Full-field stress and deformation dynamic digital reconstruction method and device in thin-wall part assembling process and electronic equipment

Abstract

The invention relates to the technical field of thin-wall part assembly stress and deformation measurement, in particular to a full-field stress and deformation dynamic digital reconstruction method and device in the thin-wall part assembly process and electronic equipment, and the method comprises the steps: collecting surface deformation data in the thin-wall part assembly process by using a three-dimensional digital correlation system; constructing a finite element model in the thin-wall part assembly process, and performing iterative correction on the finite element model based on the acquired surface deformation data to realize correspondence between the actually measured surface deformation data and a finite element simulation stress field to obtain stress field data of the thin-wall part; constructing a training sample by using the surface deformation data and the stress field data, and training the deep learning model by using the training sample to obtain a dynamic stress field prediction model; and acquiring real-time surface deformation data in the assembling process of the thin-wall part to be assembled by using a three-dimensional digital related system, and inputting the real-time surface deformation data into the dynamic stress field prediction model to obtain a dynamic stress field of the thin-wall part to be assembled. By applying the scheme provided by the invention, the defect that the internal stress distribution in the assembly process cannot be accurately represented in a traditional measurement method can be effectively overcome, accurate measurement of the stress field in the assembly process of the thin-wall part is realized, and reliable technical support is provided for optimization of the assembly accuracy of the thin-wall part.

Description

Technical Field

[0001] The present invention relates to the technical field of measurement of assembly stress and deformation of thin-walled parts, and specifically to a method, device and electronic equipment for dynamically digitally reconstructing the full-field stress and deformation during the assembly process of thin-walled parts. Background Art

[0002] During the assembly process, thin-walled parts are prone to deformation. After deformation, it will lead to dimensional errors and shape deviations, and directly affect the assembly accuracy and stability of the thin-walled structure. Traditional assembly methods often rely on manual experience and lack real-time monitoring means. Especially for the internal stress and deformation of thin-walled parts, accurate dynamic measurement cannot be carried out. For this reason, the present invention provides a method for measuring the deformation and stress of thin-walled parts in a full-field, real-time and non-contact manner, which is used to monitor the stress change during the assembly process of thin-walled parts in real time, and provides important technical support for improving the assembly technology of thin-walled parts. Summary of the Invention

[0003] Aiming at the deficiencies of the prior art, the purpose of the present invention is to provide a method, device and electronic equipment for dynamically digitally reconstructing the full-field stress and deformation during the assembly process of thin-walled parts based on the combination of three-dimensional digital image correlation, finite element simulation and deep learning model, aiming to accurately reconstruct the stress field distribution during the assembly process of thin-walled parts.

[0004] The technical solution adopted by the present invention to solve its technical problems is: A method for dynamically digitally reconstructing the full-field stress and deformation during the assembly process of thin-walled parts, including:

[0005] S1. Using a three-dimensional digital correlation system to collect surface deformation data during the assembly process of thin-walled parts;

[0006] S2. Constructing a finite element model of the thin-walled part assembly process, and based on the collected surface deformation data, iteratively correcting the finite element model to achieve the correspondence between the measured surface deformation data and the finite element simulation stress field, and obtaining the stress field data of the thin-walled part;

[0007] S3. Using the surface deformation data and stress field data to construct training samples, and using the training samples to train a deep learning model to obtain a dynamic stress field prediction model;

[0008] S4. Using a three-dimensional digital correlation system to collect real-time surface deformation data during the assembly process of the thin-walled part to be assembled, and inputting the real-time surface deformation data into the dynamic stress field prediction model to obtain the dynamic stress field of the thin-walled part to be assembled.

[0009] As a preference, a further technical solution of the present invention is: S2 includes:

[0010] Using the finite element model to simulate the geometric shape, material properties and boundary conditions of the thin-walled part assembly structure;

[0011] The iterative correction of the finite element model is completed by calculating the difference between the surface deformation data and the deformation data at the corresponding positions in the finite element model, and iteratively correcting the assembly parameters;

[0012] Calculate the stress field data based on the corrected finite element model.

