KR102093327B1 - Method for modelling factory process using artificial intelligence - Google Patents

Abstract

Provided is a factory process modeling method capable of simply performing process modeling of a corresponding factory with high accuracy. The factory process modeling method comprises: (a) a data collection and training step; and (b) a step of performing factory process modeling.

Description

Method for modeling factory process using artificial intelligence}

The present invention relates to a method for modeling a factory process, and relates to a method for modeling a factory process simply and accurately by collecting data on the factory process modeling and learning artificial intelligence.

In the assembly plant, it is important to check the entire process of the plant through the movement path of each item. If the product is accumulated or the movement speed is lowered at a specific location or there is a location where a defect occurs, the overall efficiency of the factory is lowered.

From this point of view, a smart factory to which IT technology and the like have been introduced was introduced. Smart Factory collects and analyzes data by installing sensors, etc. in facilities and machines in the factory, so that all situations in the factory can be seen at a glance and controlled according to the purpose. In particular, it can bring about a powerful performance improvement in connection with automation equipment, and is more advantageous in an assembly process in which various parts are used.

However, smart factories cannot be introduced into all factories. As it is a large-scale factory, it is necessary to improve the process, but in a factory in a less developed country or a small factory, it is still in the process of improving the process in light of the experience of skilled workers.

If it is difficult to introduce a smart factory, to improve efficiency, it is necessary to model and confirm the current state of the factory process. Factory process modeling refers to modeling the movement paths and conditions of goods (moving, stopping, assembling, painting, packaging, etc.).

However, most of the current process modeling of the plant is done manually. Each person tracks the movement path of factory items, inputs the status, and models them. Not only does it take a lot of time, it has the disadvantage that it is not accurate and needs to be repeated every time a process is improved in the future and is not cost-effective.

Look at the related prior art.

Korean Patent Publication No. 10-2005-0022762 suggests a method for identifying and improving the process of a manufacturing plant. It is a technology that collects data necessary for efficiency by creating parts layout, work process diagram, assembly instruction, etc. and creating a layout of the workplace.

Korean Patent Registration No. 10-1846793 utilizes manufacturing equipment control data and monitoring data to understand the process of manufacturing equipment.

All of the above patents include the disadvantage that a person must record and manage to perform the entire process modeling of the factory. In particular, Korean Patent Registration No. 10-1846793 cannot identify the accumulation of parts, etc. because the process is understood from the viewpoint of the manufacturing equipment, not from the viewpoint of the flow of parts.

(Patent Document 1) KR 10-2005-0022762 A

(Patent Document 2) KR 10-1846793 B

(Patent Document 3) KR 10-2018-0054354 A

The present invention has been devised to solve the above problems.

Even in factories where smart factories are difficult to apply, we would like to propose a method that can perform process modeling of the factories with ease and high accuracy by simply attaching sensors to several items and placing multiple antennas in the factories.

One embodiment of the present invention for solving the above problems, (a) data collection and learning step; And (b) performing factory process modeling, wherein the (a) data collection and learning step comprises: (a1) three or more antennas (AP ₀ ) in a factory (F ₀ ) for data collection. ) Is installed and the installation location (AP ₀ 1, AP ₀ 2, AP ₀ 3) is stored in the location confirmation unit 110, the sensor (S) is attached to a plurality of articles (G ₀ ); (a2) three or more antennas AP ₀ periodically receiving signal strengths from the sensor S as the article G ₀ moves according to the factory process; (a3) The position confirmation unit 110, the received signal strength and the installation position (AP ₀ 1, AP ₀ 2, AP ₀ 3) of the three or more antennas (AP ₀ ), using the article (G ₀ ) Periodically checking the location; (a4) receiving, by the status input unit 120, the position of the article G ₀ , which is periodically checked, the status of the article to which the sensor is attached; -Here, the state of the input item is any one selected from a group of states of a preset item (a5) The data collection unit 130 checks the identifier (ID) and step (a3) for each item (G ₀ ). Collecting a data set including the location over time and the state input in step (a4); (a6) steps (a1) to (a5) are repeated in a factory for collecting multiple data; And (a7) the AI learning unit 150 using the data set collected in the step (a6) to perform artificial intelligence learning, wherein the (b) factory process modeling step comprises: (b1 ) Three or more antennas (APs) are installed in the factory (F) to be modeled, and the installation locations (AP1, AP2, AP3) are stored in the location confirmation unit 210, and the sensor (S) is applied to a plurality of articles (G). ) Is attached; (b2) three or more antennas AP periodically receiving signal strengths from the sensors S as the article G moves according to the factory process; (b3) The position checking unit 210 periodically checks the position of the article G by using the received signal strength and the installation positions AP1, AP2, and AP3 of three or more antennas AP. step; (b4) The state estimator 220 periodically checks in the step (b3) by applying the position of the article G identified in the step (b3) to the result learned in the step (a7). Estimating a state for each position of the article G; (b5) Data set unit 230, for each article (G), a data set including an identifier (ID), a position over time identified in step (b3), and a state estimated in step (b4) setting (data set); And (b6) the factory process modeling execution unit 260 performing the factory process modeling of the factory F to be modeled using the data set set in the step (b5), in the step (b6). The performed process modeling provides a factory process modeling method, which includes the movement path and state of the article G.

