JP2021189466A - Factor estimation system and device - Google Patents

Abstract

To provide a system and a device for accurately estimating factors of defects occurring at an actual manufacturing site even if measures against the factors of the defects are not always one-to-one (do not match) and even when on-site measures are not always accurate.SOLUTION: Including defect patterns 107, a factor narrowing section 102 indicating the correspondence between observed values and locations requiring defect-measures, a 4M current flow determination section 104 that determines which of 4M influences the factor most based on a current flow, and a model re-learning section 105 that retrains the trained statistical model, factor estimation is made highly accurate based on defect-measures taken by on-site users and 4M information 101.SELECTED DRAWING: Figure 1

Description

The present invention relates to a technique for estimating the cause of a predetermined event such as a defect occurring in a product in a process. Among them, in particular, the present invention relates to a technique for estimating factors that cause product defects that occur on a production line.

Deliverables in some process may have defects that cannot achieve the initially expected, planned, specifications, quality, and accuracy. For example, defective products may occur on the production line. Patent Documents 1 and 2 have been proposed as countermeasures against this.

Patent Document 1 describes the defect rate required from the defect record, the countermeasure content and the countermeasure time required from the defect record, for the purpose of supporting the judgment of the effectiveness of the countermeasure and the selection of the countermeasure. Manage by associating. Of the countermeasures implemented in the past, those that did not work, those that had a remarkable effect, and those that were new, and the defect rate before and after the execution time of the countermeasure, correspond to the output device. Output with "" (see summary).

Further, Patent Document 2 states that "any one of jigs / equipment, manufacturing methods, parts / materials, work sets, manufacturing systems, time information, position information, and state information, which are elements in the manufacture of products, or these. The storage means of the information processing device is an element in the manufacturing of the product, in response to the problem of providing the information processing device that presents the position information and the state information of the current product associated with them by using the combination. A search means that uniquely identifies a certain jig / equipment, manufacturing method, parts / materials, work set, and manufacturing system, stores time information, position information, and status information in association with the product identification code. Is the position of the current product associated with the search target stored in the storage means, with any one or a combination of the element identification code, the time information, the position information, and the state information as the search target. The information and the state information are searched, and the presenting means presents the search result by the search means. "

Japanese Unexamined Patent Publication No. 08-297699 Japanese Unexamined Patent Publication No. 2011-242830

Patent Document 1 describes the mechanism of the manufacturing defect analysis support system. However, in the manufacturing defect analysis support system of Patent Document 1, the countermeasures against the causes of defects occurring at the actual manufacturing site are not necessarily one-to-one (mismatch), and the countermeasures at the site are not always accurate. In the absence of this, incorrect output can be counterproductive to assistance. For example, if a defect dependent on a manufacturing device has appeared for a long time, but if a component (Material) factor is suspected and countermeasures are continued at the site, incorrect information will be accumulated in the countermeasure history. In this case, in a Human-in-the-loop (HITL) type system in which a user such as a worker inputs a countermeasure result, learns it, and reflects it in the estimated content, erroneous learning occurs. Therefore, it is difficult to apply it in the field as it is.

Further, in Patent Document 2, the manufacturing elements such as jigs and equipment and the status of related products are output. However, in Patent Document 2, although it is possible to grasp the current state of the product, the analysis of the cause of the defect is not considered.

Therefore, it is an object of the present invention to provide a more accurate factor analysis technique for defects and the like.

The present invention for solving the above problems includes the following aspects.

In a factor estimation system that estimates the cause of defects in events that occur in a process using a learning model, input of multiple peripheral data indicating the status of the elements of the process is accepted, and the degree of fluctuation of the accepted peripheral data is determined. It has a tidal current determination unit that executes trend analysis for analyzing the fluctuation trend shown, and a model re-learning unit that periodically executes learning of a learning model stored in a storage device based on the executed trend analysis. Then, in the factor narrowing unit, the peripheral data that affects the factor is specified from a plurality of peripheral data by using the learning model learned in the model re-learning unit, and the tidal current determination unit uses the fluctuation trend. Further, the analysis can be performed using the peripheral data specified by the factor narrowing down unit.

Here, the factor narrowing unit may be possessed by either the factor estimation system or a terminal device connected to the factor estimation system via a network.

The present invention also includes a computer program product (the program itself or a medium in which the program itself is stored) for causing the factor estimation system to function as a computer device, and a method thereof.

According to the present invention, it is possible to estimate the cause of the defect in the process more accurately, and it is possible to deal with the defect more appropriately.

Issues, configurations and effects other than those described above will be clarified by the following description of the embodiments.