[0013] Preferably, the process of calculating the stress field data based on the corrected finite element model includes:

[0014] During the assembly process of the thin-walled part, the surface roughness will cause significant changes in the stress field, which will in turn affect the accuracy of the entire assembly process. In order to more accurately reflect the influence of roughness, the stress modeling of the finite element model considering surface roughness is as follows:

[0015] σ contact (r)=∫ S K(r,r′)·δ(r - r′)dS (1)

[0016] Among them, σ contact (r) represents the stress at the contact point r, K(r,r′) is the contact stiffness matrix related to roughness, δ(r - r′) is the influence function of surface irregularity on contact stress, and the integration region S covers the contact surface;

[0017] In response to the influence of surface roughness, a roughness weighting function W(r) is further introduced to describe the contribution of roughness at different positions, and the stress modeling is improved to:

[0018] σ contact (r)=∫ S K(r,r′)·δ(r - r′)·W(r)dS (2)

[0019] W(r)=1 + C·(‖r‖ α ) (3)

[0020] Among them, C and α are constants, representing the weighted influence of roughness on contact stress at different positions, ||r|| represents the local scale of the surface point, which is related to the distance from the contact center;

[0021] In addition, in the assembly of thin-walled parts, the contact stress field not only depends on the large-scale geometry but also is affected by the microscopic-scale surface roughness; based on this, through a cross-scale modeling method, the contact stresses at different scales are combined, and a scale correlation function is introduced to express the stress transfer effect between different scales, which is expressed as follows:

[0022]

[0023] Among them, γ i and βi is a coefficient, L i is the size of the i-th scale, and n is the number of scales considered;

[0024] The scale correlation function is used to describe the stress transfer law from the macroscale to the microscale, connecting the contact effects at different scales of the thin-walled part. The stress modeling after introducing the scale correlation function is expressed as:

[0025] σ contact (r) = ∫ S K(r, r′)·δ(r - r′)·W(r)·S scale (r)dS (5).

[0026] Preferably, the constant C is used to control the amplitude of the influence of roughness on the contact stress. It is usually related to the material and surface quality, reflecting the overall hardness and stiffness characteristics of the surface of the thin-walled part. The calculation process is as follows:

[0027] Use a high-precision surface measurement instrument to obtain the three-dimensional topography data of the surface of the thin-walled part;

[0028] Select thin-walled part samples with different roughness grades, and measure the stress distribution in the contact area for the thin-walled part samples through loading tests or contact tests;

[0029] Obtain stress-deformation data based on the stress distribution, and calculate the value of C by the weighted average method.

[0030] Preferably, the constant α is used to control the amplitude of the influence of roughness on the contact stress. It is usually related to the material and surface quality, reflecting the overall hardness and stiffness characteristics of the surface of the thin-walled part. The calculation process is as follows:

[0031] Use a high-precision surface measurement instrument to obtain the three-dimensional topography data of the surface of the thin-walled part;

[0032] Based on the three-dimensional topography data, model the surface topography through fractal dimension calculation to obtain the fractal dimension D of the surface;

[0033] Conduct a loading test on the established model, and analyze the stress change during the contact process of the rough surface through the contact stress distribution under different roughness conditions. Finally, determine the value of α based on regression analysis.

[0034] Preferably, the deep learning model is a multi-task stress field deep learning framework constructed based on the Unet deep learning structure;

[0035] During the training process, the surface deformation data of the thin-walled part is used as input data, processed through the shared encoding layer, and the shared encoding layer mechanism is used to map different input data to a unified feature space to capture the features of the input data;

[0036] After being processed by the shared encoding layer, the captured features are distributed to multiple independent decoder branches. Each decoder branch focuses on predicting a specific stress component or deformation pattern. Each decoder has an independent network structure to decode the features of a specific assembly task and reconstruct the features into the final output.

