A method of pipe forming based on pattern recognition

Abstract

The present application relates to a kind of pipeline forming method based on pattern recognition, belong to pipeline forming technical field, solve the problem that existing cannot identify rebound amount and detect pipeline defect and lead to pipeline processing efficiency and quality are not high.Present application includes: simulation analysis is carried out to three-dimensional pipeline model, output preset parameter;According to the pipeline parameter of pipeline model and preset parameter, preliminary test is carried out to pipe bending machine, and after updating and supplementing preset parameter, it is constructed into sample and is placed into sample database;A plurality of types of parameter samples are constructed to train corresponding pattern recognition model respectively, predict pipeline category and rebound amount;Data acquisition component is installed on pipe bending machine, and the perception test is carried out to pipe bending machine using pattern recognition model, and decision rule is extracted;Update sample database and train pattern recognition model again, optimize decision rule;When pipeline category is not qualified, reference data is predicted according to decision rule, compensates pipe bending machine process data to complete pipeline forming.Present application realizes the improvement of pipeline quality and efficiency.

Description

Technical Field

[0001] This invention relates to the field of pipe forming technology, and in particular to a pipe forming method based on pattern recognition. Background Technology

[0002] Pipe bending technology is widely used in the production of pipes in various industries such as central air conditioning, automobile industry, aerospace industry, and shipbuilding. The quality of pipe bending directly affects the structural rationality, safety, and reliability of products in these industries.

[0003] Current pipeline forming processing equipment mainly revolves around pipeline routing measurement. In the process of pipeline model input-processing-measurement-comparison measurement-compensation-processing-measurement-output, theoretical and actual measurement data comparison and feedback are used to focus on the entire bend direction target. However, during the processing, the pipeline is prone to wrinkles, cracks and elliptical distortion.

[0004] Current automated pipe bending methods often use factors such as front and rear clamp pressure, rotation, clamping die position, conduit diameter, conduit thickness, and bending radius as fixed quantities or experimental conditions, while bending angle and mandrel position are used as variables, springback as a dependent variable, and wrinkling and cracking as quality rejection items. After pipe forming, inspectors visually check for breakage, wrinkling, excessive thinning, etc., and then use gauges to further inspect the bending angle of products that have passed this step, thereby preventing defective products from being shipped out as much as possible. This inspection is a process of distinguishing between qualified and defective pipes, but it cannot detect and quickly identify wrinkling and cracks in real time during pipe processing, increasing the pipe scrap rate, reducing pipe accuracy and quality, and is also not conducive to fully automated production. Summary of the Invention

[0005] Based on the above analysis, the present invention aims to provide a pipe forming method based on pattern recognition to solve the problem of low pipe processing efficiency and quality caused by the inability to identify springback and detect pipe defects.

[0006] This invention provides a pipeline forming method based on pattern recognition, comprising the following steps:

[0007] Simulation analysis of the three-dimensional pipeline model is performed, and preset parameters are output to the pipe bending machine;

[0008] A preliminary test of the pipe bending machine was conducted based on the pipe parameters and preset parameters of the pipe model. After updating and supplementing the preset parameters based on the test results, a sample was constructed and placed into the sample database.

[0009] Based on the sample database, various types of parameter samples are constructed to train corresponding pattern recognition models to predict pipeline type and rebound amount;

[0010] A data acquisition component is installed on the pipe bending machine. A pattern recognition model is used to conduct a perception experiment on the pipe bending machine and extract decision rules. The sample database is updated based on the perception data at each stage, and the pattern recognition model is retrained to optimize the decision rules.

[0011] During pipe processing, the process data acquired by the data acquisition component is fed into the pattern recognition model to predict the pipe category. When the pipe category is not qualified, reference data is predicted according to the decision rules to compensate the pipe bending machine process data and complete the pipe forming.

[0012] Based on further improvements to the above method, simulation analysis is performed on the three-dimensional pipeline model, and preset parameters are output. These parameters include: using the basic information of the three-dimensional pipeline model as input for simulation analysis; simulating defects such as wrinkling, cracking, and elliptical distortion based on the initialized process parameters; adjusting the process parameters until the defects are resolved; and using the adjusted process parameters as preset parameters. The process parameters include: clamp pressure, pressure block pressure, auxiliary thrust, core ball diameter, core ball thickness, core ball distance, number of core balls, pipe clamp friction coefficient, pipe pressure block friction coefficient, pipe anti-wrinkle friction coefficient, and pipe core ball friction coefficient.

[0013] Further improvements to the above method involve updating and supplementing preset parameters based on test results to construct a sample. This involves removing unreasonable preset parameters from the pipe bending machine test, supplementing the new process parameters set in the test, and determining the springback amount when the pipe is bent to each pre-fabricated angle based on the new process parameters and reasonable preset parameters. The new process parameters, reasonable preset parameters, pre-fabricated angles, and springback amount are then used to construct the sample.

[0014] Further improvements to the above method involve constructing various types of parameter samples based on a sample database. These samples, representing different levels of defects, are generated through simulation and input from the database. Specifically, stress-strain samples, dynamic parameter samples, geometric curvature samples, and anti-wrinkle block position samples are obtained through motion simulation and pipe bending tests. Stress-strain samples include, but are not limited to, principal strain, secondary strain, and principal stress. Dynamic parameter samples include, but are not limited to, clamp pressure, clamping block pressure, and auxiliary thrust. Geometric curvature samples include, but are not limited to, bending direction, cross-section, and Gaussian curvature. Anti-wrinkle block position samples include, but are not limited to, anti-wrinkle block X-axis position, anti-wrinkle block Y-axis position, and clamping block pressure.