An example of the functional configuration of the factor estimation system in Example 1. An example of a production line to which an embodiment of the present invention is applied. An example of a defective pattern when a work is dropped An example of the display contents in the factor display / countermeasure input function unit (display / input terminal) in Example 1. An example of the functional configuration of a conventional factor estimation system Deformation example of defect factor display area An example of a flowchart showing the processing flow of the model re-learning unit in the embodiment of the present invention. An example of a trained statistical model An example of a production line having a supply device to which the embodiment of the present invention is applied. An example of attaching a camera to the hand at the tip of a robot An example of the output of a camera attached to the hand at the tip of the robot An example of a trained statistical model according to an embodiment of the present invention An example of the functional configuration of the factor estimation system of Example 2 An example of the display contents in the estimation factor location display / factor display / countermeasure input function unit (display / input terminal) in Example 2. An example of the hardware configuration of the factor estimation system in the embodiment of the present invention. An example of a functional configuration when the factor estimation system in the embodiment of the present invention is implemented as a server device. An example of a functional configuration when the factor estimation system in the embodiment of the present invention is implemented as a cloud system.

Hereinafter, Example 1 of the present invention will be described. In the first embodiment, factor estimation for defects in the production line (manufacturing process) will be described as an example. Therefore, as an example of peripheral data, it is assumed that the degree of variation (4M trend) of 4M (Man, Machine, Material, Method) indicating the manufacturing element is used. In this embodiment, the 4M trend is used, but other fluctuation trends may be used.

An example of a production line to which this embodiment is applied is shown in FIG. In FIG. 2, the work 206 is moving on the belt conveyor 205. Then, when the work 206 reaches a predetermined position, the robot 203 picks up the work 206 with the hand at the tip of the robot and stores it in the storage box 204. In this production line, this series of movements is repeated.

An enlarged view of the vicinity of the hand at the tip of the robot is shown in 207 of FIG. This is an example of a situation when a gap occurs between the work 206 (grasping object) and the hand at the tip of the robot 203. It shows a state in which a gap occurs between the work 206 (grasping object) and the hand at the tip of the robot 203, the work 206 cannot be fully grasped by the hand at the tip of the robot 203, and the work falls. As an example of a defect in the production line of this embodiment, the drop of this work can be mentioned as an example.

When this defect occurs, the operator 201 confirms the operating status of the robot 203 and the belt conveyor 205, and confirms the gripping state of the gripped object which is the work 206. As a result, when a defect (error) occurs, the content and countermeasures are reported via the display / input terminal 202. Further, the worker 201 can obtain information on the defect via the display / input terminal 202.

In order to explain the manufacturing process and the contents of 4M, first, the factor estimation technique in the prior art of the present embodiment will be described, and then the present embodiment will be described with reference to the drawings.

FIG. 5 is an example showing the functional configuration of the factor estimation system 500 utilizing the conventional Human-in-the-loop (HITL) machine learning used in the production line shown in FIG. 2 and the like. In this embodiment as well, HITL machine learning will be utilized in the same manner.

The factor estimation system 500 utilizing HITL machine learning accepts a defective pattern 107 and has a learned statistical model unit 106, a factor narrowing unit 102, a factor display / countermeasure input function unit 103, and a model relearning unit 105.

The factor estimation system 500 is realized by a so-called computer system. Therefore, the hardware configuration is similar to the factor estimation system 100 of this embodiment described later (see FIG. 15). That is, the input / output unit 13 for receiving the defective pattern 107 is provided. Then, each function of the factor estimation system 500, the factor narrowing unit 102, and the model re-learning unit 105 can be realized by the processing unit 11. The trained statistical model unit 106 can be realized by the storage unit 12 for storing the trained statistical model 501, the database 16 connected via the communication path 15 such as a bus, or the internal network 14. Further, the factor display / countermeasure input function unit 103 can be realized by the display / input terminal 202 connected via wireless communication. In this embodiment, the display / input terminal 202 is configured via wireless communication, but may be configured to be connected via a communication path 15 or an internal network 14.

Hereinafter, the functions of each configuration will be described. The factor estimation system 500 of the prior art does not have the 4M power flow determination unit 104 and the 4M power flow knowledge unit 108 as compared with the factor estimation system 100 of this embodiment, and the information handled is partially different. Therefore, even in the configuration of the conventional factor estimation system 500, some of the functions of this embodiment will be described. Therefore, in the explanation of the factor estimation system 100 of this embodiment described later, the differences from the factor estimation system 500 will be mainly described.

First, the factor narrowing-down unit 102 is a portion for selecting the cause of the defect based on the information of the defect pattern 107. As an example, when the defective pattern is "PICK NG" of the robot 203, the learned statistical model unit 106, which will be described later, is searched, and "mismatch between the gripping parameter of the hand at the tip of the robot and the size of the work" is selected as the defective factor. Extract.

The factor display / countermeasure input function unit 103 is a part that displays the cause of the defect narrowed down by the factor narrowing unit 102 and inputs the countermeasure result at the site. This is the content displayed on the display / input terminal 202 shown in FIG. 2, and an example of this is shown in FIG.

This display includes a factor display area 403 indicating the cause of the defect narrowed down by the factor narrowing unit 102. In FIG. 4, as an example of the factor display area 403, “mismatch between the gripping parameter of the hand at the tip of the robot and the size of the work” is displayed as a defect factor.