[0037] Preferably, in order to further improve the performance of the model on different assembly tasks, a transfer learning strategy is combined at the initial stage of model training. Using the pre-trained data from similar assembly tasks helps the deep learning model quickly adapt to the current task. In the case of less training data, by transferring the features of similar assembly tasks, the learning speed is accelerated and the prediction accuracy is improved. The process of transfer learning is expressed as:

[0038]

[0039] Among them, represents the features in task i, T represents the transfer operation, are the parameters of the shared encoding layer pre-trained from task j; through the operation of transfer learning, the deep learning model can transfer and transform the features of similar assembly tasks into features available for the current task;

[0040] The deep learning model after transfer learning is trained on the data of the current task to further optimize the prediction of stress components and deformations. During the training process, the deep learning model dynamically adjusts the weight coefficients between different tasks through an adaptive loss function to ensure that the learning effect of each task is reasonably controlled. The expression of the adaptive loss function is:

[0041]

[0042] Among them, L total is the total loss, is the loss function of task k, and y k are the predicted value and the true value of task k respectively, α k is the weight coefficient of task k, and N is the number of tasks.

[0043] The present invention also discloses a device for dynamic digital reconstruction of full-field stress and deformation during the assembly process of thin-walled parts, including:

[0044] An acquisition module for using a three-dimensional digital correlation system to acquire the surface deformation data during the assembly process of thin-walled parts;

[0045] A construction module for constructing a finite element model of the assembly process of thin-walled parts and, based on the acquired surface deformation data, iteratively correcting the finite element model to achieve the correspondence between the measured surface deformation data and the finite element simulation stress field, and obtaining the stress field data of the thin-walled parts;

[0046] A training module, configured to construct training samples by using surface deformation data and stress field data, and train a deep learning model by using the training samples to obtain a dynamic stress field prediction model;

[0047] A reconstruction module, configured to collect real-time surface deformation data during the assembly process of the thin-walled part to be assembled by using a three-dimensional digital correlation system, and input the real-time surface deformation data into the dynamic stress field prediction model to obtain the dynamic stress field of the thin-walled part to be assembled.

[0048] The present invention also discloses an electronic device, including a processor, a communication interface, a memory, and a communication bus. Among them, the processor, the communication interface, and the memory complete communication with each other through the communication bus;

[0049] The memory is used to store executable instructions of the processor;

[0050] The processor, when executing the instructions stored in the memory, implements the full-field stress and deformation dynamic digital reconstruction method in the above-mentioned thin-walled part assembly process.

[0051] The present invention adopting the above technical solution, compared with the prior art, has the prominent feature that it can effectively overcome the defect that the internal stress distribution during the assembly process cannot be accurately characterized in the traditional measurement method, realize the accurate measurement of the stress field during the assembly process of the thin-walled part, and provide a reliable technical support for optimizing the assembly accuracy of the thin-walled part. Description of the Drawings

[0052] Figure 1 is a schematic flow chart of the full-field stress and deformation dynamic digital reconstruction method during the assembly process of the thin-walled part in the embodiment of the present invention;

[0053] Figure 2 is a schematic block diagram of the principle of the full-field stress and deformation dynamic digital reconstruction method during the assembly process of the thin-walled part in the embodiment of the present invention;

[0054] Figure 3 is a schematic structural diagram of a multi-task stress field deep learning framework constructed based on the Unet deep learning structure in the embodiment of the present invention;

[0055] Figure 4 is a schematic diagram of the full-field stress and deformation measurement during the welding assembly process of the thin-walled part in the embodiment of the present invention;

[0056] Figure 5 is a schematic diagram of the full-field stress and deformation measurement during the riveting assembly process of the thin-walled part in the embodiment of the present invention;

[0057] Figure 6 is a schematic diagram of the full-field stress and deformation measurement during the interference fit assembly process of the thin-walled part in the embodiment of the present invention;

[0058] Figure 7 It is a schematic structural diagram of the full-field stress and deformation dynamic digital reconstruction device during the assembly process of thin-walled parts in the embodiments of the present invention;

[0059] Figure 8 It is a schematic structural diagram of the electronic device in the embodiments of the present invention.