[0015] Based on further improvements to the above methods, the pattern recognition model employs one or more of Bayesian decision models, clustering models, and deep neural network models, including: predicting the pipeline category based on stress-strain samples, dynamic parameter samples, geometric curvature samples, and anti-wrinkle block position samples; predicting the rebound amount based on dynamic parameter samples; and the pipeline categories include: front wrinkle, back wrinkle, external crack, internal crack, elliptical distortion, and qualified.

[0016] Further improvements to the above method involve using a pattern recognition model to conduct a perception test on the pipe bending machine and extracting decision rules. These rules include: acquiring process data using data acquisition components installed on the pipe bending machine before the perception test; using the pattern recognition model to predict the springback amount based on the process data and compensating for the pipe parameters; during the perception test, measuring the actual springback amount when bending to the pre-designed angle according to the pipe parameters; if the difference between the actual springback amount and the predicted springback amount exceeds a threshold, adding the actual springback amount and process data to the sample database; otherwise, continuing the bending process; predicting the pipe category based on the acquired process data during the bending process; if the pipe category is not qualified, adjusting the process data until the pipe category is qualified, and continuing the bending process until each corner is bent to the design angle; and extracting decision rules based on the process parameter adjustment method.

[0017] Based on further improvements to the above method, the decision rules include: when the pipe type is front wrinkle, increase the clamp pressure and auxiliary thrust, and decrease the pressure block pressure; when the pipe type is back wrinkle and internal crack, increase the clamp pressure and pressure block pressure, and decrease the auxiliary thrust; when the pipe type is external crack, increase the auxiliary thrust, and decrease the clamp pressure and pressure block pressure; when the pipe type is elliptical distortion, increase the auxiliary thrust and decrease the pressure block pressure.

[0018] Based on a further improvement of the above method, the method also includes optimizing the process data of the pipe bending machine according to the collision detection results between the three-dimensional pipe model and the pipe bending machine. Specifically, before the collision detection, a pre-specified springback amount is obtained based on the material of the three-dimensional pipe model to compensate the pipe parameters.

[0019] Based on a further improvement of the above method, when the pipeline parameters are Cartesian coordinate XYZ data, the coordinates of the intersection of the pipeline axes and the center point of the pipeline end are compensated using the following formula:

[0020]

[0021] Where (X1, Y1, Z1) represents the coordinates of the starting point P1 of the pipeline axis, (X... t ,Y t Z t P represents the coordinates of the intersection of the pipeline axes and the center point of the pipeline's end. t , t≥2; ΔC s This indicates the amount of rebound during the review.

[0022] Based on further improvements to the above method, when the pipeline category is not qualified, reference data is predicted according to the decision rules, including:

[0023] Obtain process data for predicting pipeline categories. Adjust the process data according to the decision rules corresponding to the pipeline categories and the preset difference to obtain the test data. Input the test data into the pattern recognition model to predict the new pipeline category. If the new pipeline category is qualified, the test data is used as reference data. Otherwise, continue to adjust the process data according to the corresponding decision rules until the predicted new pipeline category is qualified.

[0024] Compared with the prior art, the present invention can achieve at least one of the following beneficial effects:

[0025] 1. By constructing a sample database and multiple pattern recognition models, and in conjunction with an improved pipe bending machine, defects such as wrinkles, cracks and elliptical distortions in multiple stages can be automatically identified. Based on decision rules, reference data can be predicted, which facilitates rapid parameter compensation during processing, outputs qualified pipes, reduces pipe scrap rate, and improves pipe accuracy and quality.

[0026] 2. This study explores and researches multi-dimensional intelligent analysis by using front and rear clamping block pressure, rotation, clamping die position, conduit diameter, conduit thickness, and bending radius as variables, supplemented by bending angle and mandrel position. Wrinkling, cracking, and springback of the pipeline are used as dependent variables. Through intelligent analysis methods, parameter accumulation and quality problem prediction are conducted, and parameter optimization methods are provided. This reduces the experience gap between pipeline development and first-time processing, helping to minimize the repeated problems, incomplete considerations, and chaotic quality adjustment directions caused by the intellectual labor involved in controlling wrinkling, cracking, and springback during production, thus significantly improving the automation and intelligence of pipeline forming. This helps ensure the progress of scientific research and production and reduces production costs.

[0027] In this invention, the above-described technical solutions can be combined with each other to achieve more preferred combinations. Other features and advantages of this invention will be set forth in the following description, and some advantages may become apparent from the description or be learned by practicing the invention. The objects and other advantages of this invention can be realized and obtained from what is particularly pointed out in the description and drawings. Attached Figure Description

[0028] The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Throughout the drawings, the same reference numerals denote the same parts.

[0029] Figure 1 This is a flowchart of a pipeline forming method based on pattern recognition in an embodiment of the present invention;

[0030] Figure 2 This is a schematic diagram of the circular grid on the surface of the pipe bending section in the pipe bending machine in an embodiment of the present invention;

[0031] Figure 3 This is a schematic diagram of the FLC curve of the pipe bending machine in an embodiment of the present invention;

[0032] Figure 4 This is a schematic diagram of data category division based on FLC curves in an embodiment of the present invention;

[0033] Figure 5 This is a schematic diagram of an existing pipe bending machine.