The on-site countermeasure recording area 404 is the content actually taken by the worker 201 on the site, and receives the content input from the worker 201 via the input interface. In the example of FIG. 4, "readjustment of the gripping parameter of the hand at the tip of the robot" is input.

Next, returning to FIG. 5, the model re-learning unit 105 will be described. The model re-learning unit 105 relearns the trained statistical model 501 stored in the trained statistical model unit 106 based on the contents input via the defective pattern 107 and the field countermeasure recording area 404.

Here, the method of re-learning in the model re-learning unit 105 will be described with reference to FIG. 7. FIG. 7 is an example of a flowchart showing the processing flow of the model re-learning unit 105.

First, when this process is started in step S701, the model re-learning unit 105 reads out the trained statistical model 501 from the trained statistical model unit 106 in step S702. FIG. 8 shows an example of this trained statistical model 501. As shown in FIG. 8, the trained statistical model 501 is tabular data that stores defective patterns, countermeasures, and corresponding weights.

Returning to FIG. 7, the explanation of the processing contents is continued. In step S703, the model re-learning unit 105 searches for a bad pattern sequence for the trained statistical model 501 and searches for the robot's PICK NG. That is, the information corresponding to the error content shown in FIG. 6 described later is searched.

Next, in step S704, the model re-learning unit 105 determines whether the searched defective pattern matches. That is, the model re-learning unit 105 determines whether the trained statistical model 501 includes PICK NG. Since it is included (YES) in this example, the process proceeds to step S706.

If the model re-learning unit 105 determines that the trained statistical model 501 does not contain PICK NG (NO), the process proceeds to step S705. In step S705, the model re-learning unit 105 adds the robot's PICK NG to the bad row of the trained statistical model 501.

Further, in step S706, the model re-learning unit 105 searches for the countermeasure sequence of the trained statistical model 501 and searches for the readjustment of the gripping parameter of the hand at the tip of the robot. This means exploring the readjustment of retention parameters associated with the bad pattern PICK NG.

Next, in step S707, the model re-learning unit 105 determines whether or not the countermeasure columns match. In the above example, the process proceeds to step S709 because of a match (YES). If they do not match (NO), the process proceeds to step S708, and the model re-learning unit 105 adds readjustment of the grip parameter of the hand at the tip of the robot to the countermeasure row.

Next, in step S709, the model re-learning unit 105 increments the weight of the readjustment of the grip parameter of the hand at the tip of the robot by +1. Then, in step S710, the model re-learning unit 105 saves the trained statistical model 501. This completes the process (S711).

As described above, the weight of # 13 of the trained statistical model is incremented by +1 by the processing of the model re-learning unit 105.

Next, the factor narrowing unit 102 uses the trained statistical model 501 whose weights have been updated above, and uses it as reference information when narrowing down the factors. Specifically, when the defective pattern 107 is input to the factor narrowing unit 102, and when there are a plurality of corresponding factors in the trained statistical model 501, the factors in the row with the heavier weight are selected and narrowed down.

An example of abnormal HITL machine learning was to update the weighting from the actual results in the field based on the factor countermeasure table, but the realization of HITL is not limited to this. For example, there is an example of realizing by the idea of a neural network that takes a bad pattern as an input and takes a countermeasure as an output and optimizes the transition parameter (weighting) of the neurons between them. In addition, it is realized by the idea of optimizing the threshold branching from the apex in the binary tree that takes the bad pattern as input, searches the condition in the binary tree (binary tree), and finally reaches the node representing the countermeasure. There is a way. As described above, HITL can be realized by general multivariate analysis methods, statistical methods, and machine learning methods. This also applies to the present embodiment described later.

The problems in the conventional factor estimation system 500 described so far will be described below.

FIG. 9 is an example of a production line having a supply device 901. The production line 900 is a production line in which a supply device 901 for supplying a work 206 is added to the production line 200 shown in FIG.

The feeder 901, for example, carries the work 206 at a constant speed and places the work 206 in a fixed position on the belt conveyor 205. As an example, the work 206 is placed in the center of the belt conveyor 205. Here, it is assumed that the speed of carrying the work 206 of the supply device 901 has changed for some reason. For example, due to deterioration of the motor of the supply device 901. The work 206 is not set at a fixed position on the belt conveyor 205.

Therefore, a gap occurs between the work 206 and the hand at the tip of the robot 203, the work 206 cannot be fully grasped by the hand at the tip of the robot 203, and the work falls. Here, in the conventional factor estimation system 500, as shown in FIG. 2, the defective pattern "robot PICK NG" is narrowed down due to the discrepancy between the gripping parameter of the hand at the tip of the robot and the size of the work. Therefore, the operator 201 hits the display / input terminal 202 with the readjustment of the gripping parameter of the hand at the tip of the robot as a countermeasure. Or, depending on the weighting situation, it is estimated that the wear of the tip of the robot hand is a factor, and the replacement of the tip member of the hand is taken as a countermeasure at the site.

