Knowledge-driven conduit inner diameter roll forming digital twinning method, device and equipment

Abstract

The invention belongs to the technical field of digital twinning, and particularly relates to a knowledge-driven digital twinning method, device and equipment for conduit inner diameter rolling forming. Aiming at the problem that the forming quality is difficult to predict in real time and the process is difficult to optimize dynamically in the prior art, the method comprises the following steps: acquiring a macroscale data set in real time by constructing a digital twinborn model of a physical entity; inputting the data into a knowledge-driven machine learning model, and predicting a process time sequence and two-dimensional / three-dimensional field quantity parameters; evaluating the shape precision, the connection strength and the fatigue performance of the conduit based on the field quantity parameters, and automatically generating an optimization target after comparison with a preset standard; and the optimal process parameters are dynamically output by solving the target, and the forming equipment is regulated and controlled in real time. Final products are guaranteed to meet high-standard requirements, rejection rate is reduced, and product consistency and reliability are improved.

Description

Technical Field

[0001] This invention belongs to the field of digital twin technology, specifically relating to a knowledge-driven method, apparatus and system for forming a digital twin of the inner diameter of a conduit by rolling. Background Technology

[0002] With the rapid growth in demand for high-performance precision conduit connectors from industries such as aerospace, automotive manufacturing, and petrochemicals, higher requirements are being placed on the precision of conduit fitting forming, high-pressure resistance, connection strength, fatigue resistance, and sealing performance. The non-flared inner diameter roll forming method, due to its plastic forming principle, offers advantages such as high pressure resistance, high connection strength, good sealing, and good fatigue resistance, making it an important technical approach to meet these stringent requirements.

[0003] However, in existing technologies, the inner diameter roll forming process of conduits involves complex physical phenomena and highly nonlinear material behavior. Traditional simulation methods struggle to accurately simulate the complex interactions during the actual forming process, resulting in insufficient accuracy in predicting key performance indicators such as the shape accuracy, internal stress-strain distribution, connection strength, and fatigue life of the formed conduit. The selection of process parameters relies heavily on experience or tedious offline trial-and-error optimization, leading to low efficiency. More critically, existing technologies lack the ability to predict forming quality in real time and dynamically adjust process parameters during processing. This makes it difficult to guarantee the final product's shape accuracy and optimize connection strength and fatigue performance in real-time during actual production, thus limiting product quality stability and reliability under extreme conditions. Summary of the Invention

[0004] To address the shortcomings of existing technologies in accurately predicting the forming process of inner diameter roll forming, which makes it difficult to optimize the processing in real time, the present invention aims to provide a knowledge-driven digital twin method, apparatus, and equipment for inner diameter roll forming of conduits.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: A first aspect of the present invention provides a knowledge-driven digital twin method for tubing inner diameter rolling forming, comprising: Construct a digital twin model of the physical entity; based on the digital twin model, acquire macroscopic datasets in real time during the inner diameter roll forming process of the conduit; the physical entity includes the inner diameter roll forming equipment and the conduit; the macroscopic dataset includes the material data and geometric data of the conduit, as well as the process timing data of the inner diameter roll forming process of the conduit. Macroscale datasets are input into a knowledge-driven machine learning model to obtain real-time prediction data; the real-time prediction data includes predicted process time series data, two-dimensional field parameters, and three-dimensional field parameters. The performance of the conduit is determined based on real-time prediction data, including shape accuracy, connection strength, and fatigue performance. The performance is compared with preset standards. When the comparison results do not meet the preset conditions, the optimization objectives of the processing roll forming process are determined, including minimizing shape accuracy error and maximizing strength and fatigue performance. Solve for the optimization objective to obtain the optimal process parameters; the optimal process parameters include at least one of load, torque, speed and displacement.

[0006] Furthermore, in the step of inputting the macro-scale dataset into the knowledge-driven machine learning model to obtain real-time prediction data, the knowledge-driven machine learning model is trained in the following manner: Multimodal simulation data is generated based on a pre-trained GAN model; the multimodal simulation data includes: training process time series data, training two-dimensional image data, training two-dimensional field parameters, training three-dimensional geometric data, and training three-dimensional field parameters. The analysis and identification model is trained based on multimodal simulation data and corresponding material data. The trained analysis and identification model and the GAN model are then combined into a knowledge-driven machine learning model.

[0007] Furthermore, in the step of generating multimodal simulation data based on the pre-trained GAN model, the GAN model is trained as follows: A finite element simulation model of the catheter is constructed, and simulation processing is performed based on the finite element simulation model to obtain simulation results; wherein, the material physical characterization of the catheter is based on physical knowledge during the construction of the finite element simulation model. An initial dataset is extracted from the simulation results; the initial dataset includes: material data of the conduit, initial simulation geometric data, initial simulation process timing data, initial two-dimensional simulation field parameters, and initial three-dimensional simulation field parameters. The initial simulation geometric data, initial two-dimensional simulation field parameters, and initial three-dimensional simulation field parameters are used to generate a sample space through DOE; The DOE space discrete field quantity dataset is obtained by discretizing the sample space. The first multimodal data is extracted from the DOE spatial discrete field quantity dataset. The material data, the first multimodal data, and the initial simulation process time series data are used to construct the second multimodal data. The second multimodal data includes: material data, initial simulation process time series data, discretized two-dimensional simulation image data and corresponding discretized two-dimensional simulation field quantity parameters, and discretized three-dimensional simulation geometric data and corresponding discretized three-dimensional simulation field quantity parameters. The second multimodal data is input into the GAN model for pre-training. The GAN model includes a generator and a discriminator. The generator generates comparison data based on physical knowledge and random noise as input. The discriminator performs a similarity test between the second multimodal data and the comparison data to determine their realism. When the comparison data and the second multimodal data reach a preset similarity value, the GAN model is considered to have completed pre-training. The comparison data includes: comparison process time series data, comparison two-dimensional image data and corresponding comparison two-dimensional field parameters, comparison three-dimensional geometric data and corresponding comparison three-dimensional field parameters.

