Method for real-time monitoring and prediction of state during pipe bending process, and digital twin system

Abstract

Disclosed in the present invention are a method for real-time monitoring and prediction of a state during a pipe bending process, and a digital twin system. A gyroscope, a force sensor, etc. are mounted on a bending die, a pressing die and other dies of a pipe bending device to implement real-time collection of data such as the position, speed and acceleration of each die, and a pressure and friction force between each die and a pipe; a displacement sensor is mounted on a wiper die to implement wrinkling monitoring; a camera is mounted above the pipe to implement pipe wall state collection; and a gyroscope is mounted on a tail end of a mandrel to implement unloading springback monitoring of the pipe. The collected state information of the dies of the pipe bending device and that of the pipe is processed by means of a data processing system, and is then subjected to spatio-temporal fusion by means of a spatio-temporal fusion conversion module based on multi-task learning, so as to realize the construction of a digital twin model, and finally realize real-time monitoring and prediction of a state during the pipe bending process. In the present invention, the states of dies of a pipe bending device and the bending state of a pipe can be collected, and real-time monitoring and prediction of a state during a pipe bending process can be realized, and thus the precision and pipe bending quality of the pipe bending device can be improved.

Description

A method and digital twin system for real-time monitoring and prediction of pipe bending process status Technical Field

[0001] The present invention relates to the field of digital twins, specifically a method and a digital twin system for real-time monitoring and prediction of pipe bending process status. Background Art

[0002] Bent pipe components are formed by bending straight hollow pipes through bending equipment coupled with various dies. They are widely used in high-tech fields such as aviation and aerospace. The most precise and widely used bending equipment is CNC bending equipment, which uses a bending die, clamping die, pressing die, anti-wrinkle die, and mandrel. However, traditional bending methods use open-loop control and lack intelligence. This makes it impossible to obtain data on the actual operating status of the bending equipment dies and the deformation process during the bending process. To achieve higher-quality pipes, manual bending and offline measurement are required, relying on experience, resulting in low efficiency and difficulty in ensuring accuracy.

[0003] Digital twin systems use physical geometry models, real-time state data, and historical operational data to create a twin model in virtual space. This allows for the simulation and evolution of multiple timescales and multiple physical fields within the twin system, and provides online guidance and control of the actual scenario through a feedback system. With the application of digital twin systems in process manufacturing, mapping physical models to virtual models through data acquisition and predicting future states through digital twin modeling have become crucial foundational tasks.

[0004] To map the actual physical process of the full tube bending process to a virtual twin model and predict its future state, a reasonable data acquisition system must be constructed to enable online, real-time status monitoring and future state prediction of both the tube bending equipment mold state and the tube bending process. Current digital twin models for machining processes only model the equipment or workpiece. However, the workpiece's state monitoring and future state are affected by the equipment's state. Considering only a single dimension of information weakens the generalization capability of the digital twin model. Therefore, we propose a data acquisition method and digital twin system for real-time monitoring and prediction of the tube bending process state.

[0005] Summary of the Invention

[0006] In order to solve the problems in the background technology, the present invention discloses a real-time monitoring and prediction method and digital twin system for the status of the pipe bending process, which effectively realizes multi-directional online real-time monitoring and prediction of the mold status of the pipe bending equipment and the pipe bending process, and can improve the intelligence level of the pipe bending process and improve the quality of pipe bending forming.

[0007] The technical solution adopted in the present invention is as follows:

[0008] 1. A digital twin system for real-time monitoring and prediction of pipe bending process status

[0009] It includes pipe bending equipment, pipe fittings, data acquisition system, data processing system and twin model system. The data acquisition system is used to collect the status data of each mold of the pipe bending equipment and the deformation data of the pipe fittings during the bending process. The data processing system is used to preprocess the data collected by the data acquisition system. The twin model system predicts the mold status and pipe bending status of the pipe bending equipment at the current and future moments based on the data preprocessed by the data processing system.

[0010] The pipe bending equipment includes a bending die, an insert, a clamping die, a pressing die, an anti-wrinkle die, a booster trolley, a core shaft, and a core ball; the tail of the pipe fitting is clamped by the chuck of the booster trolley, the insert and the clamping die clamp the other end of the pipe fitting, and the middle is clamped by the pressing die and the anti-wrinkle die; a core ball and a core shaft for supporting the pipe wall are arranged in the pipe fitting, and multiple core balls are hinged on the core shaft after being connected in series, and the bending die and the insert are fixedly connected; the bending die, the insert, and the clamping die rotate synchronously, and apply torque to the pipe fitting through the synergistic action of pressure and friction. As the rotation angle increases, the pipe fitting undergoes plastic deformation; during the bending process, the anti-wrinkle die and the core shaft remain stationary, and the multiple core balls swing at a certain angle as the axial shape of the pipe fitting changes; during the bending process, the pressing die and the booster trolley move forward at a certain speed, and provide forward power to the unbent part of the pipe fitting through pressure and friction, so as to avoid defects such as fracture and cross-section collapse that cause failure of the pipe fitting;

[0011] The data acquisition system includes a pipe bending equipment mold state monitoring module and a pipe bending process monitoring module; the pipe bending equipment mold state monitoring module collects the state data of each pipe bending equipment mold, and the pipe bending process monitoring module collects the deformation data of the pipe during the pipe bending process;

[0012] The data processing system includes a data filtering and denoising module, a time series preprocessing module, a pipe bending process comprehensive information model, and a historical information storage module;

[0013] The twin model system includes a spatiotemporal fusion conversion module based on multi-task learning and a twin model visualization presentation module.