In any case, in the production line 900 of FIG. 9, although the change in the transport speed of the work 206 due to the deterioration of the motor of the supply device 901 is a factor, the worker 201 is the hand at the tip of the robot or the hand and the work 206. I suspect that the relationship is a factor and take measures. As described above, at the manufacturing site where the worker 201 works, the countermeasures against the factors may not always have a one-to-one relationship (does not match), or the countermeasures at the site may not always be accurate. For example, even though machine-dependent defects have been appearing for a long time, we suspect a component factor and continue to take countermeasures. In addition, as another example, the part where the defective pattern appears and the part requiring countermeasures may be separated. For this reason, since learning is continued with erroneous measures, system support may have an adverse effect, and the conventional factor estimation system 500 does not solve these problems.

As described above, by using the factor estimation system 500 utilizing the conventional HITL machine learning, which is shown as an example in Fig. 5, at the manufacturing site, the following problems can be solved when multiple factors can be considered for the defective pattern. was there. Previously, the worker 201 estimated the factors from his experience and took countermeasures, so he became a personalized person and needed experience before he could complete the defect countermeasures with a short TAT. In addition, there is a problem that it takes time to complete the defect countermeasures. On the other hand, in this embodiment, by accumulating information on factors and countermeasures at the site, it is possible to take countermeasures based on past results, and it is possible to take countermeasures against defects with a short TAT regardless of the experience of the worker. And said.

Next, the present embodiment that solves these problems will be described. FIG. 1 shows an example of the functional configuration of the factor estimation system 100 of this embodiment.

The factor estimation system 100 of this embodiment receives various information such as a defective pattern 107, a human operation, a log, and sensor data from a sensor. As described above, the factor estimation system 100 has a 4M power flow determination unit 104 and a 4M power flow knowledge unit 108 with respect to the factor estimation system 500.

Here, the hardware configuration of the factor estimation system 100 will be described. FIG. 15 shows an example of the hardware configuration of the factor estimation system 100. As described above, the factor estimation system 100 can be realized by a computer system such as a server device. Therefore, the factor estimation system 100 has a processing unit 11, a storage unit 12, and an input / output unit 13. The factor estimation system 100 is connected to the internal network 14 via the communication path 15 connected to the input / output unit 13. The internal network 14 means that it is used in the production lines 200 and 900, and when the factor estimation system 100 is implemented on the cloud, a wide area network such as the Internet may be used.

Then, the factor estimation system 100 can be connected to the sensor 17 of the production line via the internal network 14 and acquire the sensor data via the input / output unit 13. Further, the factor estimation system 100 is connected to the database 16 via the communication path 15 connected to the input / output unit 13. Therefore, the factor estimation system 100 can access the trained statistical model 501, the log, and the 4M power flow knowledge via the input / output unit 13. Further, the factor estimation system 100 can be connected to the factor display / countermeasure input function unit 103, that is, the display / input terminal 202 via wireless communication. Although the display / input terminal 202 is configured via wireless communication, it may be configured to be connected via a communication path 15 or an internal network 14. Further, the display / input terminal 202 can be realized as a terminal device such as a smartphone, a tablet, or a personal computer.

Here, the processing unit 11 can be realized by a calculation unit such as a CPU, and has a factor narrowing unit 102, a 4M power flow determination unit 104, and a model relearning unit 105. These can be realized by an operation according to a program stored in the storage unit 12 such as a memory. That is, the factor narrowing unit 102, the 4M power flow determination unit 104, and the model relearning unit 105 can be implemented as a computer program.

Next, returning to FIG. 1, the contents will be described. In the following, the contents that overlap with the factor estimation system 500 described with reference to FIG. 5 will be omitted.

The defective pattern 107 is information indicating a production state when the product is not produced according to a preset setting or a design, for example, in a production line of a factory. FIG. 3 shows an example of the defective pattern 107 when the work is dropped. FIG. 3 shows, for example, the operation log information output by the robot 203 as an example of the defective pattern 107 in the production line 200 (or the production line 900, the same applies hereinafter). It is generated as text data with the date and time of the state of being grasped by the hand at the tip of the robot and not being bitten. In the defect log 301 when a defect occurs in the production lines 200 and 900, the robot 203 detects that the work 206 cannot be grasped by the hand at the tip of the robot 203 and the work has fallen, and the character string "PICK NG" is used. This is an example of recording a bad condition in a log. The defective pattern 107 is information indicating the production state when the product is not produced according to the preset settings or the design, and may be indicated by the character string OK / NG as in the defective log 301. The above is an example of the defective pattern 107 when the work is dropped. In addition, camera images and moving images showing the appearance of defects, and sensor information obtained by capturing some physical quantity with a sensor are also examples of defective patterns.

In the production line 200 of FIG. 2, when a workpiece falls, the production is stopped in the production line 200, and the worker 201 is notified of the state (not shown in FIG. 2) by turning on a colored light or by e-mail. Will be notified by.