[0034] Figure 6 This is a schematic diagram of the pipe bending machine structure in an embodiment of the present invention;

[0035] Figure 7 This is a magnified schematic diagram of the internal laser head structure in an embodiment of the present invention;

[0036] Figure 8 This is a flowchart illustrating the pipeline forming process based on twin space in an embodiment of the present invention;

[0037] Figure 9 This is a schematic diagram of a pipeline forming method based on twin space in an embodiment of the present invention;

[0038] Figure label:

[0039] 1-Mandrel; 2-Bundle clamp; 3-Anti-wrinkle block; 4-Anti-wrinkle block bracket; 5-Bending wheel; 6-Equipment motor; 7-Chuck; 8-Chuck frame; 9-Pressure block frame; 10-Pressure block; 11-In-tube laser head; 12-Auxiliary push drive cylinder; 13-X-direction anti-wrinkle drive cylinder; 14-Y-direction anti-wrinkle drive cylinder; 15-Pipeline; 16-Indentation laser head; 17-Indentation laser head bracket; 18-Chuck drive cylinder; 19-First pressure block drive cylinder; 20-Second pressure block drive cylinder; 101-First ball head; 102-Second ball head. Detailed Implementation

[0040] Preferred embodiments of the present invention will now be described in detail with reference to the accompanying drawings, which form part of this application and are used together with the embodiments of the present invention to illustrate the principles of the present invention, but are not intended to limit the scope of the present invention.

[0041] A specific embodiment of the present invention discloses a pipeline forming method based on pattern recognition, such as... Figure 1 As shown, it includes the following steps:

[0042] S1. Perform simulation analysis on the three-dimensional pipeline model and output preset parameters to the pipe bending machine;

[0043] S2. Conduct preliminary tests on the pipe bending machine based on the pipe parameters and preset parameters of the pipe model. After updating and supplementing the preset parameters based on the test results, construct a sample and put it into the sample database.

[0044] S3. Based on the sample database, construct various types of parameter samples to train corresponding pattern recognition models to predict pipeline type and rebound amount;

[0045] S4. Install data acquisition components on the pipe bending machine, use the pattern recognition model to conduct a perception test on the pipe bending machine, and extract decision rules; update the sample database based on the perception data at each stage, retrain the pattern recognition model, and optimize the decision rules.

[0046] S5. During pipe processing, the process data acquired by the data acquisition component is input into the pattern recognition model to predict the pipe category. When the pipe category is not qualified, reference data is predicted according to the decision rules to compensate the pipe bending machine process data and complete the pipe forming.

[0047] To address defects generated during the bending process, this embodiment integrates simulation analysis, pipe bending tests, sampling perception, pattern recognition and judgment, decision rules, and parameter adjustments. This reduces the experience gap between the initial processing and the first processing of the pipeline, and helps to reduce the impact of repeated problems, incomplete considerations, and chaotic quality adjustment directions caused by the intellectual labor involved in controlling wrinkles, cracks, and springback in production on efficiency and effectiveness. It also greatly improves the automation and intelligence of pipeline forming.

[0048] In step S1, the three-dimensional pipeline model is simulated and analyzed, and preset parameters are output. This includes: using the basic information of the three-dimensional pipeline model as input for simulation analysis, simulating defects such as wrinkling, cracking, and elliptical distortion based on the initialized process parameters, and then adjusting the process parameters until the defects are resolved. The adjusted process parameters are used as preset parameters. The process parameters include: clamp pressure, pressure block pressure, auxiliary thrust, core ball diameter, core ball thickness, core ball distance, number of core balls, pipe clamp friction coefficient, pipe pressure block friction coefficient, pipe anti-wrinkle friction coefficient, and pipe core ball friction coefficient.

[0049] Specifically, the basic information of the designed three-dimensional pipeline model includes: bending radius r, pipeline diameter D, thickness t, surface curvature K, straight segment length Y, bending angle C, spatial rotation angle B, and material (titanium alloy, aluminum alloy, stainless steel), which serve as inputs for simulation analysis of the simplified pipe bending machine model.

[0050] First, initialize the process parameters, including: chuck pressure F1, pressure block pressure F2, auxiliary thrust F3; core ball diameter D1, core ball thickness H1, core ball distance H2; number of core balls N; tube chuck friction coefficient f1, tube pressure block friction coefficient f2, tube anti-wrinkle friction coefficient f3, and tube core ball friction coefficient f4.

[0051] For example, f1 = 0.5, f2 = 0.2, f3 = 0.2, f4 = 0.1; for pipelines made of aluminum alloy and stainless steel, H1 = 20, H2 = 2; for pipelines made of titanium alloy, H1 = 10, H2 = 1.

[0052] The initial values ​​of the clamp pressure F1, the block pressure F2, and the auxiliary thrust F3 are set using the following formulas:

[0053]

[0054] Among them, Y min This represents the minimum straight segment length in the pipeline; for example, it is set to 65. s This indicates the yield strength of the material.

[0055] The mandrel diameter is initially set using the following formula:

[0056] D1=D-2t-0.2n Formula (2)

[0057] Where n represents the number of simulations, initially n=1.

[0058] The mandrel structure is initialized based on the pipe diameter and bending angle using the following formula:

[0059]

[0060] Where N=0 indicates no mandrel is used; N=0.5 indicates a spherical mandrel is used without movable balls; N=1,2,3 indicates a chain ball mandrel is used, and N is the number of movable balls.

[0061] Based on the initial process parameters, when defects such as wrinkling, cracking, or elliptical distortion are detected in the simulation, the process parameters are adjusted until the defects are resolved. After multiple simulations, the process parameters for pipe forming are obtained and used as preset parameters.

[0062] Preferably, the difference is used to increase or decrease the value after rounding down to 0.1 times the preset value. When wrinkles appear, the chuck pressure F1, the block pressure F2, the auxiliary thrust F3, the core ball thickness H1, the core ball distance H2, and the core rod diameter D1 are increased. When cracks appear, the adjustment is performed in the opposite way to the wrinkles.

