An Online Monitoring Method and Tool for the Forming Quality of Stamped Workpieces

Abstract

The present invention belongs to the field of stamping forming process monitoring, and specifically relates to an online monitoring method and tool for the forming quality of stamping workpieces, as well as a corresponding visualization method for the forming quality and its stamping equipment. The online monitoring method includes the following steps: S01: Obtain the process energy map of the current stamping workpiece; S02: Real-time collect the equipment parameters of the stamping workpiece during the processing; S03: Calculate the real-time process energy according to the real-time stamping depth and real-time punching force collected during the processing; S04: Fit a corresponding state trajectory in the process energy map according to the state changes of the real-time stamping depth and real-time process energy; S05: Judge whether the state trajectory passes through the fracture zone and whether it terminates in the wrinkling zone, and then obtain the evaluation result of the forming quality. The online monitoring tool is the corresponding data processing equipment. The present invention solves the problem that the prior art lacks an online evaluation means for the forming quality of stamping workpieces.

Description

Technical Field

[0001] The present invention belongs to the field of stamping forming process monitoring, and particularly relates to an on-line monitoring method and tool for the forming quality of stamping workpieces, as well as a visualization method for the forming quality of corresponding stamping workpieces and a stamping device therefor. Background Art

[0002] In the stamping processing technology, how to monitor the forming quality during the processing has always been a difficult problem faced by those skilled in the art. Due to the closed nature of the stamping die, the stamping workpiece processing is usually completed quickly in a closed processing environment; it is technically difficult to directly measure the forming quality of the stamping workpiece, and it is even more impossible to on-line monitor the forming quality of the stamping workpiece.

[0003] In the existing stamping processing technological process, in order to ensure the production quality, it is often necessary for the front-end production department to process samples according to a pre-established production plan. Then the samples are sent to the back-end QA department, and quality inspectors detect defects in the forming quality of the processed stamping workpiece samples. Then, the inspection results of the QA department are fed back to the front-end production department, and the production department decides whether to adjust the production process according to the quality inspection results.

[0004] This process of manual quality inspection, feedback, and modification of process parameters is long, inefficient, time-consuming and laborious, thus reducing the production efficiency and the yield rate of the enterprise, and increasing the production cost of the workpiece. Therefore, how to design a scheme at the production end that can on-line monitor the forming quality of stamping workpieces is becoming a technical problem urgently to be solved by those skilled in the art. Summary of the Invention

[0005] In order to solve the problem that the existing technology lacks means for on-line evaluating the forming quality of stamping workpieces, the present invention provides an on-line monitoring method and tool for the forming quality of stamping workpieces, as well as a visualization method for the forming quality of corresponding stamping workpieces and a stamping device therefor.

[0006] The present invention is implemented by adopting the following technical solutions:

[0007] An on-line monitoring method for the forming quality of a stamping workpiece, comprising the following steps:

[0008] S01: Obtain a process energy map of the current stamping workpiece under specified process conditions.

[0009] The process energy map in the present invention is a pre-generated map with the stamping depth h p and the process energy E as the abscissa and ordinate respectively, and the maximum thickening rate Δ p and the maximum thinning rate Δ MTC under any stamping depth h MTNIs the state attribute of the coordinate point, and divides the wrinkle area and the rupture area range according to the state attribute of the coordinate point.

[0010] S02: Real-time collect the equipment parameters during the processing of the stamping workpiece, including the stamping depth h p and the stamping force F.

[0011] S03: According to the real-time stamping depth h collected during the processing real and the real-time stamping force F real Calculate the real-time process energy E real :

[0012]

[0013] S04: According to the state changes of the real-time stamping depth h real and the real-time process energy E real during the current stamping workpiece processing, fit a corresponding state trajectory in the process energy map.

[0014] S05: Make the following judgments according to the position of the state trajectory during the current stamping workpiece processing in the process energy map:

[0015] (i) When the state trajectory neither passes through the rupture area nor terminates in the wrinkling area, it is determined that the current stamping workpiece is completely qualified.

[0016] (ii) When the state trajectory only passes through the rupture area, it is determined that the current stamping workpiece has ruptured.

[0017] (iii) When the state trajectory only terminates in the wrinkling area, it is determined that the current stamping workpiece has local wrinkling.

[0018] (iv) When the state trajectory passes through the rupture area and terminates in the wrinkling area, it is determined that the current stamping workpiece has ruptured and has local wrinkling.

[0019] The present invention also includes an on-line monitoring tool for the forming quality of stamping workpieces, which uses the on-line monitoring method for the forming quality of stamping workpieces as described above to monitor the processing process of the target stamping workpiece on the stamping equipment and generate a quality evaluation result of the processed target stamping workpiece. The on-line monitoring tool includes: a reference database, a data acquisition unit, a trajectory generation unit, and a quality determination unit.

[0020] The reference database stores the process energy maps of each stamping workpiece under specified process conditions.

[0021] The data acquisition unit is used to real-time collect the stamping depth h p and the stamping force F during the processing of the stamping equipment, and according to the real-time stamping depth h collected during the processing real and the real-time stamping force Freal Calculate the real-time process energy E real .

[0022] The trajectory generation unit is used to query the reference database to obtain the process energy map corresponding to the current stamping workpiece under the current process conditions, and then according to the real-time stamping depth h real and the real-time process energy E real during the processing of the current stamping workpiece, fit a corresponding state trajectory in the process energy map according to the state changes of

[0023] The quality determination unit is used to generate a corresponding quality evaluation result according to the position of the state trajectory in the process energy map: (i) When the state trajectory neither passes through the fracture zone nor terminates in the wrinkling zone, it is determined that the current stamping workpiece is completely qualified. (ii) When the state trajectory only passes through the fracture zone, it is determined that the current stamping workpiece has cracked. (iii) When the state trajectory only terminates in the wrinkling zone, it is determined that the current stamping workpiece has local wrinkling. (iv) When the state trajectory passes through the fracture zone and terminates in the wrinkling zone, it is determined that the current stamping workpiece has cracked and has local wrinkling.

[0024] The present invention also includes a visualization method for the forming quality of a stamping workpiece, which is used to visually present the change process of the forming quality of the stamping workpiece during the processing of the target stamping workpiece in combination with the aforementioned online monitoring method for the forming quality of the stamping workpiece:

[0025] I. Before processing the stamping workpiece:

[0026] Query the process energy map corresponding to the current stamping workpiece according to the preset equipment processing parameters.

[0027] II. During the processing of the stamping workpiece:

[0028] (1) Present the first screen and the second screen according to the process energy map.

[0029] Among them, the first screen uses the stamping depth h p as the abscissa and the process energy E as the ordinate; and represents the value of the corresponding maximum thickening rate Δ MTC through the hue difference of the pixel points within the constraint area. The boundary of the wrinkling zone is also divided in the first screen.

[0030] Among them, the second screen uses the stamping depth h p as the abscissa and the process energy E as the ordinate; and represents the value of the corresponding maximum thinning rate Δ MTN through the hue difference of the pixel points within the constraint area. The boundary of the fracture zone is also divided in the second screen.

[0031] (2) Determine the real-time stamping depth h corresponding to the processing process according to the feedback signal of the stamping equipment collected in real timereal and the real-time process energy E real ; this is used as the stamping state variable U.

[0032] (3) In the first and second screens, the position change of the stamping state variable U is displayed in real time, and a motion trajectory Ut is fitted.

[0033] III. After the stamping workpiece is processed:

[0034] Using any one or more interaction means, the evaluation result of the forming quality of the current stamping workpiece is presented to the user.

[0035] A stamping device includes a device body and an interaction component. The stamping device also includes an online monitoring tool for the forming quality of the stamping workpiece as described above. The online monitoring tool is used to generate a corresponding quality evaluation result during the stamping process of any stamping workpiece according to the operation parameters of the device body collected in real time.

[0036] The interaction component is used to present the forming quality of the stamping workpiece to the user in real time according to the visualization method of the forming quality of the stamping workpiece as described above.

[0037] The technical solution provided by the present invention has the following beneficial effects:

[0038] Based on the constructed process energy map, the present invention designs a method and tool for online monitoring of the processing process of any stamping workpiece in a stamping device. Using the solution designed by the present invention, the state trajectory of the processing process on the process energy map can be determined by real-time collection of the punching force and punching depth of the stamping device, and then whether the stamping workpiece has wrinkling and cracking defects can be determined according to the position relationship between the state trajectory and the cracking area and wrinkling area in the process energy map, and further the forming quality of the processed stamping workpiece can be evaluated.

