CN116653346A

CN116653346A - Online monitoring method and tool for forming quality of stamping workpiece

Info

Publication number: CN116653346A
Application number: CN202310819891.2A
Authority: CN
Inventors: 黄海鸿; 甘雷; 李磊; 刘志峰
Original assignee: Hefei University of Technology
Current assignee: Hefei University of Technology
Priority date: 2023-07-05
Filing date: 2023-07-05
Publication date: 2023-08-29

Abstract

The invention belongs to the field of stamping forming process monitoring, and particularly relates to an online monitoring method and tool for forming quality of a stamping workpiece, a corresponding visualization method for forming quality and stamping equipment thereof. The online monitoring method comprises the following steps: s01: acquiring a process energy map of a current stamping workpiece; s02: collecting equipment parameters of a stamping workpiece in the processing process in real time; s03: calculating real-time process energy according to the real-time stamping depth and the real-time stamping force acquired in the processing process; s04: fitting a corresponding state track in the process energy map according to the real-time stamping depth and the state change of the real-time process energy; s05: judging whether the state track passes through the cracking zone and is stopped at the wrinkling zone, and further obtaining the evaluation result of the forming quality. The online monitoring tool is corresponding data processing equipment. The invention solves the problem that the prior art lacks an online evaluation means for the forming quality of the stamping workpiece.

Description

Online monitoring method and tool for forming quality of stamping workpiece

Technical Field

The invention belongs to the field of stamping forming process monitoring, and particularly relates to an online monitoring method and tool for stamping workpiece forming quality, a corresponding visualization method for stamping workpiece forming quality and stamping equipment thereof.

Background

In the stamping process, how to monitor the forming quality of the process has been a challenge for those skilled in the art. Because of the closeness of the stamping die, stamping workpiece processing is typically accomplished rapidly in a closed processing environment; the method is technically difficult to realize direct measurement of the forming quality of the stamping workpiece, and the forming quality of the stamping workpiece cannot be monitored on line.

In the existing stamping process flow, in order to ensure the production quality, a front-end production department often needs to process samples according to a preset production plan. And then sent to the QA department at the rear end, and quality inspectors detect the forming quality of the processed stamping workpiece sample. And then the checking result of the QA department is fed back to the front-end production department, and the production department decides whether the production process needs to be adjusted according to the quality checking result.

The manual quality inspection, feedback and modification process parameters are long in flow process, low in efficiency, time-consuming and labor-consuming, production efficiency and yield of enterprises are further reduced, and production cost of workpieces is improved. Therefore, how to design a scheme capable of monitoring the forming quality of the stamping workpiece in real time at the production end is becoming a technical problem to be solved by those skilled in the art.

Disclosure of Invention

In order to solve the problem that the prior art lacks means for on-line evaluation of the forming quality of a punched workpiece. The invention provides an on-line monitoring method and tool for forming quality of a stamping workpiece, a corresponding visualization method for forming quality of the stamping workpiece and stamping equipment thereof.

The invention is realized by adopting the following technical scheme:

an online monitoring method for forming quality of a stamping workpiece comprises the following steps:

s01: and obtaining a process energy map of the current stamping workpiece under the specified process condition.

The process energy map in the invention is pre-generated to the stamping depth h _p And the process energy E is respectively an abscissa and an ordinate, and is arbitraryDepth of press h _p And maximum thickening ratio delta under process energy E conditions _MTC And a maximum thinning rate delta _MTN And the coordinate points are state attributes, and the patterns of the wrinkling area and the cracking area range are divided according to the state attributes of the coordinate points.

S02: collecting equipment parameters of a stamping workpiece in the processing process in real time, wherein the equipment parameters comprise stamping depth h _p And a pressing force F.

S03: according to the real-time stamping depth h acquired in the processing process _real And a real-time punching force F _real Calculating the real-time process energy E _real ：

S04: according to the real-time stamping depth h of the current stamping workpiece processing process _real And real-time process energy E _real Fitting a corresponding state trace in the process energy map.

S05: the following judgment is made according to the position of the state track of the current stamping workpiece processing process in the process energy spectrum:

and (i) determining that the currently punched workpiece is fully qualified when the state trajectory neither passes through the rupture zone nor terminates in the crumpling zone.

(ii) determining that the currently punched workpiece is broken when the state trajectory passes through only the breaking zone.

(iii) determining that there is localized wrinkling of the currently punched workpiece when the state trajectory terminates only in the crumpling zone.

(iv) determining that the currently punched workpiece is broken and that there is localized wrinkling when the state trajectory passes through the breaking zone and terminates at the wrinkling zone.

The invention also comprises an on-line monitoring tool for the forming quality of the stamping workpiece, which adopts the on-line monitoring method for the forming quality of the stamping workpiece, monitors the processing process of the target stamping workpiece on the stamping equipment, and generates the quality evaluation result of the processed target stamping workpiece. The on-line monitoring tool includes: the system comprises a reference database, a data acquisition unit, a track generation unit and a quality judgment unit.

The reference database stores process energy maps of each stamping workpiece under specified process conditions.

The data acquisition unit is used for acquiring stamping depth h of stamping equipment in the processing process in real time _p And stamping force F, and according to the real-time stamping depth h acquired in the processing process _real And a real-time punching force F _real Calculating the real-time process energy E _real 。

The track generation unit is used for inquiring the reference database, acquiring a process energy map corresponding to the current stamping workpiece under the current process condition, and then according to the real-time stamping depth h of the current stamping workpiece in the processing process _real And real-time process energy E _real Fitting a corresponding state trace in the process energy map.

The quality judging unit is used for generating a corresponding quality evaluation result according to the position of the state track in the process energy map: and (i) determining that the currently punched workpiece is fully qualified when the state trajectory neither passes through the rupture zone nor terminates in the crumpling zone. (ii) determining that the currently punched workpiece is broken when the state trajectory passes through only the breaking zone. (iii) determining that there is localized wrinkling of the currently punched workpiece when the state trajectory terminates only in the crumpling zone. (iv) determining that the currently punched workpiece is broken and that there is localized wrinkling when the state trajectory passes through the breaking zone and terminates at the wrinkling zone.

The invention also comprises a visual method of the forming quality of the stamping workpiece, which is used for visually representing the change process of the forming quality of the stamping workpiece by combining the online monitoring method of the forming quality of the stamping workpiece in the processing process of the target stamping workpiece:

1. Before the stamping workpiece is processed:

inquiring the energy map of the corresponding process of the current stamping workpiece according to preset equipment processing parameters.

2. During the processing of the stamping workpiece:

(1) And presenting the first picture and the second picture according to the process energy map.

Wherein the first picture is pressed with a depth h _p The abscissa is the process energy E, and the ordinate is the ordinate; and representing the corresponding maximum thickening rate delta by hue difference of pixel points in the constraint area _MTC Is a value of (2). The first frame also defines the boundary of the crumpling zone.

Wherein the second picture is pressed with a depth h _p The abscissa is the process energy E, and the ordinate is the ordinate; and the corresponding maximum thinning rate delta is represented by the hue difference of the pixel points in the constraint area _MTN Is a value of (2). The boundary of the rupture zone is also divided in the second picture.

(2) According to the feedback signals of the stamping equipment acquired in real time, determining the real-time stamping depth h corresponding to the machining process _real And real-time process energy E _real The method comprises the steps of carrying out a first treatment on the surface of the This is used as a press state variable U.

(3) And displaying the position change of the stamping state variable U in the first picture and the second picture in real time, and fitting a motion track Ut.

3. After the stamping workpiece is machined:

and (3) presenting the evaluation result of the forming quality of the current stamping workpiece to a user by adopting any one or more interaction means.