The worker 201 confirms the information of the production line 200 and the defect log 301, and recognizes that the work cannot be grasped by the hand at the tip of the robot 203 and the work has fallen. Here, it is assumed that the worker 201 estimates the cause of the drop, which is a defect, from the defect log 301 in the log stored in the database 16. For example, there was an error in the input information of the CAD (Computer Aided Design) software that designed the production line 200, and the size of the work 206 was slightly small, causing the work to fall. In this case, the operator 201 readjusts the gripping parameter of the hand at the tip of the robot 203. After the readjustment is complete, production line 200 resumes production.

The display / input terminal 202 may display the state of manufacturing defects in such a production line 200, or may record the cause of the defect estimated by the worker 201 and the contents of countermeasures at the site. Here, the display content of the display / input terminal 202 is the same as that in FIG. 4 described above. The details will be described below. The defect occurrence location display area 401 displays the result of receiving the defect log 301 by another computer (not shown), labeling it as a defect of the robot 203, and transmitting it to the display / input terminal 202. Like the defect occurrence location display unit 401, the defect occurrence time display area 402 receives the defect log 301 on another computer (not shown), displays the time information and error details, and displays the result of transmission to the input terminal 202. ing.

Further, in the factor display area 403, as an example, the worker 201 confirms the error content "PICK NG" displayed in the defect occurrence time display area 402, and "the grip parameter of the hand at the tip of the robot and the size of the work". Here is an example of typing "mismatch". A modified example of the factor display area 403 will be described with reference to FIG.

FIG. 6 is a diagram showing a modified example of the defect factor display area in the display / input terminal 202. The display / input terminal 202 uses the defect factor list display area 1501 instead of the factor display area 403 in FIG. The display / input terminal 202 searches the database 16 from the information in the defect occurrence location display area 401 and the defect occurrence time display area 402, and estimates the cause that the grip parameter of the hand at the tip of the robot and the work size do not match. do. In addition, this estimation may be executed by the processing unit 11 of the factor estimation system 100.

Then, the display / input terminal 202 displays "mismatch between the gripping parameter of the hand at the tip of the robot and the size of the work" in the defect factor list display area 1501. As a result, the worker 201 can confirm the display contents of the defect factor list display area 1501 and input a check in the check box or the like.

Returning to FIG. 4, the display contents will be explained. As an example, the on-site countermeasure recording area 404 displays the contents actually taken by the worker 201 on the site and recorded via the display / input terminal 202. As another example of the on-site countermeasure recording area 404, the processing unit 11 of the factor estimation system 100 or the display / input terminal 202 stores the countermeasure contents in the form of a list in advance. Then, an example is also conceivable in which the processing unit 11 or the display / input terminal 202 displays the list in the field countermeasure recording area 404 in the form of a list so that the worker 201 can select it. The display / input terminal 202 stores the information in the display / input terminal 202 or the database 16 after the input and selection of the worker 201 in the on-site countermeasure recording area 404 is completed. For this purpose, the display / input terminal 202 accepts the selection of the OK button from the worker 201.

Utilizing the information input at these sites, the factor estimation system 100 in this embodiment estimates the cause for the occurrence of the defective pattern 107. Further, the factor estimation system 100 displays the factor estimation result on the display / input terminal 202 as a method for facilitating countermeasures. As a result, the worker 201 inputs the actual countermeasure result to the display / input terminal 202 with reference to the contents of the display / input terminal 202. Hereinafter, returning to FIG. 1, this factor estimation system 100 will be described using an example used for HITL.

The 4M information 101 shown in FIG. 1 is, for example, an element of the production line 900, and is 4M information (Man, Machine, Material, Method) indicating 4M. This 4M information 101 can also be realized by combining sensor data from sensors on the production line, operations on the production line such as the worker 201, and logs. In this embodiment, it is assumed that in the production line 900 of FIG. 9, a sensor for acquiring the 4M information 101 is attached to the hand at the tip of the robot 203.

FIG. 10 shows an example of attaching the camera 1000 to the hand at the tip of the robot. It is assumed that the camera 1000 sequentially acquires, for example, the positional relationship between the position of the hand and the work 206. Here, FIG. 11 shows an example of the output of the camera 1000 attached to the hand at the tip of the robot. 〇 in the figure shows the positional relationship between the hand and the work sensed by the camera 1000 for each grip by the robot 203. If there is a 〇 in the place where centering is written, the positional relationship between the hand and the work is just in the middle, and if it deviates from that, the 〇 moves up or down. In FIG. 11, as the number of grips increases, the position of ◯ also expands from the centering, and the amount of deviation is kept constant from a certain place, and the trend occurs before the timing of grip failure occurrence. You can see that it is.

Therefore, in this embodiment, the 4M tidal current determination unit 104 utilizes the 4M tidal current knowledge from the defective pattern and the 4M trend of the 4M information 101 to estimate the potential cause of the defect and generate a weight. For example, when the 4M power flow determination unit 104 receives the output of the camera 1000, the deviation between the hand at the tip of the robot 203 and the work 206 keeps a certain amount of deviation from the occurrence of gripping failure. Judge. From this, it is presumed that the 4M power flow determination unit 104 is not the cause of the wear of the tip of the robot hand as the cause of the defect.