[0063] In step S2, preliminary tests, such as single-pipe bending tests, are conducted on the pipe bending machine based on the YBC pipe parameters and preset parameters of the pipe model. During the test, wrinkling and cracking are predicted according to the FLC curves (Forming Limit Curves) in the FLD diagram (Forming Limit Diagrams). Unreasonable preset parameters (unachievable preset parameters) are removed based on the actual situation, and new process parameters are set based on the actual situation or based on reasonable preset parameters to obtain pipe test results data for each stage that are smooth and without cracks, including: the springback amount ΔC when the pipe is bent to each pre-set angle.

[0064] New process parameters, reasonable preset parameters, prefabrication angles, and springback amounts are constructed as samples and placed into a sample database.

[0065] In step S3, constructing various types of parameter samples based on the sample database involves generating samples with different levels of defects through simulation based on the sample database.

[0066] Preferably, wrinkles are formed by simulating the positional movement of each finite element node, based on the pipeline stress state. A pipe bending test is conducted using curves and principal stress values ​​in the FLD diagram. The distance from any point to the boundary curve can be used as the criterion for determining the presence of cracks, thereby obtaining samples with different degrees of defects.

[0067] Furthermore, in simulations and experiments, various types of samples are collected and calculated for qualified and defective pipelines, corresponding to different pipeline categories. Among them, stress-strain samples include, but are not limited to: principal strain, secondary strain, and principal stress; dynamic parameter samples include, but are not limited to: clamp pressure, clamping block pressure, and auxiliary thrust; geometric curvature samples include, but are not limited to: bending direction, cross section, and Gaussian curvature; anti-wrinkle block position samples include, but are not limited to: anti-wrinkle block X-direction position, anti-wrinkle block Y-direction position, and clamping block pressure; pipeline categories include: front wrinkle, back wrinkle, external crack, internal crack, elliptical distortion, and qualified.

[0068] As shown in Tables 1-4, each type of sample contains various information, and different numbers are used to represent different pipeline categories.

[0069] Table 1 Sample of Stress-Strain Parameters

[0070] Serial Number <![CDATA[Principal strain ε1]]> <![CDATA[Secondary strain ε2]]> <![CDATA[Principal stress σ3]]> Piping categories 1 <![CDATA[ε 11 ]]> <![CDATA[ε 21 ]]> <![CDATA[σ 31 ]]> 1 2 <![CDATA[ε 12 ]]> <![CDATA[ε 22 ]]> <![CDATA[σ 32 ]]> 2 3 <![CDATA[ε 13 ]]> <![CDATA[ε 23 ]]> <![CDATA[σ 33 ]]> 3 4 <![CDATA[ε 14 ]]> <![CDATA[ε 24 ]]> <![CDATA[σ 34 ]]> 4 …… …… …… …… …… i <![CDATA[ε 1i ]]> <![CDATA[ε 2i ]]> <![CDATA[σ 3i ]]> 6

[0071] Table 2 Sample of Power Parameters

[0072] Serial Number <![CDATA[Chuck pressure F1]]> <![CDATA[Clamping block pressure F2]]> <![CDATA[Auxiliary pushing force F3]]> Piping categories 1 F11 F21 F31 1 2 F12 F22 F32 2 3 F13 F23 F33 3 4 F14 F24 F34 4 …… …… …… …… …… j F1j F2j F3j 6

[0073] Table 3 Geometric Curvature Samples

[0074] Serial Number Bending towards k1 Section K2 Gaussian curvature K3 Piping categories 1 K11 K21 K31 1 2 K12 K22 K32 2 3 K13 K23 K33 3 4 K14 K24 K34 4 …… …… …… …… …… p K1p K2p K3p 6

[0075] Table 4 Sample of anti-wrinkle block locations

[0076] Serial Number X-direction position X1 Y-direction position Y2 Clamping pressure F2 Piping categories 1 X11 Y21 F21 1 2 X12 Y22 F22 2 3 X13 Y23 F23 3 4 X14 Y24 F24 4 …… …… …… …… …… q X1q Y2q F2q 6

[0077] Specifically, stress-strain samples are obtained through the following steps:

[0078] The surface of the bent portion of the pipe in the pipe bending machine is marked with a circular grid with a diameter d0 of 0.2–1.5 mm using laser marking, chemical etching, or electro-etching. Figure 2 As shown; defects are generated through pipe bending tests, and image processing and analysis are performed using a grid strain automatic measurement system to obtain the FLC curve in the FLD diagram; based on the FLC curve, the principal strain, secondary strain, and principal stress data and their corresponding categories are obtained, and a stress-strain sample is constructed. For example, as... Figure 3 and Figure 4 As shown.

[0079] In existing technologies, the forming limit curve FLC in the FLD diagram is based on cupping, hole enlargement, and other tests performed on sheet metal. It forms a strip-shaped region or curve based on the local instability limit strain and engineering strain or actual strain under different strain paths, reflecting the local forming limit of the sheet metal under unidirectional and bidirectional tensile stress. In contrast, this embodiment obtains the limit strain force when different problems occur in the pipeline through pipe bending tests, facilitating the differentiation of pipeline types.

[0080] The dynamic parameter samples and anti-wrinkle block position samples were obtained by installing a data acquisition component on the pipe bending machine. The geometric curvature sample was calculated using the curvature formula based on the surface dimensions of the bending part of the pipe bending machine.

[0081] Based on a sample database, multiple pattern recognition models are trained. The pattern recognition models employ one or more of Bayesian decision models, clustering models, and deep neural network models to predict pipeline type and rebound amount.

[0082] A pattern recognition model is trained for each type of parameter sample. At the same time, features of different dimensions from different types of parameter samples are aggregated through a fully connected linear layer neural network and then fused to train a pattern recognition model, which is suitable for scenarios that comprehensively consider various types of data.

