Residual stress analysis method and device, computer storage medium and terminal

Abstract

The invention discloses a residual stress analysis method and device, a computer storage medium and a terminal.The method comprises the steps that a preset number of initial residual stress distribution data of a thin-walled workpiece blank is generated, and each initial residual stress distribution data is simulated through a finite element model; rough machining deformation field data caused by redistribution of the residual stress after the blank material is removed of each initial residual stress distribution data are obtained; in order to solve the problem that complexity in the manufacturing process needs to be considered when initial residual stress field data is obtained from rough machining deformation field data inversion, a deep neural network model used for determining the mapping relation between initial residual stress data and rough machining deformation data is established according to mapping relation data; rough machining deformation field data of actual machining is input into the deep neural network model, complex theoretical analysis and derivation are bypassed, and on the basis of not damaging the structure of the blank, the initial residual stress field prediction of the thin-walled workpiece blank is achieved.

Description

Technical Field

[0001] This article relates to intelligent manufacturing technology, and in particular to a method, apparatus, computer storage medium and terminal for residual stress analysis. Background Technology

[0002] In recent years, driven by the demand for lightweighting, the aerospace industry has used a large number of large thin-walled structural components. These components are often characterized by complex structures, large dimensions, high material removal rates, and low stiffness, making them prone to deformation after machining. During the machining of thin-walled components, the release and redistribution of residual stress within the blank are the main factors causing machining deformation. Therefore, determining the initial residual stress field data of the blank is of great significance for the analysis of machining deformation of thin-walled structural components.

[0003] In related technologies, residual stress measurement methods can be mainly divided into destructive methods and non-destructive methods. Destructive methods destroy the structure of the blank and indirectly measure the initial residual stress field data by the deformation caused by the redistribution of residual stress. Non-destructive methods include X-ray diffraction and neutron diffraction, which generally have limited measurement depth and cannot obtain three-dimensional initial residual stress field data.

[0004] In summary, how to determine the initial residual stress field data without damaging the blank structure has become a problem that needs to be solved. Summary of the Invention

[0005] This application provides a method for residual stress analysis, including: Simulation generates a preset number of initial residual stress distribution data for thin-walled part blanks; Each time an initial residual stress is applied to all nodes of the finite element model based on an initial residual stress distribution data, a finite element simulation is performed, and a set of mapping relationship data corresponding to each initial residual stress distribution data is extracted. The finite element model is a pre-established model of rough machining of a thin-walled part blank, and the set of mapping relationship data includes the initial residual stress field data and rough machining deformation field data corresponding to the initial residual stress distribution data. Based on a preset number of mapping relationship data, a deep neural network model is trained to determine the mapping relationship between the initial residual stress field data and the roughing deformation field data; By inputting the roughing deformation field data of the actual blank of the thin-walled part into the deep neural network model, the predicted value of the initial residual stress field of the blank of the thin-walled part is obtained.

[0006] On the other hand, embodiments of this application also provide a computer storage medium storing a computer program, which, when executed by a processor, implements the above-described method for residual stress analysis.

[0007] Furthermore, embodiments of this application also provide a terminal, including: a memory and a processor, wherein the memory stores a computer program; wherein, The processor is configured to execute computer programs in memory; When the computer program is executed by the processor, it implements the residual stress analysis method described above.

[0008] Furthermore, embodiments of this application also provide an apparatus for residual stress analysis, comprising: a generation unit, a simulation unit, a training unit, and an application processing unit; wherein, The generation unit is set to: simulate and generate a preset number of initial residual stress distribution data for thin-walled part blanks; The simulation unit is set up as follows: each time an initial residual stress is applied to all nodes of the finite element model based on an initial residual stress distribution data, a finite element simulation is performed, and a set of mapping relationship data corresponding to each initial residual stress distribution data is extracted. The finite element model is a pre-established model of rough machining of a thin-walled part blank, and the set of mapping relationship data includes the initial residual stress field data and rough machining deformation field data corresponding to the initial residual stress distribution data. The training unit is set to: train a deep neural network model to determine the mapping relationship between the initial residual stress field data and the roughing deformation field data based on a preset number of sets of mapping relationship data; The processing unit is configured to input the roughing deformation field data of the actual blank of the thin-walled part into the deep neural network model to obtain the predicted value of the initial residual stress field of the blank of the thin-walled part.

