KR102548808B1 - Artificial intelligence monitoring system for press device which can monitor the malfunction of the press devices for various types of molds based on deep learning - Google Patents

Abstract

We provide a monitoring system for artificial intelligence press devices that can monitor malfunctions of press devices for multiple types of molds based on deep learning. The monitoring system includes a slide performing vertical linear motion, a bolster positioned at the bottom of the slide and equipped with a lower mold that mates with an upper mold mounted on the slide, a frame supporting the slide and the bolster, and driving force to the slide through rotational motion. A monitoring system for a press device built in a press device including a drive unit for providing a pressurizing slide to the bolster side to monitor the pressure balance and pressing force of the slide, the frame being fixed to the frame and bent by the pressure applied to the bolster by the slide A plurality of load cells that measure the physical quantity and calculate it as digital data, an encoder that measures the rotation angle of the driving unit, converts the digital data received from the plurality of load cells into load data, receives the rotation angle of the driving unit from the encoder, and receives the rotation angle It includes a control unit that generates a load data graph for each load, but the plurality of load cells calculates digital data for each unit rotation angle of the drive unit when the slide is located at the bottom dead center, and the control unit receives and converts load data received from the plurality of load cells and It is characterized in that the abnormality judgment related to the pressing balance or pressing force of the slide is performed based on the artificial intelligence abnormality detection model based on the neural network.

Description

A monitoring system for artificial intelligence press devices that can monitor malfunctions of press devices for various types of molds based on deep learning DEEP LEARNING}

The present invention relates to a monitoring system for a press device, and more particularly, to a monitoring system for an artificial intelligence press device capable of monitoring malfunctions of a press device for multiple types of molds based on deep learning.

The press device is a device that presses the mold after arranging the plate material in a mold to form the plate material to correspond to the shape of the mold, such as bending, shearing, and multi-sided shrinkage. Press devices can be divided into hydraulic presses and mechanical presses according to the structure that generates compression force, and various types of press devices customized for each purpose and characteristics, such as crank press, knuckle press, toggle press, friction press, etc. It has been developed and widely used in industry.

Depending on the load of the processing agent used in the press device, the mold may be overloaded or worn, so a system for monitoring the load (load) of the slide is required. In order to measure the load of the slide, the load of the slide is indirectly measured using pneumatic pressure inside the press device. The method has a problem in that the internal pressure is unstable and the measurement error is large because the rod of the slide cannot be directly measured.

Furthermore, the conventional mold monitoring system not only has low analysis accuracy for the sensed data, but also requires data analysis and diagnosis by skilled experts for more precise analysis, which reduces the time and cost of analysis and press maintenance. There has been a problem that occupies a large proportion of

The present invention has been made to solve the above problems, and the abnormality detection model comprehensively considers load data associated with different frames and performs abnormality detection, thereby enabling more precise monitoring of the press device based on artificial intelligence. It is providing a monitoring system for press equipment.

In addition, an additional object of the present invention is a press that can minimize the measurement error of the pressing force by measuring the bending deformation of the frame that is deformed by directly receiving energy due to the pressing of the slide and calculating the pressing force of the slide through the measured physical quantity. It is to provide a monitoring system for devices.

The technical problems of the present invention are not limited to the technical problems mentioned above, and other technical problems not mentioned will be clearly understood by those skilled in the art from the description below.

A monitoring system for a press device according to an embodiment for solving the above problems includes a slide performing vertical linear motion; a bolster positioned below the slide and equipped with a lower mold that mates with an upper mold mounted on the slide; a frame supporting the slide and the bolster; And a monitoring system for a press device built in a press device including a driving unit for providing a driving force to the slide through rotational motion to press the slide toward the bolster and monitoring the pressing balance and pressing force of the slide, wherein the monitoring system is fixed to the frame. a plurality of load cells mounted thereon to measure the physical quantity of the frame bent by the pressure applied to the bolster by the slide and calculate it as digital data; an encoder measuring a rotational angle of the drive unit; and a controller that converts the digital data received from the plurality of load cells into load data, receives the rotation angle of the drive unit from the encoder, and generates a graph of load data for each rotation angle, When is positioned at the bottom dead center, the digital data is calculated for each unit rotation angle of the drive unit, and the controller calculates the load data received from the plurality of load cells and converted based on the neural network-based artificial intelligence anomaly detection model. An abnormality judgment related to the pressing balance or pressing force of the slide is performed.

The plurality of load cells is an even number, and the plurality of load cells include a first load cell mounted on one side of the frame and a second load cell mounted on the other side of the frame, wherein the first load cell is connected to the slide to the bolster. The second load cell measures the physical quantity of one side of the frame bent by the pressure applied and calculates it as first digital data, and the second load cell measures the physical quantity of the other side of the frame bent by the pressure applied to the bolster by the slide and obtains second digital data. It is characterized by calculating with data.

The control unit converts the first digital data received from the first load cell into first load data for each rotation angle, and converts the second digital data received from the second load cell into second load data for each rotation angle. and generating two-dimensional third load data by arranging the first load data and the second load data on a data plane formed by the rotation angle axis and the load data axis, and convolving the third load data. It is characterized in that an abnormality judgment related to the pressing balance or pressing force of the slide is performed by inputting an abnormality detection model based on a neural network.