A stamping apparatus includes an apparatus body and an interaction assembly. The stamping equipment also comprises an on-line monitoring tool for the forming quality of the stamping workpiece, wherein the on-line monitoring tool is used for generating a corresponding quality evaluation result in the stamping processing process of any stamping workpiece according to the operation parameters of the equipment body acquired in real time.

The interactive component is used for presenting the forming quality of the stamping workpiece to a user in real time according to the visual method of the forming quality of the stamping workpiece.

The technical scheme provided by the invention has the following beneficial effects:

the invention designs a method and a tool capable of carrying out on-line monitoring on the processing process of any stamping workpiece in stamping equipment based on the constructed process energy map. By utilizing the scheme of the invention, the state track of the processing process on the process energy map can be determined by collecting the stamping force and the stamping depth of the stamping equipment in real time, and then whether the stamping workpiece is wrinkled and broken defect is determined according to the position relation between the state track and the breaking area and the wrinkling area in the process energy map, so that the forming quality of the processed stamping workpiece is evaluated.

The online monitoring method for the forming quality of the stamping workpiece provided by the invention realizes the effect of synchronously evaluating the forming quality in the processing process. And the quality inspection flow in the large-scale stamping process can be greatly shortened, the working efficiency of the batch stamping process of the stamped workpieces is improved, and the workload and the cost of product quality inspection are reduced. Has extremely high practical value and economic value, and is suitable for popularization and application in industry.

Drawings

Fig. 1 is a flow chart of steps of a process energy spectrum generation method provided in embodiment 1 of the present invention.

Fig. 2 is a schematic diagram of a prediction network based on a Stacking integrated learning framework constructed in embodiment 1 of the present invention.

Fig. 3 is a schematic diagram of the process of constructing and training a prediction network in embodiment 1 of the present invention.

Fig. 4 is a block diagram of a process energy spectrum generation system for monitoring the forming quality of a stamping workpiece according to embodiment 2 of the present invention.

Fig. 5 is a flowchart of the steps of an online monitoring method for forming quality of a stamping workpiece according to embodiment 4 of the present invention.

Fig. 6 is a schematic diagram of an on-line monitoring tool for forming quality of a punched workpiece provided in embodiment 5 of the present invention.

Fig. 7 is a sample picture of a door panel in a class selected during the performance test stage.

FIG. 8 is a plot of data density scatter of the predicted results of thickness variation for each alternative model during performance testing. In FIG. 8, the portion (a) shows the maximum thinning rate Δ _MTN Is a prediction result and an error of the same; in FIG. 8, the portion (b) shows the maximum thickening ratio delta _MTC Is a prediction result and an error of the same. Color bars in the figure represent the density of the data; the dashed line in the figure is called zero error line table Showing that the measurement results coincide with the predicted results.

Fig. 9 is a pearson correlation coefficient distribution diagram of the predicted results of each alternative model during the performance test.

FIG. 10 is a plot of data density scatter of the predicted results of thickness variation for each base model during performance testing. In FIG. 10, the portion (a) shows the maximum thinning rate Δ _MTN Is a prediction result and an error of the same; in FIG. 10, the portion (b) shows the maximum thickening ratio delta _MTC Is a prediction result and an error of the same.

FIG. 11 is a graph of process energies constructed using the inventive protocol during performance testing.

FIG. 12 is a graph of process energy spectrum constructed by different schemes during performance testing versus thickness variation monitoring of training sets at different stamping depths; wherein, part (a) in fig. 12 shows the monitoring result of the thickness variation of the punched workpiece at the punching depth of 5 mm; part (b) in fig. 12 shows the monitoring result of the change in the thickness of the punched workpiece at a punching depth of 23 mm.

FIG. 13 is a graph of process energy spectra constructed in different schemes during performance testing versus thickness variation monitoring of stamped workpieces at 9.5mm, 15.5mm and 27.5mm stamping depths. Wherein, part (a) in fig. 13 shows the monitoring result of the thickness variation of the punched workpiece at the punching depth of 9.5 mm; part (b) in fig. 13 shows the monitoring result of the change in the thickness of the punched workpiece at a punching depth of 15.5 mm; part (c) in fig. 13 shows the monitoring result of the thickness change of the punched workpiece at 27.5 mm.

FIG. 14 shows the average absolute percentage error of the process energy spectrum constructed by different schemes versus the monitoring results at different punching depths during the performance test. Wherein, part (a) in fig. 14 shows the average absolute percentage error of the monitoring result of the maximum thinning rate of the punched workpiece; part (b) in fig. 14 shows the average absolute percentage error of the monitoring result of the maximum thickening ratio of the punched workpiece.

Fig. 15 shows the mean square error of the process energy spectrum constructed by different schemes versus the monitoring results at different stamping depths during the performance test. In fig. 15, (a) shows the mean square error of the monitoring result of the maximum thinning rate of the punched workpiece; part (b) in fig. 15 shows the mean square error of the monitoring result of the maximum thickening ratio of the press work.

FIG. 16 is a graph of process energy constructed based on thin-plate spline interpolation during performance testing.

Fig. 17 is a pearson correlation coefficient profile of process energy versus thickness variation at different stamping depths during performance testing.

Detailed Description

The present invention will be described in further detail with reference to the drawings and examples, in order to make the objects, technical solutions and advantages of the present invention more apparent. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the invention.

Example 1

The embodiment provides a process energy spectrum generation method for online monitoring of forming quality of a stamping workpiece, wherein the process energy spectrum is used for reflecting stamping depth h of a specified stamping workpiece in the stamping process _p And the mapping relation between the process energy E and the thickness variation TV. The image also uses different subareas to punch the workpiece at different punching depths h _p And the forming quality under the process energy E condition.

As shown in fig. 1, the method for generating an energy map of a process provided in this embodiment includes the following steps:

s1: measuring a plurality of equipment parameters and workpiece thickness D of each target stamping workpiece of the same type under the specified process conditions; the device parameters include: depth of press h _p And a pressing force F.

S2: respectively calculating the process energy E and the maximum thickening rate delta corresponding to each group of equipment parameters _MTC And a maximum thinning rate delta _MTN The method comprises the steps of carrying out a first treatment on the surface of the Wherein,,

in the above, h _max Representing the maximum stamping depth of equipment corresponding to the current target stamping workpiece; d (D) _i Punch representing specified measuring points in a punched workpieceThickness before pressing; d (D) _i+1 Representing the post-press thickness of a specified measurement point in the punched workpiece.

What needs to be specifically stated is: depth of press h _p Refers to the whole pressing depth of a pressing head of the stamping equipment in the stamping process; and maximum thickening ratio delta _MTC And a maximum thinning rate delta _MTN The deformation conditions of the stamping workpiece at different positions before and after stamping are detected, and the deformation conditions of the position with the largest thickening amplitude and the largest thinning degree on the same stamping workpiece are determined and calculated.

S3: stamping depth h corresponding to each group of equipment parameters _p The process energy E and the thickness variation TV as a set of metadata constitute the original dataset. Wherein the thickness variation TV includes a maximum thickening ratio delta _MTC And a maximum thinning rate delta _MTN 。

The original dataset a is represented as: a= (E, TV), where E is the measurement process energy matrix of jx 2:

E＝(h _p ,E _p )

in the above, h _p Is a J×1 stamping depth vector, h _p ＝[h _p,1 ,h _p,2 ,...,h _p,J ] ^-1 ；E _p Is a J×1 process energy vector, E _p ＝[E _p,1 ,E _p,2 ,...,E _p,J ] ^-1 The method comprises the steps of carrying out a first treatment on the surface of the TV is a j×2 thickness variation matrix expressed as:

TV＝(Δ _MTN ,Δ _MTC )

in the above, delta _MTN A maximum thinning rate vector of J×1, Δ _MTN ＝[Δ _MTN,1 ,Δ _MTN ,2,...,Δ _MTN,J ] ^-1 ；Δ _MTC Is the maximum thickening rate vector J multiplied by 1, delta _MTC ＝[Δ _MTC,1 ,Δ _MTC ,2,...,Δ _MTC,J ] ^-1 。

S4: selecting M machine learning models as alternative models to punch depth h _p And process energy E as input and thickness variation TV as output, pre-training and pre-testing the candidate model with partial metadata randomly selected from the original dataset. Alternative model package in this embodiment Includes SVM, RF, GBDT, XGBoost, KNN and LSTM.