[0083] Preferably, a corresponding pattern recognition model is trained based on stress-strain samples, dynamic parameter samples, geometric curvature samples, and anti-wrinkle block position samples in the sample database to predict the pipeline category; a pattern recognition model is trained based on the dynamic parameter samples to predict the springback level, with each springback level corresponding to a springback amount, thereby obtaining the springback amount; a fusion model recognition model is trained based on the dynamic parameter samples, geometric curvature samples, and anti-wrinkle block position samples to predict the pipeline category.

[0084] In step S4, the pipe bending machine is improved by conducting a deeper perception test. The perception test involves inputting the pipe parameters of the pipe model into the pipe bending machine and performing pipe bending processing and adjustment based on the automatically collected process data pattern recognition model.

[0085] It should be noted that, Figure 5 It is an existing pipe bending machine, including: mandrel 1, bundle clamp 2, anti-wrinkle block 3, 4 is anti-wrinkle block support, bending wheel 5, equipment motor 6, chuck 7, chuck frame 8, pressure block frame 9, pressure block 10, first ball head 101, second ball head 102 and pipe 15.

[0086] In this embodiment, in order to acquire and monitor process data in real time, Figure 5 The data acquisition component, displacement control component, and indentation monitoring component were installed on the basis of the pipe bending machine, such as... Figure 6 As shown.

[0087] Specifically, the data acquisition components include: an in-tube laser head 11, a clamp drive cylinder 18, a first pressure block drive cylinder 19, a second pressure block drive cylinder 20, and an auxiliary push drive cylinder 12.

[0088] The laser head 11 inside the tube is connected to the front end of the outermost ball head (second ball head 102) on the mandrel via a ball hinge. In conjunction with the core feeding and pulling action, it detects the surrounding working surface, obtains the inner diameter and inner surface dimensions of the tube, and determines the position of the anti-wrinkle block by measuring its front end. For example... Figure 7 The enlarged view of the laser head 11 inside the tube shows that it is formed by embedding 8 optical fibers into the mechanical base in eight directions to form an 8-directional four-dimensional probe, which can measure the surrounding area. At each point, it can measure the position of a point on a curved surface with a diameter of 0.2 to 1 mm.

[0089] It should be noted that data measurement includes, but is not limited to, laser measurement, and can employ mechanical or brush-type sensors for measuring electrical signals.

[0090] The chuck drive cylinder 18 is mounted on the chuck frame 8, the first pressure block drive cylinder 19 and the second pressure block drive cylinder 20 are mounted on the Y-axis of the pressure block frame 9, and the auxiliary push drive cylinder 12 is mounted on the X-axis of the pressure block frame 9, or it can be mounted on the jaws of the tube clamp 2. The drive cylinders are equipped with pressure sensors, which may be located inside, at the front and rear ends of the drive cylinders, or inside, outside, and at contact points of the chuck 7, chuck frame 8, and tubing 15. During the movement of the drive cylinders, the pressure sensors acquire the chuck pressure F1, pressure block pressure F2, and auxiliary push force F3.

[0091] It should be noted that in commonly used equipment, the chuck pressure and drive stroke are fixed; only the chuck position can be manually fine-tuned. Therefore, fluctuations in pressure often lead to relative slippage between the tubing and the chuck. Often, even after manual parameter adjustments, wrinkles and cracks appear in the tubing during automatic bending, resulting in wasted time, effort, and materials, and hindering the adjustment of forming parameters. Therefore, this invention adds a pressure sensor to the chuck frame 8. A pressure sensor is installed at the force transmission end of the hydraulic or motor system to collect the actual pressure value during bending. Based on the collected process data, samples are constructed, patterns are recognized, and decisions are made.

[0092] Specifically, the displacement control component includes an X-axis anti-wrinkle drive cylinder 13 and a Y-axis anti-wrinkle drive cylinder 14 mounted on the anti-wrinkle block bracket 4. During pipeline processing, the X-axis anti-wrinkle drive cylinder 13 and the Y-axis anti-wrinkle drive cylinder 14 are moved to compensate for displacement based on the deviation between the anti-wrinkle block position measured by the laser head 11 inside the pipe and the reference data provided by the twin space.

[0093] Specifically, the indentation monitoring component includes an indentation laser head bracket 17 and an indentation laser head 16. The indentation laser head bracket 17 is fixed to the clamp 7, and the indentation laser head 16 is installed and fixed through the holes in the indentation laser head bracket. It measures the position of the pipeline at a distance of 0.1 mm from the front end of the clamp 7. The structure is the same as that of the laser head 11 inside the pipeline, with eight optical fibers embedded in the mechanical base in eight directions to form an 8-directional four-dimensional probe. When the position deviation is greater than the pipeline threshold, it is determined that an indentation has occurred, and the monitoring terminal issues an alarm. For example, the pipeline threshold is 0.1 times the pipeline wall thickness, and the multiple is set based on the results of multiple tests.

[0094] It should be noted that the diameters of the first ball head 101 and the second ball head 102 are less than or equal to the inner diameter of the pipe minus 0.3 mm and greater than or equal to the inner diameter of the pipe minus 0.7 mm. In existing equipment, when multiple ball heads are used, the diameters of the ball heads are the same. In this embodiment, the ball heads are designed with progressively smaller diameters, and the diameter of the second ball head 102 is smaller than that of the first ball head 101. For example, the diameter of the first ball head 101 is designed to be 30.7 mm and the diameter of the second ball head 102 is designed to be 30.5 mm. Furthermore, if more than two ball heads are used, a stepped design is selected. For example, three ball heads are respectively one 30.7 mm ball head, one 30.5 mm ball head, and one 30.3 mm ball head, and six ball heads are respectively two 30.7 mm ball heads, two 30.5 mm ball heads, and two 30.3 mm ball heads.