[0009] This embodiment simulates and generates a preset number of initial residual stress distribution data for a thin-walled part blank. For each initial residual stress distribution data, a finite element model is used for simulation to obtain the roughing deformation field data caused by the redistribution of residual stress after material removal, corresponding to the simulated initial residual stress distribution data. To address the need to consider more complex issues in the blank manufacturing process when inverting the initial residual stress field data from the roughing deformation field data of the thin-walled part blank, this embodiment establishes a deep neural network model based on the extracted preset number of mapping relationship data to determine the mapping relationship between the initial residual stress data and the roughing deformation data, providing technical support for predicting the initial residual stress field during the roughing of thin-walled parts. By inputting the actual roughing deformation field data of the thin-walled part blank into the deep neural network model, the complex theoretical derivation for determining the initial residual stress field data based on the roughing deformation field data is bypassed. This achieves the prediction of the initial residual stress field of the thin-walled part blank without damaging the blank structure, providing data support for the roughing deformation analysis of thin-walled parts.

[0010] Other features and advantages of this application will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing the application. Other advantages of this application can be realized and obtained by means of the solutions described in the description and the accompanying drawings. Attached Figure Description

[0011] The accompanying drawings are used to provide an understanding of the technical solutions of this application and constitute a part of the specification. They are used together with the embodiments of this application to explain the technical solutions of this application and do not constitute a limitation on the technical solutions of this application.

[0012] Figure 1 This is a flowchart of the residual stress analysis method according to an embodiment of the present disclosure; Figure 2 This is a structural block diagram of the apparatus for residual stress analysis according to an embodiment of this disclosure; Figure 3 This is a flowchart illustrating a method used in an application example of this disclosure. Figure 4 This is a schematic diagram of the initial residual stress distribution data in an embodiment of this disclosure; Figure 5 This is a schematic diagram of the simulation results of Embodiment 1 of this disclosure; Figure 6 This is a schematic diagram of another simulation result of an embodiment of this disclosure; Figure 7 This is a schematic diagram of a three-dimensional convolutional neural network according to an embodiment of the present disclosure; Figure 8 This is a schematic diagram showing the stress data cloud map mapped by the deep neural network in the embodiments of this disclosure and the initial residual stress distribution state generated by finite element simulation; Figure 9 This is a schematic diagram of the root mean square error and coefficient of determination of the predicted values ​​in an embodiment of this disclosure. Detailed Implementation

[0013] This application describes several embodiments, but these descriptions are exemplary and not restrictive, and it will be apparent to those skilled in the art that many more embodiments and implementations are possible within the scope of the embodiments described herein. Although many possible combinations of features are shown in the drawings and discussed in the detailed description, many other combinations of the disclosed features are also possible. Unless specifically limited, any feature or element of any embodiment may be used in combination with, or may replace, any feature or element of any other embodiment.

[0014] This application includes and contemplates combinations of features and elements known to those skilled in the art. The embodiments, features, and elements disclosed in this application can also be combined with any conventional features or elements to form unique inventive solutions. Any feature or element of any embodiment can also be combined with features or elements from other inventive solutions to form another unique inventive solution. Therefore, it should be understood that any feature shown and / or discussed in this application can be implemented individually or in any suitable combination. Therefore, the embodiments are not limited except by the limitations imposed by the appended claims and their equivalents. Furthermore, various modifications and changes can be made within the scope of the appended claims.

[0015] Furthermore, in describing representative embodiments, the specification may have presented methods and / or processes as a specific sequence of steps. However, the method or process should not be limited to the specific order of steps described herein, to the extent that it does not depend on such a specific order. As will be understood by those skilled in the art, other sequences of steps are also possible. Therefore, the specific order of steps set forth in the specification should not be construed as a limitation of the claims. Moreover, the claims concerning the method and / or process should not be limited to the steps performed in the written order, and those skilled in the art will readily understand that these orders can be varied and still remain within the spirit and scope of the embodiments of this application.