Further, since the constant deviation amount is kept in FIG. 11, the 4M tidal current determination unit 104 estimates as follows using the 4M tidal current knowledge. The 4M power flow determination unit 104 estimates that there is a problem with the hardware of the robot 203 itself, or that some problem is continuously occurring in the supply device 901, for example, at the stage before that.

Here, there is a case where the worker 201 makes an erroneous factor estimation and takes an erroneous countermeasure. However, in the 4M power flow determination unit 104, by combining the 4M trend of the 4M information 101 and the knowledge about the production line 900 such as the 4M power flow knowledge, it is possible to lower the priority of clearly different factors. .. Further, the 4M power flow determination unit 104 estimates the problem of the supply device 901 as a cause of the manufacturing defect caused by other than the acquisition target (robot 203) of the defective pattern 107 and the 4M information 101.

By the above processing, the model re-learning unit 105 of this embodiment can be re-learned. The content of this re-learning is based on the content described in FIG. 7, and only the differences will be described below.

In the model re-learning unit 105 of this embodiment, the model re-learning is performed using the factors estimated by the 4M power flow determination unit 104, the defective pattern, and the 4M information 101. In the example of this embodiment, the defective pattern is the robot "PICK NG". As a factor, the 4M power flow determination unit 104 determines that it is not a factor of wear on the tip of the robot hand because it keeps a certain amount of deviation in FIG. Therefore, the model re-learning unit 105 relearns by reducing the weight of the factor with the tip wear.

In addition, as a factor, it is conceivable to increase the weight of the factor of the problem of the hardware of the robot 203 itself. However, as an example, when log information from the robot 203 is sequentially collected by 4M information 101, if the problem of the robot hardware itself is not reported, the factor of the problem of the robot hardware itself is a factor. The weight does not have to be increased either. Therefore, in step S709, the model re-learning unit 105 may have some trouble in the supply device 901. Therefore, the model re-learning unit 105 re-learns the model by adding +1 to the weight of the row existing in the supply device 901 as a factor. Do learning. As a result, the retrained trained statistical model shown in FIG. 12 is generated. FIG. 12 shows an example of the trained statistical model 1201 using the processing result of the 4M power flow determination unit 104 using the 4M information 101.

From the above, it is estimated by the factor estimation system 100 of this embodiment that the change in the transport speed of the work 206 is the cause of the poor grip due to the deterioration of the motor of the supply device 901 in the production line 900. become. That is, in the factor estimation system 100, the supply device 901 can be correctly estimated as a cause of failure.

In this embodiment, even if the countermeasures against the factors existing at the manufacturing site do not always have a one-to-one relationship (does not match) and the countermeasures at the site are not always accurate, it is possible to respond more accurately. For example, in this embodiment, it is possible to suppress the case where a machine-dependent defect has appeared for a long time, but the component (Material) factor is suspected and countermeasures are continued, or the location where the defect pattern appears and the location requiring countermeasures are separated. ..

Next, Example 2 in which the function of displaying the estimated factor location is provided for Example 1 will be described. In this embodiment, the worker 201 is highly efficient in detail factors by presenting the defect factor location to the worker 201 from the fluctuation degree (trend) of the peripheral data (for example, 4M (Man, Machine, Material, Method)). The estimation is feasible. Furthermore, in this embodiment, it is possible to catch the event with a short-sighted eye, estimate an erroneous factor, and suppress the trial of countermeasures.

FIG. 13 is an example of the functional configuration of the factor estimation system 1300 of this embodiment. In the factor estimation system 1300, the factor display / countermeasure input function unit 103 of the factor estimation system 100 of the first embodiment has been changed to the estimation factor location display / factor display / countermeasure input function 1301. Then, the estimation factor location display / factor display / countermeasure input function 1301 is connected to the 4M power flow determination unit 104 and accepts the input. The output destination of the estimation factor display / factor display / countermeasure input function 1301 is the model re-learning unit 105. The output from the estimation factor location display / factor display / countermeasure input function 1301 to the model re-learning unit 105 may be output from the 4M power flow determination unit 104. The estimated factor display / factor display / countermeasure input function 1301 can be realized by a terminal device such as a smartphone, a tablet, or a personal computer, similarly to the factor display / countermeasure input function unit 103.

FIG. 14 shows an example of the display contents in the estimation factor location display / factor display / countermeasure input function 1301 (display / input terminal). The estimated factor location display / factor display / countermeasure input function 1301 displays the output of the 4M power flow determination unit 104, that is, the estimated factor location display from the 4M trend, in the display content of the factor display / countermeasure input function unit 103. It has a region 1302. Others have a defect factor list display area 1501 and a site countermeasure recording area 404 as in the first embodiment, but the acquisition source of these information is different from that of the first embodiment. These will be described below.