In addition, the step (b5) further includes the step of further setting an additional data set by further using information included in the set data set, and the step (b6) comprises: the factory process modeling performing unit (000) It is preferable to perform the factory process modeling using the data set and the additional data set set in the step (b4).

In addition, it is preferable that the additional data set includes at least one of a previous average position, speed, process type (Ptype), type of object, type of adjacent object, and distance of adjacent object.

In addition, the additional data set includes a process type (Ptype), and it is preferable that the process type (Ptype) is determined according to whether different articles G are in different locations and then changed to the same location. .

In addition, each article (G) is divided into one of a general object (0) and a painted object (1), and when different general objects (0) are in different positions and changed to the same position, the process type is When set to "Assembly", the process type is preferably set to "Paint" when the normal object 0 and the painted object 1 are in different positions and then changed to the same position.

In addition, the additional data set includes a previous average position, and the previous average position is preferably set using an average value of X coordinates and Y coordinates, which are positions from a plurality of previous viewpoints, based on a time point to be checked. Do.

In addition, the additional data set includes speed, and the speed is based on a time point to be checked, and uses a distance between both time points and a moving distance calculated using a position at a current time point and a position at a different time point. It is preferably set.

In addition, the additional data set includes a type of object, and the type of object is preferably a preset value for each group of items G.

In addition, the additional data set includes the types of adjacent objects, and the types of the adjacent objects are identified by using the position of the item G and the item closest to the position of the item to be checked, and then the object of the identified item It is preferably set by checking the type of.

In addition, the additional data set further includes the distance of adjacent objects, and the distance of the adjacent objects is different from the position of the object to be checked using the position of the object G, and the type of the object is different from the close items. It is preferred that after selecting the closest article as the article, it is set by calculating the distance between the location of the selected article and the location of the article to be checked.

In addition, steps (a2) to (a3) are performed using an indoor positioning technique, and steps (a7) are preferably performed using deep learning.

In addition, the state input in the step (a4) preferably includes “move” and “stop”.

According to the present invention, a factory process modeling is possible by periodically attaching a sensor to a moving object at any factory and installing an antenna to periodically check the location of the article through signal strength.

After modeling is performed, the attached sensors and antennas can be collected and used at different plants or at different times, and it is cost-effective in that the present invention requires little manpower.

Since various information is calculated and estimated by artificial intelligence learning, modeling accuracy is high even when no manpower is required, and the accuracy gradually increases as modeling is performed multiple times and data is accumulated.

In addition to the position and condition over time, additional data such as the previous average position, speed, and process type can also be learned and estimated, and accordingly, the modeling accuracy is further increased and various data for efficient operation of the plant can be identified.

1 is a schematic diagram of a system for carrying out a method according to the invention.
2 is a schematic view of a factory in which the method according to the invention is carried out.
3 shows a triangulation method for confirming a position through signal strength in the method according to the present invention.
4 shows an example of a data set of a method according to the invention.
5 is a view for explaining an additional data set in the method according to the present invention.
6 is a view for explaining the "previous average position" included in the additional data in the method according to the present invention.
7 is a view for explaining the "speed" included in the additional data in the method according to the present invention.
8 is a view for explaining a "process type (Ptype)" included in additional data in the method according to the present invention.
9 is a view for explaining the "distance of adjacent objects" included in the additional data in the method according to the present invention.
10 shows an additional data set in the method according to the invention.
11 is an illustration of factory process modeling performed with the method according to the present invention.

Hereinafter, a method according to the present invention will be described with reference to the drawings.

The method according to the present invention is largely divided into a data collection and learning step and a factory process modeling step.