[0095] It should be noted that the process data collected by the in-tube laser head 11, the indentation laser head 16, and various pressure sensors are transmitted to the memory and placed into the monitoring database via cable connection or other wired or wireless transmission methods.

[0096] It is worth noting that the method in this embodiment also includes: optimizing the process data of the pipe bending machine based on the collision detection results between the three-dimensional pipeline model and the pipe bending machine.

[0097] In existing technologies, operators first rely on experience to determine which pipe models should be checked for software interference. Then, they use the pipe bending machine's built-in software or 3D motion simulation software to perform interference checks and provide feedback to the designers regarding the pipe models exhibiting interference. When there are many pipes and the operators lack experience, it is necessary to simulate all pipe models, which is time-consuming. Furthermore, existing software interference detection does not consider springback detection, resulting in low accuracy.

[0098] This embodiment pre-specifies the springback amount (i.e., the springback amount used to check for interference) for each type of pipeline based on the pipeline material. Based on the Cartesian coordinate XYZ data of the pipeline model, the coordinates of the pipeline's axis intersection and end center point are compensated for the springback amount. A two-dimensional interference analysis method for pipeline bending, disclosed in patent CN202211714754.4, is used. This includes: based on the pipeline XYZ data compensated for the springback amount, simplifying straight segments and arc segments into two-dimensional spatial distance curves and two-dimensional spatial angular distance curves; simplifying the bending wheel of the pipe bending machine and the pipeline into two-dimensional obstacle curves; determining whether the two-dimensional spatial distance curves and two-dimensional spatial angular distance curves interfere with the two-dimensional obstacle curves; if interference exists, it is considered a pipeline model to be confirmed; after compensating for the springback amount in the YBC pipeline data of the pipeline model to be confirmed, three-dimensional interference detection is performed using three-dimensional motion simulation software, and the results are compared with the two-dimensional analysis results to demonstrate the credibility of the interference position, thereby improving detection efficiency and accuracy.

[0099] Specifically, the rebound amount is checked by compensating for the coordinates of the intersection of the pipeline axes and the center point of the pipeline end using the following formula:

[0100]

[0101] Where (X1, Y1, Z1) represents the coordinates of the starting point P1 of the pipeline axis, (X... t ,Y t Z t P represents the coordinates of the intersection of the pipeline axes and the center point of the pipeline's end. t , t≥2; ΔC s This indicates the amount of rebound during the review.

[0102] Pipeline models exhibiting interference are fed back to designers for process optimization, including: optimizing pipe bending machine parameters, changing the bending start point, and replacing the pipe bending machine. Once it is confirmed that there is no interference between the pipeline models and the pipe bending machine, a perception test of pipeline processing is conducted. The perception test involves inputting the pipeline parameters from the pipeline model into the pipe bending machine and using a pattern recognition model in twin space to perform actual pipe bending processing and process parameter adjustments.

[0103] Furthermore, based on the improved and optimized process data of the pipe bending machine, a pattern recognition model was used to conduct a perception experiment on the pipe bending machine to extract decision rules, including:

[0104] ① Before the sensing test, process data is obtained from the data acquisition components installed on the pipe bending machine. Based on the process data, the rebound amount is predicted using a pattern recognition model to compensate for the pipeline parameters.

[0105] Specifically, the clamp pressure, clamp block pressure, and auxiliary thrust are obtained from the process data to construct a dynamic parameter sample, and the springback amount is predicted. The pipeline parameters are then compensated based on the predicted springback amount. For example, when the pipeline parameters use CNC machining coordinates YBC, the C-axis is compensated for the predicted springback amount (the value can be positive or negative).

[0106] ② In the perception test, when the pipeline is bent to the pre-designed angle according to the pipeline parameters, the actual rebound amount is measured. If the difference between the actual rebound amount and the predicted rebound amount exceeds the difference threshold, the actual rebound amount and process data are added to the sample database; otherwise, the bending process continues. During the bending process, the pipeline category is predicted based on the collected process data. If the pipeline category is not qualified, the process data is adjusted until the pipeline category is qualified, and the bending process continues until each corner is bent to the design angle. Decision rules are extracted based on the adjustment method of the process parameters.

[0107] It should be noted that the prefabricated angle is the angle set according to a preset variational ratio before the design angle is reached. For example, if the design angle is 90 degrees and the variational ratio is 1 / 2, then the prefabricated angle is 45 degrees. This embodiment also considers the springback amount and will compensate for the prefabricated angle based on the predicted springback amount. For example, if the springback amount is 1 degree, then the prefabricated angle is 46 degrees.

[0108] When bending to the pre-set angle, the actual springback is large. Therefore, the process parameters are compensated to control bending accuracy by increasing the clamp pressure and pressure block pressure while reducing the auxiliary thrust, in accordance with the rule of increasing the clamp pressure and pressure block pressure and reducing the auxiliary thrust. In existing technologies, bending is performed step-by-step after bending is complete, and the processed pipe is then transferred to a straightening or scrapping process, making further processing on the pipe bending machine impossible.

[0109] After multiple trials and verifications of the process data adjustment method, decision rules were extracted to ensure that the pipeline category becomes qualified when predicted again, thus providing accurate reference data in the actual pipe bending process and improving the pipe bending accuracy.

[0110] Preferably, the decision rules extracted in this embodiment include: when the pipeline type is front wrinkle, increase the clamp pressure, auxiliary thrust, and pipe clamp friction coefficient, and decrease the pressure block pressure; when the pipeline type is back wrinkle and internal crack, increase the clamp pressure and pressure block pressure, and decrease the auxiliary thrust; when the pipeline type is external crack, increase the auxiliary thrust and decrease the clamp pressure and pressure block pressure; when the pipeline type is elliptical distortion, increase the auxiliary thrust and decrease the pressure block pressure.