The controller converts the first digital data received from the first load cell into first load data, converts the second digital data received from the second load cell into second load data, and rotates the drive unit. A first angle-load data graph including the first load data information for each angle and a second angle-load data graph including the second load data information for each rotation angle of the drive unit are generated, It is characterized in that the pressure balance of the slide is determined by comparing the first angle-load data graph and the second angle-load data graph.

The control unit calculates third load data by summing the first load data and the second load data, and compares the third load data with predetermined reference load data to determine a defect of the press device. to be

The control unit calculates third load data by summing the first load data and the second load data, and generates a third angle-load data graph including the third load data information for each rotation angle of the drive unit. And, by comparing the third angle-load data graph and the reference angle-load data graph, it is characterized in that a defect in the pressing force of the slide is determined.

The monitoring system for the press device further includes a sensor unit that is fixedly mounted to the bolster and measures a load value applied to the bolster by the slide, and the control unit determines the load value of the slide from the sensor unit before press molding. Receives load data including information and digital data from the load cell, respectively, stores a look-up table including the digital data corresponding to the load data, and uses the look-up table to obtain the information received from the load cell. It is characterized in that the load data is converted by using digital data as an input address.

Each of the plurality of load cells may include a plurality of strain gauges.

The bottom dead center portion is characterized in that the rotation angle of the driving unit is within the range of 150 degrees to 210 degrees.

According to the embodiments of the present invention, it is possible to minimize the measurement error of the pressing force by measuring the bending deformation of the frame that is deformed by directly receiving energy due to the pressing of the slide and calculating the pressing force of the slide through the measured physical quantity. .

In addition, as a plurality of load cells mounted (installed) on the frame are composed of an even number, it is possible to monitor whether the pressure balance of the slide is uniform on the left and right sides or up and down.

In addition, when the slide is located at the bottom dead center, the load data transmitted to each frame for each rotation angle of the driving unit is calculated in real time and output as an angle-load data graph, thereby outputting an angle-load data graph according to the type of mold. and can learn. Therefore, even if the type of mold is changed, reliable monitoring can be performed because it is possible to determine whether or not the press device is defective based on the learned angle-load data.

Effects according to the technical spirit of the present disclosure are not limited to the effects mentioned above, and other effects not mentioned will be clearly understood by those skilled in the art from the description below.

1 is a schematic perspective view of a press device according to an embodiment.
2 is a schematic perspective view of a press device according to another embodiment.
3 is a schematic block diagram of a monitoring system for a press device according to an embodiment.
4 is an exemplary look-up table for explaining a method for calculating digital data values according to physical quantities measured by a load cell of a monitoring system for a press device according to an embodiment.
5 is a graph showing an example of load data for each rotation angle for explaining the operating principle of the monitoring system for a press device according to an embodiment.
6 is an exemplary diagram for explaining a deep learning-based monitoring method according to an embodiment.
7 and 8 are exemplary diagrams for explaining a deep learning-based monitoring method according to another embodiment.
9 is an exemplary diagram for explaining a deep learning-based monitoring method according to another embodiment.
10 is an exemplary diagram for explaining a deep learning-based monitoring method according to another embodiment.

Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings. Advantages and features of the present invention, and methods of achieving them, will become clear with reference to the detailed description of the following embodiments taken in conjunction with the accompanying drawings. However, the present invention is not limited to the embodiments disclosed below and may be implemented in various different forms, only the present embodiments make the disclosure of the present invention complete, and common knowledge in the art to which the present invention belongs. It is provided to completely inform the person who has the scope of the invention, and the present invention is only defined by the scope of the claims. Like reference numbers designate like elements throughout the specification.

Unless otherwise defined, all terms (including technical and scientific terms) used in this specification may be used in a meaning commonly understood by those of ordinary skill in the art to which the present invention belongs. In addition, terms defined in commonly used dictionaries are not interpreted ideally or excessively unless explicitly specifically defined.

Also, in this specification, the singular form may also include the plural form unless otherwise specified in the phrase. As used herein, "comprises" and/or "comprising" means that a stated component, step, operation, and/or element is present in the presence of one or more other components, steps, operations, and/or elements. or do not rule out additions.

1 is a schematic perspective view of a press device according to an embodiment.

Referring to FIG. 1, a press apparatus 1 according to an embodiment of the present invention uses two or more molds corresponding to each other, places a workpiece such as metal or plastic between the molds, and then presses the molds. It may be a mechanical device for forming a workpiece such as bending, shearing, cross-sectional shrinkage, etc. to correspond to the shape of the mold. The press device 1 can be divided into a hydraulic press device and a mechanical press device according to a structure that generates a compressive force, and the present invention is not limited to a specific driving method, and any of the conventional hydraulic and mechanical press devices can be applied. can

The press device 1 according to one embodiment may include a slide 120, a bolster 130, a frame 110, and a driving unit 140.

The slide 120 performs a linear motion up and down, and a first mold (or upper mold) may be mounted thereon. The bolster 130 is located below the slide 120, and a second mold (or lower mold) mating with the first mold mounted on the slide 120 may be mounted. Although not limited thereto, the first mold may be referred to as an upper mold and the second mold may be referred to as a lower mold.

The driving unit 140 may provide a driving force to the slide 120 so that the first mold presses toward the second mold. The driving unit 140 may include a connecting rod connected to the crankshaft and the slide 120 performing rotational motion. The slide 120 connected through the connecting rod may move up and down linearly by the rotational motion of the crankshaft.