1, the data collection and the learning phase is the data collection and the learning module 100 is performed by the data collection plant (F ₀₎ from the collecting data for an artificial intelligence learning and based on it ai for which It is a step to perform learning.

The step of performing the factory process modeling is a step of performing the factory process modeling by checking the data at the factory F to perform real modeling based on the artificial intelligence learned in the data collection and learning step.

Note, however, that these two steps need not be explicitly distinguished and can be performed together. That is, it is natural that the factory process modeling of the factory for data collection (F ₀ ) is performed even in the data collection and learning stage, and conversely, the result confirmed in the factory process modeling stage is used as the data of AI learning in the data collection and learning stage. Will be used.

On the other hand, the factory for data collection (F ₀ ) is preferably the same as the target factory (F) to model the real factory process. As will be described later, the present invention estimates data such as states through learning by artificial intelligence, because the accuracy of the estimated data of artificial intelligence set based on data of the same plant is high. For example, if you want to model a factory process in an auto parts factory (F), it is preferable that the factory for data collection (F ₀ ) is another auto parts factory, and if you want to model a factory process in a food factory (F), collect data It is preferable that the dragon factory F0 is other food factories.

Data collection and learning steps will be described with reference to FIGS. 2 to 4.

First, three or more antennas (AP _0; access point) are installed in the factory F ₀ for data collection, and the installation location (AP ₀ 1, AP ₀ 2, AP ₀ 3) is stored in the location confirmation unit 110 And, the sensor (S) is attached to a plurality of articles (G ₀ ), the preparation for collecting data is performed. The sensor S may be a real time location system (RTLS) tag for applying indoor positioning technology. If the size of the article G ₀ is too small, the article may be stored in a basket and a sensor S may be attached to the basket.

2 schematically shows a factory F ₀ for data collection, in which six antennas are shown installed, but the number is not limited. It is sufficient as long as it is an appropriate number capable of receiving a signal from the sensor S. In particular, it is preferable that three antennas are arranged to receive signals wherever the position of the product G ₀ is.

Now, when the factory is running, the goods G ₀ start moving. As the article G ₀ moves according to the factory process, three or more antennas AP ₀ periodically receive signal strength from the sensor S.

The positioning unit 110 periodically uses the received signal strength and the installation positions (AP ₀ 1, AP ₀ 2, AP ₀ 3) of three or more antennas AP ₀ to periodically position the position of the article G ₀ . Confirm with. As shown in FIG. 3, when the signal strength from the sensor S is confirmed, the distances d1, d2, and d3 spaced apart from the antenna AP ₀ can be known, and the triangulation method of the article G ₀ Location can be checked. The checking period can be set appropriately, and may be in real time or may be several seconds.

Next, the status input unit 120 receives the status of the article to which the sensor is attached, for each location of the article G ₀ that is periodically checked. The input state of the item may be any one selected from a group of preset items. It may include "move", "stop", "painting", "packaging", etc. may be further included.

Now, the data collection unit 130 may collect, for each article G ₀ , a data set including an identifier ID, a location over time, and a status over time. An example of the collected data set is shown in FIG. 4. The identifier ID may be matched through the sensor S identifier in the sensor S attaching step, and may be input in advance.

When such a process is repeated in a number of factories, a plurality of data sets are collected, and the artificial intelligence learning unit 150 may perform artificial intelligence learning using the same. The artificial intelligence used here may be deep learning. For example, if the positions of products A and B are the same at time L, the status is [assemble]. As the accuracy of artificial intelligence increases due to the accumulation of a large number of data, it is possible to estimate the state with only time and location.

Next, with reference to FIGS. 5 to 11, a step of performing factory process modeling in a factory F, which is an object to be actually modeled, using artificial intelligences learned in this process will be described.

Similar to the data collection and learning steps described above, three or more antennas AP are installed in the factory F to be modeled, and the installation locations AP1, AP2, and AP3 are stored in the location confirmation unit 210, A sensor S is attached to a plurality of articles G.

Likewise, as the product G moves according to the factory process, three or more antennas AP periodically receive signal strength from the sensor S, and the location confirmation unit 210 receives the received signal. By using the intensity and the installation positions AP1, AP2, and AP3 of three or more antennas AP, the position of the article G can be periodically checked.

Unlike the data collection and learning steps described above, there is no need to enter a status. This is because it is possible to estimate the state by artificial intelligence. If the state estimation unit 220 knows the identifier (ID) of the first identified item (G), the time, and the time-dependent location, the artificial intelligence learning unit 150 applies the result learned by the article to be periodically checked. The state for each position of (G) is estimated.