[0039] The online monitoring method for the forming quality of the stamping workpiece provided by the present invention realizes the effect of synchronously evaluating the forming quality during the processing process. Furthermore, it can greatly shorten the quality inspection process in the large-scale stamping process, improve the work efficiency in the batch processing process of stamping workpieces, and reduce the work load and cost of product quality inspection. It has extremely high practical value and economic value and is suitable for popularization and application in the industry. Description of the Drawings

[0040] Figure 1 It is a flowchart of the steps of a method for generating a process energy map provided in Embodiment 1 of the present invention.

[0041] Figure 2 It is an architecture diagram of a prediction network based on the Stacking integrated learning framework built in Embodiment 1 of the present invention.

[0042] Figure 3 Schematic flow chart of the construction and training of the prediction network in Embodiment 1 of the present invention.

[0043] Figure 4 Block diagram of a system for generating a process energy map for monitoring the forming quality of stamping workpieces provided in Embodiment 2 of the present invention.

[0044] Figure 5 Step flow chart of an on - line monitoring method for the forming quality of stamping workpieces provided in Embodiment 4 of the present invention.

[0045] Figure 6 Schematic diagram of a principle of an on - line monitoring tool for the forming quality of stamping workpieces provided in Embodiment 5 of the present invention.

[0046] Figure 7 Sample picture of a door inner panel - like part selected in the performance test stage.

[0047] Figure 8 Data density scatter plot of the predicted results of the thickness change of each alternative model during the performance test. Among them, Figure 8 Part (a) represents the predicted results and errors of the maximum thinning rate Δ MTN ; Figure 8 Part (b) represents the predicted results and errors of the maximum thickening rate Δ MTC . The color bars in the figure represent the density of the data; the dashed lines in the figure are called zero - error lines indicating that the measurement results coincide with the predicted results.

[0048] Figure 9 Pearson correlation coefficient distribution plot of the predicted results of each alternative model during the performance test.

[0049] Figure 10 Data density scatter plot of the predicted results of the thickness change of each base model during the performance test. Among them, Figure 10 Part (a) represents the predicted results and errors of the maximum thinning rate Δ MTN ; Figure 10 Part (b) represents the predicted results and errors of the maximum thickening rate Δ MTC .

[0050] Figure 11 Process energy map constructed by using the solution of the present invention during the performance test.

[0051] Figure 12 Monitoring results of the thickness change of the training set at different stamping depths by process energy maps constructed by different solutions during the performance test; among them, Figure 12 Part (a) represents the monitoring results of the thickness change of the stamping workpiece at a stamping depth of 5 mm; Figure 12Part (b) in the figure shows the monitoring results of the thickness change of the stamping workpiece at a stamping depth of 23 mm.

[0052] Figure 13 During the performance test, the monitoring results of the thickness change of the stamping workpiece at stamping depths of 9.5 mm, 15.5 mm, and 27.5 mm by the process energy maps constructed with different schemes are shown. Among them, Figure 13 Part (a) in the figure shows the monitoring results of the thickness change of the stamping workpiece at a stamping depth of 9.5 mm; Figure 13 Part (b) in the figure shows the monitoring results of the thickness change of the stamping workpiece at a stamping depth of 15.5 mm; Figure 13 Part (c) in the figure shows the monitoring results of the thickness change of the stamping workpiece at 27.5 mm.

[0053] Figure 14 During the performance test, the mean absolute percentage error of the monitoring results at different stamping depths by the process energy maps constructed with different schemes is shown. Among them, Figure 14 Part (a) in the figure shows the mean absolute percentage error of the monitoring results of the maximum thinning rate of the stamping workpiece; Figure 14 Part (b) in the figure shows the mean absolute percentage error of the monitoring results of the maximum thickening rate of the stamping workpiece.

[0054] Figure 15 During the performance test, the mean square error of the monitoring results at different stamping depths by the process energy maps constructed with different schemes is shown. Among them, Figure 15 Part (a) in the figure shows the mean square error of the monitoring results of the maximum thinning rate of the stamping workpiece; Figure 15 Part (b) in the figure shows the mean square error of the monitoring results of the maximum thickening rate of the stamping workpiece.

[0055] Figure 16 During the performance test, the process energy map constructed based on thin plate spline interpolation is shown.

[0056] Figure 17 During the performance test, the Pearson correlation coefficient distribution diagram of the process energy and thickness change at different stamping depths is shown. Detailed implementation manners

[0057] In order to make the objectives, technical solutions, and advantages of the present invention clearer and more understandable, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0058] Embodiment 1

[0059] This embodiment provides a method for generating a process energy map for on-line monitoring of the forming quality of stamping workpieces. The process energy map in the present invention is an image reflecting the mapping relationship between the stamping depth h p 、process energy E and thickness variation TV during the stamping process of a specified stamping workpiece. The forming quality of the stamping workpiece under different stamping depths h p and process energy E conditions is also divided by different partitions in this image.

[0060] As Figure 1 shown, the method for generating the process energy map provided in this embodiment includes the following steps:

[0061] S1: Measure a large number of equipment parameters and workpiece thickness D of each target stamping workpiece of the same type under specified process conditions; the equipment parameters include: stamping depth h p and punching force F.

[0062] S2: Calculate the process energy E, maximum thickening rate Δ MTC and maximum thinning rate Δ MTN corresponding to each set of equipment parameters respectively; where

[0063]

[0064] In the above formula, h max represents the maximum stamping depth of the equipment corresponding to the current target stamping workpiece; D i represents the thickness before stamping at the specified measurement point on the stamping workpiece; D i+1 represents the thickness after stamping at the specified measurement point on the stamping workpiece.

[0065] It should be particularly noted that: the stamping depth h p refers to the overall downward pressing depth of the punch of the stamping equipment during the stamping process; while the maximum thickening rate Δ MTC and the maximum thinning rate Δ MTN are determined by detecting the deformation conditions of different positions of the stamping workpiece before and after stamping, and calculating the deformation conditions of the positions with the largest thickening amplitude and the largest thinning degree on the same stamping workpiece.

[0066] S3: Take the stamping depth h p , process energy E and thickness variation TV corresponding to each set of equipment parameters as a set of metadata to form an original data set. Among them, the thickness variation TV includes the maximum thickening rate Δ MTC and the maximum thinning rate Δ MTN .

[0067] The original data set A is expressed as: A=(E,TV), where E is a measurement process energy matrix of J×2:

[0068] E = (h p , E p )

[0069] In the above formula, h p is a stamping depth vector of J×1, h p = [h p,1 , h p,2 ,..., h p,J -1 ; E p is a process energy vector of J×1, E p = [E p,1 , E p,2 ,..., E p,J -1 ; TV is a thickness change matrix of J×2, expressed as:

[0070] TV = (Δ MTN , Δ MTC )

[0071] In the above formula, Δ MTN is a maximum thinning rate vector of J×1, Δ MTN = [Δ MTN,1 , Δ MTN , 2,..., Δ MTN,J -1 ; Δ MTC is a maximum thickening rate vector of J×1, Δ MTC = [Δ MTC,1 , Δ MTC , 2,..., Δ MTC,J -1 .

[0072] S4: Select M machine learning models as alternative models. Using the stamping depth h p and process energy E as inputs and the thickness change amount TV as the output, pre-train and pre-test the alternative models using a part of the metadata randomly selected from the original dataset. The alternative models in this embodiment include SVM, RF, GBDT, XGBoost, KNN, and LSTM.

[0073] S5: Conduct correlation analysis and prediction accuracy evaluation on the prediction results of each alternative model in the pre-test stage.

[0074] Perform Pearson correlation analysis on the alternative models; among them, the Pearson correlation coefficient r xy of the alternative model is calculated as follows:

[0075]

[0076] In the above formula, TV p,i,x and TV​​​​p,i,y The predicted results of thickness changes for the x-th and y-th alternative models respectively; and are the average values of the predicted results of thickness changes for the x-th and y-th alternative models respectively; I is the total number of predicted results.