[0008] Furthermore, a finite element simulation model of the conduit is constructed, and simulation processing is performed based on the finite element simulation model to obtain simulation results, including: In finite element simulation software, a three-dimensional model of the duct considering the complex constraints of the tooling is constructed, and a mesh model considering the complex constraints of the tooling is constructed based on the three-dimensional model. Simulations of the duct under rolling forming and bending fatigue conditions were performed based on a mesh model, and simulation results were obtained. The simulation results include material data, geometric data, and process timing data, as well as the corresponding two-dimensional and three-dimensional field parameters.

[0009] Furthermore, the second multimodal data is input into the GAN model for pre-training, including: The generator uses physical knowledge as constraints and random noise as input to generate data to be compared. The discriminator compares the process timing data to be compared with the initial simulation process timing data to obtain the first comparison result; the discriminator compares the two-dimensional image data and the two-dimensional field parameters to be compared with the discretized two-dimensional simulation image data and the discretized two-dimensional simulation field parameters to obtain the second comparison result; the discriminator compares the three-dimensional geometric data and the three-dimensional field parameters to be compared with the discretized three-dimensional simulation geometric data and the discretized three-dimensional simulation field parameters to obtain the third comparison result. When the first comparison result, the second comparison result, and the third comparison result all meet the preset range, the GAN model pre-training is determined to be complete. The first comparison result includes the similarity between the process time series data to be compared and the initial simulated process time series data; the second comparison result includes the similarity between the two-dimensional image data to be compared and the discretized two-dimensional simulated image data, and the similarity between the two-dimensional field parameters to be compared and the discretized two-dimensional simulated field parameters; the third comparison result includes the similarity between the three-dimensional geometric data to be compared and the discretized three-dimensional simulated geometric data, and the similarity between the three-dimensional field parameters to be compared and the discretized three-dimensional simulated field parameters.

[0010] Furthermore, the analysis and identification model is trained based on multimodal simulation data and corresponding material data, including: The analysis and recognition models include GNN, CNN, and LSTM models. The CNN model is trained based on material data, two-dimensional image data for training, and two-dimensional field parameters for training; the GNN model is trained based on material data, three-dimensional geometric data for training, and three-dimensional field parameters for training; and the LSTM model is trained based on material data and process time series data for training.

[0011] In a second aspect, the present invention provides a knowledge-driven digital twin device for tubing inner diameter rolling forming, comprising: The model building module is used to construct digital twin models of physical entities; based on the digital twin models, macroscopic datasets of the inner diameter rolling forming process of conduits are acquired in real time; the physical entities include the conduit inner diameter rolling forming equipment and the conduit; the macroscopic datasets include the material data and geometric data of the conduit, as well as the process timing data of the inner diameter rolling forming process of the conduit. The prediction module is used to input macro-scale datasets into a knowledge-driven machine learning model to obtain real-time prediction data; the real-time prediction data includes predicted process time series data, two-dimensional field parameters, and three-dimensional field parameters. The optimization module is used to determine the performance of the conduit based on real-time prediction data. The performance includes shape accuracy, connection strength, and fatigue performance. The performance is compared with preset standards. When the comparison results do not meet the preset conditions, the optimization objectives of the processing roll forming process are determined. The optimization objectives include minimizing shape accuracy error and maximizing strength and fatigue performance. The solver module is used to solve the optimization objective and obtain the optimal process parameters; the optimal process parameters include at least one of load, torque, speed and displacement.

[0012] Furthermore, in the prediction module, the knowledge-driven machine learning model is trained as follows: Multimodal simulation data is generated based on a pre-trained GAN model; the multimodal simulation data includes: training process time series data, training two-dimensional image data, training two-dimensional field parameters, training three-dimensional geometric data, and training three-dimensional field parameters. The analysis and identification model is trained based on multimodal simulation data and corresponding material data. The trained analysis and identification model and the GAN model are then combined into a knowledge-driven machine learning model.

[0013] Furthermore, in the step of generating multimodal simulation data based on the pre-trained GAN model, the GAN model is trained as follows: A finite element simulation model of the catheter is constructed, and simulation processing is performed based on the finite element simulation model to obtain simulation results; wherein, the material physical characterization of the catheter is based on physical knowledge during the construction of the finite element simulation model. An initial dataset is extracted from the simulation results; the initial dataset includes: material data of the conduit, initial simulation geometric data, initial simulation process timing data, initial two-dimensional simulation field parameters, and initial three-dimensional simulation field parameters. The initial simulation geometric data, initial two-dimensional simulation field parameters, and initial three-dimensional simulation field parameters are used to generate a sample space through DOE; The DOE space discrete field quantity dataset is obtained by discretizing the sample space. The first multimodal data is extracted from the DOE spatial discrete field quantity dataset. The material data, the first multimodal data, and the initial simulation process time series data are used to construct the second multimodal data. The second multimodal data includes: material data, initial simulation process time series data, discretized two-dimensional simulation image data and corresponding discretized two-dimensional simulation field quantity parameters, and discretized three-dimensional simulation geometric data and corresponding discretized three-dimensional simulation field quantity parameters. The second multimodal data is input into the GAN model for pre-training. The GAN model includes a generator and a discriminator. The generator generates comparison data based on physical knowledge and random noise as input. The discriminator performs a similarity test between the second multimodal data and the comparison data to determine their realism. When the comparison data and the second multimodal data reach a preset similarity value, the GAN model is considered to have completed pre-training. The comparison data includes: comparison process time series data, comparison two-dimensional image data and corresponding comparison two-dimensional field parameters, comparison three-dimensional geometric data and corresponding comparison three-dimensional field parameters.

[0014] In a third aspect, the present invention also provides an electronic device including a computer-readable storage medium, the electronic device including a processor and a memory, the computer-readable storage medium storing at least one instruction, which, when executed by the processor, implements the method described in any one of the preceding claims.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: This solution constructs a digital twin model and acquires macroscopic datasets in real time. Combined with a knowledge-driven machine learning model, it can predict key process timing data and two-dimensional and three-dimensional field parameters reflecting the local and global stress-strain states of the conduit in real time. Based on the real-time predicted field parameters, it automatically extracts and evaluates the shape accuracy, connection strength, and fatigue performance indicators of the conduit. By comparing the extracted performance indicators with preset standards in real time, and automatically determining optimization targets when standards are not met, it solves for the optimal process parameters and provides real-time feedback to control the forming equipment, forming a closed loop of perception, analysis, decision-making, and execution. This allows for dynamic adjustment of process parameters to ensure that the final product meets high standards, reduces scrap rates, and improves product consistency and reliability. This invention combines digital twins, machine learning, and optimization control, reducing the inefficiency of traditional experience-based and offline trial-and-error optimization, and achieving intelligent monitoring and proactive optimization of the forming process.