Now, the data setting unit 230 sets, for each article G, a data set including an identifier ID, a location over time, and an estimated state.

Next, the factory process modeling execution unit 260 performs factory process modeling of the factory F to be modeled using the set data set. The process modeling performed here includes the movement path and state of the article G. An example is shown in FIG. 11.

11 simplifies the movement path of the article G. Modeling without simplification is possible, but in this case, a path using XY coordinates in which the article G is actually moved in the factory F will be modeled. In FIG. 11, the English language described above of each node means a state.

In another embodiment of the present invention, it is possible to set additional data sets. As shown in FIG. 5, the existing data set described above includes only the identifier (ID), the position over time (X and Y coordinates), and the estimated state, but in this embodiment, the previous average position, speed, It includes any one or more of a process type (Ptype), a type of object, a type of adjacent object, and a distance of the adjacent object. The accuracy of the plant process modeling can be increased by using these data, and various information for improving plant efficiency can be provided.

6 explains how to set the "previous average position".

As in the illustrated example, the previous average position of the object A is the position at a plurality of previous viewpoints (t _k-1 , t _k-2 , t _k-3 ), based on the viewpoint t _k to be checked. It is set using the average values of the X and Y coordinates (P _k-1 , P _k-2 , and P _k-3 ).

If you set "previous average position" and learn AI, factory process modeling accuracy increases. For example, even if the positions of different items at the current point of view are all the same coordinate L, the item having the previous average position L1 is a part of the right arm, so it will move to the coordinate L3, and the previous average position is L2. Since the article is a part of the left arm, the process can be modeled to move to coordinate L4.

7 explains how to set the "speed". Here, speed includes current speed, past speed, and future speed.

The speed is set based on a time point to be checked, using a movement distance calculated using a position at a current time point and a position at a different time point, and a time between both time points. For example, in the example shown in FIG. 7, the current speed of object A can be set to d (P _k-1 , P _k ) / (t _k -t _k-1 ), and the past speed is d (P _k-3 , P _k ) / (t _k -t _k-3 ), and future speed can be set to d (P _{k + 3} , P _k ) / (t _{k + 3} -t _k ). have.

Setting "speed" and learning AI will increase plant process modeling accuracy. For example, if the future velocity of object A is zero, it can be learned to enter the assembly process.

8 illustrates a method of setting the "process type (Ptype)".

The process type may be determined according to whether different articles G are in different positions and then changed to the same position.

In the illustrated example, each article (G) is divided into either a general object (0) or a painted object (1), the process when the different general object (0) is in a different location and changed to the same location The Ptype may be set to "Assembly", and the process type Ptype may be set to "Painting" when the normal object 0 and the painted object 1 are in different positions and then changed to the same position. This is only an example, and in addition to these examples, it is possible to set various process types (Ptypes) by using a change in the type and position that are classified for each article.

When setting the "Ptype" and learning AI, more accurate estimation of the state is possible, so the accuracy of the factory process modeling increases.

The "type of object" may be a preset value for each group of articles G. A part, B part, and C part are given according to the type.

The " kind of adjacent object " is set by confirming the item closest to the position of the object to be checked using the position of the object G, and then confirming the type of the object of the identified object.

The "distance of the adjacent object" is to select the closest item to the item having a different kind of object from among items close to the position of the item to be checked using the position of the item G, and then to check the position of the selected item. It is set by calculating the distance between the positions of the articles. As illustrated in FIG. 9, the “□” which is the same kind among the adjacent objects of the object A is excluded, and since “○” is the closest, the distance of the adjacent object is set to d.

An example of a data set including an additional data set set in this way is illustrated in FIG. 10.

As described above, the present specification has been described with reference to the embodiments illustrated in the drawings so that those skilled in the art can easily understand and reproduce the present invention, but these are merely exemplary, and those skilled in the art can make various modifications and equivalents from the embodiments of the present invention. It will be understood that embodiments are possible. Therefore, the protection scope of the present invention should be defined by the claims.