[0111] As the sensing test proceeds, the sample database is continuously updated, including: analyzing the sample data in the sample database based on the sensing test data, updating sample data with the same parameters but inconsistent pipeline types, and removing sample data corresponding to parameters that cannot be achieved in the sensing test; and supplementing the sample database with sensing test data from each processing stage.

[0112] After the sample database is updated, the pattern recognition model is continuously trained, and the predicted pipe category becomes more and more accurate. The difference between the predicted rebound amount and the actual rebound amount also becomes smaller and smaller. Then, during the pipe processing in step S5, the process data obtained by the data acquisition component is input into the pattern recognition model to predict the pipe category. When the pipe category is not qualified, reference data is predicted according to the decision rules to compensate the pipe bending machine process data.

[0113] Specifically, the process data for predicting pipeline categories is obtained. Based on the decision rules corresponding to the pipeline categories, the process data is adjusted according to a preset difference to obtain the test data. The test data is then input into the pattern recognition model to predict new pipeline categories. If the new pipeline category is qualified, the test data is used as reference data. Otherwise, the process data is adjusted according to the corresponding decision rules until the predicted new pipeline category is qualified.

[0114] Preferably, if the new pipeline category is still not qualified after reaching the maximum number of automatic predictions, manually input process data is received and predictions are made until the pipeline category is qualified, thus obtaining reference data. In other words, this embodiment provides the function of manually inputting process data, and the manually input process data and actual pipe bending conditions can also be added to the sample database.

[0115] Based on the improved pipe bending machine, in another embodiment, such as Figure 8 and Figure 9 As shown, a twin space is constructed to store and analyze data, and to train pattern recognition and decision-making rules, obtaining various pattern recognition models for prediction. In the pre-, during, and post-pipe processing stages, based on the twin space and in conjunction with an improved pipe bending machine, the system automatically identifies the presence and specific categories of defects based on collected process data. It also provides reference compensation data for the decision-making methods adopted for each defect, thus replacing existing manual experience and continuous trial and error. The twin space can be deployed on a monitoring terminal or a cloud server.

[0116] Specifically, the tubing forming method based on twin space:

[0117] (1) Obtain the three-dimensional pipeline model of the design.

[0118] (2) ① Extract basic information such as the thickness t, surface curvature K, straight segment length Y, bending angle C, spatial rotation angle B, and material of the pipeline model, and input it into the simulation analysis as analysis input to obtain the pipeline preset parameters; ② Based on the pipeline preset parameters, initially preset the YBC pipeline parameters for the pipe bending machine, further analyze the wrinkle and crack rules and limits, and input them into the twin space. Combined with collision detection, optimize the process parameters of the pipe bending machine.

[0119] (3) ① Design multiple pattern recognition models to perform overall analysis and simulation of the pipe bending process, parameters of the pipe bending forming equipment, experimental and perception information, and pattern decision samples. Establish multi-dimensional and multi-level analysis collaboration to compare each stage, and store the experience data, samples, decisions and results of each stage; ② Based on the three-dimensional pipeline model, transmit the pipeline parameters to the pipe bending machine as experimental input. Based on the twin space sample database and pattern recognition model, conduct pipeline perception experiments, verify and concretize the twin space model, update the sample database, optimize process parameters, and finally obtain reasonable samples that are close to the actual working conditions.

[0120] (4) Based on the updated sample database, the pattern recognition model is retrained, the decision rules are optimized, and decisions are made on the parameters of the pipe bending machine.

[0121] (5) Based on the parameters after the decision, the pipeline to be formed is processed and detected in real time. When the pipeline has defects, the reference data is predicted according to the twin space, and the process parameters of the pipe bending machine are adjusted. After completion, the pipeline is inspected, tested, and virtually assembled. The compensation parameters are optimized and mechanical compensation is eliminated in subsequent processing to achieve precision manufacturing.

[0122] (6) Delivery of pipelines.

[0123] Compared with existing technologies, this embodiment provides a pattern recognition-based pipe forming method. By constructing a sample database and multiple pattern recognition models, and in conjunction with an improved pipe bending machine, it automatically identifies defects such as wrinkles, cracks, and elliptical distortions at multiple stages. Based on decision rules, it predicts reference data, facilitating rapid parameter compensation during processing, outputting qualified pipes, reducing pipe scrap rates, and improving pipe accuracy and quality. It explores and studies multi-dimensional intelligent analysis by using front and rear clamping block pressure, rotation, clamping die position, conduit diameter, conduit thickness, and bending radius as variables, supplemented by bending angle and mandrel position. Furthermore, it uses pipe wrinkling, cracking, and springback as dependent variables, employing intelligent analysis methods to accumulate parameters, predict quality problems, and provide parameter optimization methods. This reduces the experience gap between pipe formation and first-time processing, helping to minimize the repeated problems, incomplete considerations, and chaotic quality adjustment directions caused by the intellectual labor involved in controlling wrinkling, cracking, and springback during production, thus significantly improving the automation and intelligence of pipe forming, helping to ensure research and production progress, and reducing production costs.

[0124] Those skilled in the art will understand that all or part of the processes of the methods described in the above embodiments can be implemented by a computer program instructing related hardware, and the program can be stored in a computer-readable storage medium. The computer-readable storage medium may be a disk, optical disk, read-only memory, or random access memory, etc.