[0016] This solution utilizes pre-trained GAN models to generate multimodal simulation data including process timing, two-dimensional images and field quantities, and three-dimensional geometry and field quantities. This effectively expands the training dataset and overcomes the problems of high cost and limited data volume in physical experiments or high-fidelity simulations.

[0017] This approach uses GAN multimodal simulation data to train and analyze the recognition model, which helps to learn the inherent laws of complex forming processes more comprehensively, thereby improving the accuracy and robustness of real-time prediction.

[0018] In the data preparation stage for GAN model training, material physics characterization was performed based on physical knowledge, and a finite element simulation model was constructed for simulation, ensuring the physical basis of the initial dataset.

[0019] The generator of the GAN model is constrained by physical knowledge, and the generated comparison data are consistent with physical laws. The discriminator judges the realism similarity through multi-dimensional similarity to ensure that the generated synthetic data is highly similar to the physical simulation data in multiple modes and at multiple levels, thereby providing a high-quality, physically reliable multimodal simulation data source.

[0020] By constructing a 3D model and corresponding mesh model in finite element software that considers the complex constraints of tooling, the limitations of simplified modeling are overcome, and the complex interaction boundary conditions between the mold and the tube are reflected more realistically in the actual forming process.

[0021] The simulation not only targets the rolling forming condition but also the bending fatigue condition, which can obtain more comprehensive field parameter data reflecting the fatigue resistance of the guide tube. Attached Figure Description

[0022] The accompanying drawings, which form part of this application, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an undue limitation of the invention. In the drawings: Figure 1 This is a flowchart of a knowledge-driven digital twin method for tubing inner diameter rolling forming according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the architecture of a knowledge-driven digital twin system for tubing inner diameter rolling forming according to an embodiment of the present invention; Figure 3 This is a diagram illustrating the training principle of the knowledge-driven machine learning model in this embodiment of the invention. Figure 4 This is a flowchart simulating the rolling forming process of the conduit in an embodiment of the present invention; Figure 5 This is a schematic diagram of the optimized feedback control module in an embodiment of the present invention; Figure 6 This is a schematic diagram of the architecture of a knowledge-driven digital twin device for rolling and forming the inner diameter of a conduit, according to an embodiment of the present invention. Figure 7 This is a structural block diagram of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0023] The present invention will now be described in detail with reference to the accompanying drawings and embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other.

[0024] The following detailed description is exemplary and intended to provide further detailed explanation of the invention. Unless otherwise specified, all technical terms used in this invention have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains. The terminology used in this invention is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention.

[0025] Example 1 GAN, Generative Adversarial Network.

[0026] GNN, Graph Neural Network.

[0027] CNN stands for Convolutional Neural Network.

[0028] LSTM, Long Short-Term Memory.

[0029] Reference Figure 1 A knowledge-driven digital twin method for rolling the inner diameter of a conduit includes steps 100 to 500.

[0030] Step 100: Construct a digital twin model of the physical entity; wherein the physical entity includes the conduit inner diameter rolling forming equipment and the conduit; based on the digital twin model, acquire the macroscopic scale dataset in real time during the conduit inner diameter rolling forming process; wherein the macroscopic scale dataset includes the material data and geometric data of the conduit, as well as the process timing data during the conduit inner diameter rolling forming process.

[0031] In one optional embodiment, material data may include the type and properties of the conduit's material. Geometric data may include two-dimensional image data and three-dimensional geometric data of the conduit during processing. Process timing data may include the torque, rotational speed, displacement, and load of the conduit inner diameter rolling forming equipment during processing.

[0032] Step 200: Input the macro-scale dataset into the knowledge-driven machine learning model to obtain real-time prediction data; the real-time prediction data includes predicted process time series data, two-dimensional field parameters and three-dimensional field parameters; wherein, the two-dimensional field parameters are the distribution of the stress field and strain field of the local duct on the two-dimensional image, and the three-dimensional field parameters are the overall distribution of the stress field and strain field of the global duct on the three-dimensional image.

[0033] Step 300: Determine the performance of the conduit based on real-time prediction data, including shape accuracy, connection strength, and fatigue performance; compare the performance with preset standards, and when the comparison results do not meet the preset conditions, determine the optimization objectives for the processing roll forming process, including minimizing shape accuracy error and maximizing strength and fatigue performance.

[0034] Specifically, the optimization objective of this scheme is a multi-objective optimization. The goal of the optimization is to minimize the error value of shape accuracy and maximize the limit values ​​of strength performance and fatigue performance.

[0035] Specifically, based on real-time predicted process timing data, two-dimensional field parameters, and three-dimensional field parameters, the shape accuracy, connection strength, and fatigue performance of the conduit are determined. The shape accuracy, connection strength, and fatigue performance are compared with the corresponding preset standards to obtain the comparison results. When the comparison results do not meet the preset conditions, the optimization objectives of the processing are determined. Among them, the optimization objectives include minimizing the shape accuracy error and maximizing the strength and fatigue performance.

[0036] Step 400: Solve for the optimal process parameters that satisfy the optimization objective in the optimizer; wherein the optimal process parameters include at least one of load, torque, speed and displacement; control the inner diameter roll forming equipment of the conduit in real time according to the optimal process parameters.

[0037] In one embodiment, the knowledge-driven machine learning model in step 200 is obtained according to steps 201 to 207.

[0038] Step 201: Construct a finite element simulation model of the catheter, perform simulation processing based on the finite element simulation model, and obtain simulation results; wherein, in the process of constructing the finite element simulation model, the material physical characterization of the catheter is performed based on physical knowledge.

[0039] In one optional implementation, step 201 specifically involves: constructing a three-dimensional model of the conduit in finite element simulation software, constructing a mesh model based on the three-dimensional model, and performing simulations of the conduit under rolling forming and bending fatigue conditions based on the mesh model to obtain simulation results; wherein, the simulation results may include material data, geometric data, and process timing data, as well as corresponding two-dimensional field parameters and three-dimensional field parameters, etc.