The slide 120 may linearly move up and down according to the rotational angle of the crankshaft. In an exemplary embodiment, the angle of rotation of the crankshaft is 0 degrees when the slide 120 is at its highest position from the bolster 130, and the slide 120 is at its lowest position from the bolster 130. When (or when the slide 120 descends and contacts the bolster 130), the rotation angle of the crankshaft may be 180 degrees. That is, the slide 120 disposed on the upper part of the bolster 130 and located at the highest distance from the bolster 130 in the upper direction, rotates the crankshaft from 0 degrees to 180 degrees in the lower direction on the bolster 130 side. It moves in a straight line, and the bolster 130 can be further pressed while the crankshaft rotates at 180 degrees or more. Accordingly, the first mold mounted on the slide 120 may press toward the second mold mounted on the bolster 130 .

In the present specification, a portion (position) at which pressure is substantially formed by the slide 120 moving downward may be defined as a 'bottom dead center portion'. For example, as the slide 120 moves downward by the rotational motion of the crank shaft of the drive unit 140, the first mold of the slide 120 contacts the second mold or workpiece disposed below. In this case, pressure may be formed, and a portion where the pressure is formed may be referred to as a 'bottom dead center portion'. Accordingly, when the rotational angle of the crankshaft is within a preset angle range, the slide 120 may be said to be located at the bottom dead center portion. In an exemplary embodiment, when the rotation angle of the crankshaft is in the range of 150 degrees to 210 degrees, the slide 120 may be positioned at the bottom dead center portion. As will be described later, when the slide 120 is located at the bottom dead center, the load cell 150 included in the monitoring system 2 for the press device (see FIG. 3) moves the frame 110 by the pressure of the slide 120. A physical quantity that is deformed as it is bent can be measured and converted into a digital data value. The control unit 170 included in the monitoring system 2 for the press device (see FIG. 3 ) may analyze the calculated digital data value and output corresponding load data, and the press device 1 based on the load data malfunctions or wear of parts (eg, molds) can be monitored.

The frame 110 is positioned between the slide 120 and the bolster 130, and may support the slide 120 and the bolster 130. When the slide 120 presses toward the bolster 130, a predetermined physical force may be applied to the frame 110 by the pressure applied by the slide 120. Accordingly, bending may occur in the frame 110 due to the pressure applied by the slide 120 toward the bolster 130 .

The frame 110 may include a plurality of frames disposed spaced apart from each other between the slide 120 and the bolster 130 . A plurality of frames included in the frame 110 may be an even number.

In one embodiment, the frame 110 may include a first frame 111 disposed on one side between the slide 120 and the bolster 130 and a second frame 112 disposed on the other side opposite to the one side. there is. The first frame 111 and the second frame 112 may be spaced apart from each other. In an exemplary embodiment, the first frame 111 is disposed on the right between the slide 120 and the bolster 130, and the second frame 112 is disposed on the left between the slide 120 and the bolster 130. It can be.

2 is a schematic perspective view of a press device according to another embodiment.

Referring to FIG. 2, the press device 1_1 of FIG. 2 is different from the press device 1 of FIG. 1 in that the frame 110_1 includes first to fourth frames 111_1, 112_1, 113_1, and 114_1. there is

As will be described later, the press device 1_1 according to this embodiment uses the first to fourth load cells 151_1, 152_1, 153_1, and 154_1 to the first to fourth frames 111_1, 112_1, 113_1, and 114_1. Since each applied load can be compared, it is possible to more precisely monitor whether the slide 120 is pressed up, down, left, and right in a balanced manner compared to the press device 1 of FIG.

3 is a schematic block diagram of a monitoring system for a press device according to an embodiment.

Referring to Figure 3, the monitoring system 2 for the press device is built in the press device 1 for processing and molding the workpiece to determine the pressure balance of the slide 120, and can calculate the pressing force of the slide 120. And, based on this, it may be a system that can monitor malfunctions of the press device 1 or wear of parts (eg, molds).

According to one embodiment, the monitoring system 2 for the press device includes a load cell 150 installed in the frame 110 of the press device 1 and a sensor unit 160 installed in the bolster 130 of the press device 1. ), an encoder 180 and a controller 170 may be included. The monitoring system 2 for the press device may further include Ethernet and IoT.

1 and 3, the load cell 150 measures the physical quantity of the frame 110 deformed by the pressure applied while the slide 120 moves toward the bolster 130, and the measured frame 110 ) It may be a sensing means for obtaining basic data capable of calculating load data for the pressing force of the slide 120 based on the physical quantity of the slide 120.

The load cell 150 may be installed in the frame 110 of the press device 1 to measure the physical quantity deformed according to the bending of the frame 110 and calculate it as a digital data value. The digital data value calculated by the load cell 150 may be output to the controller 170.

Specifically, according to the rotation angle of the drive unit 140, the slide 120 linearly moves toward the bolster 130, and pressure molding may be performed. When the slide 120 moves toward the bolster 130 and is located at the bottom dead center, the slide 120 can press the bolster 130, and the frame 110 is bent by the pressing force of the slide 120. can The load cell 150 may measure a physical quantity deformed according to bending of the frame 110 , that is, an amount of compression/extension applied to the frame 110 . Specifically, when the slide 120 is located at the bottom dead center, the load cell 150 measures the physical quantity deformed as the frame 110 is bent, and the applied to the frame 110 based on the measured physical quantity Load data including load information may be calculated.