[0060] Explanation of reference numerals: 1. Computer; 2. Synchronous trigger; 3. Binocular camera; 4. Thin-walled part. Specific embodiments

[0061] The following further elaborates the present invention in combination with specific embodiments, and the purpose is only to better understand the content of the present invention. Therefore, the examples given do not limit the protection scope of the present invention.

[0062] As Figure 1 and Figure 2 shown, this embodiment provides a full-field stress and deformation dynamic digital reconstruction method during the assembly process of thin-walled parts, including:

[0063] S1. Use a three-dimensional digital correlation system to collect surface deformation data during the assembly process of thin-walled parts;

[0064] S2. Build a finite element model of the thin-walled part assembly process, and based on the collected surface deformation data, iteratively correct the finite element model to achieve the correspondence between the measured surface deformation data and the finite element simulation stress field, and obtain the stress field data of the thin-walled part;

[0065] S3. Use the surface deformation data and stress field data to construct training samples, and use the training samples to train a deep learning model to obtain a dynamic stress field prediction model;

[0066] S4. Use a three-dimensional digital correlation system to collect real-time surface deformation data during the assembly process of the thin-walled part to be assembled, and input the real-time surface deformation data into the dynamic stress field prediction model to obtain the dynamic stress field of the thin-walled part to be assembled.

[0067] In the implementation, referring to Figure 4 , Figure 5 and Figure 6 , the three-dimensional digital correlation system 3D-DIC includes at least a binocular camera 3 for collecting deformation images on the surface of the thin-walled part during the assembly process, a synchronous trigger 2 to control the acquisition frequency of the binocular camera 3, and image processing is performed on the collected deformation images based on the computer 1 to obtain surface deformation data; the assembly process of the thin-walled part 4 includes welding, riveting, and interference fit connection.

[0068] Specifically included in S2: Use the finite element model to simulate the geometric shape, material properties, and boundary conditions of the thin-walled part assembly structure;

[0069] The iterative correction of the finite element model is completed by calculating the difference between the surface deformation data and the deformation data at the corresponding positions in the finite element model, and iteratively correcting the assembly parameters;

[0070] Calculate the stress field data based on the corrected finite element model.

[0071] Specifically, the process of calculating the stress field data based on the corrected finite element model includes:

[0072] During the assembly process of the thin-walled part, surface roughness will cause significant changes in the stress field, which will in turn affect the accuracy of the entire assembly process. In order to more accurately reflect the influence of roughness, the stress modeling of the finite element model considering surface roughness is as follows:

[0073] σ contact (r) = ∫ S K(r, r′)·δ(r - r′)dS (1)

[0074] Among them, σ contact (r) represents the stress at the contact point r, K(r, r′) is the contact stiffness matrix related to roughness, δ(r - r′) is the influence function of surface irregularity on the contact stress, and the integration region S covers the contact surface;

[0075] In response to the influence of surface roughness, a roughness weighting function W(r) is further introduced to characterize the contribution of roughness at different positions, and the stress modeling is improved to:

[0076] σ contact (r) = ∫ S K(r, r′)·δ(r - r′)·W(r)dS (2)

[0077] W(r) = 1 + C·(‖r‖ α ) (3)

[0078] Among them, C and α are constants, representing the weighted influence of roughness on the contact stress at different positions, ||r|| represents the local scale of the surface point, which is related to the distance from the contact center;

[0079] In addition, in the assembly of thin-walled parts, the contact stress field not only depends on the large-scale geometry, but is also affected by the micro-scale surface roughness; based on this, through a cross-scale modeling method, the contact stresses at different scales are combined, and a scale correlation function is introduced to express the stress transfer effect between different scales, as shown below:

[0080]

[0081] Among them, γ i and β i are coefficients, L iis the size of the i-th scale, and n is the number of scales considered;

[0082] The scale correlation function is used to describe the stress transfer law from the macroscale to the microscale, connecting the contact effects at different scales of the thin-walled part. The stress modeling after introducing the scale correlation function is expressed as:

[0083] σ contact (r) = ∫ S K(r, r′)·δ(r - r′)·W(r)·S scale (r)dS (5).