[0016] Figure 1 This is a flowchart of the residual stress analysis method according to an embodiment of the present disclosure, such as... Figure 1 As shown, for each thin-walled component to be analyzed, the following are included: Step 101: Simulate and generate a preset number of initial residual stress distribution data for the thin-walled part blank; Step 102: After applying initial residual stress to all nodes of the finite element model based on an initial residual stress distribution data each time, perform finite element simulation and extract a set of mapping relationship data corresponding to each initial residual stress distribution data. The finite element model is a pre-established model of rough machining of a thin-walled part blank. The set of mapping relationship data includes the initial residual stress field data and rough machining deformation field data corresponding to the initial residual stress distribution data. Step 103: Based on a preset number of mapping relationship data, train a deep neural network model to determine the mapping relationship between the initial residual stress field data and the roughing deformation field data; Step 104: Input the roughing deformation field data of the actual blank of the thin-walled part into the deep neural network model to obtain the predicted value of the initial residual stress field of the blank of the thin-walled part.

[0017] This embodiment simulates and generates a preset number of initial residual stress distribution data for a thin-walled part blank. For each initial residual stress distribution data, a finite element model is used for simulation to obtain the roughing deformation field data caused by the redistribution of residual stress after material removal, corresponding to the simulated initial residual stress distribution data. To address the need to consider more complex issues in the blank manufacturing process when inverting the initial residual stress field data from the roughing deformation field data of the thin-walled part blank, this embodiment establishes a deep neural network model based on the extracted preset number of mapping relationship data to determine the mapping relationship between the initial residual stress data and the roughing deformation data, providing technical support for predicting the initial residual stress field during the roughing of thin-walled parts. By inputting the actual roughing deformation field data of the thin-walled part blank into the deep neural network model, the complex theoretical derivation for determining the initial residual stress field data based on the roughing deformation field data is bypassed. This achieves the prediction of the initial residual stress field of the thin-walled part blank without damaging the blank structure, providing data support for the roughing deformation analysis of thin-walled parts.

[0018] In one exemplary instance, embodiments of this disclosure simulate and generate a predetermined number of initial residual stress distribution data for a thin-walled part blank, including: Based on the residual stress distribution characteristics of the thin-walled blank, the initial residual stress distribution data is generated by simulation.

[0019] In this embodiment of the disclosure, the residual stress of the pre-stretched plate blank exhibits a layered distribution, that is, in the thickness direction ( (Direction) Determined height , Residual stress in the direction The distribution is uniform, along... Directional residual stress The initial residual stress distribution exhibits an "M"-shaped curve distribution; therefore, the simulation generates random initial residual stress distribution data. , The curve is mainly generated by adding random noise to an "M"-shaped curve.

[0020] In one exemplary instance, the preset quantity in this embodiment includes a value greater than or equal to 10; in other words, this embodiment simulates and generates initial residual stress distribution data for more than 10 stress fields; this embodiment can use related technologies, including Matlab data processing software, to generate the initial residual stress distribution data.

[0021] In one exemplary embodiment, this disclosure describes training a deep neural network model to determine the mapping relationship between initial residual stress field data and roughing deformation field data based on a preset number of mapping relationship data sets, including: A training set containing a first preset ratio and a test set containing a second preset ratio are constructed based on the mapping relationship data of a preset number of groups; A deep neural network model was established and trained using roughing deformation field data from the training set as input and initial residual stress field data as output. The trained deep neural network model is tested using a test set to obtain a deep neural network model that meets the pre-set prediction accuracy.

[0022] In one exemplary instance, this disclosure embodiment can refer to related technologies, write a Python script to establish a finite element model of a thin-walled component, calculate the initial residual stress value of each node on the finite element model by interpolation, save it to a CSV file, and load the initial residual stress distribution data into the model by modifying the initial condition keywords in the finite element software ABAQUS in the script.