S5: and carrying out correlation analysis and prediction accuracy evaluation on the prediction results of the alternative models in the pre-test stage.

Carrying out correlation analysis on the alternative model by adopting pearson correlation analysis; wherein, the pearson correlation coefficient r of the alternative model _xy The calculation formula of (2) is as follows:

in the above, TV _p,i,x With TV _p,i,y Thickness variation prediction results of the xth and the y alternative models are respectively obtained;and->Average values of thickness variation prediction results of the xth and the y alternative models respectively; i is the total number of predicted outcomes.

The prediction precision of each alternative model adopts average absolute percentage error e _MA And mean square error e _MS For evaluation, the calculation formula is as follows:

wherein c _i ' and c _i Respectively measuring and monitoring data; n is the number of monitored data.

S6: selecting n machine learning models with weak correlation as a base model, wherein n is less than M; and (3) taking the machine learning model with highest prediction precision as a meta model, and building a prediction network based on a Stacking integrated learning framework. The model architecture of the predictive network is shown in fig. 2.

Specifically, in the prediction network based on the Stacking integrated learning framework actually constructed in this embodiment, the base model selects XGBoost, LSTM, SVM and KNN, and the meta model selects XGBoost. In the built prediction network, the input of each base model is the stamping depth h _p And process energy E, output as respective predicted thickness variation TV _p,i I=1, …, n; the input of the metamodel is the output of each base model, and the output of the metamodel is the final predicted thickness variation TV.

S7: and (3) carrying out two-stage training on the prediction network by using the original data set, and reserving the prediction network after training as a required thickness variation prediction model.

As shown in fig. 3, the two-stage training process of the predictive network is as follows:

the first stage:

(1) Dividing the original data set into n sub-data sets, A ₁ ～A _n 。

(2) The n-1 sub-data sets are used as training sets, and the last one is used as testing set.

(3) The training set and the testing set are utilized to train and test each basic model in the first stage.

(4) According to the training result, n prediction results TV corresponding to each metadata in the original data set are obtained _p,1 ～TV _p,n 。

And a second stage:

(5) Predicting results TV of thickness change matrix TV and corresponding n base models in original dataset _p,1 ～TV _p,n Combined into new metadata.

(6) A new fitting dataset B is constructed with the new metadata: b= (TV) _p,1 ,…,TV _p,n TV), wherein TV _p,1 ～TV _p,n And J×2 thickness change matrices each composed of the prediction results of the 1 st to nth base models. The length of J is the group number of the equipment parameters.

(7) And dividing the fitting data set B into a training set and a testing set, and training and testing the meta-model in the second stage.

S8: at a stamping depth h _p And process energy E is respectively an abscissa and an ordinate to draw a blank process energy map, then a thickness variation prediction model is utilized to predict the thickness variation TV of each point in the process energy map, and the maximum thickening rate delta in the thickness variation TV is utilized _MTC Maximum thinning rate delta _MTN And its color map colors the empty process energy spectrum.

The process of drawing the energy spectrum is as follows:

s81: establish a stamping depth h _p The horizontal axis is the vertical axis of the process energy E, and only comprises a blank coordinate system of one quadrant and four quadrants.

S82: and setting the abscissa of any point in the fourth quadrant as the opposite number of the original value, and taking the axis of the ordinate as the Y axis, so that the same coordinate values in the two quadrants are symmetrical about the Y axis.

S83: according to the stamping depth h in the process condition of the target stamping workpiece _p And the constraint relation between the process energy E, and generating two closed areas to be filled symmetrical about the Y axis in a blank coordinate system.

S84: and taking the abscissa and the ordinate of any position in the region to be filled as input, and generating the thickness variation TV of the corresponding point by using the thickness variation prediction model.

S85: based on the preset color mapping relation, the maximum thickening rate delta contained in the thickness variation TV of each coordinate point is respectively generated _MTC And a maximum thinning rate delta _MTN And coloring the pixel points of the to-be-filled areas in the first quadrant and the fourth quadrant according to the corresponding color information.

In the process energy map designed in this embodiment, the maximum thickening ratio delta is based on _MTC Determining whether the workpiece is wrinkled or not and based on the maximum reduction rate delta _MTN And judging whether the workpiece is broken or not, wherein the maps are arranged in two symmetrical quadrants under the same coordinate system. In practice, the technician can also respectively present related map information through the two images; this is still part of the solution provided by the present invention.

The embodiment scheme is mainly by colorThe hue difference characterizes the maximum thickening rate delta corresponding to each coordinate point _MTC Or maximum thinning rate delta _MTN For example, the larger the value, the closer the pixel value of the corresponding coordinate point is to red, and the smaller the value, the closer the pixel value of the corresponding coordinate point is to purple.

Of course, in other embodiments, in order to exhibit a maximum thickening ratio delta _MTC Or maximum thinning rate delta _MTN The difference in the values of (a) may be otherwise applied to the maximum thickening ratio delta _MTC Or maximum thinning rate delta _MTN Visual display is performed, and further the later division of the areas and boundaries of the wrinkling area and the cracking area is facilitated. For example, in a three-dimensional coordinate system, the maximum thickening ratio delta _MTC Or maximum thinning rate delta _MTN As the data of two ends of the Z axis, a curved surface with concave and convex is constructed, and then different partitions are distinguished according to the peak and the valley of the curved surface.

S9: partitioning the colored process energy spectrum, and dividing the maximum thickening rate delta in the process energy spectrum _MTC Marking the areas corresponding to all the pixel points exceeding the wrinkling threshold as wrinkling areas; the maximum thinning rate delta in the process energy spectrum is calculated _MTN The areas corresponding to all the pixel points exceeding the rupture threshold are marked as rupture areas.

The process energy map is partitioned as follows:

s91: judging the maximum thickening rate delta corresponding to each pixel point in the colored process energy map _MTC Whether a preset wrinkling threshold is exceeded, and making the following decision:

(1) When the maximum thickening ratio delta _MTC And if the first state attribute of the current pixel point exceeds the crinkling threshold value, marking the first state attribute of the current pixel point as crinkling.

(2) When the maximum thickening ratio delta _MTC If the first state attribute is within the crinkling threshold, the first state attribute of the current pixel point is marked as 'safe'.

S92: judging the maximum thinning rate delta corresponding to each pixel point in the colored process energy map _MTN Whether a preset rupture threshold is exceeded, and making the following decision:

(1)when the maximum thinning rate delta _MTN And if the rupture threshold value is exceeded, marking the second state attribute of the current pixel point as 'rupture'.

(2) When the maximum thinning rate delta _MTN And if the second state attribute is within the rupture threshold, marking the second state attribute of the current pixel point as 'safe'.

S93: partitioning the colored region in the first quadrant and the fourth quadrant based on the first state attribute and the second state attribute, respectively:

dividing all coordinate points marked with crinkles in a first quadrant into crinkling areas, and dividing the rest coordinate points into safety areas; and dividing all coordinate points marked with 'rupture' in the fourth quadrant into rupture areas, and dividing the rest into safety areas.

Example 2

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

As shown in fig. 4, the process energy map generation system facing the punch workpiece forming quality monitoring in the present embodiment includes: the system comprises a data acquisition module, a data set generation module, a prediction network training module and a map drawing module.