[0077] The prediction accuracy of each alternative model is evaluated using the mean absolute percentage error e MA and the mean squared error e MS as follows, and the calculation formulas are as follows:

[0078]

[0079]

[0080] where c i ' and c i are the measured and monitored data respectively; N is the number of monitored data.

[0081] S6: Select n machine learning models with relatively weak correlations as the base models, where n < M; select the machine learning model with the highest prediction accuracy as the meta-model, and build a prediction network based on the Stacking ensemble learning framework. The model architecture of the prediction network is as Figure 2 shown.

[0082] Specifically, in this embodiment, the base models in the prediction network actually built based on the Stacking ensemble learning framework are XGBoost, LSTM, SVM, and KNN, and the meta-model is XGBoost. In the built prediction network, the input of each base model is the stamping depth h p and the process energy E, and the output is the predicted thickness change amount TV p,i , i = 1,..., n; the input of the meta-model is the output of each base model, and the output of the meta-model is the finally predicted thickness change amount TV.

[0083] S7: Use the original data set to train the prediction network in two stages, and retain the trained prediction network as the required thickness change prediction model.

[0084] As Figure 3 shown, the two-stage training process of the prediction network is as follows:

[0085] The first stage:

[0086] (1) Divide the original data set into n sub-data sets, A1 to A n .

[0087] (2) Use n - 1 of the sub-data sets as the training set and the last one as the test set.

[0088] (3) Use the training set and the test set to conduct the first-stage training and testing on each base model.

[0089] (4) According to the training results, obtain the n prediction results TV corresponding to each metadata in the original dataset. p,1 ~TV p,n .

[0090] Second stage:

[0091] (5) Combine the thickness change matrix TV in the original dataset with the prediction results TV p,1 ~TV p,n of the corresponding n base models to form new metadata.

[0092] (6) Use the new metadata to form a new fitting dataset B: B = (TV p,1 , …, TV p,n , TV), where TV p,1 ~TV p,n are respectively J×2 thickness change matrices formed by the prediction results of the 1st to the nth base models. The length of J is the number of groups of equipment parameters.

[0093] (7) Divide the fitting dataset B into a training set and a test set, and conduct the second-stage training and testing on the meta-model.

[0094] S8: Draw a blank process energy map with the stamping depth h p and the process energy E as the horizontal and vertical coordinates respectively. Then use the thickness change prediction model to predict the thickness change amount TV of each point in the process energy map, and color the blank process energy map according to the maximum thickening rate Δ MTC , the maximum thinning rate Δ MTN in the thickness change amount TV and its color mapping.

[0095] The drawing process of the process energy map is as follows:

[0096] S81: Establish a blank coordinate system with the stamping depth h p as the abscissa and the process energy E as the ordinate, and only including the first and fourth quadrants.

[0097] S82: Set the abscissa of any point in the fourth quadrant to the opposite of the original value, and use the axis where the ordinate is located as the Y-axis, so that the same coordinate values in the two quadrants are symmetric about the Y-axis.

[0098] S83: According to the constraint relationship between the stamping depth h p and the process energy E in the process conditions of the target stamping workpiece, generate two closed areas to be filled that are symmetric about the Y-axis in the blank coordinate system.

[0099] S84: Using the abscissa and ordinate of any position in the area to be filled as inputs, generate the thickness change amount TV of the corresponding point by using the thickness change amount prediction model.

[0100] S85: Based on the preset color mapping relationship, respectively according to the maximum thickening rate Δ MTC and the maximum thinning rate Δ MTN contained in the thickness change amount TV of each generated coordinate point, color the pixel points in the area to be filled in the first quadrant and the fourth quadrant.

[0101] In the process energy map designed in this embodiment, the map for determining whether the workpiece wrinkles based on the maximum thickening rate Δ MTC and the map for determining whether the workpiece ruptures based on the maximum thinning rate Δ MTN are set in two symmetric quadrants in the same coordinate system. In fact, those skilled in the art can also present the relevant map information through two images respectively; this still belongs to a part of the solution provided by the present invention.

[0102] The solution of this embodiment mainly uses the hue difference of colors to represent the numerical value of the maximum thickening rate Δ MTC or the maximum thinning rate Δ MTN corresponding to each coordinate point. For example, the larger the numerical value, the closer the pixel value of the corresponding coordinate point is to red, and the smaller the numerical value, the closer the pixel value of the corresponding coordinate point is to purple.

[0103] Of course, in other embodiments, in order to reflect the numerical difference between the maximum thickening rate Δ MTC or the maximum thinning rate Δ MTN , other methods can also be used to visually display the maximum thickening rate Δ MTC or the maximum thinning rate Δ MTN , which is helpful for later dividing the wrinkling area and the rupture area and their boundaries. For example, in a three-dimensional coordinate system, using the maximum thickening rate Δ MTC or the maximum thinning rate Δ MTN as the data at both ends of the Z-axis to construct a concave and convex surface, and then distinguishing different partitions according to the "peaks" and "valleys" of the surface.

[0104] S9: Partition the colored process energy map, mark the area corresponding to all pixel points whose maximum thickening rate Δ MTC exceeds the wrinkling threshold in the process energy map as the wrinkling area; mark the area corresponding to all pixel points whose maximum thinning rate Δ MTN exceeds the rupture threshold in the process energy map as the rupture area.

[0105] The partitioning process of the process energy map is as follows:

[0106] S91: Determine whether the maximum thickening rate Δ corresponding to each pixel point in the colored process energy map exceeds a preset wrinkling threshold, and make the following decisions: MTC If so,

[0107] (1) When the maximum thickening rate Δ MTC exceeds the wrinkling threshold, mark the first state attribute of the current pixel point as "wrinkling".

[0108] (2) When the maximum thickening rate Δ MTC is within the wrinkling threshold, mark the first state attribute of the current pixel point as "safe".

[0109] S92: Determine whether the maximum thinning rate Δ corresponding to each pixel point in the colored process energy map exceeds a preset cracking threshold, and make the following decisions: MTN If so,

[0110] (1) When the maximum thinning rate Δ MTN exceeds the cracking threshold, mark the second state attribute of the current pixel point as "cracking".

[0111] (2) When the maximum thinning rate Δ MTN is within the cracking threshold, mark the second state attribute of the current pixel point as "safe".

[0112] S93: Based on the first state attribute and the second state attribute, partition the colored regions in the first quadrant and the fourth quadrant respectively:

[0113] Divide all the coordinate points marked "wrinkling" in the first quadrant into the wrinkling area, and the rest into the safe area; divide all the coordinate points marked "cracking" in the fourth quadrant into the cracking area, and the rest into the safe area.

[0114] Embodiment 2

[0115] This embodiment provides a generation system for a process energy map for monitoring the forming quality of a stamping workpiece, which adopts the generation method of the process energy map for monitoring the forming quality of a stamping workpiece as in Embodiment 1, and generates a process energy map for characterizing the mapping relationship between the forming quality of the target stamping workpiece and the equipment parameters according to the sample data of the equipment parameters of the target stamping workpiece under specified process conditions.

[0116] As Figure 4 shown, the generation system for the process energy map for monitoring the forming quality of a stamping workpiece in this embodiment includes: a data acquisition module, a data set generation module, a prediction network training module, and a map drawing module.

[0117] Among them, the data acquisition module is used to collect the equipment parameters and the workpiece thickness D when the stamping equipment processes stamping workpieces under specified process conditions, as well as the constraint conditions of the equipment parameters. The equipment parameters include the stamping depth h p and the stamping force F. The data acquisition module also calculates the corresponding process energy E, the maximum thickening rate Δ MTC and the maximum thinning rate Δ MTN .

[0118] The dataset generation module takes the stamping depth h p , the process energy E, and the thickness change amount TV corresponding to each set of equipment parameters as a set of metadata to form the original dataset. Among them, the thickness change amount TV includes the maximum thickening rate Δ MTC and the maximum thinning rate Δ MTN .

[0119] The prediction network training module runs a prediction network based on the Stacking ensemble learning framework with XGBoost, LSTM, SVM, and KNN as the base models and XGBoost as the meta-model; the prediction network training module is used to perform two-stage training on the prediction network using the original dataset, and then obtain a thickness change prediction model with the stamping depth h p and the stamping force F as the input and the maximum thickening rate Δ MTC and the maximum thinning rate Δ MTN as the output.