[0014] The pipe bending equipment mold status monitoring module includes:

[0015] Bend die gyroscope, embedded in the bend die surface, used to measure the bend die rotation angle, bending velocity and angular acceleration;

[0016] Several bend die temperature sensors are evenly distributed on the bend die and pass through the bend die from top to bottom, used to measure the temperature of the bend die;

[0017] The clamping die force sensor has an arc-shaped sensing surface and is embedded in the inner surface of the clamping die, that is, the contact surface with the pipe fitting, and the sensor sensing surface is completely in contact with the pipe fitting. The clamping die force sensor is used to measure the force between the clamping die and the pipe fitting, including the pressure between the clamping die and the pipe fitting and the friction between the clamping die and the pipe fitting.

[0018] The die gyroscope is embedded in the die surface and is used to measure the feed displacement, velocity and acceleration of the die;

[0019] The die force sensor has an arc-shaped sensing surface and is embedded in the inner surface of the die, that is, the contact surface with the pipe fitting, and the sensor sensing surface is completely in contact with the pipe fitting. The die force sensor is used to measure the force between the die and the pipe fitting, including the pressure between the die and the pipe fitting and the friction between the die and the pipe fitting.

[0020] Several die temperature sensors are evenly distributed on the die and pass through the die from top to bottom, for measuring the temperature of the die;

[0021] The booster trolley gyroscope is embedded in the outer surface of the booster trolley and is used to measure the feed displacement, speed and acceleration of the booster trolley;

[0022] The booster trolley force sensor has an arc-shaped sensing surface and is embedded in the inner surface of the chuck of the booster trolley, that is, the contact surface with the pipe fitting, and the sensor sensing surface is completely in contact with the pipe fitting; the booster trolley force sensor is used to measure the force between the booster trolley and the pipe fitting, including the pressure between the booster trolley and the pipe fitting, and the friction between the booster trolley and the pipe fitting.

[0023] The pipe bending process monitoring module includes:

[0024] The anti-wrinkle mold displacement sensor is embedded in the anti-wrinkle mold's curved surface and located at the very bottom of the curved surface, where the anti-wrinkle mold contacts the innermost concave side of the straight pipe section of the pipe. The anti-wrinkle mold displacement sensor probe contacts the pipe, with the probe axis perpendicular to the axis of the straight pipe section. It is used to monitor wrinkling during pipe bending, that is, to measure the displacement of wrinkle corrugations when the pipe is bent and wrinkled.

[0025] The core ball end gyroscope is installed at the tail of the core ball link to monitor the core ball status during the pipe bending process and the rebound angle of the pipe when the pipe is unloaded after bending.

[0026] A camera, mounted above the pipe using a camera bracket, monitors the pipe's bending state, including cross-sectional distortion, in real time. This camera, a depth camera, is mounted horizontally to the pipe's bending plane and can measure the deformation of the exposed portion of the pipe that is not in contact with the mold during bending.

[0027] All gyroscopes are high-precision six-axis gyroscopes that can measure displacement, angle, speed, angular velocity, acceleration, and angular acceleration; all force sensors are multi-dimensional force sensors that can measure pressure and friction; all temperature sensors use thermocouple temperature acquisition probes. The temperature sensor is only used during the heating bending process and can be selected not to be used when bending at room temperature.

[0028] 2. A method for real-time monitoring and prediction of pipe bending process status

[0029] Step 1) Data acquisition: The multi-sensor data of the data acquisition system is used to collect the state data of the pipe bending equipment and the deformation data of the pipe in real time during the pipe bending process;

[0030] Step 2) Data processing: The data collected by the data acquisition system is pre-processed by the data processing system, specifically:

[0031] The data collected by the data acquisition system is filtered and denoised by the data filtering and denoising module, and then the time stamps are unified by the time series preprocessing module to convert the data into data with the same time stamp and the same time interval. The data processed by the time series preprocessing module is integrated to obtain the comprehensive information model IM of the pipe bending process. The historical information storage module performs structured storage on the data processed by the comprehensive information model of the pipe bending process.

[0032] The present invention adopts Kalman filtering algorithm to filter the data and perform denoising to avoid adverse effects on the data caused by inherent vibration of the pipe bending equipment;

[0033] Step 3) Prediction and visualization of bending equipment and fittings status:

[0034] The spatiotemporal fusion conversion module based on multi-task learning predicts the status information of the pipe bending equipment and the bending status of the pipe fittings at the current and future moments according to the data preprocessed by the data processing system;

[0035] The twin model visualization module uses Unity to visualize the current and future status of the bending equipment mold and the pipe bending process, and then conducts feedback control on the bending equipment to improve the forming quality.