The data acquisition module is used for acquiring equipment parameters and workpiece thickness D of the stamping equipment when the stamping equipment processes the stamped workpiece under the specified process conditions, and constraint conditions of the equipment parameters. The equipment parameters include the stamping depth h _p And a pressing force F. The data acquisition module also calculates corresponding process energy E and maximum thickening rate delta according to the equipment parameters _MTC And a maximum thinning rate delta _MTN 。

The data set generation module is used for generating stamping depth h corresponding to each group of equipment parameters _p The process energy E and the thickness variation TV as a set of metadata constitute the original dataset. Wherein the thickness variation TV includes a maximum thickeningRate delta _MTC And a maximum thinning rate delta _MTN 。

The prediction network training module is internally provided with a prediction network which takes XGBoost, LSTM, SVM and KNN as base models and XGBoost as meta model and is based on a Stacking integrated learning framework; the prediction network training module is used for performing two-stage training on the prediction network by utilizing the original data set, so as to obtain a stamping depth h _p And pressing force F as input, with maximum thickening ratio delta _MTC And a maximum thinning rate delta _MTN And predicting the model for the output thickness variation.

The map drawing module includes a white map generating unit, a pixel information generating unit, a coloring unit, an attribute marking unit, and a partitioning unit. The white graph generating unit is used for: first establish a stamping depth h _p The horizontal axis is the vertical axis of the process energy E, and only comprises a blank coordinate system of one quadrant and four quadrants. And setting the abscissa of any point in the fourth quadrant as the opposite number of the original value, and taking the axis of the ordinate as the Y axis, so that the same coordinate values in the two quadrants are symmetrical about the Y axis. And finally, generating two closed areas to be filled symmetrical about the Y axis in a blank coordinate system according to constraint conditions in process conditions of stamping the workpiece, and obtaining a blank process energy map. The pixel information generating unit is used for inputting the abscissa and the ordinate of each pixel point in the region to be filled into the thickness variation prediction model to generate a maximum thickening rate delta corresponding to each pixel point _MTC And a maximum thinning rate delta _MTN . The coloring unit is used for making the maximum thickening rate delta of each pixel point according to the preset color mapping relation _MTC And a maximum thinning rate delta _MTN Respectively converting the color values into corresponding color values; and then filling color pixels in the areas to be filled on two sides of the blank process energy map by utilizing the two color values respectively to obtain the color process energy map. The attribute marking unit is used for judging the maximum thickening rate delta of each pixel point in the color process energy map according to a preset wrinkling threshold value and a preset cracking threshold value _MTC And a maximum thinning rate delta _MTN The method comprises the steps of carrying out a first treatment on the surface of the And assign a corresponding first state attribute or second state attribute. The partition unit is used for generating a first pixel point according to the first pixel pointThe state attribute or the second state attribute partitions the color process energy spectrum, and divides the area where the pixel points exceeding the wrinkling threshold value and the cracking threshold value are located into a wrinkling area and a cracking area; thereby obtaining the required process energy spectrum.

Example 3

The present embodiment also includes a data processing apparatus comprising 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 process energy map generating system facing the forming quality monitoring of the stamping workpiece as in the embodiment 2 is created, and then the process energy map of the target stamping workpiece under the specified process condition is generated according to the collected equipment parameters and the constraint conditions of the target stamping workpiece under the specified process condition.

The data processing apparatus provided in this embodiment is essentially a computer apparatus, which may be an intelligent terminal, a tablet computer, a notebook computer, a desktop computer, a rack-mounted server, a blade server, a tower server, or a rack-mounted server (including an independent server, or a server cluster formed by a plurality of servers) capable of executing a program, or the like.

The computer device of the present embodiment includes at least, but is not limited to: a memory, a processor, and the like, which may be communicatively coupled to each other via a system bus.

In this embodiment, the memory (i.e., readable storage medium) includes flash memory, hard disk, multimedia card, card 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 a computer device, such as a hard disk or memory of the computer device. In other embodiments, the memory may also be an external storage device of a computer device, such as a plug-in hard disk, smart Media Card (SMC), secure Digital (SD) Card, flash Card (Flash Card) or the like, which are provided on the computer device. Of course, the memory may also include both internal storage units of the computer device and external storage devices. In this embodiment, the memory is typically used to store an operating system and various application software installed on the computer device. In addition, the memory can be used to temporarily store various types of data that have been output or are to be output.

The processor may be a central processing unit (Central Processing Unit, CPU), controller, microcontroller, microprocessor, or other data processing chip in some embodiments. The processor is typically used to control the overall operation of the computer device. In this embodiment, the processor is configured to execute the program code stored in the memory or process the data.

Example 4

Using the method of example 1, a process energy map for each stamped workpiece under the specified process conditions can be generated. The embodiment further provides an on-line monitoring method for forming quality of a stamping workpiece by utilizing the generated process energy spectrum, as shown in fig. 5, the method comprises the following steps:

The process energy map in the invention is pre-generated to the stamping depth h of the processing course _p And the process energy E is respectively an abscissa and an ordinate, and the stamping depth h is arbitrary _p And maximum thickening ratio delta under process energy E conditions _MTC And a maximum thinning rate delta _MTN And the coordinate points are state attributes, and the patterns of the wrinkling area and the cracking area range are divided according to the state attributes of the coordinate points.

S04: according to the real-time stamping depth h of the current workpiece processing process _real And real-time process energy E _real Fitting a corresponding state trace in the process energy map.

What needs to be specifically stated is: the process energy maps used in this embodiment should be pre-generated by a technician according to sample parameters of the target stamped workpiece under specified process conditions, each process energy map corresponding one-to-one to a particular type of stamped workpiece and its specified process conditions. Process conditions include material properties and equipment parameters; the material properties comprise material labels, shapes, specifications and the like of plates adopted in the processing process of the stamping workpiece; the equipment parameters refer to initial parameters preset by punching equipment in the processing process of the punched workpiece. When any one of the type of the punched workpiece and the process conditions is adjusted, the corresponding process energy map needs to be updated.

In step S01, the method for dividing the wrinkling region and the cracking region in the process energy spectrum is as follows:

at a stamping depth h _p The process energy E is plotted on the abscissa as the ordinate as at least one blank coordinate system.

(II) according to the stamping depth h in the current process conditions of stamping the workpiece _p Constraint relation between the energy E and the process energy E in a blank coordinate systemA closed region to be filled is generated.

And (III) generating a thickness variation TV corresponding to each coordinate point in the region to be filled by using a trained thickness variation prediction model.

The thickness variation prediction model is input as the stamping depth h _p And process energy E, output as thickness variation TV; the thickness variation TV includes a maximum thickening ratio delta _MTC And a maximum thinning rate delta _MTN 。

(IV) predetermining the maximum thickening rate delta of the current stamping workpiece in the processing process under the specified process conditions according to expert experience _MTC Upper limit alpha of (2) _MTC And a maximum thinning rate delta _MTN Upper limit beta of (2) _MTN 。

(V) meeting delta in the filled region _MTC ＞α _MTC The region constituted by all coordinate points of (a) is divided into crumple zones, and the boundary of the crumple zone and the remaining portion (safety zone) is formed.

(VI) meeting delta in the filled region _MTN ＞β _MTN The region constituted by all coordinate points of (a) is divided into a fracture region, and a boundary of the fracture region and the remaining portion (safety region) is formed.

When the online monitoring function of the forming quality of the stamping workpiece is realized, the defect partition information in the process energy spectrum is mainly utilized, so that very accurate crinkling areas and boundaries of the rupture areas and the safety areas are required. Color information in the map is not needed in the stamping workpiece forming quality monitoring process, and the color information is used for facilitating users to intuitively know different partition boundaries.