[0084] Among them, the constant C is used to control the amplitude of the influence of roughness on the contact stress. It is usually related to the material and surface quality, reflecting the overall hardness and stiffness characteristics of the surface of the thin-walled part. The calculation process is as follows:

[0085] Use high-precision surface measurement instruments (such as white light interferometers, surface roughness meters, etc.) to obtain the three-dimensional topography data of the surface of the thin-walled part;

[0086] Select thin-walled part samples with different roughness grades, and measure the stress distribution in the contact area for the thin-walled part samples through loading tests or contact tests (such as spherical indentation tests or micro-contact mechanics tests);

[0087] Obtain stress-deformation data based on the stress distribution, and calculate the value of C by the weighted average method.

[0088] The constant α is used to control the amplitude of the influence of roughness on the contact stress. It is usually related to the material and surface quality, reflecting the overall hardness and stiffness characteristics of the surface of the thin-walled part. The calculation process is as follows:

[0089] Use high-precision surface measurement instruments to obtain the three-dimensional topography data of the surface of the thin-walled part;

[0090] Based on the three-dimensional topography data, model the surface topography through fractal dimension calculation to obtain the fractal dimension D of the surface;

[0091] Conduct a loading test on the established model, and analyze the stress changes during the contact process of the rough surface through the contact stress distribution under different roughness conditions. Finally, determine the value of α based on regression analysis.

[0092] In implementation, refer to Figure 3, the deep learning model is a multi-task stress field deep learning framework based on the Unet deep learning structure; the multi-task stress field deep learning framework (Multi-task Stress Field Learning Model, MTSL-Unet), the model uses the surface local deformation as a separate input, shares the encoding layer, and at the same time uses exclusive decoding layers to output multiple stress components of multiple contact interfaces;

[0093] During the training process, the surface deformation data of the thin-walled part is used as the input data, processed through the shared encoding layer, and the shared encoding layer mechanism is used to map the data of different inputs (such as different stress components or different assembly conditions) to a unified feature space to capture the features of the input data; this sharing mechanism greatly improves the model's multi-task ability in dealing with complex assembly processes, especially in the extraction of complex features in multiple aspects such as stress distribution and deformation.

[0094] After being processed by the shared encoding layer, the captured features will be sent to multiple independent decoder branches. Each decoder branch focuses on the prediction of a specific stress component or deformation mode. Each decoder has an independent network structure to decode the features of a specific assembly task and reconstruct the features into the final output. Through this design, MTSL-Unet can accurately process different stress components in the assembly process, ensuring that each task can be optimized in decoding and prediction.

[0095] In implementation, in order to further improve the model's performance on different assembly tasks, a transfer learning strategy is combined in the initial stage of model training, using pre-trained data from similar assembly tasks to help the deep learning model quickly adapt to the current task, so that in the case of less training data, by transferring the features of similar assembly tasks, the learning speed is accelerated and the prediction accuracy is improved. The process of transfer learning is expressed as:

[0096]

[0097] Among them, represents the features in task i (the surface local deformation of the thin-walled part measured by 3D-DIC), T represents the transfer operation, are the parameters of the shared encoding layer pre-trained from task j (the stress fields in different directions at different connection interfaces inside the thin-walled part in finite element simulation); through the operation of transfer learning, the deep learning model can transfer and transform the features of similar assembly tasks into features available for the current task;

[0098] The deep learning model after transfer learning is trained with the data for the current task to further optimize the prediction of stress components and deformations. During the training process, the deep learning model dynamically adjusts the weight coefficients between different tasks through an adaptive loss function to ensure that the learning effect of each task is reasonably controlled. The expression of the adaptive loss function is:

[0099]

[0100] where L total is the total loss, is the loss function of task k, and y k are the predicted value and the true value of task k respectively, α k is the weight coefficient of task k, and N is the number of tasks.