[0036] In the step 2):

[0037] The pipe bending process comprehensive information model IM includes the total data DataMach of the pipe bending equipment mold status monitoring part and the total data DataTube of the pipe bending process monitoring part, wherein each data is a time series data with equal time intervals. The pipe bending process comprehensive information model IM includes the total data DataMach of the pipe bending equipment mold status monitoring part and the total data DataTube of the pipe bending process monitoring part, that is,

[0038] IM = {DataMach, DataTube}

[0039] Among them, the total data DataMach of the mold status monitoring part of the pipe bending equipment is composed of the bending die rotation angle θbend, the bending die bending speed ωbend, the bending die angular acceleration αbend, the bending die temperature Tbend, the pressure between the clamping die and the tube Pclamp-tube, the friction between the clamping die and the tube Fclamp-tube, the die displacement dpress, the die speed vpress, the die acceleration αpress, the pressure between the die and the tube Ppress-tube, the friction between the die and the tube Fpress-tube, the die temperature Tpress, the booster trolley displacement dboost, the booster trolley speed vboost, the booster trolley acceleration αboost, the pressure between the booster trolley and the tube Pboost-tube, and the friction between the booster trolley and the tube Fboost-tube, that is,

[0040] DataMach = {θ bend, ω bend, α bend, T bend, P clamp-tube, F clamp-tube, d pressure, v pressure, α pressure, P pressure-tube, F pressure-tube, T pressure, d boost, v boost, α boost, P boost-tube, F boost-tube}

[0041] The total data of the pipe bending process monitoring part DataTube consists of the wrinkle corrugation displacement dwrinkle, the pipe bending angle θtube, and the pipe cross-section deformation data εtube, that is,

[0042] DataTube = {d wrinkle, θ tube, ε tube}

[0043] All data in the integrated information model IM of the pipe bending process are in the form of time series.

[0044] In step 3), the spatiotemporal fusion conversion module based on multi-task learning includes three parts: an input layer, a private-shared layer, and a task output layer;

[0045] The input layer receives data of the integrated information model IM of the pipe bending process, including the total data DataMach of the pipe bending equipment mold status monitoring part and the total data DataTube of the pipe bending process monitoring part;

[0046] The private-shared layer includes a shared LSTM module and two private LSTM modules, the two private LSTM modules are respectively an auxiliary task private LSTM module and a main task private LSTM module; the shared LSTM module receives two parts of data, namely, DataMach, the total data of the mold state monitoring part of the pipe bending equipment, and DataTube, the total data of the pipe bending process monitoring part, for shared common feature extraction and mining the interaction relationship between the pipe bending equipment and the bent pipe; the auxiliary task private LSTM module and the main task private LSTM module receive DataMach, the total data of the mold state monitoring part of the pipe bending equipment, and DataTube, the total data of the pipe bending process monitoring part, for extracting the temporal evolution law of the mold state of the pipe bending equipment and the temporal evolution law of the pipe bending process;

[0047] The task output layer includes an auxiliary task Dense module, a feature fusion Concatenate module and a main task Dense module; the output results of the auxiliary task private LSTM module and the shared LSTM module are added element by element and then input into the auxiliary task Dense module; the output results of the main task private LSTM module and the shared LSTM module are added element by element, and then input into the feature fusion Concatenate module together with the output results of the auxiliary task Dense module for series splicing to achieve feature fusion, and the fused data is input into the main task Dense module; the auxiliary task Dense module and the main task Dense module respectively output the prediction results of the bending equipment status and the prediction results of the pipe fitting bending forming as the final output results of the module.

[0048] The spatiotemporal fusion conversion module based on multi-task learning is trained through a joint loss function and adapted to the multi-task learning scenario. The joint loss function is defined as the weighted sum of the auxiliary task loss L auxiliaries of the tube bending equipment mold state and the main task loss L main main of the tube bending process state:

[0049] L=L main+αL auxiliary

[0050] Among them, L is the total loss of the model, and the weight α is determined according to the empirical method.

[0051] In the step 3):

[0052] Based on the bending equipment information and the bending state of the pipe fittings before the current moment, the state of the bending equipment mold and the deformation state of the pipe fittings at the current moment and in the future are predicted;

[0053] The state of the pipe bending equipment mold includes an abnormal state of the pipe bending equipment, such as vibration of the pipe bending equipment;

[0054] The predicted deformation state of the pipe fitting includes predicting the distortion defect of the cross section of the pipe fitting.

[0055] The pipe bending equipment can compensate online based on the wrinkle ripples directly measured by the sensor and the cross-sectional distortion defects obtained through prediction. The compensation can be achieved by speeding up or slowing down the speed of the bending die, pressing die, and booster trolley, as well as increasing or decreasing the pressure of the pressing die and booster trolley on the pipe.

[0056] The pipe bending equipment can bend and compensate again online based on the springback angle of the pipe measured after bending unloading.