The load cell 150 may be installed at any position of the frame 110 located between the slide 120 and the bolster 130. The load cell 150 may include a plurality of load cells disposed spaced apart from each other. In one embodiment, the plurality of load cells included in the load cell 150 may be an even number. For example, the plurality of load cells included in the load cell 150 may be two or four. As the number of load cells 150 installed in the frame 110 is even, it is possible to monitor whether the pressure balance of the slide 120 is uniform on both left and right sides or up and down, so that monitoring accuracy can be improved.

In an exemplary embodiment, the load cell 150 may include a first load cell 151 installed on the first frame 111 and a second load cell 152 installed on the second frame 112 . On the other hand, although the figure shows that load cells are installed in each of the first frame 111 and the second frame 112 spaced apart from each other, it is not limited thereto. For example, the frame 150 includes only one frame between the slide 120 and the bolster 130, but the load cell 150 may be installed on one side and the other side of the one frame, respectively. One side where the load cell 150 is installed may be the right side, and the other side may be the left side. As another example, the load cell 150_1 is installed in the first to fourth frames 111_1, 112_1, 113_1, and 114_1, respectively, as shown in FIG. 154_1) may be included.

Hereinafter, a case in which the load cell 150 includes two load cells, that is, first and second load cells 151 and 152 will be described as an example.

The first and second load cells 151 and 152 may each include a strain gauge and a controller. The first and second load cells 151 and 152 may each include a plurality of strain gauges, that is, first to fourth strain gauges. The first to fourth strain gauges may form a full bridge circuit among Wheatstone bridges. When the frame 150 is bent as the slide 120 is pressurized, the strain gauge of the load cell 150 installed in the frame 110 may be tensioned or compressed to change resistance, and the strain gauge may change resistance based on the resistance change value. The bridge voltage displacement value can be measured.

The controller may include an analog digital converter and ADC convert the bridge voltage displacement value into digital data. The digital data calculated by the controller may be 24 bits, a value ranging from 0 to 16,777,215. The controller of the load cell 150 may transmit the calculated digital data to the controller 170.

That is, the load cell 150 uses a strain gauge according to the bending of the frame 110 in real time as the rotational angle of the drive unit 140 changes within the range where the slide 120 is located at the bottom dead center portion. A bridge voltage displacement value may be measured based on the resistance change value, digital data may be calculated by ADC conversion of the bridge voltage displacement value, and the calculated digital data may be output to the control unit 170 .

The controller 170 may calculate load data applied to the frame 150 by analyzing digital data received (received) from the load cell 150 . An experimentally set look-up table may be stored in the controller 170, and the controller 170 converts the calculated digital data value into a load data value corresponding to the calculated digital data value based on the look-up table. can The look-up table may store load data, which is a size of a load applied to a frame corresponding to digital data calculated from the controller. The look-up table may be created through a calibration step and stored in the control unit 170 before performing pressure molding using the press device 1 .

Meanwhile, since a signal of a strain gauge generally changes linearly according to an increase in load or pressure, the look-up table may have a linear distribution or a proportional distribution. Accordingly, precise measurement may be possible even in a low load environment in which the pressing of the press device 1 is small.

The control unit 170 may convert load data corresponding to the digital data value received from the load cell 150. The load data may be load data including load size information applied to the frame 150 . The encoder 180 may measure a rotational angle of the driving unit 140 . Accordingly, the control unit 170 may generate a load data graph for each rotation angle based on the rotation angle of the drive unit 140 received from the encoder 180 and the converted load data. That is, the control unit 170 may generate (or output) a rotation angle-load data graph including load data information converted by the control unit 170 according to a unit rotation angle of the drive unit 140 within the bottom dead center portion. .

The control unit 170 may monitor and determine malfunction of the press device 1 or wear of parts based on the rotation angle-load data. A detailed description thereof will be described later.

The controller 170 may be a micro controller unit (M.C.U.). The control unit 170 may control the load measurement according to the pressing force of the slide 120 by the load transmitted to the load cell 150 in the control environment of the microcontroller unit.

In an exemplary embodiment in which the load cell 150 includes a first load cell 151 and a second load cell 152, each of the load cells 151 and 152 has a frame 111 and 112 bent as the slide 120 is pressed. ) is ADC converted based on the transformed physical quantity to calculate digital data values, which can be transmitted to the control unit 170, respectively.

As the plurality of load cells 151 and 152 measure the degree of bending of each frame 111 and 112 located on the left and right sides between the slide 120 and the bolster 130, when the slide 120 descends When driving pressure molding), it is possible to monitor whether the force is equally distributed 1:1 on the left and right sides, and whether the pressure is balanced, and accordingly, the processing precision of the workpiece can be improved.

The sensor unit 160 may directly measure the pressure or load applied to the bolster 130 by the slide 120 . The pressure or load value measured by the sensor unit 160 may be transmitted to the controller 170 . As will be described later, the sensor unit 160 may be used to generate and store a look-up table in a calibration step prior to performing a machining process using the press device 1 .

4 is an exemplary look-up table for explaining a method for calculating digital data values according to physical quantities measured by a load cell of a monitoring system for a press device according to an embodiment. 5 is a graph showing an example of load data for each rotation angle for explaining the operating principle of the monitoring system for a press device according to an embodiment.

Hereinafter, the operation of the monitoring system 2 for the press device 1 according to an embodiment will be described with reference to FIGS. 3 to 5 .

Before the press device 1 is driven for pressing, the sensor unit 160 and the load cell 150 are used to look- You can perform a calibration step to save the up table.