[0040] Simulations of the roller forming and bending fatigue conditions of the guide tube are performed, including the following steps 2010 to 2012.

[0041] Step 2010: Simulation modeling processing.

[0042] Specifically, this includes: constructing a three-dimensional model of the duct in finite element simulation software, and constructing a mesh model based on the three-dimensional model.

[0043] In one alternative embodiment, a 3D model of the conduit is obtained by considering the complex constraints of the tooling in finite element simulation software. A mesh model considering the complex constraints of the tooling is then constructed based on the 3D model. This consideration of the complex constraints during simulation modeling facilitates accurate finite element simulation of the conduit's inner diameter rolling forming process.

[0044] It should be noted that the simulation process of the inner diameter roll forming connection of the conduit involves spinning the inner wall of the conduit under the action of a roll forming bulge (tooling). The roll forming bulge is an assembly of a mandrel, rollers, and a cage. Relying on the linkage of mandrel drive, cage constraint, and roller forming, the conduit is connected to the sleeve under the spinning force of the circumferentially moving rollers within the inner diameter. The complex constraints of the tooling in the inner diameter roll forming process of the conduit are mainly reflected in: the complex motion characteristics in a confined space, the complex boundary condition constraints of the cage on the rollers, and the non-uniform flow of the conduit material. Driven by the mandrel, the rollers undergo a combined rotation and revolution motion within the confined space inside the high-strength titanium tube; the rollers, mandrel, cage, and inner wall of the conduit are under complex constraints with multiple boundary conditions. These complex constraints of the tooling need to be considered when constructing the finite element simulation model.

[0045] Step 2011, simulation stage processing (taking roll forming as an example).

[0046] To simulate the actual process as completely as possible, the rolling forming process was constructed according to the process steps of the inner diameter of the conduit. Each stage included: rolling preparation stage, rolling forming stage, tooling unloading stage, and connection strength testing stage. Based on the process steps of these four stages, the simulation process of the rolling forming process was created as four stages, and finite element analysis was performed on each of the four stages based on the mesh model.

[0047] It should be noted that the process steps of inner diameter roll forming of the conduit include: relying on the mandrel, cage, and three rollers of the inner diameter roll forming equipment, as well as the complex boundary constraints of the conduit and sleeve structures, the conduit achieves plastic connection and forming between the conduit and sleeve under the spin pressure of the circumferentially moving rollers. Under the rotational extrusion action of the three rollers, the conduit undergoes plastic flow under extrusion force, pressing into the groove on the inner wall of the sleeve, and interfering with the groove on the inner surface of the sleeve. After the plastic flow deformation of the conduit is unloaded, the conduit and sleeve rebound together due to their different elastic moduli. During the extrusion process, due to the constraints of the sleeve components and the difference in elastic-plastic mechanical properties between the conduit and sleeve materials, the conduit material has lower strength and better plasticity than the sleeve material. Therefore, by pressing the conduit into the groove of the sleeve through inner diameter roll forming, mechanical connection and sealing can be achieved.

[0048] Step 2012: Output simulation results.

[0049] The simulation results can be obtained through the simulation stage processing in step 2011.

[0050] Step 202: Extract the initial dataset from the simulation results; The initial dataset includes: material data of the conduit, initial simulation geometric data, initial simulation process timing data, initial two-dimensional simulation field parameters, and initial three-dimensional simulation field parameters; The initial simulation geometry data includes: initial two-dimensional simulation images and initial three-dimensional simulation geometry. The initial two-dimensional simulation images correspond to the initial two-dimensional simulation field parameters, and the initial three-dimensional simulation geometry corresponds to the initial three-dimensional simulation field parameters. The initial two-dimensional simulation field parameters include: the distribution of the simulated local stress field and the simulated local strain field on the initial two-dimensional simulation image; the initial three-dimensional simulation field parameters include: the overall distribution of the simulated global stress field and the simulated global strain field on the initial three-dimensional simulation geometry.

[0051] Step 203: Generate a sample space from the initial simulation geometric data, initial two-dimensional simulation field parameters, and initial three-dimensional simulation field parameters using DOE (Design of Experiments).

[0052] Step 204: Discretize the sample space to obtain the DOE space discrete field quantity dataset.

[0053] Step 205: Extract the first multimodal data from the DOE spatial discrete field quantity dataset, and construct the second multimodal data from the material data, the first multimodal data, and the initial simulation process time series data, such as... Figure 3 As shown.

[0054] In this scheme, the second multimodal data serves as the training set for the pre-training of the GAN model. The GAN model is mainly used to enhance the data required for training the generative analysis and recognition model. Therefore, the second multimodal data may include: material data, initial simulation process time series data, discretized two-dimensional simulation image data and corresponding discretized two-dimensional simulation field parameters, and discretized three-dimensional simulation geometric data and corresponding discretized three-dimensional simulation field parameters.

[0055] Step 206: Input the second multimodal data into the GAN model for pre-training; wherein, the GAN model includes a generator and a discriminator; the generator generates data to be compared with physical knowledge as a constraint and random noise as input; the discriminator judges the realism similarity between the second multimodal data and the data to be compared, and when the data to be compared and the second multimodal data reach a preset similarity value, it is determined that the GAN model has completed pre-training.

[0056] Specifically, the data to be compared includes: process timing data to be compared, two-dimensional image data to be compared and corresponding two-dimensional field parameters to be compared, and three-dimensional geometric data to be compared and corresponding three-dimensional field parameters to be compared.

[0057] Step 206 includes steps 2061 to 2063: Step 2061: The generator generates data to be compared, constrained by physical knowledge and with random noise as input.

[0058] Step 2062: The discriminator compares the process timing data to be compared with the initial simulation process timing data to obtain a first comparison result; the discriminator compares the two-dimensional image data to be compared and the two-dimensional field parameters to be compared with the discretized two-dimensional simulation image data and the discretized two-dimensional simulation field parameters to obtain a second comparison result; the discriminator compares the three-dimensional geometric data to be compared and the three-dimensional field parameters to be compared with the discretized three-dimensional simulation geometric data and the discretized three-dimensional simulation field parameters to obtain a third comparison result.