[0125] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A pipeline forming method based on pattern recognition, characterized in that, Includes the following steps: Simulation analysis of the three-dimensional pipeline model is performed, and preset parameters are output to the pipe bending machine; A preliminary test of the pipe bending machine was conducted based on the pipe parameters and preset parameters of the pipe model. After updating and supplementing the preset parameters based on the test results, a sample was constructed and placed into the sample database. Based on the sample database, various types of parameter samples are constructed to train corresponding pattern recognition models to predict pipeline type and rebound amount; A data acquisition component is installed on the pipe bending machine. A pattern recognition model is used to conduct a perception experiment on the pipe bending machine and extract decision rules. The sample database is updated based on the perception data at each stage, and the pattern recognition model is retrained to optimize the decision rules. During pipe processing, the process data acquired by the data acquisition component is fed into the pattern recognition model to predict the pipe category. When the pipe category is not qualified, reference data is predicted according to the decision rules to compensate the pipe bending machine process data and complete the pipe forming. Based on the collision detection results between the 3D pipeline model and the pipe bending machine, the process data of the pipe bending machine is optimized. Specifically, before the collision detection, the pre-specified springback amount is obtained based on the material of the 3D pipeline model, and the pipeline parameters are compensated accordingly. When the pipeline parameters are Cartesian coordinate XYZ data, the coordinates of the intersection of the pipeline axes and the center point of the pipeline end are compensated using the following formula: in, Indicates the starting point of the pipeline axis coordinate, Indicates the coordinates of the intersection of the pipeline axes and the center point of the pipeline end. , ; This indicates the amount of rebound during the review.

2. The pipeline forming method based on pattern recognition according to claim 1, characterized in that, The simulation analysis of the three-dimensional pipeline model and the output of preset parameters include: using the basic information of the three-dimensional pipeline model as input for simulation analysis, simulating defects such as wrinkling, cracking and elliptical distortion according to the initialized process parameters, and then adjusting the process parameters until the defects are resolved. The adjusted process parameters are used as preset parameters. The process parameters include: clamp pressure, pressure block pressure, auxiliary thrust, core ball diameter, core ball thickness, core ball distance, number of core balls, pipe clamp friction coefficient, pipe pressure block friction coefficient, pipe anti-wrinkle friction coefficient and pipe core ball friction coefficient.

3. The pipeline forming method based on pattern recognition according to claim 2, characterized in that, The process of updating and supplementing preset parameters based on test results to construct a sample involves removing unreasonable preset parameters from the pipe bending machine test, supplementing the new process parameters set in the test, and determining the springback amount when the pipe is bent to each pre-fabricated angle based on the new process parameters and reasonable preset parameters. The new process parameters, reasonable preset parameters, pre-fabricated angles, and springback amount are then used to construct the sample.

4. The pipeline forming method based on pattern recognition according to claim 3, characterized in that, The construction of various types of parameter samples based on the sample database involves generating samples with different defect levels through simulation supplementation based on the sample database. Specifically, stress-strain samples, dynamic parameter samples, geometric curvature samples, and anti-wrinkle block position samples are obtained through motion simulation and pipe bending tests. The stress-strain samples include, but are not limited to, principal strain, secondary strain, and principal stress. The dynamic parameter samples include, but are not limited to, clamp pressure, pressure block pressure, and auxiliary thrust. The geometric curvature samples include, but are not limited to, bending direction, cross-section, and Gaussian curvature. The anti-wrinkle block position samples include, but are not limited to, anti-wrinkle block X-axis position, anti-wrinkle block Y-axis position, and pressure block pressure.

5. The pipeline forming method based on pattern recognition according to claim 4, characterized in that, The pattern recognition model employs one or more of Bayesian decision models, clustering models, and deep neural network models, including: predicting the pipeline category based on stress-strain samples, dynamic parameter samples, geometric curvature samples, and anti-wrinkle block position samples; and predicting the rebound amount based on dynamic parameter samples. The pipeline categories include: front wrinkle, back wrinkle, external crack, internal crack, elliptical distortion, and qualified.

6. The pipeline forming method based on pattern recognition according to claim 1 or 5, characterized in that, The process of using a pattern recognition model to conduct a perception test on the pipe bending machine and extract decision rules includes: acquiring process data based on the data acquisition components installed on the pipe bending machine before the perception test; using the pattern recognition model to predict the springback amount based on the process data and compensating for the pipe parameters; during the perception test, when the pipe is bent to the pre-designed angle according to the pipe parameters, measuring the actual springback amount; if the difference between the actual springback amount and the predicted springback amount exceeds the difference threshold, then the actual springback amount and process data are added to the sample database; otherwise, the pipe bending continues; during the bending process, the pipe category is predicted based on the acquired process data; if the pipe category is not qualified, the process data is adjusted until the pipe category is qualified, and the pipe bending continues until each corner is bent to the design angle; and decision rules are extracted based on the adjustment method of the process parameters.

7. The pipeline forming method based on pattern recognition according to claim 6, characterized in that, The decision rules include: when the pipe type is front wrinkle, increase the clamp pressure and auxiliary thrust, and decrease the pressure block pressure; when the pipe type is back wrinkle and internal crack, increase the clamp pressure and pressure block pressure, and decrease the auxiliary thrust; when the pipe type is external crack, increase the auxiliary thrust, and decrease the clamp pressure and pressure block pressure; when the pipe type is elliptical distortion, increase the auxiliary thrust and decrease the pressure block pressure.

8. The pipeline forming method based on pattern recognition according to claim 7, characterized in that, When the pipeline category is not qualified, reference data is predicted according to the decision rules, including: Obtain process data for predicting pipeline categories. Adjust the process data according to the decision rules corresponding to the pipeline categories and a preset difference to obtain test data. Input the test data into the pattern recognition model to predict new pipeline categories. If the new pipeline category is qualified, the test data is used as reference data. Otherwise, continue to adjust the process data according to the corresponding decision rules until the predicted new pipeline category is qualified.

Images