In a stamping workpiece forming quality monitoring process, a monitored state trajectory of the stamping process as it passes through a fracture zone is typically indicative of the stamping workpiece having been fractured. Since the fracture is of an irreversible defect type, the fracture defect still exists even if the state trajectory eventually passes through the fracture zone. When the monitored state trajectory of the stamping process passes through the crumpling zone, it is generally indicative that crumpling has occurred at the current time of stamping the workpiece. However, since the wrinkling is a reversible defect type, it is generally considered that the stamped workpiece is eventually free of wrinkling defects as long as the state trajectory eventually passes through the wrinkling zone. The summary is: judging whether the stamping workpiece is broken only needs to analyze whether the state track of the machining process passes through the breaking area, and judging whether the stamping workpiece is wrinkled needs to analyze whether the end point of the machining track falls into the wrinkling area.

Example 5

Based on the on-line monitoring method for forming quality of the stamping workpiece provided in embodiment 4, this embodiment also provides an on-line monitoring tool for forming quality of the stamping workpiece, which is used for monitoring the processing process of the target stamping workpiece on the stamping device by adopting the on-line monitoring method for forming quality of the stamping workpiece in embodiment 4, and generating the quality evaluation result of the processed target stamping workpiece.

As shown in fig. 6, the online monitoring tool provided in this embodiment includes: the system comprises a reference database, a data acquisition unit, a track generation unit and a quality judgment unit.

The reference database stores process energy maps of each stamping workpiece under specified process conditions. Each process energy map comprises a label representing the type and process condition of the punched workpiece corresponding to the current process energy map.

The track generation unit is used for inquiring the reference database and acquiring a process energy map of the label information corresponding to the type of the current stamping workpiece and the process conditions thereof. Then according to the real-time stamping depth h of the current stamping workpiece processing process _real And real-time process energy E _real Fitting a corresponding state trace in the process energy map.

Example 6

The method for visualizing the forming quality of the stamping workpiece provided in this embodiment is used in combination with the method for online monitoring the forming quality of the stamping workpiece provided in embodiment 4 in the target stamping workpiece processing process, to evaluate the forming quality of the stamping workpiece and to visualize the quality evaluation process, and specifically comprises the following steps:

1. before the stamping workpiece is processed:

and inquiring the energy map of the corresponding process of the current stamping workpiece according to preset processing parameters.

During each stamping workpiece machining process, the type of the machined stamping workpiece and specific machining process conditions are input into the stamping equipment by a technician, and the stamping equipment can retrieve corresponding process energy maps generated by pre-testing from a reference database.

2. During the stamping process of the stamping workpiece:

(2) Determining real-time stamping corresponding to the processing process according to the feedback signals of the stamping equipment acquired in real timeDepth h _real And real-time process energy E _real The method comprises the steps of carrying out a first treatment on the surface of the This is used as a press state variable U.

(3) Displaying the position change of the stamping state variable U in the first picture and the second picture in real time, and fitting a motion track Ut; the motion track is a state track reflecting the forming quality in the processing process of the stamping workpiece.

In the actual machining process, the stamping equipment determines the stamping depth h of the machining process according to the feedback information of the related sensor of the machining process _p And the pressing force F, and the corresponding process energy E is calculated according to the pressing force F. Stamping equipment records real-time stamping depth h _real And real-time process energy E _real And displays it on the process energy map to form a "status point". With the continuous development of the processing process, the stamping depth h _p And the process energy E is changed continuously, the state point is also moved continuously on the process energy map, and a corresponding 'state track' can be obtained by recording the displacement change of the state point. The forming quality of the current stamping workpiece can be determined by combining the relation between the state track and the rupture zone and the wrinkling zone in the process energy map.

5. After the processing of the stamping workpiece is finished, any one or more interaction means are adopted, and the evaluation result of the forming quality of the current stamping workpiece is presented to a user.

Example 7

The embodiment provides stamping equipment, which comprises an equipment body and an interaction component. The interaction component comprises a display, a loudspeaker and an input device; the display is used for displaying image interaction information representing the forming quality of the stamping workpiece in the processing process; the loudspeaker is used for playing audio interactive information representing the forming quality of the stamping workpiece in the processing process; the input device is used for inputting manual operation instructions to the stamping device.

The stamping equipment also comprises an on-line monitoring tool for the forming quality of the stamping workpiece as in the embodiment 5, and the on-line monitoring tool is used for generating a corresponding quality evaluation result in the stamping processing process of any stamping workpiece according to the operation parameters of the equipment body acquired in real time.

The interactive assembly was used in the method of visualizing the forming quality of a stamped workpiece according to example 6, presenting the forming quality of the stamped workpiece to a user in real time.

In the more optimized scheme of the embodiment, the stamping equipment further comprises a feedback early warning unit, wherein the feedback early warning unit acquires a quality evaluation result generated by the online monitoring tool after the processing of any stamping workpiece is completed, and sends a drive instruction representing shutdown to a main controller of the stamping equipment when the processed stamping workpiece is evaluated to be in a non-fully qualified state, so that the stamping equipment is driven to shutdown; and refusing to respond to the operation instruction of the equipment before the preset processing parameters of the equipment are not adjusted.

Performance testing

In order to verify the method for generating the process energy spectrum for monitoring the forming quality of the stamping workpiece provided in the embodiment 1 of the invention, and the effectiveness of the overall scheme for monitoring the forming quality of the stamping process by utilizing the generated process energy spectrum. The embodiment also carries out an actual stamping experiment on the scheme, and the experimental process is as follows:

1. Construction of the Process energy Spectrum of the present embodiment

1.1, the experiment uses the door inner plate shown in figure 7 as a stamping workpiece, and adopts AA5052 aluminum alloy as a material thereof, so as to construct a process energy spectrum. The profile of the process energy map uses 16 sets of punch speed and blank holder force combinations to form sample parameters as the punch process conditions. The abscissa of the process energy spectrum is the stamping depth, the stamping depth of 1mm to 28mm is selected as the value range of the abscissa of the process energy spectrum at intervals of 1mm, so that the stability and the applicability of the monitoring precision under different stamping depths can be verified conveniently. In addition, each stamping experiment was lubricated with 16ml of lubricating oil.

And 1.2, after process energy and thickness change data acquisition is completed, starting to construct a prediction network based on a Stacking integrated learning framework. In this embodiment, a more commonly used machine learning model such as a support vector machine (Support Vector Machines, SVM), extreme gradient lifting (EXtreme Gradient Boosting, XGBoost), gradient lifting decision tree (Gradient Boosting Decision Tree, GBDT), random Forest (RF), K-nearest neighbor algorithm (K-Nearest Neighbor Classification, KNN) and Long Short-Term Memory (LSTM) are selected as the candidate models. The principle profile of these alternative models is shown in table 1.

Table 1: brief description of principles of alternative models

To fairly compare the predicted performance of the candidate models, their superparameters are set to default values as follows during the training process.

Table 2: super-parameters of each model and prediction errors thereof

1.3, in this embodiment, the relevant samples obtained by the pre-stamping experiment are measured, so as to obtain the original data, and the required original data set is formed. 70% of the original data set is randomly extracted as a training set, and the other 30% is used as a test set to obtain the prediction results of different alternative models, and the prediction performance of the different alternative models is evaluated.

First, average absolute percentage error e is adopted respectively _MA And mean square error e _MS To evaluate the prediction accuracy of each candidate model. The prediction results of the alternative models are shown in fig. 8.

It can be observed in connection with fig. 8 that: compared with other alternative models, the thickness change prediction result of the XGBoost is closer to a zero error line, and the error also shows that the XGBoost has the smallest prediction error, so that the XGBoost is selected as a meta-model based on the Stacking integrated learning thickness change prediction model.

And after the prediction results of the alternative models are obtained, carrying out model correlation analysis through Pearson correlation analysis. The correlation coefficients of the alternative models are shown in fig. 9.