[0059] Step 2063: When the first comparison result, the second comparison result, and the third comparison result all meet the preset range, the pre-training of the GAN model is determined to be complete.

[0060] Specifically, the first comparison result includes: the similarity between the process timing data to be compared and the initial simulation process timing data; the second comparison result includes: the similarity between the two-dimensional image data to be compared and the discretized two-dimensional simulation image data, and the similarity between the two-dimensional field parameters to be compared and the discretized two-dimensional simulation field parameters; the third comparison result includes: the similarity between the three-dimensional geometric data to be compared and the discretized three-dimensional simulation geometric data, and the similarity between the three-dimensional field parameters to be compared and the discretized three-dimensional simulation field parameters.

[0061] When the similarity scores in the first, second, and third comparison results all meet the preset range, it is determined that the accuracy of the comparison data generated by the GAN model at this point meets the requirements. In other words, when the comparison data generated by the generator gradually approaches the real second multimodal data, the pre-trained GAN model is obtained.

[0062] In one optional embodiment, the preset range of each similarity in step 2063 can be set according to the accuracy requirements (the specific value is not limited in this invention), for example, 90%.

[0063] Step 207: Perform transfer learning on the pre-trained GAN model to obtain an optimized GAN model; generate multimodal simulation data based on the optimized GAN model; wherein, the multimodal simulation data includes: training process time series data, training two-dimensional image data, training two-dimensional field parameters, training three-dimensional geometric data, and training three-dimensional field parameters; train the analysis and recognition model based on the multimodal simulation data and the corresponding material data, and combine the trained analysis and recognition model and the optimized GAN model into a knowledge-driven machine learning model.

[0064] The analysis and identification model is mainly used to take the macroscopic scale dataset of the conduit during the processing as input and output real-time predicted process time series data, two-dimensional field parameters and three-dimensional field parameters.

[0065] In one alternative implementation, the analysis and recognition model includes at least a GNN model, a CNN model, and an LSTM model.

[0066] In the process of analyzing and recognizing the training of the model: the CNN model is trained based on material data, training two-dimensional image data, and training two-dimensional field parameters; the GNN model is trained based on material data, training three-dimensional geometric data, and training three-dimensional field parameters; and the LSTM model is trained based on material data and training process time series data.

[0067] After the GNN, CNN, and LSTM models are all trained, the analysis and recognition model can generate real-time predicted process time series data, two-dimensional field parameters, and three-dimensional field parameters based on the macroscopic dataset of the conduit during the processing.

[0068] The process timing data is mainly used to characterize the inner diameter rolling forming process of the conduit, and the process timing data can be displayed in real time on the interactive server.

[0069] Example 2 Reference Figure 2 A knowledge-driven digital twin system for tubing inner diameter rolling forming includes: On the physical entity side, this includes the conduit inner diameter roll forming equipment and sensors; the conduit inner diameter roll forming equipment can be a pipe fitting inner diameter roll forming machine and tooling. Sensors are used to collect macroscopic datasets of the conduit during the conduit inner diameter roll forming process; these macroscopic datasets can include process time-series data, geometric data, and material data. Optionally, the macroscopic dataset can be real-time data (online) or historical data (offline); real-time data can be used for real-time prediction of process time-series data, two-dimensional field parameters, and three-dimensional field parameters; historical data can be combined with simulation results to serve as the data basis for extracting the initial dataset.

[0070] The macroscopic dataset (real-time data) collected by the sensor is input into the knowledge-driven machine learning model, which outputs real-time predicted process time series data, two-dimensional field parameters, and three-dimensional field parameters.

[0071] The optimized feedback control module acquires real-time predicted process timing data, two-dimensional field parameters, and three-dimensional field parameters, and further extracts the shape accuracy, connection strength, and fatigue performance of the conduit. Based on these parameters, it determines whether to perform optimized control of the processing. For example, the shape accuracy, connection strength, and fatigue performance are compared with corresponding preset standards to obtain comparison results. When the comparison results do not meet the preset conditions, the process parameters of the conduit inner diameter rolling forming equipment are optimized to obtain optimal process parameters. These optimal process parameters include load, torque, speed, and displacement. The conduit inner diameter rolling forming equipment is then controlled in real time based on these optimal process parameters.

[0072] The simulation module is used to construct a finite element simulation model of the conduit, perform simulation processing based on the finite element simulation model, and obtain simulation results.

[0073] On the physics side, constraints are applied to the finite element model based on physical principles. An initial dataset is extracted from the simulation results. Based on this dataset, initial simulation geometric data, initial two-dimensional simulation field parameters, and initial three-dimensional simulation field parameters are determined. These initial geometric data, initial two-dimensional simulation field parameters, and initial three-dimensional simulation field parameters are then used to generate a sample space through Design of Experiments (DOE). This sample space is then discretized to obtain a discrete field quantity dataset in DOE space, which is then constrained using physical principles.

[0074] Specifically, physics knowledge can be acquired in the following ways: The residual stress, texture, and thickness anisotropy coefficient of the catheter are determined as physical knowledge. Residual stress data is obtained from residual stress testing, the thickness anisotropy coefficient is obtained from digital image correlation (DIC) testing, and texture data is obtained from electron microscopy. These residual stress, texture, and thickness anisotropy coefficients constitute the physical knowledge. The residual stress, texture, and thickness anisotropy coefficients are then normalized, and the normalized test data are input as attribute data into the finite element simulation model. Based on these physical knowledge, the catheter is characterized using material physics, further describing the material properties. This method incorporates physical knowledge into the simulation process.

[0075] The embodiments of the present invention constrain the finite element simulation model through physical knowledge, thereby conforming to the physical specifications of plastic deformation during subsequent model training, reducing model illusion, and improving the accuracy and generalization ability of model prediction.

[0076] Specifically, residual stress is the self-balancing internal stress that remains within an object after the removal of external forces or uneven temperature fields. Machining and strengthening processes can both induce residual stress. Residual stress can be used to assess the fatigue strength and wear resistance of parts. Texture refers to the process by which the orientation of grains changes due to alterations in the crystal structure during material deformation. In roll forming, texture assessment primarily evaluates material properties, representing only the material's mechanical behavior and connection quality. The thickness anisotropy coefficient of a conduit refers to the anisotropic characteristics of the material along its thickness direction, reflecting the differences in mechanical properties in different directions. For roll forming, the thickness anisotropy coefficient assessment represents the deformation behavior during the forming process and the performance of the connection points.

