to predict the yield strength and springback of the complex component to be predicted.

[0109] On the other hand, an embodiment of the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to execute a method for rapid springback prediction of a complex component based on CNN-LSTM provided in the above-mentioned embodiments, the method comprising: obtaining three-dimensional model images of complex components with different structural features, and the yield strength and springback of a single structural feature of the complex component under different process conditions, and constructing a training data set, wherein the process conditions include material properties, creep aging forming process and forming mold surface; pre-training the CNN-LSTM model based on the training data set to obtain the trained CNN-LSTM model; obtaining the three-dimensional model image and process conditions of the complex component to be predicted, and applying the trained CNN-LSTM model according to the three-dimensional model image and the process conditions to predict the yield strength and springback of the complex component to be predicted.

[0110] It should be understood that, although the steps in the flowchart of the accompanying drawings are displayed in sequence as indicated by the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least a part of the steps in the flowchart of the accompanying drawings may include multiple sub-steps or multiple stages, and these sub-steps or stages are not necessarily executed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be executed in turn or alternately with other steps or at least a part of the sub-steps or stages of other steps.

[0111] The above is only a partial implementation of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A fast springback prediction method for complex components based on CNN-LSTM, characterized in that: The method comprises: Obtaining three-dimensional model images of complex components with different structural features, as well as the yield strength and springback of a single structural feature of the complex component under different process conditions, to construct a training data set, wherein the process conditions include material properties, creep aging forming process, and forming die surface; The three-dimensional model image includes structural dimensions of rib height and skin thickness; Pre-training the CNN-LSTM model based on the training data set to obtain the trained CNN-LSTM model; The pre-training of the CNN-LSTM model based on the training data set to obtain the trained CNN-LSTM model includes: Applying a CNN model to extract structural features of the complex component based on the three-dimensional model images with different structural features; Taking the structural features and the process conditions as LSTM model inputs, applying the LSTM model to output the predicted yield strength and predicted springback of the complex component; Calculate the error between the predicted yield strength, the predicted springback amount and the yield strength and the springback amount in the training data set, and adjust the parameters of the CNN model and the LSTM model according to the error until the training termination condition is met, thereby obtaining the trained CNN-LSTM model; Wherein, the application of the CNN model to extract the structural features of the complex component according to the three-dimensional model images with different structural features includes: Converting the three-dimensional model images of the complex component of different structural sizes into a pixel matrix; The multi-layer convolution in the CNN model is used to extract the priority features and non-priority features in the three-dimensional model image, wherein the convolution kernel , the features of two one-dimensional convolution combinations The extraction formula is: in, is the index of the output sequence, are the elements of the input sequence, is the convolution kernel weight; Normalizing all the extracted features through a normalization layer, compressing the features and reducing the dimension through a pooling layer, performing statistics on adjacent pixels, and fusing all the features to obtain the structural features for the complex component image recognition; Obtain a three-dimensional model image and process conditions of the complex component to be predicted, and apply the trained CNN-LSTM model according to the three-dimensional model image and the process conditions to predict the yield strength and springback of the complex component to be predicted.

2. The method according to claim 1, characterized in that The three-dimensional model images of complex components with different structural features are obtained, as well as the yield strength and springback of a single structural feature of the complex component under different process conditions, to construct a training data set, including: Obtain 3D model images of complex components with different structural features; Establish macro- and micro-constitutive equations of creep and yield strength under different process conditions; Perform creep aging forming finite element simulation at the specimen level and embed the macro and micro constitutive equations of the material to obtain the yield strength and springback of the single structural feature of the complex component under different process conditions; The three-dimensional model images, yield strength and springback under different process conditions are combined to construct a training data set.

3. The method according to claim 2, characterized in that The macro-micro constitutive equations of the material for establishing creep amount and yield strength under different process conditions include: Construct a creep aging forming experiment using a standard creep specimen; Through experiments, the macro-micro constitutive equations of materials are established to predict the creep amount and yield strength under different process conditions by different time, temperature and stress: in, is the creep strain; is the applied creep stress, , , , , , , , , , , is the material constant; is the average radius of the spherical precipitates, , , They represent precipitation strengthening, solid solution strengthening and yield strength changes respectively.

4. The method according to claim 1, characterized in that The method of applying the trained CNN-LSTM model according to the three-dimensional model image and the process conditions to predict the yield strength and springback of the complex component to be predicted includes: Extracting structural features of the complex component to be predicted by applying a CNN model according to the three-dimensional model image; The LSTM model is applied to predict the springback and yield strength of the complex component to be predicted according to the structural characteristics and the process conditions.

5. The method according to claim 4, characterized in that The applying of the LSTM model to predict the springback amount and yield strength of the complex component to be predicted according to the structural characteristics and the process conditions includes: Inputting the structural features, the material properties, the creep aging forming process and the forming die surface into the input gate of the LSTM model; The structural features, material properties, creep aging forming process and forming die surface of part of the input are removed through the forget gate of the LSTM model, and new input is input through the input gate. After the unit state is updated, the corresponding new yield strength and springback of the complex component to be predicted are output from the output gate.

6. A fast springback prediction device for complex components based on CNN-LSTM, characterized in that: Executing the method according to claim 1, the device comprises: A training data acquisition unit is used to acquire three-dimensional model images of complex components with different structural features, and the yield strength and springback of a single structural feature of the complex component under different process conditions, to construct a training data set, wherein the process conditions include material properties, creep aging forming process and forming die profile; A model training unit, used for pre-training the CNN-LSTM model based on the training data set to obtain the trained CNN-LSTM model; The model prediction unit is used to obtain a three-dimensional model image and process conditions of the complex component to be predicted, and apply the trained CNN-LSTM model according to the three-dimensional model image and the process conditions to predict the yield strength and springback of the complex component to be predicted.

7. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, a fast springback prediction method for complex components based on CNN-LSTM is implemented as described in any one of claims 1 to 5.

8. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, a fast springback prediction method for complex components based on CNN-LSTM is implemented as described in any one of claims 1 to 5.

Images