[0057] Beneficial effects of the present invention:

[0058] The present invention sets corresponding sensors at appropriate positions of the pipe bending equipment and the pipe bending site to collect the time series of the operation status of the pipe bending equipment and the deformation status of the pipe fittings, and processes the data through the data processing system to obtain a digital twin data model of the pipe bending process, thereby realizing the integration of the state data of the pipe bending equipment and the pipe fittings in the pipe bending process, and modeling the integrated time series information through the spatiotemporal fusion conversion module based on multi-task learning. The multi-task learning comprehensively considers the mutual influence between the state of the pipe bending equipment and the pipe fittings in the pipe bending process, and can realize more accurate pipe forming by considering the influence of the future state of the pipe bending equipment information on the future state of the pipe fittings. The forming state prediction of the pipe bending equipment and pipe fittings at the current moment is predicted by using the data before the current moment, which compensates for the time lag problem caused by the data processing process, improves the real-time performance of the digital twin model, and can predict the future state at the same time, providing a basis for the digital twin system to optimize the pipe bending process in real time through decision-making. The real-time monitoring state and future prediction results are finally visualized through the twin model, effectively realizing the multi-dimensional online real-time monitoring and future prediction of the pipe bending equipment mold state and the pipe bending process, which can improve the intelligence level of the pipe bending process and improve the quality of pipe bending. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] FIG1 is a schematic diagram of a pipe bending process according to the present invention;

[0060] FIG2 is a simplified schematic diagram of the pipe bending equipment of the present invention;

[0061] FIG3 is a simplified schematic diagram of the overall structure of the system of the present invention;

[0062] FIG4 is a cross-sectional view of the overall structure of the system of the present invention;

[0063] FIG5 is a schematic diagram of the installation of the anti-wrinkle mold displacement sensor of the present invention;

[0064] FIG6 is a structural diagram of the anti-wrinkle mold of the present invention;

[0065] FIG7 is a system principle diagram of the present invention;

[0066] Figure 8 is a schematic diagram of the spatiotemporal fusion conversion module based on multi-task learning.

[0067] Figure: 1. Pipe bending equipment, 2. Pipe fittings, 3. Data acquisition system, 4. Data processing system, 5. Twin model system, 6. Bending die, 7. Insert, 8. Clamping die, 9. Pressing die, 10. Anti-wrinkle die, 11. Booster trolley, 12. Mandrel, 13. Core ball, 14. Bending die gyroscope, 15. Bending die temperature sensor, 16. Clamping die force sensor, 17. Clamping die temperature sensor, 18. Pressing die gyroscope, 19. Pressing die force sensor, 20. , die temperature sensor, 21. Booster trolley gyroscope, 22. Booster trolley force sensor, 23. Anti-wrinkle mold displacement sensor, 24. Core ball end gyroscope, 25. Camera bracket, 26. Camera, 41. Data filtering and denoising module, 42. Time series preprocessing module, 43. Comprehensive information model of bending process, 44. Historical task storage module, 51. Spatiotemporal fusion conversion module based on multi-task learning, 52. Twin model visualization presentation module. DETAILED DESCRIPTION

[0068] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0069] As shown in Figures 1 to 7, the present invention includes pipe bending equipment 1, pipe fittings 2, a data acquisition system 3, a data processing system 4, and a twin model system 5.

[0070] As shown in FIG2 , the pipe bending equipment 1 includes a bending die 6 , an insert 7 , a clamping die 8 , a pressing die 9 , an anti-wrinkle die 10 , a booster trolley 11 , a core shaft 12 , and a core ball 13 .

[0071] As shown in Figures 3 and 4, the data acquisition system 3 includes a tube bending equipment mold state monitoring part and a tube bending process monitoring part, wherein the tube bending equipment mold state monitoring part includes a bending die gyroscope 14, a bending die temperature sensor 15, a clamping die force sensor 16, a clamping die temperature sensor 17, a pressing die gyroscope 18, a pressing die force sensor 19, a pressing die temperature sensor 20, a booster trolley gyroscope 21, and a booster trolley force sensor 22; the tube bending process monitoring part includes an anti-wrinkle die displacement sensor 23, a core ball end gyroscope 24, a camera bracket 25 and a camera 26, wherein all gyroscopes are high-precision six-axis gyroscopes that can measure displacement, angle, velocity, angular velocity, acceleration, and angular acceleration; all force sensors are multi-dimensional force sensors; and the temperature sensor uses a thermocouple temperature acquisition probe. The temperature sensor is only used during the heating bending process and can be selected not to be used when bending at room temperature.

[0072] As shown in Figures 3 and 4, in the mold status monitoring section of the pipe bending equipment, a bend die gyroscope 14 is embedded in the upper surface of the bend die 6 and can accurately measure the rotation angle, bending velocity, and angular acceleration of the bend die 6. Several bend die temperature sensors 15 are embedded in the bend die 6, penetrating the upper and lower surfaces of the bend die 6 to ensure that the temperature of the contact portion between the bend die 6 and the pipe fitting 2 is measured. A clamping die force sensor 16 is embedded in the contact surface with the pipe fitting 2. The contact portion of the clamping die force sensor 16 with the pipe fitting 2 forms an arc surface, sharing a circular arc with the contact surface between the clamping die 8 and the pipe fitting 2. The clamping die force sensor 16 is at least a two-dimensional force sensor, one dimension of which measures the pressure between the clamping die 8 and the pipe fitting 2, and the other dimension measures the friction between the clamping die 8 and the pipe fitting 2. Because the clamping die 8 rotates synchronously with the bend die 6, no gyroscope is installed in the clamping die 8. The die gyroscope 18 is embedded in the outer surface of the die 9 and can accurately measure the feed displacement, velocity, and acceleration of the die 9. The die force sensor 19 is embedded in the surface in contact with the pipe 2. The contact portion of the die force sensor 19 with the pipe 2 is an arc surface, co-circular with the contact surface between the die 9 and the pipe 2. The die force sensor 19 is at least a two-dimensional force sensor, one dimension of which measures the pressure between the die 9 and the pipe 2, and the other dimension measures the friction between the die 9 and the pipe 2. Several die temperature sensors 20 are evenly embedded in the die 9, extending across the upper and lower surfaces of the die 9 to ensure that the temperature of the contact portion between the die and the pipe 2 is measured. The booster trolley gyroscope 21 is embedded in the outer surface of the booster trolley 11 and can accurately measure the feed displacement, speed and acceleration of the booster trolley 11. The booster trolley force sensor 22 is embedded in the contact surface between the chuck of the booster trolley 11 and the pipe fitting 2. The contact part between the booster trolley force sensor 22 and the pipe fitting 2 is an arc surface, and is a cocircular arc surface with the contact surface between the chuck of the booster trolley 11 and the pipe fitting 2. The booster trolley force sensor 22 is at least a two-dimensional force sensor, one dimension of which measures the pressure between the booster trolley 11 and the pipe fitting 2, and the other dimension measures the friction between the booster trolley 11 and the pipe fitting 2.