[0101] In addition, in terms of quantitatively evaluating the accuracy of stress field reconstruction, the absolute mean error and the relative mean error can be used to evaluate the difference between the prediction result and the actual stress field. At the same time, the peak signal-to-noise ratio (PSNR) is used to quantitatively analyze the similarity between the generated stress field image and the true stress field. The higher the PSNR value, the higher the similarity between the prediction result and the actual stress field, which is used to prove the accuracy and reliability of the method of the present invention in stress field reconstruction.

[0102] See Figure 7 The present invention also discloses a device for dynamic digital reconstruction of full-field stress and deformation during the assembly process of thin-walled parts, including:

[0103] An acquisition module, configured to acquire surface deformation data during the assembly process of thin-walled parts by using a three-dimensional digital correlation system;

[0104] A construction module, configured to construct a finite element model of the assembly process of thin-walled parts, and based on the acquired surface deformation data, iteratively correct the finite element model to realize the correspondence between the measured surface deformation data and the finite element simulation stress field, and obtain the stress field data of the thin-walled parts;

[0105] A training module, configured to construct training samples by using the surface deformation data and the stress field data, and use the training samples to train a deep learning model to obtain a dynamic stress field prediction model;

[0106] A reconstruction module, configured to acquire real-time surface deformation data during the assembly process of the thin-walled part to be assembled by using a three-dimensional digital correlation system, and input the real-time surface deformation data into the dynamic stress field prediction model to obtain the dynamic stress field of the thin-walled part to be assembled.

[0107] An embodiment of the present invention also provides an electronic device, such as Figure 8As shown, it includes a processor 001, a communication interface 002, a memory 003, and a communication bus 004. Among them, the processor 001, the communication interface 002, and the memory 003 complete mutual communication through the communication bus 004.

[0108] The memory 003 is used to store computer programs.

[0109] When the processor 001 is used to execute the program stored in the memory 003, it realizes the full-field stress and deformation dynamic digital reconstruction method in the above-mentioned thin-walled part assembly process, including:

[0110] S1. Use a three-dimensional digital correlation system to collect surface deformation data during the thin-walled part assembly process.

[0111] S2. Construct a finite element model of the thin-walled part assembly process, and based on the collected surface deformation data, iteratively correct the finite element model to achieve the correspondence between the measured surface deformation data and the finite element simulation stress field, and obtain the stress field data of the thin-walled part.

[0112] S3. Use the surface deformation data and stress field data to construct training samples, and use the training samples to train a deep learning model to obtain a dynamic stress field prediction model.

[0113] S4. Use a three-dimensional digital correlation system to collect real-time surface deformation data during the assembly process of the thin-walled part to be assembled, and input the real-time surface deformation data into the dynamic stress field prediction model to obtain the dynamic stress field of the thin-walled part to be assembled.

[0114] Applying the solution provided by the present invention can effectively overcome the defect that the internal stress distribution during the assembly process cannot be accurately characterized in the traditional measurement method, realize the accurate measurement of the stress field during the thin-walled part assembly process, and provide a reliable technical support for optimizing the assembly accuracy of the thin-walled part.

[0115] The communication bus mentioned in the above electronic device can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This communication bus can be divided into an address bus, a data bus, a control bus, etc. For the sake of representation, only a thick line is used in the figure, but it does not mean that there is only one bus or one type of bus.

[0116] The communication interface is used for communication between the above electronic device and other devices.

[0117] The memory may include a Random Access Memory (RAM), or may also include a non-volatile memory (NVM), such as at least one disk memory. Optionally, the memory may also be at least one storage device located far from the aforementioned processor.

[0118] The aforementioned processor may be a general-purpose processor, including a Central Processing Unit (CPU), a Network Processor (NP), etc.; it may also be a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field-Programmable Gate Array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0119] In the above embodiments, it may be implemented in whole or in part by software, hardware, firmware, or any combination thereof. When implemented using software, it may be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the processes or functions described in the embodiments of the present invention are generated in whole or in part. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions may be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions may be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (such as coaxial cable, optical fiber, Digital Subscriber Line (DSL)) or wireless (such as infrared, wireless, microwave, etc.) means. The computer-readable storage medium may be any available medium that can be accessed by a computer, or a data storage device such as a server, data center, etc. that includes one or more integrated available media. The available medium may be a magnetic medium (such as a floppy disk, hard disk, magnetic tape), an optical medium (such as a DVD), or a semiconductor medium (such as a Solid State Disk (SSD)).