[0077] An interactive server is used to display an interactive interface, which can show real-time predicted process timing data.

[0078] Reference Figure 3 The knowledge-driven machine learning model comprises two parts: a GAN model and an analysis and recognition model. The GAN model primarily provides the dataset for training the analysis and recognition model. Once trained, the analysis and recognition model is mainly used to predict the two-dimensional and three-dimensional field parameters of the duct in real time, as well as predict process time-series data, based on macroscopic-scale datasets.

[0079] Specifically, the GAN model is pre-trained based on the second multimodal data. The GAN model includes a generator and a discriminator. The generator generates data to be compared by taking physical knowledge as constraints and noisy data as input. The discriminator is used to judge the similarity between the data to be compared and the second multimodal data. The process is iterated continuously. When the model convergence condition is reached, the pre-trained GAN model is obtained.

[0080] The pre-trained GAN model undergoes transfer learning processing to obtain an optimized GAN model.

[0081] Reference Figure 4 The simulation module primarily executes three stages: simulation modeling, simulation processing, and simulation result output. Simulation modeling includes constructing a 3D model of the conduit in the finite element simulation software and building a mesh model based on the 3D model. Simulation processing includes simulating the conduit under rolling forming and bending fatigue conditions based on the mesh model, obtaining the simulation results. Simulation result output follows the same workflow as outputting the simulation results.

[0082] Reference Figure 5 The optimized feedback control module includes: The service performance evaluation module is used to extract the shape accuracy, connection strength, and fatigue performance of the conduit based on real-time predicted process timing data, two-dimensional field parameters, and three-dimensional field parameters; compare the shape accuracy, connection strength, and fatigue performance with corresponding preset standards to obtain comparison results; when the comparison results do not meet the preset conditions, determine the optimization target of the processing; among which, the optimization targets include minimizing the shape accuracy error and maximizing the strength and fatigue performance; The solver is used to find the optimal process parameters that satisfy the optimization objective; the optimal process parameters include at least one of load, torque, speed and displacement; the inner diameter roll forming equipment of the conduit is controlled in real time according to the optimal process parameters.

[0083] The optimized feedback control module of this invention can find the optimal process parameters and send them to the conduit inner diameter roll forming equipment. This allows for real-time control of at least one of the equipment's frequency, power, torque, speed, and displacement, thus enabling the manufacture of conduits that meet service performance requirements. This achieves closed-loop control of online prediction of conduit forming and real-time optimization of process parameters, overcoming the shortcomings of existing roll forming technologies where conduit performance does not meet standards.

[0084] Example 3 Reference Figure 6 The present invention, embodiment 3, discloses a knowledge-driven digital twin device for tubing inner diameter rolling forming, comprising: The model building module is used to construct digital twin models of physical entities; based on the digital twin models, macroscopic datasets of the inner diameter rolling forming process of conduits are acquired in real time; the physical entities include the conduit inner diameter rolling forming equipment and the conduit; the macroscopic datasets include the material data and geometric data of the conduit, as well as the process timing data of the inner diameter rolling forming process of the conduit. The prediction module is used to input macro-scale datasets into a knowledge-driven machine learning model to obtain real-time prediction data; the real-time prediction data includes predicted process time series data, two-dimensional field parameters, and three-dimensional field parameters. The optimization module is used to determine the performance of the conduit based on real-time prediction data. The performance includes shape accuracy, connection strength, and fatigue performance. The performance is compared with preset standards. When the comparison results do not meet the preset conditions, the optimization objectives of the processing roll forming process are determined. The optimization objectives include minimizing shape accuracy error and maximizing strength and fatigue performance. The solver module is used to solve the optimization objective and obtain the optimal process parameters; the optimal process parameters include at least one of load, torque, speed and displacement.

[0085] Furthermore, in the prediction module, the knowledge-driven machine learning model is trained as follows: Multimodal simulation data is generated based on a pre-trained GAN model; the multimodal simulation data includes: training process time series data, training two-dimensional image data, training two-dimensional field parameters, training three-dimensional geometric data, and training three-dimensional field parameters. The analysis and identification model is trained based on multimodal simulation data and corresponding material data. The trained analysis and identification model and the GAN model are then combined into a knowledge-driven machine learning model.

[0086] Furthermore, in the step of generating multimodal simulation data based on the pre-trained GAN model, the GAN model is trained as follows: A finite element simulation model of the catheter is constructed, and simulation processing is performed based on the finite element simulation model to obtain simulation results; wherein, the material physical characterization of the catheter is based on physical knowledge during the construction of the finite element simulation model. An initial dataset is extracted from the simulation results; the initial dataset includes: material data of the conduit, initial simulation geometric data, initial simulation process timing data, initial two-dimensional simulation field parameters, and initial three-dimensional simulation field parameters. The initial simulation geometric data, initial two-dimensional simulation field parameters, and initial three-dimensional simulation field parameters are used to generate a sample space through DOE; The DOE space discrete field quantity dataset is obtained by discretizing the sample space. The first multimodal data is extracted from the DOE spatial discrete field quantity dataset. The material data, the first multimodal data, and the initial simulation process time series data are used to construct the second multimodal data. The second multimodal data includes: material data, initial simulation process time series data, discretized two-dimensional simulation image data and corresponding discretized two-dimensional simulation field quantity parameters, and discretized three-dimensional simulation geometric data and corresponding discretized three-dimensional simulation field quantity parameters. The second multimodal data is input into the GAN model for pre-training. The GAN model includes a generator and a discriminator. The generator generates comparison data based on physical knowledge and random noise as input. The discriminator performs a similarity test between the second multimodal data and the comparison data to determine their realism. When the comparison data and the second multimodal data reach a preset similarity value, the GAN model is considered to have completed pre-training. The comparison data includes: comparison process time series data, comparison two-dimensional image data and corresponding comparison two-dimensional field parameters, comparison three-dimensional geometric data and corresponding comparison three-dimensional field parameters.