[0023] In one exemplary instance, the initial residual stress field data in the mapping relationship data of this disclosure embodiment includes the following results of residual stress redistribution simulated by a finite element model: Stress components in six directions; Correspondingly, the roughing deformation field data extracted from the mapping relationship data includes data in the following three directions: ; in, This represents the normal stress along the X direction. This represents the normal stress along the Y direction. This represents the normal stress along the Z direction. This represents the shear stress in the XY plane. This represents the shear stress in the XZ plane. This represents the shear stress in the YZ plane. This represents the displacement along the X direction. This represents the displacement along the Y direction. This represents the displacement along the Z direction.

[0024] In this embodiment of the disclosure, in order to organize the dataset into a form suitable for convolution processing, it is also necessary to export the three-dimensional coordinate data of each node of the finite element model.

[0025] The deep neural network model disclosed in this embodiment is not limited to a specific architecture or type, and can be any model capable of mapping three-dimensional data.

[0026] In one exemplary instance, the deep neural network model of this disclosure includes: Convolutional Neural Networks (CNN), Graph Neural Networks (GNN), or combinations thereof.

[0027] In one exemplary instance, the evaluation criteria for the prediction accuracy of the deep neural network described in this disclosure include: Root Mean Square Error (RMSE): ; Coefficient of determination The expression is ; In the formula, This represents the total number of nodes in the finite element model. and Respectively The actual and predicted values ​​of each stress component; This represents the average value of the true stress components.

[0028] In one exemplary instance, the process type and parameters, workpiece material and dimensions, etc., during the roughing process of thin-walled parts in this disclosure embodiment are consistent with the finite element model used to generate the dataset. The sampling points for roughing deformation measurement are as dense as possible to enhance the reliability of the data after interpolation to the finite element mesh nodes.

[0029] This disclosure also provides a computer storage medium storing a computer program, which, when executed by a processor, implements the above-described method for residual stress analysis.

[0030] This disclosure also provides a terminal, including: a memory and a processor, wherein the memory stores a computer program; wherein, The processor is configured to execute computer programs in memory; The computer program is executed by the processor to implement the residual stress analysis method described above.

[0031] Figure 2 This is a structural block diagram of the apparatus for residual stress analysis according to an embodiment of this disclosure, as shown below. Figure 2 As shown, it includes: a generation unit, a simulation unit, a training unit, and an application processing unit; wherein, The generation unit is set to: simulate and generate a preset number of initial residual stress distribution data for thin-walled part blanks; The simulation unit is set up as follows: each time an initial residual stress is applied to all nodes of the finite element model based on an initial residual stress distribution data, a finite element simulation is performed, and a set of mapping relationship data corresponding to each initial residual stress distribution data is extracted. The finite element model is a pre-established model of rough machining of a thin-walled part blank, and the set of mapping relationship data includes the initial residual stress field data and rough machining deformation field data corresponding to the initial residual stress distribution data. The training unit is set to: train a deep neural network model to determine the mapping relationship between the initial residual stress field data and the roughing deformation field data based on a preset number of sets of mapping relationship data; The processing unit is configured to input the roughing deformation field data of the actual blank of the thin-walled part into the deep neural network model to obtain the predicted value of the initial residual stress field of the blank of the thin-walled part.

[0032] In one exemplary instance, the generation unit of this disclosure embodiment is configured as follows: Based on the residual stress distribution characteristics of the thin-walled blank, a preset number of initial residual stress distribution data are generated through simulation.

[0033] In one exemplary instance, the preset quantity in this disclosure embodiment includes a value greater than or equal to 10.

[0034] In one exemplary instance, the training unit of this disclosure embodiment is configured as follows: A training set containing a first preset ratio and a test set containing a second preset ratio are constructed based on the mapping relationship data of a preset number of groups; A deep neural network model was established and trained using roughing deformation field data from the training set as input and initial residual stress field data as output. The trained deep neural network model is tested using a test set to obtain a deep neural network model that meets the pre-set prediction accuracy.