[0120] The atlas drawing module includes a white drawing generation unit, a pixel information generation unit, a coloring unit, an attribute marking unit, and a partitioning unit. The white drawing generation unit is used to: first establish a blank coordinate system with the stamping depth h p as the abscissa and the process energy E as the ordinate, and only including the first and fourth quadrants. Then set the abscissa of any point in the fourth quadrant to the opposite of the original value, with the ordinate axis as the Y-axis, so that the same coordinate values in the two quadrants are symmetric about the Y-axis. Finally, according to the constraint conditions in the process conditions of the stamping workpiece, two closed areas to be filled that are symmetric about the Y-axis are generated in the blank coordinate system, that is, a blank process energy atlas is obtained. The pixel information generation unit is used to input the abscissa and ordinate of each pixel point in the area to be filled into the thickness change prediction model to generate the maximum thickening rate Δ MTC and the maximum thinning rate Δ MTN corresponding to each pixel point. The coloring unit is used to map each pixel point's maximum thickening rate Δ MTC and the maximum thinning rate Δ MTNThey are respectively converted into a corresponding color value; then, the two color values are used to fill the areas to be filled on both sides of the blank process energy map with color pixels to obtain a colored process energy map. The attribute marking unit is used to judge the maximum thickening rate Δ MTC and the maximum thinning rate Δ MTN of each pixel point in the colored process energy map according to the preset wrinkling threshold and cracking threshold; and assign a corresponding first state attribute or second state attribute. The partitioning unit is used to partition the colored process energy map according to the first state attribute or the second state attribute of each pixel point, and divide the areas where the pixel points exceeding the wrinkling threshold and the cracking threshold are located into a "wrinkling area" and a "cracking area" respectively; thus, the required process energy map is obtained.

[0121] Embodiment 3

[0122] This embodiment further includes a data processing device, which includes a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, a generation system of the process energy map for monitoring the forming quality of the stamping workpiece as in Embodiment 2 is created, and then, according to the equipment parameters and their constraint conditions of the target stamping workpiece under the specified process conditions collected, the process energy map of the target stamping workpiece under the specified process conditions is generated.

[0123] The data processing device provided in this embodiment is essentially a computer device, and the computer device can be an intelligent terminal capable of executing programs, a tablet computer, a notebook computer, a desktop computer, a rack server, a blade server, a tower server, or a cabinet server (including an independent server, or a server cluster composed of multiple servers), etc.

[0124] The computer device of this embodiment at least includes, but is not limited to: a memory and a processor that can communicate with each other through a system bus.

[0125] In this embodiment, the memory (i.e., the readable storage medium) includes flash memory, hard disk, multimedia card, card-type memory (e.g., SD or DX memory, etc.), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the memory may be an internal storage unit of the computer device, such as the hard disk or memory of the computer device. In other embodiments, the memory may also be an external storage device of the computer device, such as a plug-in hard disk, Smart Media Card (SMC), Secure Digital (SD) card, Flash Card, etc. equipped on the computer device. Of course, the memory may also include both the internal storage unit and the external storage device of the computer device. In this embodiment, the memory is generally used to store the operating system and various application software installed on the computer device. In addition, the memory can also be used to temporarily store various data that have been output or will be output.

[0126] In some embodiments, the processor may be a central processing unit (CPU), a controller, a microcontroller, a microprocessor, or other data processing chips. The processor is generally used to control the overall operation of the computer device. In this embodiment, the processor is used to run the program code stored in the memory or process data.

[0127] Embodiment 4

[0128] By using the method in Embodiment 1, a process energy map of each stamping workpiece under specified process conditions can be generated. This embodiment further provides an on-line monitoring method for the forming quality of stamping workpieces by using the generated process energy map, as Figure 5 shown, the method includes the following steps:

[0129] S01: Obtain the process energy map of the current stamping workpiece under specified process conditions.

[0130] The process energy map in the present invention is a pre-generated map with the stamping depth h p and the process energy E as the abscissa and ordinate respectively, and the maximum thickening rate Δ p and the maximum thinning rate Δ MTC under any stamping depth h MTN and process energy E conditions as the coordinate points, and the wrinkling area and cracking area ranges are divided according to the state attributes of the coordinate points.

[0131] S02: Collect the equipment parameters during the machining of the stamping workpiece in real time, including the stamping depth h p and the stamping force F.

[0132] S03: Calculate the real-time process energy E based on the real-time stamping depth h real and the real-time stamping force F real collected during the machining process. real :

[0133]

[0134] S04: Fit a corresponding state trajectory in the process energy map according to the state changes of the real-time stamping depth h real and the real-time process energy E real during the current workpiece machining process.

[0135] S05: Make the following judgments based on the position of the state trajectory of the current stamping workpiece machining process in the process energy map:

[0136] (ⅰ) When the state trajectory neither passes through the fracture zone nor terminates in the wrinkling zone, it is determined that the current stamping workpiece is completely qualified.

[0137] (ⅱ) When the state trajectory only passes through the fracture zone, it is determined that the current stamping workpiece has cracked.

[0138] (ⅲ) When the state trajectory only terminates in the wrinkling zone, it is determined that the current stamping workpiece has local wrinkling.

[0139] (ⅳ) When the state trajectory passes through the fracture zone and terminates in the wrinkling zone, it is determined that the current stamping workpiece has cracked and has local wrinkling.

[0140] It should be noted specifically that: the process energy map adopted in this embodiment should be pre-generated by technicians according to the sample parameters of the target stamping workpiece under the specified process conditions, and each process energy map corresponds one-to-one to a specific type of stamping workpiece and its specified process conditions. The process conditions include material properties and equipment parameters; the material properties include the material grade, shape, specifications, etc. of the sheet used in the machining process of the stamping workpiece; the equipment parameters refer to the initial parameters preset by the stamping equipment during the machining process of the stamping workpiece. When any one of the type of stamping workpiece and the process conditions is adjusted, the corresponding process energy map needs to be updated.

[0141] In step S01, the method for dividing the wrinkling zone and the fracture zone in the process energy map is as follows:

[0142] (Ⅰ) Draw at least one blank coordinate system with the stamping depth h p as the abscissa and the process energy E as the ordinate.

[0143] (Ⅱ) Generate a closed area to be filled in a blank coordinate system according to the constraint relationship between the stamping depth h p and the process energy E in the current process conditions of the stamping workpiece.

[0144] (Ⅲ) Use a trained thickness change prediction model to generate the thickness change TV corresponding to each coordinate point in the area to be filled.

[0145] The input of the thickness change prediction model is the stamping depth h p and the process energy E, and the output is the thickness change TV; the thickness change TV includes the maximum thickening rate Δ MTC and the maximum thinning rate Δ MTN .

[0146] (Ⅳ) Predetermine the upper limit α MTC of the maximum thickening rate Δ MTC and the upper limit β MTN of the maximum thinning rate Δ MTN of the current stamping workpiece under specified process conditions according to expert experience.

[0147] (Ⅴ) Divide the area composed of all coordinate points in the filled area that satisfy Δ MTC > α MTC into the wrinkling area and form the boundary between the wrinkling area and the rest (safe area).

[0148] (Ⅵ) Divide the area composed of all coordinate points in the filled area that satisfy Δ MTN > β MTN into the cracking area and form the boundary between the cracking area and the rest (safe area).

[0149] When implementing the online monitoring function of the forming quality of the stamping workpiece in this embodiment, the defect partition information in the process energy map is mainly utilized. Therefore, very accurate boundaries between the wrinkling area and the cracking area and the safe area are required. The color information in the map is not needed in the process of monitoring the forming quality of the stamping workpiece. The role of the color information is to facilitate the user to intuitively understand the boundaries of different partitions.

[0150] During the monitoring of the forming quality of stamping workpieces, when the state trajectory of the stamping process being monitored passes through the fracture zone, it usually indicates that the stamping workpiece has cracked. Since cracking belongs to an irreversible defect type, even if the state trajectory finally passes through the fracture zone, the cracking defect still exists. When the state trajectory of the monitored stamping process passes through the wrinkling zone, it usually indicates that the stamping workpiece has wrinkled at the current moment. However, since wrinkling belongs to a reversible defect type, as long as the state trajectory finally passes through the wrinkling zone, it is generally considered that the stamping workpiece does not finally have a wrinkling defect. In summary: To determine whether a stamping workpiece has cracked, only need to analyze whether the state trajectory of the processing process passes through the fracture zone, while to determine whether a stamping workpiece has wrinkled, it is necessary to analyze whether the end point of the processing trajectory falls into the wrinkling zone.