In addition, according to the unit rotation angle of the driving unit 140 within the range where the slide 120 is located at the bottom dead center portion using the encoder 180 and the load cell 150 before the press device 1 is driven for press molding. A step of outputting and storing a reference angle-load graph RG including load data information may be further performed.

Specifically, the monitoring system 2 for the press device 1 according to the present invention presses the slide 120 to the bolster 130 before driving the press molding and uses the sensor unit 160 installed in the bolster 130. to measure the size of the punching pressure applied to the bolster 130, calculate load data based on the pressure information, output the result to the control unit 170, and use the load cell 150 installed in the frame 110 to load the frame 110 ) can be measured and output to the control unit 170 by measuring the physical quantity of deformation and calculating digital data. The control unit 170 may create and store a look-up table as shown in FIG. 4 based on the load data received from the sensor unit 160 and the digital data received from the load cell 150 .

As described above, since the load cell 150 includes a strain gauge, and the signal of the strain gauge generally changes linearly with an increase in load or pressure, the look-up table is linearly distributed or proportional as shown in FIG. may have a distribution.

In addition, in the monitoring system 2 for the press device 1 according to the present invention, when the slide 120 is positioned at the bottom dead center portion before pressing, the sensor unit 160 uses the load cell 150 to load the load. data can be converted. The encoder 180 is a reference including load data information including information on the size of a load applied to the bolster 130 for each unit rotation angle of the driving unit 140 in the range of 150 to 210 degrees of rotation angle of the driving unit 140. An angle-load graph (RG) can be output.

Subsequently, after the calibration operation is completed, the press device 1 may perform pressure molding by lowering the slide 120 by the rotational motion of the drive unit 140 .

In the monitoring system 2 for the press device 1 according to the present embodiment, while the pressure molding process is performed, when the slide 120 is positioned at the bottom dead center portion, monitoring may be performed together with the pressure molding.

Specifically, according to the rotation angle of the drive unit 140, the slide 120 linearly moves toward the bolster 130, and pressure molding may be performed. When the slide 120 is located at the bottom dead center, the frame 110 may be bent by the pressing force of the slide 120 . Therefore, within the position range of the slide 120 at the bottom dead center, the load cell 150 uses a strain gauge as the frame 110 bends in real time as the rotation angle of the driving unit 140 changes, and the resistance change value The bridge voltage displacement value may be measured based on , digital data may be calculated by ADC conversion of the bridge voltage displacement value, and the calculated digital data may be output to the control unit 170 .

In one embodiment, the rotation angle of the drive unit 140 is in the range of 150 to 210 degrees, and the amount of physical change measured by the load cell 150 is the bridge voltage displacement measured based on the resistance value of the strain gauge. value, and the bridge voltage displacement value can be converted to ADC by the controller of the load cell 150 and calculated as a digital data value. However, it is not limited thereto, and the rotation angle of the drive unit 140 determined as the bottom dead center portion may be changed according to the type of mold and the type of workpiece. As described above, the digital data value may be a 24-bit value ranging from 0 to 16,777,215.

The load cell 150 calculates the amount of bending deformation of the frame 110 as a digital data value for each unit rotation angle within the range of rotation angle of the driving unit 140 from 150 degrees to 210 degrees, and outputs it to the control unit 170 in real time. .

When the slide 120 is located at the bottom dead center, the first load cell 151 calculates the amount of physical change (amount of bending deformation) of the first frame 111 per unit rotation angle of the drive unit 140 as first digital data. and output to the control unit 170. The control unit 170 sets the first digital data received from the first load cell 151 as an input address based on a pre-stored look-up table (eg, the look-up table of FIG. 1 Load data can be converted. Accordingly, the encoder 180 is a first load including information on the size of the load applied to the first frame 111 for each rotation angle of the driving unit 140 in the range of 150 degrees to 210 degrees. A first angle-load graph G1 including data information may be output.

Specifically, in the first angle output by the encoder 180, the load graph G1 is, as shown in FIG. 5, the x-axis is the rotational angle of the drive unit 140 and the y-axis is the first frame 111 An applied first load data value may be indicated. Accordingly, the first angle output by the encoder 180-load graph G1 is applied to the first frame 111 according to a unit rotation angle (for each rotation angle of 1 degree) at a rotation angle of 150 degrees to 210 degrees A first load data value may be indicated.

Similarly, when the slide 120 is positioned at the bottom dead center, the second load cell 152 converts the amount of physical change (amount of bending deformation) of the second frame 112 per unit rotation angle of the driving unit 140 into second digital data. It can be calculated and output to the control unit 170. The control unit 170 uses the second digital data received from the second load cell 152 as an input address based on a pre-stored look-up table (eg, the look-up table of FIG. 2 Load data can be converted. Accordingly, the encoder 180 rotates the driving unit 140 in the range of 150 degrees to 210 degrees, and the rotation angle of the driving unit 140 is applied to the second frame 112 for each rotation angle. A second angle-load graph G2 including data information may be output.

The controller 170 may calculate third load data based on the first load data and the second load data. The third load data may be a sum of the first load data and the second load data. The third load data may be load data corresponding to the pressing force of the slide 120 . That is, as the slide 120 presses the first and second load data, the pressing force is applied to the first and second frames 111 and 112 in a distributed manner, and the third load data is the first and the sum of the second load data, that is, the size of the rod corresponding to the total pressing force of the slide 120 . The encoder 180 may output a third angle-load graph G3 including third load data information including information on the magnitude of the load applied as a whole by the slide 120 for each rotation angle of the drive unit 140. there is.