As can be seen from fig. 9: the correlation coefficients between XGBoost, GBDT and RF are all over 0.98, because the three machine learning models are all decision tree-based models in general, though they are somewhat different in principle, and the data prediction mechanisms have strong similarity. The modeling principle of LSTM, SVM, KNN is greatly different from the prediction mechanism, so that the correlation of the prediction result is also low. The 4 alternative models were chosen XGBoost, LSTM, SVM, KNN as base models in this experiment.

1.4 the original data set can be equally divided into 4 sub data sets { A }, based on the number of base models _i I=1,..4 } was used for training of the base model. The measured process energy and thickness variation data for each sub-data set containing 7 press depth correspondences are shown in table 3.

Table 3: process energy and thickness variation sub-data set and corresponding stamping depth thereof

Each base model was trained 4 times with the sub-data set. During each training process, one of the sub-data sets is taken as the test set and the remaining three sub-data sets are taken as training sets. SVM, XGBoost, LSTM and KNN predictive data sets are respectively TV _p,1 、TV _p,2 、TV _p,3 With TV _p,4 As shown in fig. 10.

TV set _p,1 、TV _p,2 、TV _p,3 With TV _p,4 As the input of the meta-model, XGBoost, the thickness variation dataset is measured as the output of XGBoost, and the meta-model is trained to predict the final thickness variation.

And then filling, coloring and dividing the quality area of the process energy map according to the predicted thickness change data, and finally obtaining the constructed process energy map as shown in figure 11.

2. Arrangement of control group and control experiment

2.1 Process conditions

Because the selected inner plate of the vehicle door has different forming shapes under different stamping depths, the thickness variation monitoring is carried out on the inner plate of the vehicle door with the stamping depths of 9.5mm, 15.5mm and 27.5mm in the experiment so as to verify the effectiveness of the constructed process energy map. At each stamping depth, the process energy and thickness variation of the stamped workpiece at 32 sets of process conditions as shown in table 4 were measured. In addition, the thickness change monitoring is carried out on the similar door plates with the stamping depth of 5mm and 23mm so as to explore the accuracy of the constructed process energy map on the training set data.

Table 4: monitoring stamping process conditions of a stamped workpiece

2.2 control group settings

In the verification stage, in order to embody the effectiveness and advantages of the process energy spectrum constructed in the embodiment, different prediction methods or models are also used for constructing the process energy spectrum for comparison analysis.

These prediction methods or models fall into 3 categories, namely data interpolation, machine learning models, and Stacking ensemble learning prediction models using different base models in combination with metamodels. The data interpolation is directly performed by using a thin-plate spline data interpolation technology, the machine learning model is performed by using XGBoost, LSTM and SVM, and the base model and the meta model selected from the prediction model of the Stacking integrated learning are shown in the table 5. BC1 to BC4 are mainly used for exploring the influence of a base model on the monitoring precision of a process energy spectrum constructed based on Stacking integrated learning, and M-SVM, M-LSTM and M-KNN are mainly used for exploring the influence of a meta model.

Table 5: combination of base model and meta model

3. Comparison of the Performance of the different schemes of the scheme and the control group

3.1, monitoring the thickness variation of the process energy spectrum of different schemes

The monitoring results of the thickness variation of the training set, namely, the thickness variation of the punched workpiece at the punching depth of 5mm and 23mm are shown in fig. 12, and it can be observed that the monitoring results of the process energy spectrum (i.e., SML in fig. 12) constructed by the scheme of the present embodiment are very consistent with the measurement results. The monitoring results of the thickness change of the punched workpiece with the punching depth of 9.5mm, 15.5mm and 27.5mm are shown in fig. 13, and the monitoring results of the SML still coincide with the measurement results. The average absolute percentage error and mean square error of the SML monitoring results are shown in FIG. 14 and FIG. 15, respectively, which are not more than 5.03% and 0.94%, respectively ² 。

The above data demonstrates that: the process energy spectrum constructed by the scheme provided by the invention has higher monitoring precision and stability.

Further, as shown in part (c) of fig. 13, 14 out of the 32 press works formed are broken, that is, there are 14 measured maximum thinning rates falling in the broken area. And 16 broken stamping workpieces are identified by the process energy spectrum constructed by the scheme of the invention. The 14 monitored maximum thinning rates of SML fall within the fracture zone, with only two unbroken punched workpieces being erroneously identified as broken by the process energy map, i.e., the gray circles in fig. 13 (c) are circled out. The process energy spectrum cracking recognition accuracy rate constructed by the scheme of the invention is as high as 93.75%. Similarly, 10 out of 32 formed stamping workpieces are wrinkled, 12 process energy maps constructed by the scheme of the invention are identified, and the wrinkling identification accuracy is 93.75%.

The above data demonstrates that: the process energy spectrum constructed by the invention can effectively and accurately identify wrinkling and cracking defects in the stamping workpiece.

The monitoring results of the process energy maps constructed by different prediction models are shown in fig. 12 and 13, and for the convenience of observation, the monitoring results of the process energy maps are shown by two graphs. The monitoring result of the process energy spectrum constructed by different prediction models can be seen The average absolute percentage error e of the monitoring results of the process energy maps is far from the measuring results _MA And mean square error e _MS As shown in fig. 14 and 15, the errors are higher than the errors of the process energy spectrum monitoring results constructed by the scheme. The process energy spectrum constructed by different prediction models has better monitoring precision and stability to the thickness change of the test set, and the maximum average absolute percentage error e _MA And maximum mean square error e _MS 6.64% (i.e. the result of monitoring the maximum reduction rate of a punched workpiece having a punching depth of 5mm by using a process energy spectrum constructed by LSTM) and 0.36%, respectively ² (i.e., the monitoring of the maximum thickening ratio of a punched workpiece having a punching depth of 23mm using a process energy map constructed by an SVM).

And the maximum average absolute percentage error e of the process energy spectrum constructed by different prediction models to the monitoring result of the thickness change of the stamping workpiece when the stamping depth is 9.5mm, 15.5mm and 27.5mm _MA And mean square error e _MS 13.66% (i.e. the result of monitoring the maximum thickening ratio of the process energy spectrum constructed by the SVM to the punched workpiece with the punching depth of 27.5 mm) and 7.51%, respectively ² (namely, the monitoring result of the maximum thinning rate of the stamping workpiece with the stamping depth of 27.5mm by using the process energy spectrum constructed by the SVM) is obviously higher than the error of the monitoring result of the process energy spectrum constructed by the scheme of the invention.

The above data demonstrates that: the process energy spectrum construction method for constructing the scheme provided by the invention obviously improves the monitoring precision and stability of the process energy spectrum under different process conditions.

3.2, comparing the monitoring precision of the map constructed by the scheme of the invention and the data interpolation

As can be seen from fig. 14 and 15: the monitoring error of the process energy spectrum constructed by the thin plate spline interpolation TPS on the thickness change of the stamping workpiece with different stamping depths is higher than that of the process energy spectrum constructed by the method, and the reason can be attributed to the prediction mechanism of the data interpolation technology. In data interpolation, the approximate interpolation curved surface needs to pass through all data points, namely data in a training set, so that a prediction result is easily affected by an extreme point.

Due to the high variability of the thickness variation of the punched workpiece and the process energy data, the data interpolation technology cannot realize accurate thickness variation data prediction, so that the thickness variation data predicted by thin plate spline interpolation TPS has large fluctuation and irregularity. The process energy spectrum constructed by thin-plate spline interpolation TPS is shown in fig. 16, and it can be seen that there is saw-tooth-like fluctuation in the process energy spectrum, and the boundary of the wrinkling region is also quite uneven, meaning that the data fluctuation in the process energy spectrum is large and irregular. These fluctuating, irregular predicted data can lead to reduced accuracy in monitoring unknown data, namely, data of thickness variations of the punched workpiece at punching depths of 9.5mm, 15.5mm and 27.5mm, by the process energy spectrum constructed by interpolation TPS of thin-plate splines. The uneven quality area boundary also causes the defect recognition accuracy of the constructed process energy spectrum to be reduced, as shown in (c) of fig. 13, the crack recognition accuracy and the wrinkle recognition accuracy of the process energy spectrum constructed by thin-plate spline interpolation TPS are respectively 53.13% and 40.63%, which are far lower than the defect recognition accuracy of the process energy spectrum constructed by the method of the invention.