[0151] Embodiment 5

[0152] Based on an online monitoring method for the forming quality of stamping workpieces provided in Embodiment 4, this embodiment further provides an online monitoring tool for the forming quality of stamping workpieces, which is used to monitor the processing process of a target stamping workpiece on a stamping device by using the online monitoring method for the forming quality of stamping workpieces in Embodiment 4, and generate a quality evaluation result of the processed target stamping workpiece.

[0153] As Figure 6 shown, the online monitoring tool provided in this embodiment includes: a reference database, a data acquisition unit, a trajectory generation unit, and a quality determination unit.

[0154] The reference database stores the process energy maps of each stamping workpiece under specified process conditions. Each process energy map includes a label characterizing the type of stamping workpiece and the process conditions corresponding to the current process energy map.

[0155] The data acquisition unit is used to collect the stamping depth h p and the stamping force F during the processing of the stamping device in real time, and calculate the real-time process energy E according to the real-time stamping depth h real collected during the processing process and the real-time stamping force F real real .

[0156] The trajectory generation unit is used to query the reference database to obtain the process energy map whose label information corresponds to the type of the current stamping workpiece and its process conditions. Then, according to the state changes of the real-time stamping depth h real and the real-time process energy E real during the processing of the current stamping workpiece, a corresponding state trajectory is fitted in the process energy map.

[0157] ​The quality determination unit is used to generate a corresponding quality evaluation result according to the position of the state trajectory in the process energy map: (i) When the state trajectory neither passes through the fracture zone nor terminates in the wrinkling zone, it is determined that the current stamping workpiece is completely qualified. (ii) When the state trajectory only passes through the fracture zone, it is determined that the current stamping workpiece has cracked. (iii) When the state trajectory only terminates in the wrinkling zone, it is determined that the current stamping workpiece has local wrinkling. (iv) When the state trajectory passes through the fracture zone and terminates in the wrinkling zone, it is determined that the current stamping workpiece has cracked and has local wrinkling.

[0158] Example 6

[0159] A visualization method for the forming quality of a stamping workpiece provided in this embodiment is used to evaluate the forming quality of the stamping workpiece and visually present the quality evaluation process during the processing of the target stamping workpiece, in combination with the online monitoring method for the forming quality of the stamping workpiece provided in Example 4. The specific process is as follows:

[0160] I. Before processing the stamping workpiece:

[0161] Query the corresponding process energy map of the current stamping workpiece according to the preset processing parameters.

[0162] During the processing of each stamping workpiece, the technician inputs the type of the stamping workpiece to be processed and the specific processing process conditions into the stamping equipment, and the stamping equipment will retrieve the corresponding process energy map pre-tested and generated from the reference database.

[0163] II. During the stamping process of the stamping workpiece:

[0164] (1) Present the first screen and the second screen according to the process energy map.

[0165] Among them, the first screen uses the stamping depth h p as the abscissa and the process energy E as the ordinate; and characterizes the value of the corresponding maximum thickening rate Δ MTC through the hue difference of the pixel points within the constraint region. The boundary of the wrinkling zone is also divided in the first screen.

[0166] Among them, the second screen uses the stamping depth h p as the abscissa and the process energy E as the ordinate; and characterizes the value of the corresponding maximum thinning rate Δ MTN through the hue difference of the pixel points within the constraint region. The boundary of the fracture zone is also divided in the second screen.

[0167] (2) Determine the real-time stamping depth h real and the real-time process energy E real corresponding to the processing process according to the feedback signal of the stamping equipment collected in real time; and use this as the stamping state variable U.

[0168] (3) In the first and second screens, the position change of the stamping state variable U is displayed in real time, and a motion trajectory Ut is fitted; here, the motion trajectory is the state trajectory reflecting the forming quality during the stamping workpiece processing.

[0169] During the actual processing, the stamping equipment determines the stamping depth h of the processing according to the feedback information of relevant sensors in the processing. p and the stamping force F, and calculates the corresponding process energy E accordingly. The stamping equipment records the real-time stamping depth h real and the real-time process energy E real , and displays them on the process energy map to form a "state point". As the processing progresses, the stamping depth h p and the process energy E will continuously change, and the state point will also continuously move on the process energy map. Recording the displacement change of the state point can obtain a corresponding "state trajectory". By combining the relationship between the state trajectory and the fracture zone and wrinkling zone in the process energy map, the forming quality of the current stamping workpiece can be determined.

[0170] V. After the stamping workpiece is processed, use any one or more interaction means to present the evaluation result of the forming quality of the current stamping workpiece to the user.

[0171] Embodiment 7

[0172] This embodiment provides a stamping equipment, which includes an equipment body and an interaction component. The interaction component includes a display, a speaker and an input device; the display is used to display the image interaction information characterizing the forming quality during the stamping workpiece processing; the speaker is used to play the audio interaction information characterizing the forming quality during the stamping workpiece processing; the input device is used to input manual operation instructions to the stamping equipment.

[0173] The stamping equipment also includes an online monitoring tool for the forming quality of the stamping workpiece as in Embodiment 5. The online monitoring tool is used to generate a corresponding quality evaluation result during the stamping processing of any stamping workpiece according to the operation parameters of the equipment body collected in real time.

[0174] The interaction component is used to present the forming quality of the stamping workpiece to the user in real time according to the visualization method of the forming quality of the stamping workpiece in Embodiment 6.

[0175] In a more optimized solution of this embodiment, the stamping equipment further includes a feedback and warning unit. The feedback and warning unit obtains the quality evaluation result generated by the online monitoring tool after any stamping workpiece is processed, and when the processed stamping workpiece is evaluated as not being completely qualified, issues a driving instruction representing shutdown to the main controller of the stamping equipment to drive the stamping equipment to stop; and before the preset processing parameters of the equipment are adjusted, refuses to respond to the operation instruction of the equipment.

[0176] Performance Test

[0177] In order to verify the generation method of the process energy map for stamping workpiece forming quality monitoring provided in Embodiment 1 of the present invention, and the effectiveness of the overall solution for using the generated process energy map to monitor the forming quality during the stamping process. This embodiment also conducts an actual stamping experiment on this solution, and the experimental process is as follows:

[0178] I. Construction of the process energy map in this embodiment

[0179] 1.1. In this experiment, a door inner panel-like part as shown in Figure 7 is used as the stamping workpiece, and AA5052 aluminum alloy is selected as its material to construct the process energy map. The contour of the process energy map uses 16 sets of stamping speed and blank holding force combination sample parameters as the stamping process conditions. The abscissa of the process energy map is the stamping depth. Taking 1 mm as the interval, the stamping depth from 1 mm to 28 mm is selected as the value range of the abscissa of the process energy map to facilitate verifying the monitoring accuracy stability and applicability at different stamping depths. In addition, 16 ml of lubricating oil is used for lubrication in each stamping experiment.

[0180] 1.2. After the data collection of the process energy and thickness change is completed, start constructing a prediction network based on the Stacking ensemble learning framework. In this embodiment, relatively common machine learning models such as Support Vector Machines (SVM), EXtreme Gradient Boosting (XGBoost), Gradient Boosting Decision Tree (GBDT), Random Forest (RF), K-Nearest Neighbor Classification (KNN), and Long Short-Term Memory (LSTM) are selected as alternative models. A brief introduction to the principles of these alternative models is shown in Table 1.

[0181] Table 1: Brief introduction to the principles of each alternative model

[0182]

[0183] To fairly compare the prediction performance of each alternative model, their hyperparameters are set to the following default values during the training process.

[0184] Table 2: Hyperparameters and Prediction Errors of Each Model

[0185]

[0186]

[0187] 1.3. In this embodiment, the relevant samples obtained from the pre-stamping experiment are measured to obtain the original data, which constitutes the required original dataset. 70% of the original dataset is randomly selected as the training set, and the remaining 30% is used as the test set to obtain the prediction results of different alternative models and evaluate their prediction performance.

[0188] First, the mean absolute percentage error e MA and the mean squared error e MS are respectively used to evaluate the prediction accuracy of each alternative model. The prediction results of each alternative model are as Figure 8 shown.