[0120] The above are only the preferred embodiments of the present invention that can be implemented, and do not limit the scope of the rights of the present invention. Any equivalent changes made by using the description and drawings of the present invention are included in the scope of the rights of the present invention.

Claims

1. A method for dynamic digital reconstruction of full-field stress and deformation during the assembly of thin-walled parts, characterized in that: include: S1. Collect surface deformation data of thin-walled parts during assembly using a three-dimensional digital correlation system; S2. Construct a finite element model of the thin-walled component assembly process, and iteratively correct the finite element model based on the collected surface deformation data to achieve the correspondence between the measured surface deformation data and the finite element simulation stress field, and obtain the stress field data of the thin-walled component; S3. Using surface deformation data and stress field data to construct training samples, and using the training samples to train the deep learning model to obtain a dynamic stress field prediction model; S4. Use a three-dimensional digital correlation system to collect real-time surface deformation data during the assembly process of the thin-walled parts to be assembled, input the real-time surface deformation data into a dynamic stress field prediction model, and obtain the dynamic stress field of the thin-walled parts to be assembled.

2. The method for dynamic digital reconstruction of full-field stress and deformation during the assembly of thin-walled parts according to claim 1 is characterized in that: S2 includes: Use finite element models to simulate the geometry, material properties and boundary conditions of thin-walled assembly structures; By calculating the difference between the surface deformation data and the deformation data of the corresponding position in the finite element model, the assembly parameters are iteratively corrected to complete the iterative correction of the finite element model; The stress field data are calculated based on the modified finite element model.

3. The method for dynamic digital reconstruction of full-field stress and deformation during the assembly of thin-walled parts according to claim 2, characterized in that: The process of calculating stress field data based on the modified finite element model includes: During the assembly process of thin-walled parts, surface roughness will cause significant changes in the stress field, which in turn affects the accuracy of the entire assembly process. In order to more accurately reflect the influence of roughness, the finite element model considers the stress modeling of surface roughness as follows: σ contact (r)=∫ S K(r,r′)·δ(r-r′)dS(1) Among them, σ contact (r) represents the stress at the contact point r, K(r,r′) is the contact stiffness matrix related to the roughness, δ(rr′) is the influence function of surface irregularity on the contact stress, and the integration region S covers the contact surface; In response to the influence of surface roughness, the roughness weighting function W(r) is further introduced to characterize the contribution of roughness at different positions, and the stress modeling is improved as follows: Where C and α are constants representing the weighted effect of roughness on contact stress at different locations, and |r| represents the local scale of a surface point, which is related to the distance from the contact center; In addition, in the assembly of thin-walled parts, the contact stress field depends not only on the large-scale geometry, but also on the micro-scale surface roughness. Based on this, the contact stresses at different scales are combined through the cross-scale modeling method, and the scale correlation function is introduced to express the stress transfer effect between different scales, which is expressed as follows: Among them, γ i and β i is the coefficient, L i is the size of the i-th scale, n is the number of scales considered; The scale correlation function is used to describe the stress transfer law from macroscopic to microscopic scales, connecting the contact effects of thin-walled parts at different scales. The stress modeling after the introduction of the scale correlation function is expressed as: σ contact (r)=∫ S K(r,r′)·δ(r-r′)·W(r)·S scale (r)dS(5)。 4. The method for dynamic digital reconstruction of full-field stress and deformation during assembly of thin-walled parts according to claim 3, characterized in that: The constant C is used to control the magnitude of the effect of roughness on contact stress. It is usually related to the material and surface quality and reflects the overall hardness and stiffness characteristics of the thin-walled surface. The calculation process is as follows: Use high-precision surface measuring instruments to obtain three-dimensional topographic data of thin-walled parts; Select thin-walled samples with different roughness levels, and measure the stress distribution in the contact area of ​​the thin-walled samples through loading tests or contact tests; The stress-deformation data are obtained according to the stress distribution, and the value of C is calculated by the weighted average method.