The estimated factor location display area 1302 displays a list of factor locations estimated by the 4M power flow determination unit 104. As in the example shown in the first embodiment, when the error related to the hardware of the robot 203 is not included in the 4M information 101, the malfunction of the supply device 901 is suspected more than that of the robot hardware. Therefore, in the estimation factor location display area 1302, the contents are displayed in the order of high suspicion (priority order).

Then, in the defect factor list display area 1501, the defect of the supply device having the highest possibility (high suspicion) of the defect factor is displayed in a checked form.

Further, the on-site countermeasure recording area 404 is a column in which the worker 201 confirms the estimation factor location display area 1302 and the defect factor list display area 1501 and inputs, for example, countermeasures in order of priority.

From the above, the factor estimation system 1300 of this embodiment makes it possible to deal with the fact that the countermeasures against the factors existing at the manufacturing site do not necessarily have a one-to-one relationship (does not match). In addition, in this embodiment, the countermeasures at the site are not always accurate (for example, even though machine-dependent defects have appeared for a long time, the parts (Material) factors are suspected and countermeasures are continued. It is also possible to deal with issues such as (when the countermeasures are far apart, etc.). In other words, by showing the worker 201 the location of the presumed factor from the 4M trend, it is possible to encourage efficient countermeasures.

Finally, a modification relating to the implementation of Examples 1 and 2 will be described. As described above, the factor estimation system 100 of the first embodiment and the factor estimation system 1300 of the second embodiment can be realized by a computer system. Therefore, it becomes possible to realize these as a server device or a so-called cloud system installed in a production line or an operating company thereof.

First, FIG. 16 shows the functional configuration when the factor estimation system is implemented as a server device. Although this configuration will be described using the factor estimation system 100 of the first embodiment as an example, it can be similarly implemented by the factor estimation system 1300 of the second embodiment. In FIG. 16, various devices 1601 such as robots, a server device 1602, and a factor display / countermeasure input function unit 103 are configured to connect the internal network 14 to each other. The server device 1602 is the factor estimation system 100. Therefore, this modification can be realized by the hardware configuration shown in FIG.

Next, FIG. 17 shows the functional configuration when the factor estimation system is implemented as a cloud system 1700. Although this example will also be described using the factor estimation system 100 of the first embodiment as an example, it can be similarly implemented by the factor estimation system 1300 of the second embodiment. In this example, each function is provided in the cloud device 1703. Therefore, the server device 1602 installed on the production line side may have a function of relaying the communication of various devices 1601, the factor display / countermeasure input function unit 103, and the cloud device 1703. However, the cloud device 1703 may have a 4M power flow determination unit 104 and a model relearning unit 105, and the server device 1602 may have a factor narrowing unit 102, a learned statistical model 1201, and a 4M power flow knowledge.

Further, when the factor estimation system is implemented in the cloud system 1700 shown in FIG. 17, the relationship with the hardware configuration shown in FIG. 15 is as follows. The factor estimation system 100 of FIG. 15 is mounted on the cloud device 1703. The cloud device 1703 is connected to various devices 1601, a server device 1602, and a factor display / countermeasure input function unit 103 via a wide area network such as the Internet. It is desirable that the various devices 1601, the server device 1602, and the cause display / countermeasure input function unit 103 are connected to each other via the internal network 14.

According to each of the above embodiments, the estimation accuracy can be automatically improved from the fluctuation degree (trend) of the peripheral data (for example, 4M (Man, Machine, Material, Method)).

The present invention is not limited to the above-described embodiment, and includes various modifications. For example, the above-described embodiment has been described in detail in order to explain the present invention in an easy-to-understand manner, and is not necessarily limited to the one including all the described configurations. Further, it is possible to replace a part of the configuration of one embodiment with the configuration of another embodiment, and it is also possible to add the configuration of another embodiment to the configuration of one embodiment. Further, it is possible to add / delete / replace a part of the configuration of each embodiment with another configuration.

Further, each of the above configurations, functions, processing units, processing means and the like may be realized by hardware by designing a part or all of them by, for example, an integrated circuit. Further, each of the above configurations, functions, and the like may be realized by software by the processor interpreting and executing a program that realizes each function. Information such as programs, tables, and files that realize each function can be placed in a memory, a recording device such as a hard disk or SSD (Solid State Drive), or a recording medium such as an IC card, SD card, or DVD.

In addition, the control lines and information lines indicate those that are considered necessary for explanation, and do not necessarily indicate all the control lines and information lines in the product. In practice, it can be considered that almost all configurations are interconnected.

Further, in this embodiment, a production line is illustrated as an example of the process, but it can also be applied to various plans such as distribution bases such as so-called automated warehouses, waste collection facilities, and power plants.