[0189] Combined with Figure 8 it can be observed that: The thickness change prediction result of XGBoost is closer to the zero error line compared to other alternative models, and the shown error also indicates that XGBoost has the smallest prediction error. Therefore, XGBoost is selected as the meta-model for the thickness change prediction model based on Stacking ensemble learning.

[0190] After obtaining the prediction results of each alternative model, model correlation analysis is then carried out through Pearson correlation analysis. The correlation coefficients of each alternative model are as Figure 9 shown.

[0191] From Figure 9 it can be seen that: The correlation coefficients between XGBoost, GBDT, and RF all exceed 0.98. The reason is that although these three machine learning models are slightly different in principle, they are generally decision tree-based models, and their data prediction mechanisms have strong similarities. While the modeling principles and prediction mechanisms of LSTM, SVM, and KNN are quite different, so the correlation of the prediction results is relatively low. In this experiment, 4 alternative models, namely XGBoost, LSTM, SVM, and KNN, are selected as the base models.

[0192] 1.4. According to the number of base models, the original dataset can be equally divided into 4 sub-datasets {A i, i = 1, ..., 4} are used for the training of the base models. Each sub-dataset contains the measured process energy and thickness change data corresponding to 7 stamping depths as shown in Table 3.

[0193] Table 3: Sub-datasets of process energy and thickness change and their corresponding stamping depths

[0194]

[0195] Each base model is trained 4 times using the sub-datasets. During each training process, one of the sub-datasets is used as the test set while the other three sub-datasets are used as the training sets. The prediction datasets for SVM, XGBoost, LSTM, and KNN are TV p,1 , TV p,2 , TV p,3 and TV p,4 as shown in Figure 10 .

[0196] Taking TV p,1 , TV p,2 , TV p,3 and TV p,4 as the input of the meta-model, i.e., XGBoost, and the measured thickness change dataset as the output of XGBoost, the meta-model is trained to predict the final thickness change.

[0197] Then, based on the predicted thickness change data, the process energy map can be filled, colored, and the quality regions can be divided. Finally, the constructed process energy map is as shown in Figure 11 .

[0198] II. Setting of the control group and control experiment

[0199] 2.1 Process conditions

[0200] Since the selected inner door panel parts of the class have different forming shapes at different stamping depths, in this experiment, the inner door panel parts of the class at stamping depths of 9.5 mm, 15.5 mm, and 27.5 mm are selected for thickness change monitoring to verify the effectiveness of the constructed process energy map. At each stamping depth, the process energy and thickness change of the stamping workpieces under 32 groups of process conditions as shown in Table 4 are measured. In addition, the inner door panel parts of the class at stamping depths of 5 mm and 23 mm are also selected for thickness change monitoring to explore the accuracy of the constructed process energy map on the training set data.

[0201] Table 4: Stamping process conditions for monitoring stamping workpieces

[0202]

[0203] 2.2 Setting of the control group

[0204] In the verification stage, to demonstrate the effectiveness and advantages of the process energy map constructed in this embodiment, different prediction methods or models were also used to construct the process energy map for comparative analysis.

[0205] These prediction methods or models are divided into three categories, namely data interpolation, machine learning models, and Stacking ensemble learning prediction models that use combinations of different base models and meta models. For data interpolation, the thin plate spline data interpolation technique is directly used. For machine learning models, XGBoost, LSTM, and SVM are used. The base models and meta models selected in the Stacking ensemble learning prediction model are shown in Table 5. BC1 to BC4 are mainly used to explore the influence of the base model on the monitoring accuracy of the process energy map constructed based on Stacking ensemble learning, while M-SVM, M-LSTM, and M-KNN are mainly used to explore the influence of the meta model.

[0206] Table 5: Combinations of base models and meta models

[0207]

[0208] III. Performance comparison of different solutions in this case and the control group

[0209] 3.1 Monitoring results of thickness changes of the process energy map for different solutions

[0210] The monitoring results of the thickness changes in the training set, that is, the thickness changes of the stamping workpiece at stamping depths of 5 mm and 23 mm, are as Figure 12 shown. It can be observed that the monitoring results of the process energy map constructed by the solution of this embodiment (i.e., Figure 12 SML in Figure 13 ) are in good agreement with the measurement results. The monitoring results of the thickness changes of the stamping workpiece at stamping depths of 9.5 mm, 15.5 mm, and 27.5 mm are as Figure 14 shown. The monitoring results of SML are still in good agreement with the measurement results. The mean absolute percentage error and mean square error of the SML monitoring results are as Figure 15 and 2 shown respectively, and they do not exceed 5.03% and 0.94% respectively.

[0211] The above data prove that the process energy map constructed by applying the solution provided by the present invention has high monitoring accuracy and stability.

[0212] In addition, as Figure 13As shown in part (c), 14 out of the 32 formed stamping workpieces were cracked, that is, there were 14 cases where the measured maximum thinning rate fell within the cracking zone. And the process energy map constructed by the solution of the present invention identified 16 cracked stamping workpieces. The monitored maximum thinning rates of 14 in SML fell within the cracking zone, and only two non-cracked stamping workpieces were misidentified as cracked by the process energy map, that is Figure 13 (c) The points circled by the gray circles in. The cracking recognition accuracy rate of the process energy map constructed by applying the solution of the present invention is as high as 93.75%. Similarly, 10 out of the 32 formed stamping workpieces were wrinkled, and the process energy map constructed by applying the solution of the present invention identified 12, and its wrinkling recognition accuracy rate was also 93.75%.

[0213] The above data proves that: the process energy map constructed by applying the present invention can effectively and accurately identify the wrinkling and cracking defects in stamping workpieces.

[0214] The monitoring results of the process energy maps constructed by different prediction models are as Figure 12 and Figure 13 shown. For easy observation, the monitoring results of these process energy maps are shown in two figures. It can be seen that the monitoring results of the process energy maps constructed by different prediction models are far from the measurement results. The mean absolute percentage error e MA and the mean square error e MS are respectively as Figure 14 and Figure 15 shown, and their errors are all higher than the errors of the monitoring results of the process energy map constructed by this solution. The process energy maps constructed by different prediction models have good monitoring accuracy and stability for the thickness change of the test set. The maximum mean absolute percentage error e MA and the maximum mean square error e MS are 6.64% (that is, the monitoring result of the maximum thinning rate of the stamping workpiece with a stamping depth of 5 mm by the process energy map constructed using LSTM) and 0.36% 2 (that is, the monitoring result of the maximum thickening rate of the stamping workpiece with a stamping depth of 23 mm by the process energy map constructed using SVM).

[0215] And the maximum mean absolute percentage error e MA and the mean square error e MS of the monitoring results of the process energy maps constructed by different prediction models for the thickness change of the stamping workpiece at stamping depths of 9.5 mm, 15.5 mm and 27.5 mm are 13.66% (that is, the monitoring result of the maximum thickening rate of the stamping workpiece with a stamping depth of 27.5 mm by the process energy map constructed using SVM) and 7.51% 2(That is, the monitoring result of the maximum thinning rate of a stamping workpiece with a stamping depth of 27.5 mm using the process energy map constructed by SVM) is significantly higher than the error of the monitoring result of the process energy map constructed by applying the solution of the present invention.

[0216] The above data proves that: the method for constructing a process energy map proposed by the present invention significantly improves the monitoring accuracy and stability of the process energy map under different process conditions.

[0217] 3.2. Comparison of the monitoring accuracy between the solution of the present invention and the map constructed by data interpolation

[0218] Combined with Figure 14 and Figure 15 it can be known that: the monitoring errors of the process energy map constructed by thin plate spline interpolation TPS for the thickness changes of stamping workpieces with different stamping depths are all higher than the monitoring results of the process energy map constructed by applying the method of the present invention. The reason can be attributed to the prediction mechanism of the data interpolation technique. In data interpolation, the approximate interpolation surface needs to pass through all the data points, that is, the data in the training set, resulting in the prediction results being easily affected by extreme points.