Comparing fig. 11 and 16 can be found that: the process energy spectrum constructed by the scheme of the invention has smoother color change and more uniform quality area boundary. The solution provided by the invention has such obvious advantages mainly because: the invention can integrate the advantages of each machine learning model so as to reduce fluctuation caused by data variability, thereby realizing more accurate process energy spectrum construction.

3.3, comparing the monitoring precision of the atlas constructed by the scheme of the invention and the machine learning model

As shown in fig. 14 and 15, even with XGBoost with the lowest prediction error in the machine learning model, the difference in the monitoring error of the process energy spectrum constructed on the thickness variation at different stamping depths is very obvious. The minimum monitoring error of the process energy spectrum constructed by XGBoost is 1.30 percent, namely the average absolute percentage error of the maximum thinning rate monitoring result of the punched workpiece with the punching depth of 23mm, and the maximum monitoring error is 9.42 percent, namely the average absolute percentage error of the maximum thinning rate monitoring result of the punched workpiece with the punching depth of 15.5 mm.

The reasons for the unstable monitoring accuracy of the process energy spectrum constructed by the machine learning model may be derived from the variation of the probability density distribution of the thickness variation data. In building a predictive model using a machine learning model, one key assumption is that the predictive data is from the same probability density distribution as the training data. If the distribution of training data is different from the probability density distribution of prediction data, the prediction accuracy of the machine learning model may be degraded. The data for process energy and thickness variation are highly variable as different shapes are formed at different stamping depths. Different machine learning models have different adaptations to different data sets. For example, decision tree-based machine learning models, such as XGBoost, are good at learning data laws with complex relationships, while other machine learning models may be good at learning data laws with linear relationships.

Since low correlation between machine learning model input (i.e., process energy) and output (i.e., thickness variation) also results in reduced prediction accuracy of the machine learning model, variation in correlation between process energy and thickness variation can also result in monitoring instability.

Fig. 17 further counts pearson correlation coefficients of process energy versus thickness variation at different press depths, as can be seen in conjunction with fig. 17: at a stamping depth of 5mm, the correlation between process energy and thickness variation is lower than at a stamping depth of 23mm, resulting in lower monitoring accuracy at a stamping depth of 5 mm. The reason for this phenomenon is found from fig. 14 that the reason for the low correlation of process energy and thickness variation at a punching depth of 5mm is that the deformation of the sheet material is small at this punching depth, resulting in small thickness variation. Smaller thickness variations are more susceptible to disturbances in the measurement process, such as measurement errors, resulting in lower correlation of thickness variations with process energy.

Therefore, the reason why the process energy spectrum constructed by the scheme of the invention has high precision is mainly in two aspects: on one hand, the scheme effectively neutralizes the deviation of each base model by weighting the prediction result of each machine learning model, namely the base model. A machine learning model can be seen as finding the best hypothesis, i.e. the predicted value, in some hypothesis space. When the amount of training data available is too small compared to the size of the hypothesis space, the machine learning model may find many different hypotheses in the hypothesis space, all with the same accuracy on the predicted data. By constructing an aggregate from all of these predictors for each machine learning model, the prediction model constructed based on Stacking ensemble learning may reduce the risk of choosing inaccurate predictors. On the other hand, many machine learning models work by performing some form of local search, which may be trapped in a local optimum. For example, neural network-based machine learning models, such as LSTM, employ gradient descent methods to minimize the error function of training data, while decision tree-based machine learning models, such as RF, employ greedy segmentation rules to grow the decision tree. A set constructed by running a local search from many different starting points may provide a better approximation to the measurement results, thereby reducing the risk of trapping local optima.

3.4, comparing the monitoring precision of the atlas constructed by the invention and similar schemes under the combination of different base models and meta models

As shown in fig. 14 and 15, the monitoring error of the process energy spectrum constructed by Stacking ensemble learning under different base model and meta model combinations is generally lower than that constructed by data interpolation and machine learning models. The method shows that the monitoring precision and stability of the process energy spectrum can be effectively improved by constructing the thickness variation prediction model through Stacking integrated learning. However, the monitoring errors of the process energy maps constructed by Stacking ensemble learning under the combination of different base models and meta-models are still higher than those constructed by applying the scheme of the invention.

When the prediction performance of each base model is similar, that is, when the prediction mechanisms of the base models are similar, the prediction performance of the integrated prediction model is difficult to improve regardless of the combination of the base models. In BC1 and BC3, the prediction mechanisms of XGBoost, RF and GBDT are similar, which when combined, result in redundancy of the base model and bias of the meta-model, i.e., the final prediction result is more biased toward that of XGBoost, etc. The low prediction accuracy of the Stacking ensemble learning prediction model constructed by BC2 and BC4 may be due to the low diversity of the base model, which makes it difficult for the Stacking ensemble learning prediction model to observe and learn the intrinsic law of data changes from multiple angles. For example, only two machine learning models are included in BC2, resulting in a lower predictive performance for the Stacking ensemble learning predictive model constructed with BC2 than for other Stacking ensemble learning predictive models. Neither BC2 nor BC4 contained XGBoost with optimal prediction performance, resulting in lower prediction performance for the Stacking ensemble learning prediction model built with BC2 and BC 4.

And a machine learning model with low generalization capability is used as a meta-model, so that overfitting is easy to cause, and the prediction precision of the constructed Stacking integrated learning prediction model is reduced. From fig. 8 it can be seen that XGBoost has good generalization ability. The prediction result of the base model can be effectively generalized, and the final prediction result is prevented from biasing the result of a certain base model. Therefore, the monitoring error of the process energy spectrum constructed by the Stacking integrated learning prediction model taking XGBoost as the meta-model is lower than that of the process energy spectrum constructed by the Stacking integrated learning prediction model taking SVM, LSTM and KNN as the meta-model.

To sum up: the process energy spectrum constructed by the Stacking integrated learning prediction model provided by the invention is more accurate than the process energy spectrum constructed by data interpolation, machine learning models and the like. Meanwhile, the invention selects the machine learning model with low correlation as the base model and the machine learning model with high generalization as the meta model, thereby further improving the monitoring precision and stability of the constructed process energy spectrum. The invention also realizes on-line monitoring of the forming quality of the stamping workpiece in the stamping process by applying the process energy map constructed by the invention, and provides a means for on-line evaluating the forming quality of the stamping workpiece.

The foregoing description of the preferred embodiments of the invention is not intended to be limiting, but rather is intended to cover all modifications, equivalents, and alternatives falling within the spirit and principles of the invention.