[0035] In one exemplary instance, the initial residual stress field data in the mapping data of this disclosure embodiment includes the following results of residual stress redistribution simulated by a finite element model: Stress components in six directions; Correspondingly, the roughing deformation field data extracted from the mapping relationship data includes data in the following three directions: ; in, This represents the normal stress along the X direction. This represents the normal stress along the Y direction. This represents the normal stress along the Z direction. This represents the shear stress in the XY plane. This represents the shear stress in the XZ plane. This represents the shear stress in the YZ plane. This represents the displacement along the X direction. This represents the displacement along the Y direction. This represents the displacement along the Z direction.

[0036] In one exemplary instance, the deep neural network model of this disclosure includes: Convolutional Neural Network (CNN) or Graph Neural Network (GNN).

[0037] In one exemplary instance, the evaluation criteria for the prediction accuracy of the deep neural network in this disclosure embodiment include: Root Mean Square Error (RMSE): ; Coefficient of determination The expression is ; In the formula, This represents the total number of nodes in the finite element model. and Respectively The actual and predicted values ​​of each stress component; This represents the average value of the true stress components.

[0038] The following application examples briefly illustrate the embodiments of this disclosure. These application examples are only used to illustrate the embodiments of this disclosure and are not intended to limit the scope of protection of the embodiments of this disclosure.

[0039] Application Examples This disclosure, in its embodiments, uses finite element model simulation and deep neural networks to predict and analyze the initial residual stress field of the blank during the machining process for each different thin-walled part. Figure 3 The method flowchart for this public application example includes: Step 301: Based on the residual stress distribution characteristics of the thin-walled blank, simulate and generate initial residual stress distribution data; in this embodiment, a pre-stretched plate blank is selected, whose residual stress exhibits a layered distribution, that is, in the thickness direction ( (Direction) Determined height ,exist Residual stress in the direction It is a uniform distribution, see [link / reference] Figure 4 Curve 4-1, along Directional residual stress It is distributed in an "M" shaped curve, with curve 4-2 along... Directional residual stress The distribution is in the shape of an "M" curve. Based on this "M" curve, random amplitude parameters and random noise are added to generate 1000 sets of random initial residual stress distribution data.

[0040] Step 302: Establish a finite element model of the thin-walled component by writing a Python script, and load the initial residual stress distribution data generated by simulation into the finite element model; In this embodiment of the disclosure, the initial residual stress value of each node of the finite element model is calculated by interpolation, the result is saved to a CSV file, and the initial residual stress distribution data generated by simulation is loaded into the finite element model by modifying the initial condition keywords in the ABAQUS software. Step 303: Based on each initial residual stress distribution data loaded into the finite element model, perform finite element simulation on the roughing process of the thin-walled part, and extract a set of mapping relationship data corresponding to each initial residual stress distribution data; the simulation process in this embodiment is divided into two parts: 1. After loading the initial residual stress distribution data, the residual stress is redistributed, and then the redistributed initial residual stress field data is extracted. Figure 5 This is a schematic diagram of the simulation results of Embodiment 1 of this disclosure, as shown below. Figure 5 As shown, the initial residual stress field data extracted by simulation includes Stress components in six directions; 2. Use the element birth and death method to remove the elements corresponding to the material to be removed, calculate the deformation of the part caused by the redistribution of residual stress after material removal, and then extract the rough machining deformation field data. Figure 6 This is a schematic diagram of another simulation result of an embodiment of this disclosure, as shown below. Figure 6 As shown, the roughing deformation field data extracted by simulation includes Components in three directions.

[0041] A total of 1,000 simulation operations were performed to correspond to 1,000 sets of initial residual stress distribution data. In order to organize the dataset into a form suitable for convolution processing, it is also necessary to export the three-dimensional coordinate data of each node.

[0042] Step 304: Based on the pre-determined node number index, correlate the three-dimensional coordinates, initial residual stress, and roughing deformation of the nodes in the finite element model with the extracted set of mapping relationship data. For complex models, a point cloud dataset or graph structure dataset can be created, and a corresponding deep network model can be selected for training. For the simple model in this example, it can be further simplified to 3D 10-channel data based on the coordinate position, where the 10 channels include node numbers, 6 stress components, and 3 deformation components. The node scale of this three-frame model is... Therefore, the resulting dataset size is To facilitate convolution calculations, the dataset was expanded to... The dataset is divided into training, validation, and test sets in a 6:2:2 ratio. The node numbering index in this embodiment can refer to related technologies and be numbered based on the node arrangement.