[0219] Due to the high variability of the thickness changes of stamping workpieces and the process energy data, the data interpolation technique cannot achieve accurate prediction of the thickness change data, resulting in large and irregular fluctuations in the thickness change data predicted by thin plate spline interpolation TPS. As shown in Figure 16 the process energy map constructed by thin plate spline interpolation TPS, it can be seen that there are serrated fluctuations in the process energy map, and the boundaries of its wrinkling areas are also very uneven, meaning that the data in the process energy map fluctuates greatly and irregularly. These fluctuating and irregular prediction data will lead to a decrease in the monitoring accuracy of the process energy map constructed by thin plate spline interpolation TPS for the thickness change data of stamping workpieces at unknown data, that is, stamping depths of 9.5 mm, 15.5 mm, and 27.5 mm. The uneven boundaries of the quality regions will also lead to a decrease in the defect recognition accuracy of the constructed process energy map. As shown in Figure 13 (c) in, the crack recognition accuracy and wrinkle recognition accuracy of the process energy map constructed by thin plate spline interpolation TPS are 53.13% and 40.63% respectively, which are much lower than the defect recognition accuracy of the process energy map constructed by applying the method of the present invention.

[0220] By comparing Figure 11 and Figure 16 it can be found that: the process energy map constructed by applying the solution of the present invention has a smoother color change, and the boundaries of its quality regions are also more uniform. The main reason for the obvious advantages of the solution provided by the present invention is that: the present invention can integrate the advantages of various machine learning models to reduce the fluctuations caused by data variability, thereby realizing more accurate construction of the process energy map.

[0221] 3.3. Comparison of the monitoring accuracy between the proposed solution of the present invention and the process energy maps constructed by machine learning models

[0222] As Figure 14 and Figure 15 shown, even when using XGBoost, which has the lowest prediction error among machine learning models, the monitoring error differences of the process energy maps constructed by it for the thickness changes at different stamping depths are also very obvious. The minimum monitoring error of the process energy map constructed by XGBoost is 1.30%, which is the mean absolute percentage error of the monitoring result of the maximum thinning rate of the stamping workpiece with a stamping depth of 23 mm, while the maximum monitoring error is 9.42%, which is the mean absolute percentage error of the monitoring result of the maximum thinning rate of the stamping workpiece with a stamping depth of 15.5 mm.

[0223] The reasons for the unstable monitoring accuracy of these process energy maps constructed by machine learning models may stem from the changes in the probability density distribution of the thickness change data. When using a machine learning model to establish a prediction model, a key assumption is that the prediction data comes from the same probability density distribution as the training data. If the distribution of the training data is different from the probability density distribution of the prediction data, the prediction accuracy of the machine learning model may decline. Since different shapes are formed at different stamping depths, the data of process energy and thickness change have high variability. Different machine learning models have different adaptabilities to different data sets. For example, machine learning models based on decision trees, such as XGBoost, are good at learning the data laws with complex correlation relationships, while other machine learning models may be good at learning the data laws with linear relationships.

[0224] Since the low correlation between the input (i.e., process energy) and output (i.e., thickness change) of the machine learning model will also lead to a decline in the prediction accuracy of the machine learning model, therefore, the change in the correlation between process energy and thickness change will also lead to monitoring instability.

[0225] Figure 17 Further, the Pearson correlation coefficients between process energy and thickness change at different stamping depths were statistically analyzed. Combining Figure 17 it can be found that: when the stamping depth is 5 mm, the correlation between process energy and thickness change is lower than that when the stamping depth is 23 mm, resulting in lower monitoring accuracy at a stamping depth of 5 mm. The reason for this phenomenon can be found from Figure 14 that when the stamping depth is 5 mm, the reason for the lower correlation between process energy and thickness change is that the deformation of the sheet metal is smaller at this stamping depth, resulting in smaller thickness changes. Smaller thickness changes are more susceptible to interference during the measurement process, such as measurement errors, resulting in a lower correlation between thickness change and process energy.

[0226] It can be summarized that the reasons why the process energy map constructed by the solution of the present invention has high precision are mainly in two aspects: on the one hand, the solution of the present invention effectively neutralizes the deviation of each base model by weighting the prediction results of each machine learning model, that is, the base model. A machine learning model can be regarded as finding the best hypothesis in a certain hypothesis space, that is, the predicted value. When the amount of available training data is too small compared to the size of the hypothesis space, the machine learning model can find many different hypotheses in this hypothesis space, and these hypotheses have the same accuracy in predicting data. By constructing an ensemble from all these prediction results of each machine learning model, the prediction model constructed based on Stacking ensemble learning can reduce the risk of selecting inaccurate prediction results. On the other hand, many machine learning models work by performing some form of local search and may fall into local optima. For example, machine learning models based on neural networks, such as LSTM, use the gradient descent method to minimize the error function of the training data, while machine learning models based on decision trees, such as RF, use greedy splitting rules to grow decision trees. An ensemble constructed by running local searches from many different starting points may provide a better approximation of the measurement results, thus reducing the risk of falling into local optima.

[0227] 3.4. Comparison of the monitoring accuracy of the maps constructed by the present invention with similar solutions under different combinations of base models and meta - models

[0228] As Figure 14 shown in Figure 15 the figure, the monitoring errors of the process energy maps constructed by Stacking ensemble learning under different combinations of base models and meta - models are generally lower than those of the process energy maps constructed by data interpolation and machine learning models. This shows that constructing a thickness change prediction model through Stacking ensemble learning can effectively improve the monitoring accuracy and stability of the process energy map. However, the monitoring errors of these process energy maps constructed by Stacking ensemble learning under different combinations of base models and meta - models are still higher than those of the process energy map constructed by applying the solution of the present invention.

[0229] When the prediction performance of each base model is similar, that is, when the prediction mechanisms of the base models are similar, it is difficult to improve the prediction performance of the integrated prediction model no matter how the base models are combined. In BC1 and BC3, the prediction mechanisms of XGBoost, RF, and GBDT are similar. When they are combined, it will lead to redundancy of the base models and bias of the meta-model, that is, the final prediction result is more biased towards the prediction results of XGBoost and so on. The prediction accuracy of the Stacking ensemble learning prediction models constructed in BC2 and BC4 is relatively low, probably because the diversity of the base models is relatively low, which makes it difficult for the Stacking ensemble learning prediction models to observe and learn the internal laws of data changes from multiple perspectives. For example, only two machine learning models are included in BC2, resulting in a lower prediction performance of the Stacking ensemble learning prediction model constructed with BC2 than other Stacking ensemble learning prediction models. Neither BC2 nor BC4 includes XGBoost with the best prediction performance, resulting in a lower prediction performance of the Stacking ensemble learning prediction models constructed with BC2 and BC4.

[0230] Using a machine learning model with low generalization ability as the meta-model is prone to overfitting, resulting in a decrease in the prediction accuracy of the constructed Stacking ensemble learning prediction model. According to Figure 8 it can be seen that XGBoost has good generalization ability. It can effectively generalize the prediction results of the base models and prevent the final prediction result from being biased towards the result of a certain base model. Therefore, the monitoring error of the process energy map constructed by the Stacking ensemble learning prediction model with XGBoost as the meta-model is lower than that of the process energy map constructed by the Stacking ensemble learning prediction models with SVM, LSTM, and KNN as the meta-models.

[0231] To sum up: The process energy map constructed by the Stacking ensemble learning prediction model proposed by the present invention is more accurate than the process energy map constructed by data interpolation, machine learning models, etc. At the same time, since the present invention selects machine learning models with low correlation as the base models and machine learning models with high generalization as the meta-models, the monitoring accuracy and stability of the constructed process energy map are further improved. By applying the process energy map constructed by the present invention, the present invention also realizes the online monitoring of the forming quality of the stamping workpiece in the stamping process, providing a means to online evaluate the forming quality of the stamping workpiece.