[0087] Example 4 like Figure 7 As shown, the present invention also provides an electronic device 100 for implementing a knowledge-driven digital twin method for rolling the inner diameter of a conduit; The electronic device 100 includes a memory 101, at least one processor 102, a computer program 103 stored in the memory 101 and executable on at least one processor 102, and at least one communication bus 104.

[0088] The memory 101 can be used to store the computer program 103. The processor 102 implements the steps of Embodiment 1 by running or executing the computer program stored in the memory 101 and calling the data stored in the memory 101.

[0089] The memory 101 may primarily include a program storage area and a data storage area. The program storage area may store the operating system, application programs required for at least one function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created based on the use of the electronic device 100 (such as audio data), etc. In addition, the memory 101 may include non-volatile memory, such as hard disk, RAM, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other non-volatile solid-state storage device.

[0090] At least one processor 102 may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Processor 102 may be a microprocessor or any conventional processor. Processor 102 is the control center of electronic device 100, connecting various parts of electronic device 100 via various interfaces and lines.

[0091] The memory 101 in the electronic device 100 stores multiple instructions to implement a knowledge-driven digital twin method for tubular inner diameter rolling forming, and the processor 102 can execute multiple instructions to achieve the following: Construct a digital twin model of the physical entities, including the conduit inner diameter rolling forming equipment and the conduit itself. Based on the digital twin model, acquire macroscopic datasets in real time during the conduit inner diameter rolling forming process. The macroscopic datasets include the material and geometric data of the conduit, as well as the process timing data during the conduit inner diameter rolling forming process.

[0092] Macroscale datasets are input into a knowledge-driven machine learning model to obtain real-time prediction data. The real-time prediction data includes predicted process time series data, two-dimensional field parameters, and three-dimensional field parameters. Among them, the two-dimensional field parameters are the distribution of local stress and strain fields of the duct in a two-dimensional image, and the three-dimensional field parameters are the overall distribution of global stress and strain fields of the duct in a three-dimensional image.

[0093] The performance of the conduit is determined based on real-time prediction data, including shape accuracy, connection strength, and fatigue performance. The performance is compared with preset standards. When the comparison results do not meet the preset conditions, the optimization objectives of the processing roll forming process are determined, including minimizing shape accuracy error and maximizing strength and fatigue performance.

[0094] The optimal process parameters that satisfy the optimization objective are solved in the optimizer; the optimal process parameters include at least one of load, torque, speed and displacement; the inner diameter roll forming equipment of the conduit is controlled in real time according to the optimal process parameters.

[0095] Example 5 If the modules / units integrated in the electronic device 100 are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, and read-only memory (ROM).

[0096] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0097] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0098] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0099] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0100] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0101] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A knowledge-driven digital twin method for rolling the inner diameter of a conduit, characterized in that, include: Constructing digital twin models of physical entities; Based on the digital twin model, macroscopic datasets of the inner diameter of the conduit are acquired in real time during the rolling process. The physical entities include the tubing inner diameter roll forming equipment and the tubing; the macro-scale dataset includes the material and geometric data of the tubing, as well as the process timing data of the tubing inner diameter roll forming process; Macroscale datasets are input into a knowledge-driven machine learning model to obtain real-time prediction data; the real-time prediction data includes predicted process time series data, two-dimensional field parameters, and three-dimensional field parameters. The performance of the conduit is determined based on real-time prediction data, including shape accuracy, connection strength, and fatigue performance. The performance is compared with preset standards. When the comparison results do not meet the preset conditions, the optimization objectives of the processing roll forming process are determined, including minimizing shape accuracy error and maximizing strength and fatigue performance. Solve for the optimization objective to obtain the optimal process parameters; the optimal process parameters include at least one of load, torque, speed and displacement.

2. The knowledge-driven digital twin method for tubing inner diameter rolling forming according to claim 1, characterized in that, In the step of inputting a macro-scale dataset into a knowledge-driven machine learning model to obtain real-time prediction data, the knowledge-driven machine learning model is trained in the following manner: Multimodal simulation data is generated based on a pre-trained GAN model; the multimodal simulation data includes: training process time series data, training two-dimensional image data, training two-dimensional field parameters, training three-dimensional geometric data, and training three-dimensional field parameters. The analysis and identification model is trained based on multimodal simulation data and corresponding material data. The trained analysis and identification model and the GAN model are then combined into a knowledge-driven machine learning model.

3. The knowledge-driven digital twin method for tubing inner diameter rolling forming according to claim 2, characterized in that, In the step of generating multimodal simulation data based on a pre-trained GAN model, the GAN model is trained as follows: A finite element simulation model of the catheter is constructed, and simulation processing is performed based on the finite element simulation model to obtain simulation results; wherein, the material physical characterization of the catheter is based on physical knowledge during the construction of the finite element simulation model. An initial dataset is extracted from the simulation results; the initial dataset includes: material data of the conduit, initial simulation geometric data, initial simulation process timing data, initial two-dimensional simulation field parameters, and initial three-dimensional simulation field parameters. The initial simulation geometric data, initial two-dimensional simulation field parameters, and initial three-dimensional simulation field parameters are used to generate a sample space through DOE; The DOE space discrete field quantity dataset is obtained by discretizing the sample space. The first multimodal data is extracted from the DOE spatial discrete field quantity dataset. The material data, the first multimodal data, and the initial simulation process time series data are used to construct the second multimodal data. The second multimodal data includes: material data, initial simulation process time series data, discretized two-dimensional simulation image data and corresponding discretized two-dimensional simulation field quantity parameters, and discretized three-dimensional simulation geometric data and corresponding discretized three-dimensional simulation field quantity parameters. The second multimodal data is input into the GAN model for pre-training. The GAN model includes a generator and a discriminator. The generator generates comparison data based on physical knowledge and random noise as input. The discriminator performs a similarity test between the second multimodal data and the comparison data to determine their realism. When the comparison data and the second multimodal data reach a preset similarity value, the GAN model is considered to have completed pre-training. The comparison data includes: comparison process time series data, comparison two-dimensional image data and corresponding comparison two-dimensional field parameters, comparison three-dimensional geometric data and corresponding comparison three-dimensional field parameters.

