CN113723738A

Movatterモバイル変換

Info

Publication number: CN113723738A
Application number: CN202110541572.0A
Authority: CN
Inventors: 田代和幸; 樱井祐市; 西纳修一
Original assignee: Hitachi Ltd
Current assignee: Hitachi Ltd
Priority date: 2020-05-25
Filing date: 2021-05-18
Publication date: 2021-11-30
Also published as: JP2021189466A

Abstract

Translated fromChinese

本发明的课题在于，即使在实际的制造现场产生的、针对不良的要因的对策未必是一对一的关系(不一致)、现场的对策未必正确的情况下，也高精度地推定不良的要因。为了解决上述课题，具有：不良图案(107)；要因筛选部(102)，其表示观测值与要对策部位的对应；4M潮流判定部(104)，其从潮流中判定4M中的哪个对要因最有影响；以及模型再学习部(105)，其进行学习完成统计模型的再学习，根据现场用户的对策内容和4M信息(101)，使要因推定高精度化。

An object of the present invention is to estimate the cause of the defect with high accuracy even when the countermeasures against the cause of the defect are not necessarily in a one-to-one relationship (inconsistency) and the countermeasures on the spot are not necessarily correct, occurring at the actual manufacturing site. In order to solve the above-mentioned problems, there are: a defective pattern (107); a factor screening unit (102), which indicates the correspondence between the observed value and the countermeasure-required part; most influential; and a model re-learning unit (105), which performs re-learning of the statistical model after learning, and increases the accuracy of factor estimation based on the countermeasure content of the on-site user and the 4M information ( 101).

Description

Factor estimation system and device

Technical Field

The present invention relates to a technique for estimating factors of a predetermined phenomenon such as a defect occurring in a product in a process. The present invention particularly relates to a technique for estimating a factor of defective products generated in a production line.

Background

The products in some steps may not meet the predetermined specifications, quality, and accuracy, which are originally expected. For example, a defective product may be generated in a production line. As a countermeasure against this, patent documents 1 and 2 are proposed.

Patent document 1 describes "to perform management by associating a failure rate obtained from a failure actual result with the contents of a countermeasure obtained from a countermeasure actual result and the time of the countermeasure, for the purpose of determining whether or not the effect of a countermeasure is supported, and selecting a countermeasure. Of the countermeasures executed in the past, the output device outputs the contents of the ineffective countermeasures, the effective countermeasures, and the new countermeasures, and the failure rates before and after the execution time of the countermeasures, in association with each other (refer to the abstract).

Patent document 2 describes "a problem of providing an information processing apparatus that presents position information and state information of a current product associated therewith using any 1 or a combination of a jig and a device, a manufacturing method, a component and a material, a work set, a manufacturing system, time information, position information, and state information as an element in manufacturing the product, and that stores, in association with a product identification symbol, identification symbols of the jig, the device, the manufacturing method, the component and the material, the work set, and the manufacturing system, the time information, the position information, and the state information that uniquely identify the element in manufacturing the product, and the search means searches for any 1 or a combination of the identification symbols of the element, the time information, the position information, and the state information, and a presentation unit that presents the retrieval result of the retrieval unit by retrieving the position information and the state information of the current product stored in the storage unit and associated with the retrieval target.

Documents of the prior art

Patent document 1: japanese laid-open patent publication No. H08-297699

Patent document 2: japanese patent laid-open publication No. 2011-

Disclosure of Invention

Problems to be solved by the invention

Patent document 1 describes a configuration of a manufacturing failure analysis support system. However, in the manufacturing failure analysis support system of patent document 1, if the countermeasures against the causes of the failure are not necessarily in a one-to-one relationship (disagreement) and the countermeasures at the site are not necessarily correct, the support may be adversely effected by the erroneous output. For example, even when a failure dependent on a manufacturing apparatus occurs long, and a countermeasure is continued for a reason of a suspected component (Material) in the field, wrong information is accumulated in the countermeasure history. In this case, in a Human-in-the-loop (hitl) system in which a user such as an operator inputs a countermeasure result and learns and reflects the result in the estimated content, erroneous learning is performed. Therefore, it is difficult to directly perform field application.

In addition, in patent document 2, the conditions of the manufactured elements such as jigs and devices and the products related to them are outputted. However, in patent document 2, although the current state of the product can be grasped, no consideration is given to analysis of factors of failure.

Therefore, the present invention has an object to provide a factor analysis technique for a failure or the like with higher accuracy.

Means for solving the problems

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

A factor estimation system for estimating factors of defects in a phenomenon occurring in a process using a learning model, comprising: a power flow determination unit that receives input of a plurality of pieces of peripheral data indicating conditions of elements of the process, and performs trend analysis for analyzing a fluctuation trend indicating a fluctuation degree of the received peripheral data; and a model relearning section that periodically performs learning of a learning model stored in a storage device based on the executed trend analysis. Further, the factor screening unit may specify the peripheral data affecting the factor from a plurality of peripheral data using the learning model learned by the model relearning unit, and the power flow determination unit may analyze the trend of the change using the peripheral data specified by the factor screening unit.

Here, the factor estimation system or a terminal device connected to the factor estimation system via a network may have a factor screening unit.

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

Effects of the invention

According to the present invention, the cause of a defect in the process can be estimated with higher accuracy, and a more appropriate response to the defect can be made.