[0073] As shown in Figures 3 to 6, in the pipe bending process monitoring part, a hole is opened at the bottom of the arc surface of the anti-wrinkle mold 10 (that is, the contact position with the innermost concave side of the straight pipe section of the pipe) to embed an anti-wrinkle mold displacement sensor 23. The anti-wrinkle mold displacement sensor 23 probe is extended to contact the pipe 2. When the pipe 2 is bent and wrinkled, the wrinkle corrugation displacement can be measured to monitor the wrinkling condition during the pipe bending process. The core ball end gyroscope 24 is installed at the end of the core ball 13 link, which can monitor the state of the core ball 13 during the pipe bending process, and can measure the rebound angle of the pipe 2 when the pipe 2 is unloaded after bending. The pipe bending equipment 1 can bend and compensate again online according to the measured rebound angle. The camera 26 is installed on the camera bracket 25. The camera 26 is a depth camera and is installed parallel to the bending plane of the pipe 2. It can measure the deformation state of the part of the pipe 2 that is not in contact with the mold and exposed to the outside when the pipe 2 is bent, and is used to monitor the cross-sectional deformation data of the pipe 2 during the bending deformation process in real time to observe the distortion state.

[0074] As shown in Figure 7, the digital twin model system of pipe bending equipment includes pipe bending equipment 1, pipe fittings 2, data acquisition system 3, data processing system 4, and twin model system 5. The data processing system 4 includes a data filtering and denoising module 41, a time series preprocessing module 42, a comprehensive information model of the pipe bending process 43, and a historical information storage module 44. The data filtering and denoising module 41 filters and denoises the data collected by multiple sensors in the data acquisition system 3. The present invention uses a Kalman filter algorithm to filter the data and perform denoising to avoid the adverse effects of the inherent vibration of the pipe bending equipment 1 on the data. Then, the time stamps are unified through the time series preprocessing module 42 and converted into data with the same time stamp and the same time interval. The comprehensive information model IM43 of the pipe bending process includes the total data DataMach of the mold status monitoring part of the pipe bending equipment and the total data DataTube of the pipe fitting bending process monitoring part, wherein each data is time series data with equal time intervals. The comprehensive information model IM43 of the pipe bending process includes the total data DataMach of the mold status monitoring part of the pipe bending equipment and the total data DataTube of the pipe fitting bending process monitoring part, that is,

[0075] IM = {DataMach, DataTube}

[0076] Among them, the total data DataMach of the mold status monitoring part of the pipe bending equipment is composed of the bending die rotation angle θbend, the bending die bending speed ωbend, the bending die angular acceleration αbend, the bending die temperature Tbend, the pressure between the clamping die and the tube Pclamp-tube, the friction between the clamping die and the tube Fclamp-tube, the die displacement dpress, the die speed vpress, the die acceleration αpress, the pressure between the die and the tube Ppress-tube, the friction between the die and the tube Fpress-tube, the die temperature Tpress, the booster trolley displacement dboost, the booster trolley speed vboost, the booster trolley acceleration αboost, the pressure between the booster trolley and the tube Pboost-tube, and the friction between the booster trolley and the tube Fboost-tube, that is,

[0077] DataMach = {θ bend, ω bend, α bend, T bend, P clamp-tube, F clamp-tube, d pressure, v pressure, α pressure, P pressure-tube, F pressure-tube, T pressure, d boost, v boost, α boost, P boost-tube, F boost-tube}

[0078] The total data of the pipe bending process monitoring part DataTube consists of the wrinkle corrugation displacement dwrinkle, the pipe bending angle θtube, and the pipe cross-section deformation data εtube, that is,

[0079] DataTube = {d wrinkle, θ tube, ε tube}

[0080] All data in the pipe bending process comprehensive information model IM43 is in the form of time series. Finally, the historical information storage module 44 performs structured storage on the data processed by the pipe bending process comprehensive information model 43. The twin model system 5 includes a spatiotemporal fusion conversion module 51 based on multi-task learning and a twin model visualization presentation module 52. The spatiotemporal fusion conversion module 51 based on multi-task learning is based on a private-shared multi-task learning framework and is used to receive the pipe bending process comprehensive information model IM43 and perform time series prediction on the state of the pipe bending equipment mold and the pipe bending process to achieve prediction of the current and future states of the pipe bending equipment mold state and the pipe bending process (including defect monitoring and prediction of cross-sectional distortion). Finally, the twin model visualization presentation module 52 uses Unity to visualize the state of the pipe bending equipment mold and the pipe bending process, thereby performing feedback control on the pipe bending equipment 1 to improve the forming quality.