4. The knowledge-driven digital twin method for tubing inner diameter rolling forming according to claim 3, characterized in that, A finite element simulation model of the duct is constructed, and simulation processing is performed based on the finite element simulation model to obtain simulation results, including: In finite element simulation software, a three-dimensional model of the duct considering the complex constraints of the tooling is constructed, and a mesh model considering the complex constraints of the tooling is constructed based on the three-dimensional model. Simulations of the duct under rolling forming and bending fatigue conditions were performed based on a mesh model, and simulation results were obtained. The simulation results include material data, geometric data, and process timing data, as well as the corresponding two-dimensional and three-dimensional field parameters.

5. The knowledge-driven digital twin method for tubing inner diameter rolling forming according to claim 3, characterized in that, The second multimodal data is input into the GAN model for pre-training, including: The generator uses physical knowledge as constraints and random noise as input to generate data to be compared. The discriminator compares the process timing data to be compared with the initial simulation process timing data to obtain the first comparison result; the discriminator compares the two-dimensional image data and the two-dimensional field parameters to be compared with the discretized two-dimensional simulation image data and the discretized two-dimensional simulation field parameters to obtain the second comparison result; the discriminator compares the three-dimensional geometric data and the three-dimensional field parameters to be compared with the discretized three-dimensional simulation geometric data and the discretized three-dimensional simulation field parameters to obtain the third comparison result. When the first comparison result, the second comparison result, and the third comparison result all meet the preset range, the GAN model pre-training is determined to be complete. The first comparison result includes the similarity between the process time series data to be compared and the initial simulated process time series data; the second comparison result includes the similarity between the two-dimensional image data to be compared and the discretized two-dimensional simulated image data, and the similarity between the two-dimensional field parameters to be compared and the discretized two-dimensional simulated field parameters; the third comparison result includes the similarity between the three-dimensional geometric data to be compared and the discretized three-dimensional simulated geometric data, and the similarity between the three-dimensional field parameters to be compared and the discretized three-dimensional simulated field parameters.

6. The knowledge-driven digital twin method for tubing inner diameter rolling forming according to claim 2, characterized in that, The analysis and identification model is trained based on multimodal simulation data and corresponding material data, including: The analysis and recognition models include GNN, CNN, and LSTM models. The CNN model is trained based on material data, two-dimensional image data for training, and two-dimensional field parameters for training; the GNN model is trained based on material data, three-dimensional geometric data for training, and three-dimensional field parameters for training; and the LSTM model is trained based on material data and process time series data for training.

7. A knowledge-driven digital twin device for rolling and forming the inner diameter of a conduit, characterized in that, include: The model building module is used to build digital twin models of physical entities; Based on the digital twin model, macroscopic datasets of the inner diameter of the conduit are acquired in real time during the rolling process. The physical entities include the tubing inner diameter roll forming equipment and the tubing; the macro-scale dataset includes the material and geometric data of the tubing, as well as the process timing data of the tubing inner diameter roll forming process; The prediction module is used to input macro-scale datasets into a knowledge-driven machine learning model to obtain real-time prediction data; the real-time prediction data includes predicted process time series data, two-dimensional field parameters, and three-dimensional field parameters. The optimization module is used to determine the performance of the conduit based on real-time prediction data. The performance includes shape accuracy, connection strength, and fatigue performance. The performance is compared with preset standards. When the comparison results do not meet the preset conditions, the optimization objectives of the processing roll forming process are determined. The optimization objectives include minimizing shape accuracy error and maximizing strength and fatigue performance. The solver module is used to solve the optimization objective and obtain the optimal process parameters; the optimal process parameters include at least one of load, torque, speed and displacement.

8. The knowledge-driven digital twin device for rolling and forming the inner diameter of a conduit according to claim 7, characterized in that, In the prediction module, the knowledge-driven machine learning model is trained as follows: Multimodal simulation data is generated based on a pre-trained GAN model; the multimodal simulation data includes: training process time series data, training two-dimensional image data, training two-dimensional field parameters, training three-dimensional geometric data, and training three-dimensional field parameters. The analysis and identification model is trained based on multimodal simulation data and corresponding material data. The trained analysis and identification model and the GAN model are then combined into a knowledge-driven machine learning model.

9. The knowledge-driven digital twin device for rolling and forming the inner diameter of a conduit according to claim 8, characterized in that, In the step of generating multimodal simulation data based on a pre-trained GAN model, the GAN model is trained as follows: A finite element simulation model of the catheter is constructed, and simulation processing is performed based on the finite element simulation model to obtain simulation results; wherein, the material physical characterization of the catheter is based on physical knowledge during the construction of the finite element simulation model. An initial dataset is extracted from the simulation results; the initial dataset includes: material data of the conduit, initial simulation geometric data, initial simulation process timing data, initial two-dimensional simulation field parameters, and initial three-dimensional simulation field parameters. The initial simulation geometric data, initial two-dimensional simulation field parameters, and initial three-dimensional simulation field parameters are used to generate a sample space through DOE; The DOE space discrete field quantity dataset is obtained by discretizing the sample space. The first multimodal data is extracted from the DOE spatial discrete field quantity dataset. The material data, the first multimodal data, and the initial simulation process time series data are used to construct the second multimodal data. The second multimodal data includes: material data, initial simulation process time series data, discretized two-dimensional simulation image data and corresponding discretized two-dimensional simulation field quantity parameters, and discretized three-dimensional simulation geometric data and corresponding discretized three-dimensional simulation field quantity parameters. The second multimodal data is input into the GAN model for pre-training. The GAN model includes a generator and a discriminator. The generator generates comparison data based on physical knowledge and random noise as input. The discriminator performs a similarity test between the second multimodal data and the comparison data to determine their realism. When the comparison data and the second multimodal data reach a preset similarity value, the GAN model is considered to have completed pre-training. The comparison data includes: comparison process time series data, comparison two-dimensional image data and corresponding comparison two-dimensional field parameters, comparison three-dimensional geometric data and corresponding comparison three-dimensional field parameters.

10. An electronic device comprising a computer-readable storage medium, characterized in that, The electronic device includes a processor and a memory, and the computer-readable storage medium stores at least one instruction that, when executed by the processor, implements the knowledge-driven digital twin method for duct inner diameter rolling as described in any one of claims 1 to 6.

Images