The controller 170 may determine whether or not the slide 120 is balanced left and right and pressed based on the first angle-load graph G1 and the second angle-load graph G2. Specifically, the control unit 170 compares the first angle-load graph G1 and the second angle-load graph G3 with each other, and if they differ by more than a preset error range, the press device 1 malfunctions or It can be judged that there is something wrong with the part.

In addition, the controller 170 may determine whether the pressing force of the slide 120 is normal by using the third angle-load graph G3. Specifically, the control unit 170 compares the reference angle-load graph (RG) and the third angle-load graph (G3) with each other, and if they differ by more than a preset error range, the press device 1 malfunctions or parts It can be judged that there is something wrong with However, the controller 170 compares the third load data with the reference load data, and if the difference is greater than a predetermined error range, the press device 1 may malfunction or the part may be defective.

On the other hand, the monitoring system for the press device according to the present invention measures the bending deformation of the frame that is deformed by directly receiving the energy due to the pressurization of the slide, and calculates the pressurization force of the slide through the measured physical quantity, thereby minimizing the measurement error of the pressurization force can do.

In addition, as a plurality of load cells mounted (installed) on the frame are composed of an even number, it is possible to monitor whether the pressure balance of the slide is uniform on the left and right sides or up and down.

In addition, when the slide is located at the bottom dead center, the load data transmitted to each frame for each rotation angle of the driving unit is calculated in real time and output as an angle-load data graph, thereby outputting an angle-load data graph according to the type of mold. and can learn. Therefore, even if the type of mold is changed, reliable monitoring can be performed because it is possible to determine whether or not the press device is defective based on the learned angle-load data.

Hereinafter, with reference to FIGS. 6 to 10, various embodiments of a method for performing monitoring by the monitoring system 2 for a press device (e.g., 1) using a deep learning-based abnormality detection model will be described. let it do Hereinafter, for convenience of explanation, reference numerals for the press devices (e.g., 1, 1_1) will be omitted.

6 is an exemplary diagram for explaining a deep learning-based monitoring method according to an embodiment.

As shown in FIG. 6 , the present embodiment relates to a method of monitoring a malfunction of a press device using a recurrent neural network-based abnormality detection model.

To provide convenience of understanding, the structure and operation principle of the anomaly detection model will be described first, and then the learning method of the anomaly detection model and the monitoring method using the same will be described.

As shown, the anomaly detection model according to the embodiment may be configured to include an analysis neural network 210 that analyzes input load data and an output layer 220 that outputs a prediction result on whether or not an anomaly has occurred.

The analysis neural network 210 may receive load data composed of a plurality of consecutive (or ordered) load values (ie, a sequence of load values) and analyze the load data in consideration of the order of the load values. The analysis neural network 210 may perform precise analysis by considering the order of load values input through the plurality of RNN blocks 211_1 to 211_K. Here, the load data may be, for example, load values according to rotation angles or load values according to operation time of the press device.

The output layer 220 may output a prediction result regarding whether or not an anomaly has occurred based on the analysis result of the analysis neural network 210 . That is, the output layer 220 may receive the analysis result of the analysis neural network 210 and predict whether an abnormality occurs through neural network operation. The output layer 220 may be implemented based on, for example, a fully-connected layer, but the scope of the present disclosure is not limited thereto.

The anomaly detection model illustrated in FIG. 6 may be learned using load data and label information about occurrence of anomalies. Such learning may be performed by a monitoring system for a press device (2, hereinafter abbreviated as 'monitoring system') or may be performed by a separate learning device/system. If you are skilled in the art, you will already be familiar with how to train a deep learning model (ie, update the weight parameters of the model) through backpropagation of the prediction error (ie, the difference between the prediction result and the label information) Bar, description of the learning method itself will be omitted.

When learning is completed, the control unit 170 of the monitoring system 2 may perform monitoring using the learned anomaly detection model. For example, the control unit 170 obtains load data as illustrated in FIG. 5 based on a pre-stored look-up table (eg, the look-up table of FIG. 4), and converts the obtained load data into an anomaly detection model. By inputting to , it is possible to perform a detection operation related to a malfunction of the press device. At this time, the control unit 170 may divide the acquired load data into predetermined sections (e.g., divide into one section for every 10 degrees of rotation) and input the section-specific load data to an abnormality detection model.

Meanwhile, when a plurality of load data for different frames (e.g., 111 and 112 in FIG. 3) are acquired from a plurality of load cells (e.g., 151 and 152 in FIG. 3), the anomaly detection model uses a plurality of analysis neural networks. It may be configured to include For example, the anomaly detection model includes a first analysis neural network for analyzing the first load data of the first frame, a second analysis neural network for analyzing the second load data of the second frame, and a combination of the first and second analysis neural networks. It may also be configured to include an output layer that predicts the occurrence of abnormalities by comprehensively considering the analysis results. In this case, more accurate monitoring of the press device can be performed by comprehensively considering load data associated with frames with different abnormality detection models and performing abnormality detection.

Hereinafter, a deep learning-based monitoring method according to another embodiment will be described with reference to FIGS. 7 and 8 .

As shown in FIGS. 7 and 8 , the present embodiment relates to a method for monitoring a malfunction of a press device using a convolutional neural network based abnormality detection model.