[0081] As shown in Figure 7, the spatiotemporal fusion conversion module 51 based on multi-task learning is based on a private-shared multi-task learning framework. The framework is divided into two sub-tasks: the auxiliary task of the pipe bending equipment mold state and the main task of the pipe bending process state. The main task of the pipe bending process state is used to predict the pipe forming quality, and the auxiliary task of the pipe bending equipment mold state is used to predict the pipe bending equipment state (including predicting whether the pipe bending machine is abnormal: such as vibration, temperature abnormality). Since the pipe bending equipment mold state will affect the pipe forming quality, the output results of the auxiliary tasks of the task output layer are integrated into the main task prediction to improve the robustness and accuracy of the main task of the pipe bending process state.

[0082] The private-shared multi-task learning framework includes three parts: input layer, private-shared layer, and task output layer. The input layer receives the comprehensive information model IM43 of the pipe bending process, which is divided into two parts: the total data DataMach of the mold status monitoring part of the pipe bending equipment and the total data DataTube of the pipe fitting bending process monitoring part. The private-shared layer includes a shared LSTM module and two private LSTM modules. The two private LSTM modules are respectively an auxiliary task private LSTM module and a main task private LSTM module. The shared LSTM module receives the total data DataMach of the mold status monitoring part of the pipe bending equipment and the total data DataTube of the pipe fitting bending process monitoring part, which are used for shared common feature extraction and mining the interaction relationship between pipe bending equipment and bending pipes; the two private LSTM modules receive the total data DataMach of the mold status monitoring part of the pipe bending equipment and the total data DataTube of the pipe fitting bending process monitoring part, which are used to extract the temporal evolution law of the mold status of the pipe bending equipment and the temporal evolution law of the pipe fitting bending process. The output of the shared LSTM is element-by-element added to the private LSTM outputs of the main and auxiliary tasks, respectively. The output layer includes an auxiliary task Dense module, a feature fusion Concatenate module, and a main task Dense module. The auxiliary task Dense module for the pipe bending equipment mold state only receives the auxiliary task information for the pipe bending equipment mold state processed by the private-shared layer. The auxiliary task Dense module processes the input data to obtain the final auxiliary task prediction result. The feature fusion Concatenate module receives the main task information for the pipe bending process state and the auxiliary task output results for the pipe bending equipment mold state processed by the private-shared layer and performs a concatenation operation to achieve feature fusion. The fused data is ultimately passed through the main task Dense module to generate the future prediction result for the main task. Finally, a joint loss function is used to train the model and adapt it to multi-task learning scenarios. The joint loss function is defined as the weighted sum of the auxiliary task loss L of the pipe bending equipment mold state and the main task loss L of the pipe bending process state:

[0083] L=L main+αL auxiliary

[0084] Among them, L is the total loss of the model, and the weight α can be determined according to the empirical method.

[0085] Through online training, the bending equipment information and pipe forming status before the current moment can be used to predict the bending equipment information (including abnormal operating conditions such as equipment vibration) and pipe bending status (including defects such as predicted cross-sectional distortion) at the current moment t and future moments. In addition, the influence of the future state of the bending equipment information on the future state of the pipe forming can be comprehensively considered to realize the prediction of the pipe forming status, making the prediction more accurate.

Claims

1. A digital twin system for real-time monitoring and prediction of the state of the pipe bending process, characterized in that, It includes an elbow pipe equipment (1), a pipe fitting (2), a data acquisition system (3), a data processing system (4), and a digital twin model system (5). The data acquisition system (3) is used to acquire the status data of each die of the elbow pipe equipment (1) and the deformation data of the pipe fitting during the bending process of the pipe fitting (2). The data processing system (4) is used to preprocess the data acquired by the data acquisition system (3). The digital twin model system (5) predicts the die status of the elbow pipe equipment and the bending status of the pipe fitting at the current and future moments based on the data preprocessed by the data processing system (4).

2. The digital twin system for real-time monitoring and prediction of the elbow pipe process status according to claim 1, wherein the elbow pipe equipment (1) includes a bending die (6), an insert block (7), a clamping die (8), a pressing die (9), a wrinkle prevention die (10), a boosting trolley (11), a mandrel (12), and a core ball (13); the tail of the pipe fitting (2) is clamped by the chuck of the boosting trolley (11), the insert block (7) and the clamping die (8) clamp the other end of the pipe fitting (2), and the middle is clamped by the pressing die (9) and the wrinkle prevention die (10); a core ball (13) and a mandrel (12) for supporting the pipe wall are arranged inside the pipe fitting. Multiple core balls (13) are connected in series and articulated on the mandrel (12), and the bending die (6) and the insert block (7) are fixedly connected; the bending die (6), the insert block (7), and the clamping die (8) rotate synchronously, and apply torque to the pipe fitting (2) through the combined action of pressure and friction. As the rotation angle increases, the pipe fitting undergoes plastic deformation; during the bending process, the wrinkle prevention die (10) and the mandrel (12) remain stationary, and multiple core balls (13) swing as the axis shape of the pipe fitting (2) changes; during the bending process, the pressing die (9) and the boosting trolley (11) move forward to provide forward power to the unbent part of the pipe fitting (2) through pressure and friction; the data acquisition system (3) includes an elbow pipe equipment die status monitoring module and a pipe fitting bending process monitoring module; the status data of each die of the elbow pipe equipment (1) is acquired through the elbow pipe equipment die status monitoring module, and the deformation data of the pipe fitting during the bending process of the pipe fitting (2) is acquired through the pipe fitting bending process monitoring module; the data processing system (4) includes a data filtering and denoising module (41), a time series preprocessing module (42), an elbow pipe process comprehensive information model (43), and a historical information storage module (44); the digital twin model system (5) includes a spatio-temporal fusion transformation module (51) based on multi-task learning and a digital twin model visualization presentation module (52).