100: artificial intelligence module
110: location check
120: status input
130: data collection unit
200: factory process modeling module
210: location confirmation unit
220: state estimation unit
230: data setting unit
260: factory process modeling execution unit

Claims (12)

(a) data collection and learning steps; And
(b) a factory process modeling method comprising the step of performing factory process modeling,
The (a) data collection and learning step,
(a1) Three or more antennas (AP ₀ ) are installed in the factory (F ₀ ) for data collection, and the installation location (AP ₀ 1, AP ₀ 2, AP ₀ 3) is stored in the location confirmation unit 110, Attaching the sensor S to the plurality of articles G ₀ ;
(a2) three or more antennas AP ₀ periodically receiving signal strengths from the sensor S as the article G ₀ moves according to the factory process;
(a3) The position confirmation unit 110, the received signal strength and the installation position (AP ₀ 1, AP ₀ 2, AP ₀ 3) of the three or more antennas (AP ₀ ), using the article (G ₀ ) Periodically checking the location;
(a4) receiving, by the status input unit 120, the position of the article G ₀ , which is periodically checked, the status of the article to which the sensor is attached; -Here, the state of the input item is one selected from a group of states of the preset item.
(a5) Data including the data collection unit 130, for each item G ₀ , an identifier ID, a position according to the time identified in step (a3), and a state input in step (a4). Collecting a set of data;
(a6) steps (a1) to (a5) are repeated in a factory for collecting multiple data; And
(a7) the artificial intelligence learning unit 150, using the data set collected in the step (a6), includes performing the artificial intelligence learning on "location" and "state",
The (b) factory process modeling step,
(b1) Three or more antennas AP are installed in the factory F to be modeled, and the installation positions AP1, AP2, and AP3 are stored in the location confirmation unit 210, and sensors are applied to a plurality of articles G. Step (S) is attached;
(b2) three or more antennas AP periodically receiving signal strengths from the sensors S as the article G moves according to the factory process;
(b3) The position checking unit 210 periodically checks the position of the article G by using the received signal strength and the installation positions AP1, AP2, and AP3 of three or more antennas AP. step;
(b4) The state estimator 220 applies information about the position of the article G identified in the step (b3) to the results learned about the "position" and the "state" in the step (a7). By doing so, estimating the state for each position of the article (G) periodically identified in the step (b3);
(b5) Data set unit 230, for each article (G), a data set including an identifier (ID), a position over time identified in step (b3), and a state estimated in step (b4) setting (data set); And
(b6) the factory process modeling execution unit 260 includes performing the factory process modeling of the factory F to be modeled using the data set set in step (b5),
The process modeling performed in the step (b6) includes a movement path and a state of the article G,
Factory process modeling method.
According to claim 1,
The step (b5) further includes setting an additional data set by further using information included in the set data set,
Step (b6) is,
The factory process modeling execution unit (000) is a step of performing factory process modeling using the data set and the additional data set set in step (b4),
Factory process modeling method.
According to claim 2,
The additional data set includes any one or more of the previous average position, speed, process type (Ptype), type of object, type of adjacent object, and distance of the adjacent object,
Factory process modeling method.
The method of claim 3,
The additional data set includes a process type (Ptype),
The process type (Ptype) is determined according to whether different articles (G) are in different locations and then changed to the same location,
Factory process modeling method.
The method of claim 4,
Each article (G) is divided into either a general object (0) or a painted object (1),
The process type is set to "Assembly" when different normal objects (0) are in different positions and then changed to the same position, and the normal object (0) and the painted object (1) are in different positions and then in the same position. If changed to the process type is set to "painting",
Factory process modeling method.
The method of claim 3,
The additional data set includes the previous average position,
The previous average position is set by using the average values of X coordinates and Y coordinates, which are positions from a plurality of previous viewpoints, based on a time point to be checked.
Factory process modeling method.
The method of claim 3,
The additional data set includes speed,
The speed is set based on a time point to be checked, using a movement distance calculated using a position at a current time point and a position at a different time point, and a time between both time points,
Factory process modeling method.
The method of claim 3,
The additional data set includes the type of object,
The type of the object is a preset value for each group of items G,
Factory process modeling method.
The method of claim 8,
The additional data set includes types of adjacent objects,
The type of the adjacent object is set by confirming an item closest to the position of the item to be checked using the position of the item G, and then confirming the type of the object of the identified item.
Factory process modeling method.
The method of claim 8,
The additional data set further includes the distance of adjacent objects,
The distance of the adjacent object is to determine the location of the selected item after selecting the closest item as an item having a different object type from among the items close to the position of the item to be checked using the position of the item G. Set by calculating the distance between the locations of the goods to be said,
Factory process modeling method.
According to claim 1,
Steps (a2) to (a3) are performed using indoor positioning technology,
Step (a7) is performed using deep learning,
Factory process modeling method.
According to claim 1,
The state input in the step (a4) includes "move" and "stop",
Factory process modeling method.

Images