[0043] Step 305: Use the PyTorch environment to build a three-dimensional convolutional neural network 3D U-Net, and train it with the roughing deformation field data in the training set as input and the initial residual stress field data as output. Figure 7 This is a schematic diagram of a three-dimensional convolutional neural network according to an embodiment of the present disclosure, as shown below. Figure 7 As shown, the 3D convolutional neural network takes 3-channel deformation field data as input and outputs 6-channel residual stress data. The 3D U-Net network model structure diagram is shown below. Figure 4 As shown, batch normalization is introduced after each convolution operation, and the modified linear unit ReLU is used as the activation function. During training, mean squared error is used as the loss function, and the Adam optimizer is used for optimization. At the same time, a validation set is introduced during the iteration process to monitor the convergence of the model and reduce overfitting.

[0044] Step 306: Test the 3D convolutional neural network model using a test set to obtain a 3D convolutional neural network model that meets the prediction accuracy requirements; the evaluation criteria used in this embodiment include root mean square error (RMSE), expressed as follows: Coefficient of determination The expression is: .

[0045] Figure 8 This is a schematic diagram showing the stress data cloud map mapped by the deep neural network in this embodiment and the initial residual stress distribution state generated by finite element simulation, as shown below. Figure 8 As shown, the stress data cloud map mapped by the deep neural network is basically consistent with the initial residual stress distribution generated by the finite element simulation. Figure 9 This is a schematic diagram of the root mean square error and coefficient of determination of the predicted values ​​in an embodiment of this disclosure.

[0046] Step 307: Input the roughing deformation field data of the actual thin-walled part blank into the deep neural network model to obtain the predicted value of the initial residual stress field of the thin-walled part blank. In this embodiment of the disclosure, the neural network model is applied in the actual processing. After roughing the actual workpiece, its deformation field data is measured by means of in-situ detection using a coordinate measuring machine, etc. After interpolating it onto the mesh nodes of the simulation model, it is input into the trained neural network model to obtain the predicted value of the initial residual stress field of the actual blank. It should be noted that this embodiment only demonstrates the mapping accuracy of the deep neural network to the finite element simulation results. The finite element model only simulates the influence of the redistribution of initial residual stress on processing deformation by the element removal method. In the actual processing, factors affecting processing deformation also include cutting residual stress and springback deformation. The simulation model used for training the dataset can restore the actual cutting process as much as possible. This embodiment of the disclosure does not limit the simulation model.

[0047] It will be understood by those skilled in the art that all or some of the steps, systems, or apparatuses disclosed above, and their functional modules / units, can be implemented as software, firmware, hardware, or suitable combinations thereof. In hardware implementations, the division between functional modules / units mentioned above does not necessarily correspond to the division of physical components; for example, a physical component may have multiple functions, or a function or step may be performed collaboratively by several physical components. Some or all components may be implemented as software executed by a processor, such as a digital signal processor or microprocessor, or as hardware, or as an integrated circuit, such as an application-specific integrated circuit (ASIC). Such software may be distributed on a computer-readable medium, which may include computer storage media (or non-transitory media) and communication media (or transient media). As is known to those skilled in the art, the term "computer storage medium" includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, program modules, or other data). Computer storage media include, but are not limited to, RAM, ROM, EEPROM, flash memory or other memory technologies, CD-ROM, digital versatile disc (DVD) or other optical disc storage, magnetic cartridges, magnetic tape, disk storage or other magnetic storage devices, or any other medium that can be used to store desired information and can be accessed by a computer. Furthermore, it is well known to those skilled in the art that communication media typically contain computer-readable instructions, data structures, program modules, or other data in modulated data signals such as carrier waves or other transmission mechanisms, and may include any information delivery medium.