3. The multi-sensor data acquisition and status monitoring system for digital twin in the pipe bending process according to claim 2, characterized in that, The elbow pipe equipment die status monitoring module includes: a bending die gyroscope (14), embedded on the surface of the bending die (6), for measuring the rotation angle, bending speed, and angular acceleration of the bending die (6); a plurality of bending die temperature sensors (15), evenly distributed on the bending die (6) and penetrating the bending die (6) up and down, for measuring the temperature of the bending die (6); The clamping die force sensor (16) has a sensor sensing surface in the shape of an arc surface and is embedded in the inner surface of the clamping die (8), that is, the contact surface with the pipe fitting, and the sensor sensing surface is in complete fit with the pipe fitting; the clamping die force sensor (16) is used to measure the acting force between the clamping die and the pipe fitting, including the pressure between the clamping die (8) and the pipe fitting (2) and the frictional force between the clamping die (8) and the pipe fitting (2). The pressing die gyroscope (18) is embedded in the surface of the pressing die (9) and is used to measure the feed displacement, speed and acceleration of the pressing die (9). The pressing die force sensor (19) has a sensor sensing surface in the shape of an arc surface and is embedded in the inner surface of the pressing die (9), that is, the contact surface with the pipe fitting, and the sensor sensing surface is in complete fit with the pipe fitting; the pressing die force sensor (19) is used to measure the acting force between the pressing die (9) and the pipe fitting, including the pressure between the pressing die (9) and the pipe fitting (2) and the frictional force between the pressing die (9) and the pipe fitting (2). A number of pressing die temperature sensors (20) are evenly distributed on the pressing die (9) and penetrate the pressing die (9) vertically up and down, and are used to measure the temperature of the pressing die (9). The boosting trolley gyroscope (21) is embedded in the outer surface of the boosting trolley (11) and is used to measure the feed displacement, speed and acceleration of the boosting trolley (11). The boosting trolley force sensor (22) has a sensor sensing surface in the shape of an arc surface and is embedded in the inner surface of the chuck of the boosting trolley (11), that is, the contact surface with the pipe fitting (2), and the sensor sensing surface is in complete fit with the pipe fitting (2); the boosting trolley force sensor (22) is used to measure the acting force between the boosting trolley (11) and the pipe fitting, including the pressure between the boosting trolley (11) and the pipe fitting (2) and the frictional force between the boosting trolley (11) and the pipe fitting (2).

4. The multi-sensor data acquisition and status monitoring system for digital twin in the pipe bending process according to claim 2, characterized in that, The pipe fitting bending process monitoring module includes: The anti-wrinkle die displacement sensor (23) is embedded in the arc surface of the anti-wrinkle die (10) and is located at the bottom of the arc surface, that is, the contact position between the anti-wrinkle die (10) and the innermost concave side of the straight pipe section of the pipe fitting; the probe of the anti-wrinkle die displacement sensor (23) contacts the pipe fitting (2), and the axis of the probe is perpendicular to the axis of the straight pipe section of the pipe fitting, and is used to measure the wrinkling ripple displacement when the pipe fitting (2) bends and wrinkles. The core ball end gyroscope (24) is installed at the connecting tail of the core ball (13) and is used to monitor the state of the core ball (13) during the bending process of the pipe fitting and the springback angle of the pipe fitting when the bending of the pipe fitting ends and the load is unloaded. The camera (26) is installed above the pipe fitting (2) through the camera bracket (25) and is used to monitor the bending state during the bending deformation process of the pipe fitting (2) in real time, including the cross-section distortion state.

5. A method for real-time monitoring and prediction using the system according to any one of claims 1 to 4, characterized in that, It includes: Step 1) Data acquisition: The state data of the pipe bending equipment (1) and the deformation data of the pipe fitting (2) during the pipe fitting bending process are collected in real time through multiple sensors of the data acquisition system (3). Step 2) Data processing: The data collected by the data acquisition system (3) is preprocessed by the data processing system (4). Specifically: The data filtering and denoising module (41) filters and denoises the data collected by the data acquisition system (3), and then the time series preprocessing module (42) unifies the timestamps and converts them into data with the same timestamp and the same time interval. The data processed by the time series preprocessing module (42) is integrated to obtain the comprehensive information model IM (43) of the elbow bending process, and the historical information storage module (44) structurally stores the data processed by the comprehensive information model (43) of the elbow bending process. Step 3) Prediction and visualization of the elbow bending equipment and pipe fitting status: Based on the multi-task learning spatio-temporal fusion conversion module (51), the status information of the elbow bending equipment and the bending status of the pipe fitting at the current and future moments are predicted according to the data preprocessed by the data processing system (4). The twin model visualization presentation module (52) visually presents the die status of the elbow bending equipment and the pipe bending process at the current and future moments through unity, and then performs feedback control on the elbow bending equipment.