The anomaly detection model according to this embodiment may be configured to include at least one convolution layer 241 , at least one pooling layer 242 and an output layer 243 . Those skilled in the art will already be familiar with the principle of operation of each layer 241 to 243, and a detailed description thereof will be omitted.

The convolution layer 241 may extract features required for anomaly detection by performing a convolution operation on the 2-dimensional load data 230 . The convolution operation may be, for example, a 2D convolution operation, but in some cases, a 1D convolution operation may be used to extract features by focusing more on the order of load values.

The two-dimensional load data 230 may be generated based on a plurality of load data obtained from a plurality of load cells (eg, 151 and 152 in FIG. 3 ). For example, as shown in FIG. 8 , assume that load data (hereinafter, referred to as 'first to fourth load data') of four different frames is obtained from four load cells. In this case, the control unit 170 of the monitoring system 2 arranges the first to fourth load data on a two-dimensional data plane formed by the rotation angle axis (or time axis) and the load data axis, and has a preset size. Two-dimensional load data 230 may be extracted by applying a window of . For example, the controller 170 slides a window having a preset size along the direction of the rotation angle axis (or time axis) and extracts load values corresponding to the window as one 2-dimensional load data (that is, continuously extraction), but the scope of the present disclosure is not limited thereto.

Meanwhile, a method of arranging different load data may be designed in various ways. For example, when the first load data and the second load data are data related to frames corresponding to each other (e.g., when the first load data and the second load data are data related to left and right frames, respectively, etc.) , the first load data and the second load data may be disposed adjacent to each other on the data plane (see FIG. 8). By doing so, feature extraction can be performed by focusing more on the relation between the two load data (e.g., the pressure balance relation).

It will be described with reference to FIG. 7 again.

The pooling layer 242 may perform a pooling operation to reduce the size of features extracted from the convolution layer 241 .

The output layer 243 may output a prediction result regarding whether or not an anomaly has occurred based on the extracted features. The output layer 243 receives features output from the convolution layer 241 and/or the pooling layer 242 and predicts whether an anomaly has occurred through a neural network operation. The output layer 243 may be implemented based on, for example, a fully-connected layer, but the scope of the present disclosure is not limited thereto.

The anomaly detection model illustrated in FIG. 7 may be learned using 2-dimensional load data (e.g., 230) and label information on occurrence of anomalies.

When learning is completed, the controller 170 may perform monitoring using the learned anomaly detection model. For example, the control unit 170 obtains a plurality of load data as illustrated in FIG. 5 based on a pre-stored look-up table (eg, the look-up table of FIG. 4), and obtains a plurality of load data By processing them (see FIG. 8), two-dimensional load data can be generated. In addition, the control unit 170 may perform a detection operation related to a malfunction of the press device by inputting the generated load data to the abnormality detection model.

Hereinafter, a deep learning-based monitoring method according to another embodiment will be described with reference to FIG. 9 .

As shown in FIG. 9 , the present embodiment relates to a method of monitoring a malfunction of a press device using an auto-encoder based abnormality detection model 260 .

The anomaly detection model 260 according to this embodiment may be configured to include an encoder 261 and a decoder 262 . Here, the encoder 261 may generate a latent vector by encoding the input 2D load data 251, and the decoder 262 may decode the latent vector to generate the 2D load data 252 can be reconstructed as The encoder 261 and the decoder 262 can be learned based on the reconstruction loss (i.e., the difference between the load data 251 and 252). If you are familiar with the autoencoder learning method, There will be a bar, so a detailed description thereof will be omitted.

In this embodiment, the control unit 170 of the monitoring system 2 may configure frames corresponding to each other in different pairs. For example, the controller 170 may configure a left frame and an upper frame as a first pair, and configure a right frame and a lower frame corresponding thereto as a second pair. Then, the controller 170 generates two-dimensional normal load data 251 using the normal load data of the first pair, and learns the abnormality detection model 260 using the generated normal load data 251. can make it Since the normal load data for the first pair will have similar characteristics to the normal load data for the second pair (e.g., normal pattern, load value distribution, etc. are similar), only the normal load data of one pair is used. Even so, the anomaly detection model 260 can be sufficiently learned. As described above, learning of the model may be performed by other learning devices/systems.

When learning is completed, the controller 170 may perform monitoring using the learned anomaly detection model 260 . Specifically, the controller 170 generates two-dimensional load data 253 from load data related to the second pair (or first pair), and converts the generated load data 253 to the abnormality detection model 260. Reconstructed two-dimensional load data 254 may be obtained by inputting the data. In addition, the controller 170 may calculate a reconstruction loss based on the difference between the two load data 253 and 254 and determine whether an abnormality has occurred based on the calculated reconstruction loss. For example, the control unit 170 may determine that an abnormality has occurred in the press device or the like based on the determination that the reconstruction loss exceeds a reference value. This is because data that is not well reconstructed by the autoencoder model 260 that has learned normal load data is highly likely to be load data observed when an abnormality occurs.

Hereinafter, a deep learning-based monitoring method according to another embodiment will be described with reference to FIG. 10 .

As shown in FIG. 10 , the present embodiment relates to a method for monitoring malfunctions of press devices for various types of molds using one abnormality detection model 270 .

In this embodiment, the abnormality detection model 270 may be trained to predict whether an abnormality occurs by further receiving a code value indicating the type of mold (e.g., mold identifier, mold category identifier, etc.). In this case, there is no need to build an abnormality detection model for each type of mold.