Claims

1. A method for residual stress analysis, characterized in that, include: Simulation generates a preset number of initial residual stress distribution data for thin-walled part blanks; Each time an initial residual stress is applied to all nodes of the finite element model based on an initial residual stress distribution data, a finite element simulation is performed, and a set of mapping relationship data corresponding to each initial residual stress distribution data is extracted. The finite element model is a pre-established model of rough machining of a thin-walled part blank, and the set of mapping relationship data includes the initial residual stress field data and rough machining deformation field data corresponding to the initial residual stress distribution data. Based on a preset number of mapping relationship data, a deep neural network model is trained to determine the mapping relationship between the initial residual stress field data and the roughing deformation field data; By inputting the roughing deformation field data of the actual blank of the thin-walled part into the deep neural network model, the predicted value of the initial residual stress field of the blank of the thin-walled part is obtained.

2. The method according to claim 1, characterized in that, The simulation generates a preset number of initial residual stress distribution data for the thin-walled part blank, including: Based on the residual stress distribution characteristics of the thin-walled blank, a preset number of initial residual stress distribution data are generated through simulation.

3. The method according to claim 1, characterized in that, The preset quantity can be greater than or equal to 10.

4. The method according to claim 1, characterized in that, The step of training a deep neural network model based on a preset number of mapping relationship data to determine the mapping relationship between initial residual stress field data and roughing deformation field data includes: Based on the preset number of mapping relationship data, a training set containing a first preset ratio and a test set containing a second preset ratio are constructed. Using the roughing deformation field data in the training set as input and the initial residual stress field data as output, the deep neural network model is established and trained. The trained deep neural network model is tested using a test set to obtain a deep neural network model that meets the preset prediction accuracy.

5. The method according to any one of claims 1 to 4, characterized in that, The initial residual stress field data in the mapping relationship data includes the following results of residual stress redistribution after simulation by the finite element model: Stress components in six directions; Accordingly, the roughing deformation field data extracted from the mapping relationship data includes data in the following three directions: ; in, This represents the normal stress along the X direction. This represents the normal stress along the Y direction. This represents the normal stress along the Z direction. This represents the shear stress in the XY plane. This represents the shear stress in the XZ plane. This represents the shear stress in the YZ plane. This represents the displacement along the X direction. This represents the displacement along the Y direction. This represents the displacement along the Z direction.

6. The method according to any one of claims 1 to 4, characterized in that, The deep neural network model includes: Convolutional Neural Network (CNN) or Graph Neural Network (GNN).

7. The method according to claim 6, characterized in that, The evaluation criteria for the prediction accuracy of the deep neural network include: Root Mean Square Error (RMSE): ； Coefficient of determination The expression is ; In the formula, This represents the total number of nodes in the finite element model. and Respectively The actual and predicted values ​​of each stress component; This represents the average value of the true stress components.

8. A computer storage medium storing a computer program that, when executed by a processor, implements the method for residual stress analysis as described in any one of claims 1 to 5.

9. A terminal, comprising: A memory and a processor, wherein the memory stores a computer program; wherein, The processor is configured to execute computer programs in memory; When the computer program is executed by the processor, it implements the method for residual stress analysis as described in any one of claims 1 to 5.

10. An apparatus for residual stress analysis, characterized in that, include: The system comprises a generation unit, a simulation unit, a training unit, and an application processing unit; among which, The generation unit is set to: simulate and generate a preset number of initial residual stress distribution data for thin-walled part blanks; The simulation unit is set up as follows: each time an initial residual stress is applied to all nodes of the finite element model based on an initial residual stress distribution data, a finite element simulation is performed, and a set of mapping relationship data corresponding to each initial residual stress distribution data is extracted. The finite element model is a pre-established model of rough machining of a thin-walled part blank, and the set of mapping relationship data includes the initial residual stress field data and rough machining deformation field data corresponding to the initial residual stress distribution data. The training unit is set to: train a deep neural network model to determine the mapping relationship between the initial residual stress field data and the roughing deformation field data based on a preset number of sets of mapping relationship data; The processing unit is configured to input the roughing deformation field data of the actual blank of the thin-walled part into the deep neural network model to obtain the predicted value of the initial residual stress field of the blank of the thin-walled part.

Images