[0232] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, and improvements made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. An on-line monitoring method for the forming quality of stamping workpieces, characterized in that, It includes the following steps: S01: Obtain the process energy map of the current stamping workpiece under specified process conditions; The process energy map uses the stamping depth h p and the process energy E as the abscissa and ordinate respectively. Under any stamping depth h p and process energy E, the maximum thickening rate Δ MTC and the maximum thinning rate Δ MTN are the state attributes of the coordinate points; and the ranges of the wrinkling area and the cracking area in the process energy map are divided according to the state attributes of the coordinate points; S02: Real-time collect the equipment parameters during the processing of stamping workpieces, including the stamping depth h p and the stamping force F; S03: The real-time stamping depth h real and the real-time stamping force F real collected during the processing are used to calculate the real-time process energy E real : S04: According to the real-time stamping depth h real and the real-time process energy E real of the current stamping workpiece during the machining process, fit a corresponding state trajectory in the process energy map; S05: Make the following judgments based on the position of the state trajectory of the current stamping workpiece during the processing process in the process energy map: (ⅰ) When the state trajectory neither passes through the fracture zone nor terminates in the wrinkling zone, it is determined that the current stamping workpiece is completely qualified; (ⅱ) When the state trajectory only passes through the fracture zone, it is determined that the current stamping workpiece has cracked; (ⅲ) When the state trajectory only terminates in the wrinkling zone, it is determined that the current stamping workpiece has local wrinkling; (ⅳ) When the state trajectory passes through the fracture zone and terminates in the wrinkling zone, it is determined that the current stamping workpiece has cracked and has local wrinkling.

2. The on-line monitoring method for the forming quality of stamping workpieces according to claim 1, characterized in that: In step S01, the process energy map is pre-generated by a technician according to the sample parameters of the target stamping workpiece under specified process conditions; each process energy map corresponds one-to-one to a specific type of stamping workpiece and its specified process conditions; the process energy map uses the stamping depth h p and the process energy E as the abscissa and ordinate respectively, and the maximum thickening rate Δ p and the maximum thinning rate Δ MTC under any stamping depth h MTN and process energy E conditions are the state attributes of the coordinate points; and the ranges of the wrinkling area and the cracking area in the process energy map are divided according to the state attributes of the coordinate points.

3. The on-line monitoring method for the forming quality of stamping workpieces according to claim 2, characterized in that: The process conditions include material properties and equipment parameters; the material properties include the material label, shape, and specifications of the sheet used in the processing of the stamping workpiece; the equipment parameters refer to the initial parameters preset by the stamping equipment during the processing of the stamping workpiece; When any one of the type of the stamping workpiece and the process conditions is adjusted, the corresponding process energy map needs to be updated.

4. The on-line monitoring method for the forming quality of stamping workpieces according to claim 1, characterized in that: In step S01, the method for dividing the wrinkling zone and the fracture zone in the process energy map is as follows: (Ⅰ) With the stamping depth h p as the abscissa and the process energy E as the ordinate, plot at least one blank coordinate system; (Ⅱ) Generate a closed area to be filled in the blank coordinate system according to the constraint relationship between the stamping depth h p and the process energy E in the current process conditions of the stamping workpiece; (Ⅲ) Use a trained thickness change prediction model to generate the thickness change TV corresponding to each coordinate point in the area to be filled; The input of the thickness change prediction model is the stamping depth h p and the process energy E, and the output is the thickness change TV. The thickness change TV includes the maximum thickening rate Δ MTC and the maximum thinning rate Δ MTN ; (Ⅳ) Predetermine, according to expert experience, the upper limit α of the maximum thickening rate Δ MTC during the machining process of the current stamping workpiece under specified process conditions MTC and the upper limit β of the maximum thinning rate Δ MTN MTC ; MTN MTN ; (Ⅴ) Divide the area composed of all coordinate points in the filling area that satisfy Δ MTC > α MTC into a wrinkling area; (Ⅵ) Divide the area composed of all coordinate points satisfying Δ MTN >β MTN in the filling area into a fracture area.

5. The on-line monitoring method for the forming quality of stamping workpieces according to claim 4, characterized in that: The thickness change prediction model is obtained after model training using a prediction network built based on the Stacking ensemble learning framework; the base models in the prediction network are selected as XGBoost, LSTM, SVM, and KNN; the meta-model is selected as XGBoost; In the constructed prediction network, the input of each base model is the stamping depth h p and the process energy E, and the output is the predicted thickness change TV of each p,i , i = 1, …, 4; the input of the meta-model is the output of each base model, and the output of the meta-model is the finally predicted thickness change TV.

6. The on-line monitoring method for the forming quality of stamping workpieces according to claim 5, characterized in that: The training process of the prediction network includes two stages; In the first stage, by adopting the original data set composed of the stamping depth h p , the process energy E, and the thickness change amount TV as metadata, each base model is trained; In the second stage, the output results TV of each base model are used p,i , where i = 1, …, 4, and the fitting dataset consisting of the thickness change amount TV as metadata is used to train the meta-model.

7. An on-line monitoring tool for the forming quality of stamping workpieces, characterized in that: It adopts the online monitoring method for the forming quality of the stamping workpiece as described in any one of claims 1 to 5 to monitor the processing process of the stamping workpiece on the stamping equipment and generate a quality evaluation result of the processed target stamping workpiece; the online monitoring tool includes: A reference database that stores the process energy maps of each stamping workpiece under specified process conditions; A data acquisition unit, which is used to collect the stamping depth h of the stamping equipment in real time during the processing p and the stamping force F, and calculate the real-time process energy E according to the real-time stamping depth h real and the real-time stamping force F real collected during the processing real ; A trajectory generation unit, which is configured to query the reference database to obtain a process energy map corresponding to the current stamping workpiece under the current process conditions, and then, according to the real-time stamping depth h real and the real-time process energy E real during the processing of the current stamping workpiece, fit a corresponding state trajectory in the process energy map; A quality determination unit that is used to generate a corresponding quality evaluation result according to the position of the state trajectory in the process energy map: (ⅰ) When the state trajectory neither passes through the fracture zone nor terminates in the wrinkling zone, it is determined that the current stamping workpiece is completely qualified; (ⅱ) When the state trajectory only passes through the fracture zone, it is determined that the current stamping workpiece has cracked; (ⅲ) When the state trajectory only terminates in the wrinkling zone, it is determined that the current stamping workpiece has local wrinkling; (ⅳ) When the state trajectory passes through the fracture zone and terminates in the wrinkling zone, it is determined that the current stamping workpiece has cracked and has local wrinkling.

8. A visualization method for the forming quality of a stamping workpiece, characterized in that: It is used to evaluate the forming quality of the stamping workpiece during the processing of the target stamping workpiece in combination with the online monitoring method for the forming quality of the stamping workpiece as described in any one of claims 1 to 5, and visually present the change process of the forming quality of the stamping workpiece: I. Before processing the stamping workpiece: Query the process energy map corresponding to the current stamping workpiece according to the preset equipment processing parameters; II. During the processing of the stamping workpiece: (1) Present the first screen and the second screen according to the process energy map; Among them, the first screen uses the stamping depth h p as the abscissa and the process energy E as the ordinate; and the hue difference of the pixel points within the constraint region is used to characterize the corresponding maximum thickening rate Δ MTC value; the boundary of the wrinkling area is also demarcated in the first screen; Among them, the second screen uses the stamping depth h p as the abscissa and the process energy E as the ordinate; and the hue difference of the pixel points within the constraint region is used to characterize the corresponding maximum thinning rate Δ MTN value; the boundary of the fracture zone is also demarcated in the second screen; (2) Determine the real-time stamping depth h real and the real-time process energy E real corresponding to the machining process according to the feedback signal of the stamping equipment collected in real time; and use this as the stamping state variable U; (3) The position change of the stamping state variable U is displayed in real time in the first and second screens, and a motion trajectory Ut is fitted. III. After the stamping workpiece is processed: Using any one or more interaction means, present the evaluation result of the forming quality of the current stamping workpiece to the user.

9. A stamping device, which includes a device body and an interaction component, characterized in that, The stamping equipment further includes an online monitoring tool for the forming quality of the stamping workpiece as described in claim 7. The online monitoring tool is used to generate a corresponding quality evaluation result during the stamping process of any stamping workpiece according to the operation parameters of the equipment body collected in real time. The interaction component is used to present the forming quality of the stamping workpiece to the user in real time according to the visualization method for the forming quality of the stamping workpiece as described in claim 8.

10. The stamping device according to claim 9, characterized in that: The stamping equipment further includes a feedback and warning unit. After any stamping workpiece is processed, the feedback and warning unit obtains the quality evaluation result generated by the online monitoring tool. When the processed stamping workpiece is evaluated as in a non-fully qualified state, a driving instruction representing shutdown is sent to the main controller of the stamping equipment to drive the stamping equipment to stop; and before the process parameters are adjusted, the operation instructions of the equipment are refused to be responded to.

Images