For example, assume that there are two types of molds (ie, a first mold and a second mold). In this case, the control unit 170 of the monitoring system 2 learns the abnormality detection model 270 using the code value indicating the first mold, the load data related to the first mold, and the label information regarding the occurrence of the abnormality, and The abnormality detection model 270 may be further trained using the code value indicating the two molds, the load data related to the second mold, and the label information regarding the occurrence of the abnormality. By doing so, an anomaly detection model 270 that can be applied to both types of molds can be built. Here, the mold code value can serve to prevent confusion that the abnormality detection model 270 may experience during learning by explicitly providing mold type information to the abnormality detection model 270 .

When learning is completed, the control unit 170 may perform a detection operation related to a malfunction of the press device by inputting a code value indicating the type of mold currently being processed and load data into the abnormality detection model 270 .

For reference, the anomaly detection model 270 may be a model as described with reference to FIGS. 6 to 9 or a model having a different structure.

So far, with reference to FIGS. 6 to 10 , various embodiments of a method for the monitoring system 2 to perform monitoring using a deep learning-based anomaly detection model have been described.

Those skilled in the art related to the present embodiment will be able to understand that it may be implemented in a modified form within a range that does not deviate from the essential characteristics of the above description. Therefore, the disclosed methods are to be considered in an illustrative rather than a limiting sense. The scope of the present invention is shown in the claims rather than the foregoing description, and all differences within the equivalent scope will be construed as being included in the present invention.

1: press device
2: Monitoring system for press unit
110: frame
120: slide
130: Bolster
140: driving unit
150: load cell (plural load cells)
151: first load cell
152: second load cell
160: sensor unit
170: control unit
180: encoder

Claims (9)

A slide performing up and down linear motion; a bolster positioned below the slide and equipped with a lower mold that mates with an upper mold mounted on the slide; a frame supporting the slide and the bolster; And a monitoring system for a press device built in a press device including a driving unit for providing a driving force to the slide through rotational motion to press the slide toward the bolster and monitoring the pressing balance and pressing force of the slide,
a plurality of load cells fixed to the frame and configured to measure the physical quantity of the frame bent by the pressure applied to the bolster by the slide and calculate it as digital data;
an encoder measuring a rotational angle of the drive unit; and
A control unit that converts the digital data received from the plurality of load cells into load data, receives the rotation angle of the drive unit from the encoder, and generates a load data graph for each rotation angle,
The plurality of load cells calculate the digital data for each unit rotation angle of the driving unit when the slide is located at the bottom dead center,
The controller determines an abnormality related to the pressure balance or pressure of the slide based on the load data received from the plurality of load cells and converted and an artificial intelligence abnormality detection model based on a neural network,
The plurality of load cells is an even number, and the plurality of load cells include a first load cell mounted on a left frame of the frame and a second load cell mounted on a right frame opposite to the left frame of the frame,
The first load cell measures the physical quantity of the left frame bent by the pressure applied by the slide to the bolster and calculates it as first digital data, and the second load cell measures the physical quantity of the frame bent by the pressure applied by the slide to the bolster. Measuring the physical quantity of the frame and calculating it as second digital data;
Each of the first and second load cells includes first and second strain gauges, and each of the first and second strain gauges is composed of a full Wheatstone Bridge circuit,
Each of the first and second load cells measures a bridge voltage displacement value based on a resistance change value generated as the first and second strain gauges are stretched or compressed, and then converts the bridge voltage displacement value to ADC (analog digital converter) to calculate the first and second digital data and output them to the control unit,
The control unit converts the first digital data received from the first load cell into first load data for each rotation angle, and converts the second digital data received from the second load cell into second load data for each rotation angle. and generating two-dimensional third load data by arranging the first load data and the second load data on a data plane formed by the rotation angle axis and the load data axis, and convolving the third load data. It is input to a Convolutional Neural Network-based abnormality detection model to perform an abnormality judgment related to the pressure balance or pressure of the slide,
The anomaly detection model includes a convolution layer, a pooling layer, and an output layer,
The convolution layer extracts feature data necessary for anomaly detection by performing a convolution operation on the third load data,
The pooling layer performs a pooling operation to reduce the size of the feature data extracted from the convolution layer,
The output layer outputs an abnormal occurrence prediction result related to the pressing balance or pressing force of the slide based on at least one of the feature data extracted from the convolution layer and the feature data on which a pooling operation is performed in the pooling layer,
In the third load data input to the convolution layer, the first load data and the second load data are disposed on the data plane, and the first load data and the second load data are disposed. Generated by continuously extracting load values corresponding to the window while sliding a window having a predetermined size on a plane in the direction of the rotational angle axis,
The first load data and the second load data are characterized in that arranged adjacent to each other on the data plane, AI-based monitoring system for press devices.
delete delete delete delete delete According to claim 1,
The monitoring system for the press device,
A sensor unit fixedly mounted to the bolster and measuring a load value applied to the bolster by the slide;
The control unit,
Before pressure molding, receive load data including load value information of the slide from the sensor unit and digital data from the load cell, respectively, and store a look-up table including the digital data corresponding to the load data, A monitoring system for an artificial intelligence-based press device, characterized in that the load data is converted using the digital data received from the load cell as an input address using a look-up table.
delete According to claim 1,
The bottom dead center portion is characterized in that the rotation angle of the drive unit is in the range of 150 degrees to 210 degrees, artificial intelligence-based monitoring system for press devices.

Images