6. The real-time monitoring and prediction method according to claim 5, characterized in that In the said step 2): The comprehensive information model IM of the elbow bending process includes the total data Data Mach of the die status monitoring part of the elbow bending equipment and the total data Data Tube of the pipe bending process monitoring part. Each data is a time series data with equal time intervals. The comprehensive information model IM of the elbow bending process includes the total data Data Mach of the die status monitoring part of the elbow bending equipment and the total data Data Tube of the pipe bending process monitoring part, that is IM = {Data Mach, Data Tube} Among them, the total data Data Mach of the die status monitoring part of the elbow bending equipment consists of the bending die rotation angle θ bend, the bending die bending speed ω bend, the bending die angular acceleration α bend, the bending die temperature T bend, the pressure P clamp - pipe between the clamping die and the pipe, the friction force F clamp - pipe between the clamping die and the pipe, the punch displacement d punch, the punch speed v punch, the punch acceleration α punch, the pressure P punch - pipe between the punch and the pipe, the friction force F punch - pipe between the punch and the pipe, the punch temperature T punch, the displacement d boost of the boosting trolley, the speed v boost of the boosting trolley, the acceleration α boost of the boosting trolley, the pressure P boost - pipe between the boosting trolley and the pipe, and the friction force F boost - pipe between the boosting trolley and the pipe, that is Data Mach = {θ bend, ω bend, α bend, T bend, P clamp - pipe, F clamp - pipe, d punch, v punch, α punch, P punch - pipe, F punch - pipe, T punch, d boost, v boost, α boost, P boost - pipe, F boost - pipe} The total data Data Tube of the pipe bending process monitoring part consists of the wrinkling corrugation displacement d wrinkle, the pipe bending angle θ pipe, and the pipe cross-section deformation data ε pipe, that is Data Tube = {d wrinkle, θ pipe, ε pipe} All the data in the comprehensive information model IM of the elbow bending process are in the form of time series.

7. The real-time monitoring and prediction method according to claim 2, wherein In the said step 3), the multi-task learning spatio-temporal fusion conversion module (51) includes an input layer, a private - shared layer, and a task output layer. The input layer receives data from the comprehensive information model IM(43) of the elbow bending process, including the total data Data Mach of the die state monitoring part of the elbow bending equipment and the total data Data Tube of the pipe fitting bending process monitoring part; The private - shared layer includes a shared LSTM module and two private LSTM modules. The two private LSTM modules are the auxiliary task private LSTM module and the main task private LSTM module respectively. The shared LSTM module receives the two parts of data, namely the total data Data Mach of the die state monitoring part of the elbow bending equipment and the total data Data Tube of the pipe fitting bending process monitoring part. The auxiliary task private LSTM module and the main task private LSTM module receive the total data Data Mach of the die state monitoring part of the elbow bending equipment and the total data Data Tube of the pipe fitting bending process monitoring part respectively; The task output layer includes an auxiliary task Dense module, a feature fusion Concatenate module and a main task Dense module. The output result of the auxiliary task private LSTM module and the shared LSTM module are added element - by - element and then input into the auxiliary task Dense module. The output result of the main task private LSTM module and the shared LSTM module are added element - by - element, and then jointly input into the feature fusion Concatenate module with the output result of the auxiliary task Dense module for concatenation to achieve feature fusion. The fused data is input into the main task Dense module. The auxiliary task Dense module and the main task Dense module respectively output the prediction results of the elbow bending equipment state and the pipe fitting bending forming result, as the final output results of the module.

8. The real-time monitoring and prediction method according to claim 7, characterized in that Train the spatio - temporal fusion transformation module based on multi - task learning through a joint loss function and make it adapt to the multi - task learning scenario. The joint loss function is defined as the weighted sum of the auxiliary task loss L_aux of the elbow bending equipment die state and the main task loss L_main of the pipe fitting bending process state: L = L_main+αL_aux Where L is the total loss of the model, and the weight α is determined according to the empirical method.

9. The real-time monitoring and prediction method according to claim 7, characterized in that In step 3): Predict the state of the elbow bending equipment die and the deformation state of the pipe fitting at the current moment and future moments according to the elbow bending equipment information and the pipe fitting bending state before the current moment; The state of the elbow bending equipment die includes the abnormal state of the elbow bending equipment; Predicting the deformation state of the pipe fitting includes predicting the distortion defect of the pipe fitting cross - section.

10. The real - time monitoring and prediction method according to claim 7, characterized in that The elbow bending equipment (1) can compensate online according to the wrinkling waves directly measured by the sensor and the cross - section distortion defect obtained by prediction. The compensation can be achieved by accelerating or decelerating the speeds of the bending die (6), the pressing die (9), the boosting trolley (11) and increasing or decreasing the pressure of the pressing die (9) and the boosting trolley (11) on the pipe fitting; The elbow bending equipment (1) can bend and compensate again online according to the springback angle of the pipe fitting measured after bending unloading.

Images