Claims

1. The on-line monitoring method for the forming quality of the stamping workpiece is characterized by comprising the following steps of:

s01: acquiring a process energy map of a current stamping workpiece under a specified process condition;

the process energy spectrum is measured by the stamping depth h _p And the process energy E is respectively an abscissa and an ordinate, and the stamping depth h is arbitrary _p And maximum thickening ratio delta under process energy E conditions _MTC And a maximum thinning rate delta _MTN The state attribute of the coordinate point; dividing the range of a wrinkling area and a cracking area in the process energy spectrum according to the state attribute of the coordinate point;

s02: collecting equipment parameters of a stamping workpiece in the processing process in real time, wherein the equipment parameters comprise stamping depth h _p And a pressing force F;

S04: according to the real-time stamping depth h of the current stamping workpiece processing process _real And real-time process energy E _real Fitting a corresponding state track in the process energy map;

when the state track does not pass through the rupture zone and does not end at the wrinkling zone, judging that the currently punched workpiece is completely qualified;

(ii) determining that the currently punched workpiece is broken when the state trajectory passes through only the breaking zone;

(iii) determining that there is localized wrinkling of the currently punched workpiece when the state trajectory terminates only in the crumpling zone;

2. The method for on-line monitoring of forming quality of a punched workpiece according to claim 1, wherein: in the step S01, the process energy map is pre-generated by a technician according to sample parameters of a target stamping workpiece under specified process conditions; each process energy map corresponds to a specific type of stamping workpiece one by one and designated process conditions; the process energy spectrum is measured by the stamping depth h _p And the process energy E is respectively an abscissa and an ordinate, and the stamping depth h is arbitrary _p And maximum thickening ratio delta under process energy E conditions _MTC And a maximum thinning rate delta _MTN The state attribute of the coordinate point; and dividing the range of the wrinkling area and the cracking area in the process energy spectrum according to the state attribute of the coordinate point.

3. The method for on-line monitoring of forming quality of a punched workpiece according to claim 2, wherein: the process conditions include material properties and equipment parameters; the material properties comprise material labels, shapes and specifications of plates adopted in the processing process of the stamping workpiece; the equipment parameters refer to initial parameters preset by stamping equipment in the process of stamping workpiece processing;

when any one of the type of the punched workpiece and the process conditions is adjusted, the corresponding process energy map needs to be updated.

4. The method for on-line monitoring of forming quality of a punched workpiece according to claim 1, wherein: in step S01, the method for dividing the wrinkling region and the cracking region in the process energy spectrum is as follows:

at a stamping depth h _p Drawing at least one blank coordinate system by taking the process energy E as an abscissa and the process energy E as an ordinate;

(II) according to the current punchStamping depth h in the process conditions of pressing the workpiece _p And the constraint relation between the process energy E, and generating a closed region to be filled in a blank coordinate system;

(III) generating a thickness variation TV corresponding to each coordinate point in the region to be filled by using a trained thickness variation prediction model;

The thickness variation prediction model is input into a stamping depth h _p And the process energy E, output as a thickness variation TV including a maximum thickening ratio delta _MTC And a maximum thinning rate delta _MTN ；

(IV) predetermining the maximum thickening rate delta of the current stamping workpiece in the processing process under the specified process conditions according to expert experience _MTC Upper limit alpha of (2) _MTC And a maximum thinning rate delta _MTN Upper limit beta of (2) _MTN ；

(V) meeting delta in the filled region _MTC ＞α _MTC Dividing the region formed by all coordinate points into wrinkling regions;

(VI) meeting delta in the filled region _MTN ＞β _MTN The region constituted by all coordinate points of (c) is divided into fracture regions.

5. The method for on-line monitoring of forming quality of a punched workpiece according to claim 4, wherein: the thickness variation prediction model is obtained after model training by adopting a prediction network built on the basis of a Stacking integrated learning framework; selecting XGBoost, LSTM, SVM and KNN of a base model in the prediction network; selecting XGBoost by the meta-model;

in the built prediction network, the input of each base model is the stamping depth h _p And process energy E, output as respective predicted thickness variation TV _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 final predicted thickness variation TV.

6. The method for on-line monitoring of forming quality of a punched workpiece according to claim 5, wherein: the training process of the prediction network comprises two stages;

the first stage is realized by adopting a stamping depth h _p Training each base model by taking the process energy E and the thickness variation TV as an original data set formed by metadata;

the second stage adopts the output result TV of each base model _p,i The meta model is trained by using i=1, …,4 and the thickness variation TV as fitting data sets of metadata.

7. An on-line monitoring instrument of punching press work piece shaping quality which characterized in that: the method for monitoring the forming quality of the stamping workpiece on line is characterized in that the processing process of the stamping workpiece on stamping equipment is monitored by adopting the method for monitoring the forming quality of the stamping workpiece on line according to any one of claims 1 to 5, and a quality evaluation result of the processed target stamping workpiece is generated; the on-line monitoring tool includes:

the reference database stores process energy maps of all stamping workpieces under specified process conditions;

the data acquisition unit is used for acquiring stamping depth h of stamping equipment in the machining process in real time _p And stamping force F, and according to the real-time stamping depth h acquired in the processing process _real And a real-time punching force F _real Calculating the real-time process energy E _real ；

The track generation unit is used for inquiring the reference database, acquiring a process energy map corresponding to the current stamping workpiece under the current process condition, and then according to the real-time stamping depth h of the current stamping workpiece in the processing process _real And real-time process energy E _real Fitting a corresponding state track in the process energy map;

the quality judging unit is used for generating a corresponding quality evaluation result according to the position of the state track in the process energy map: when the state track does not pass through the rupture zone and does not end at the wrinkling zone, judging that the currently punched workpiece is completely qualified; (ii) determining that the currently punched workpiece is broken when the state trajectory passes through only the breaking zone; (iii) determining that there is localized wrinkling of the currently punched workpiece when the state trajectory terminates only in the crumpling zone; (iv) determining that the currently punched workpiece is broken and that there is localized wrinkling when the state trajectory passes through the breaking zone and terminates at the wrinkling zone.

8. A visual method for forming quality of a stamping workpiece is characterized by comprising the following steps: the method is used for evaluating the forming quality of the stamping workpiece in the process of processing the target stamping workpiece by combining the online monitoring method of the forming quality of the stamping workpiece according to any one of claims 1 to 5, and visually presenting the change process of the forming quality of the stamping workpiece:

1. Before the stamping workpiece is processed:

inquiring a process energy map corresponding to the current stamping workpiece according to preset equipment processing parameters;

2. during the processing of the stamping workpiece:

(1) Presenting a first picture and a second picture according to the process energy map;

wherein the first picture is pressed with a depth h _p The abscissa is the process energy E, and the ordinate is the ordinate; and representing the corresponding maximum thickening rate delta by hue difference of pixel points in the constraint area _MTC Is a value of (2); the first picture is also divided into boundaries of wrinkling areas;

wherein the second picture is pressed with a depth h _p The abscissa is the process energy E, and the ordinate is the ordinate; and the corresponding maximum thinning rate delta is represented by the hue difference of the pixel points in the constraint area _MTN Is a value of (2); the boundary of the rupture zone is also divided in the second picture;

(2) According to the feedback signals of the stamping equipment acquired in real time, determining the real-time stamping depth h corresponding to the machining process _real And real-time process energy E _real The method comprises the steps of carrying out a first treatment on the surface of the Taking the stamping state variable U as a stamping state variable U;

(3) Displaying the position change of the stamping state variable U in the first picture and the second picture in real time, and fitting a motion track Ut;

3. after the stamping workpiece is machined:

9. A stamping device comprising a device body and an interaction assembly, wherein the stamping device further comprises an on-line monitoring tool for the forming quality of the stamping workpiece according to claim 7, and the on-line monitoring tool is used for generating a corresponding quality evaluation result in the stamping process of any stamping workpiece according to the operation parameters of the device body acquired in real time;

the interactive assembly is used for presenting the forming quality of the punched workpiece to a user in real time according to the visualization method of the forming quality of the punched workpiece.

10. The stamping apparatus of claim 9, wherein: the stamping equipment also comprises a feedback early warning unit, wherein the feedback early warning unit acquires a quality evaluation result generated by the online monitoring tool after the processing of any stamping workpiece is completed, and sends a drive instruction representing shutdown to a main controller of the stamping equipment when the processed stamping workpiece is evaluated to be in a non-fully qualified state, so as to drive the stamping equipment to shutdown; and refusing the operation instruction of the response equipment before the technological parameters are not adjusted.