101 ... 4M information, 102 ... factor narrowing down section, 103 ... factor display / countermeasure input function section, 104 ... 4M power flow judgment section, 105 ... model re-learning section, 106 ... learned statistical model section, 107 ... defective pattern, 108 ... 4M tidal current knowledge department

Claims (15)

In a factor estimation system that estimates the factors of defects that occur in products in the process using a learning model.
A storage device for storing the learning model and
A tidal current determination unit that receives input of a plurality of peripheral data indicating the status of the elements of the process and executes a trend analysis that analyzes a fluctuation trend indicating the degree of fluctuation of the received peripheral data.
A model re-learning unit that periodically executes learning of the learning model based on the executed trend analysis.
Using the learning model learned by the model re-learning unit, it has a factor narrowing unit that identifies peripheral data that affects the factor from a plurality of peripheral data.
The factor estimation system is characterized in that the tidal current determination unit further analyzes the fluctuation trend using the peripheral data specified by the factor narrowing unit. In the factor estimation system according to claim 1,
The factor estimation system is characterized in that the power flow determination unit further receives a defective pattern indicating the defective pattern and executes a trend analysis using the defective pattern. In the factor estimation system according to claim 2,
The tidal current determination unit estimates a portion that may be a defect factor of the defect from the received peripheral data, and creates weighted data in learning in the model re-learning unit based on the estimation result. A factor estimation system characterized by that. In the factor estimation system according to any one of claims 1 to 3,
In the terminal device connected to the factor estimation system, peripheral data that affects the factor specified by the factor narrowing unit is displayed, and the countermeasure result input from the user is displayed according to the displayed result. A factor estimation system characterized by outputting to the judgment unit. In the factor estimation system according to any one of claims 1 to 3,
In the terminal device connected to the factor estimation system, the peripheral data affecting the factor specified by the factor narrowing section and the result of the trend analysis executed by the trend determination section are displayed, and the displayed result is displayed. A factor estimation system characterized in that the countermeasure result input from the user is output to the model re-learning unit. A factor estimation system that is a computer device and estimates the cause of a defective event that occurs in a process using a learning model, and has a storage device that stores the learning model.
A tidal current determination unit that receives input of a plurality of peripheral data indicating the status of the elements of the process and executes a trend analysis that analyzes a fluctuation trend indicating the degree of fluctuation of the received peripheral data.
A model re-learning unit that periodically executes learning of the learning model based on the executed trend analysis.
Using the learning model learned by the model re-learning unit, it is made to function as a factor narrowing unit that identifies peripheral data that influences the factor from a plurality of peripheral data.
The tidal current determination unit is a factor estimation program characterized in that the analysis of the fluctuation trend is further executed using the peripheral data specified by the factor narrowing unit. In the factor estimation program according to claim 6,
A factor estimation program characterized by having the tidal current determination unit accept a defective pattern indicating the defective pattern and executing trend analysis using the defective pattern. In the factor estimation program according to claim 7,
The tidal current determination unit is made to estimate a portion that may be a defect factor of the defect from the received peripheral data, and based on the estimation result, the weighted data in the learning in the model re-learning unit is created. A factor estimation program characterized by that. In the factor estimation program according to any one of claims 6 to 8.
In the terminal device connected to the factor estimation system, peripheral data that affects the factor specified by the factor narrowing unit is displayed, and the countermeasure result input from the user is displayed according to the displayed result. A factor estimation program characterized by outputting to a judgment unit. In the factor estimation program according to any one of claims 6 to 8.
In the terminal device connected to the factor estimation system, the peripheral data affecting the factor specified by the factor narrowing section and the result of the trend analysis executed by the trend determination section are displayed, and the displayed result is displayed. A factor estimation program characterized in that the countermeasure result input from the user is output to the model re-learning unit. In a factor estimation system that estimates the causes of defects in events that occur in a process using a learning model.
A tidal current determination unit that receives input of a plurality of peripheral data indicating the status of the elements of the process and executes a trend analysis that analyzes a fluctuation trend indicating the degree of fluctuation of the received peripheral data.
It has a model re-learning unit that periodically executes learning of the learning model stored in the storage device based on the executed trend analysis.
In the factor narrowing section of the terminal device connected via the network, the peripheral data that influences the factor is identified from a plurality of peripheral data by using the learning model learned by the model relearning section.
The factor estimation system is characterized in that the tidal current determination unit further analyzes the fluctuation trend using the peripheral data specified by the factor narrowing unit. In the factor estimation system according to claim 11,
The factor estimation system is characterized in that the power flow determination unit further receives a defective pattern indicating the defective pattern and executes a trend analysis using the defective pattern. In the factor estimation system according to claim 12,
The tidal current determination unit estimates a portion that may be a defect factor of the defect from the received peripheral data, and creates weighted data in learning in the model re-learning unit based on the estimation result. A factor estimation system characterized by that. In the factor estimation system according to any one of claims 11 to 13.
In the terminal device, peripheral data that affects the factor specified by the factor narrowing unit is displayed, and the countermeasure result input from the user is output to the power flow determination unit according to the displayed result. A factor estimation system featuring. In the factor estimation system according to any one of claims 11 to 13.
The terminal device connected to the factor estimation system displays the peripheral data affecting the factor specified by the factor narrowing unit and the result of the trend analysis executed by the power flow determination unit, and responds to the displayed result. A factor estimation system characterized in that the countermeasure result input from the user is output to the model re-learning unit.

Images