5. The method for dynamic digital reconstruction of full-field stress and deformation during thin-walled parts assembly according to claim 3, characterized in that: The constant α is used to control the magnitude of the effect of roughness on contact stress. It is usually related to the material and surface quality and reflects the overall hardness and stiffness characteristics of the thin-walled surface. The calculation process is as follows: Use high-precision surface measuring instruments to obtain three-dimensional topographic data of thin-walled parts; The surface morphology is modeled by fractal dimension calculation based on the three-dimensional morphology data to obtain the fractal dimension D of the surface; Loading tests were carried out on the established model, and the stress changes of the rough surface during the contact process were analyzed through the contact stress distribution under different roughness conditions. Finally, the value of α was determined based on regression analysis.

6. The method for dynamic digital reconstruction of full-field stress and deformation during assembly of thin-walled parts according to claim 1, characterized in that: The deep learning model is a multi-task stress field deep learning framework built on the Unet deep learning structure; During the training process, the surface deformation data of the thin-walled part is used as input data and processed by the shared coding layer. The shared coding layer mechanism is used to map different input data into a unified feature space to capture the characteristics of the input data. After being processed by the shared encoding layer, the captured features are distributed to multiple independent decoder branches. Each decoder branch focuses on the prediction of a specific stress component or deformation mode. Each decoder has an independent network structure to decode the features of a specific assembly task and reconstruct the features into the final output.

7. The method for dynamic digital reconstruction of full-field stress and deformation during assembly of thin-walled parts according to claim 6, characterized in that: In order to further improve the performance of the model on different assembly tasks, the transfer learning strategy is combined in the early stage of model training, and pre-training data from similar assembly tasks is used to help the deep learning model quickly adapt to the current task. In the case of less training data, the learning speed is accelerated and the prediction accuracy is improved by migrating the features of similar assembly tasks. The process of transfer learning is expressed as: in, represents the features in task i, T represents the migration operation, is the parameter of the shared encoding layer obtained from pre-training of task j; through the operation of transfer learning, the deep learning model can transfer the features of similar assembly tasks and transform them into features available for the current task; The deep learning model after transfer learning is trained on the data of the current task to further optimize the prediction of stress components and deformation. During the training process, the deep learning model dynamically adjusts the weight coefficients between different tasks through an adaptive loss function to ensure that the learning effect of each task is reasonably controlled. The expression of the adaptive loss function is: Among them, L total is the total loss, is the loss function for task k, and k are the predicted value and true value of task k, respectively, k is the weight coefficient of task k, and N is the number of tasks.

8. A dynamic digital reconstruction device for full-field stress and deformation during the assembly of thin-walled parts, characterized in that: include: An acquisition module for acquiring surface deformation data during the assembly process of thin-walled parts using a three-dimensional digital correlation system; A construction module is used to construct a finite element model of the thin-walled component assembly process, and iteratively correct the finite element model based on the collected surface deformation data, so as to achieve the correspondence between the measured surface deformation data and the finite element simulation stress field, and obtain the stress field data of the thin-walled component; A training module is used to construct training samples using surface deformation data and stress field data, and to train a deep learning model using the training samples to obtain a dynamic stress field prediction model; The reconstruction module is used to collect real-time surface deformation data of the thin-walled parts to be assembled during the assembly process using a three-dimensional digital correlation system, input the real-time surface deformation data into a dynamic stress field prediction model, and obtain the dynamic stress field of the thin-walled parts to be assembled.

9. An electronic device, characterized in that: It includes a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other through the communication bus; A memory for storing processor executable instructions; A processor, for implementing the method steps described in any one of claims 1-7 when executing instructions stored in a memory.

Images