Problems, structures, and effects other than those described above will be apparent from the following description of the embodiments.

Drawings

Fig. 1 shows an example of a functional configuration of a factor estimation system in embodiment 1.

Fig. 2 shows an example of a production line to which an embodiment of the present invention is applied.

Fig. 3 shows an example of a defective pattern when the workpiece is dropped.

Fig. 4 is an example of display contents in the factor display and countermeasure input function section (display and input terminal) in embodiment 1.

Fig. 5 shows an example of a functional configuration of a conventional factor estimation system.

Fig. 6 is a modification of the defective factor display region.

Fig. 7 is an example of a flowchart showing a processing flow of the model relearning unit in the embodiment of the present invention.

Fig. 8 shows an example of the learning completion statistical model.

Fig. 9 shows an example of a production line having a supply device as an application target of the embodiment of the present invention.

Fig. 10 shows an example of mounting a camera to a robot hand at the tip of the robot.

Fig. 11 shows an example of the output of the camera of the robot hand attached to the robot tip.

Fig. 12 is an example of a learning completion statistical model according to an embodiment of the present invention.

Fig. 13 shows an example of a functional configuration of a factor estimation system according to embodiment 2.

Fig. 14 shows an example of display contents in the estimated factor part display and factor display and countermeasure input function unit (display and input terminal) in example 2.

Fig. 15 shows an example of the hardware configuration of the factor estimation system in the embodiment of the present invention.

Fig. 16 is an example of a functional configuration in a case where the factor estimation system in the embodiment of the present invention is installed as a server device.

Fig. 17 is an example of a functional configuration in a case where the factor estimation system in the embodiment of the present invention is installed as a cloud system.

Description of the reference numerals

101 … 4M information, 102 … factor screening section, 103 … factor display and countermeasure input function section, 104 … 4M power flow determination section, 105 … model relearning section, 106 … learning completion statistical model section, 107 … failure pattern, 108 … 4M power flow knowledge section.

Detailed Description

[ example 1]

Hereinafter, the present invention will be described with respect to example 1. In example 1, the following description will be given taking an example of the estimation of the cause of a failure in a production line (manufacturing process). Therefore, as an example of the peripheral data, a variation degree (4M trend) of 4M (Man, Machine, Material, Method) indicating a manufactured element is used. In the present embodiment, the 4M trend is used, but other trend may be used.

Fig. 2 shows an example of a production line to which the present embodiment is applied. In fig. 2, aworkpiece 206 is moving on abelt conveyor 205. Then, when theworkpiece 206 reaches a predetermined position, therobot 203 picks up theworkpiece 206 with a robot hand at the top end of the robot and stores it in thebin 204. In the production line, the series of actions is repeated.

In fig. 2, reference numeral 207 denotes an enlarged view of the vicinity of the hand at the robot tip. This is a diagram illustrating an example of a situation in which a workpiece 206 (a workpiece) and a hand at the tip of therobot 203 are misaligned. A case is shown in which a deviation occurs between the workpiece 206 (a holding object) and the hand at the tip of therobot 203, theworkpiece 206 is not held by the hand at the tip of therobot 203, and the workpiece falls. The falling of the workpiece is exemplified as a defect in the production line of the present embodiment.

When the failure occurs, theoperator 201 confirms the operating conditions of therobot 203 and thebelt conveyor 205, and confirms the gripping state of theworkpiece 206. As a result, when a failure (error) occurs, the content and the countermeasure are reported via the display andinput terminal 202. Further, theworker 201 can obtain information related to the failure via the display andinput terminal 202.

In order to explain the manufacturing process and the content of 4M, first, a factor estimation technique in the prior art of the present embodiment is explained, and then, the present embodiment is explained with reference to the drawings.

Fig. 5 is a diagram showing an example of a functional configuration of afactor estimation system 500 that is used in the production line shown in fig. 2 and the like and that flexibly uses conventional Human-in-the-loop (hitl) machine learning. Furthermore, even in the present embodiment, HITL machine learning is also flexibly applied.

Thecause estimation system 500 that flexibly uses HITL machine learning receives afailure pattern 107, and includes a learning completionstatistical model unit 106, acause screening unit 102, a cause display and countermeasureinput function unit 103, and amodel relearning unit 105.

Thefactor estimation system 500 is implemented by a so-called computer system. Therefore, the hardware configuration is similar to thefactor estimation system 100 of the present embodiment (see fig. 15). That is, the input/output unit 13 receives thedefective pattern 107. The functions of thefactor estimation system 500, thefactor screening unit 102, and themodel relearning unit 105 can be realized by theprocessing unit 11. The learning-completedstatistical model unit 106 can be realized by thestorage unit 12 storing the learning-completedstatistical model 501, or thedatabase 16 connected to thecommunication path 15 or theinternal network 14 via a bus or the like. The factor display/countermeasureinput function unit 103 can be realized by a display/input terminal 202 connected via wireless communication. In the present embodiment, the display andinput terminal 202 is configured to communicate wirelessly, but may be configured to be connected via thecommunication path 15 or theinternal network 14.

The functions of the respective structures will be explained below. In addition, thefactor estimation system 500 of the conventional art is compared with thefactor estimation system 100 of the present embodiment, and the information processed is partially different without the 4M powerflow determination unit 104 and the 4M powerflow knowledge unit 108. Therefore, the configuration of the conventionalfactor estimation system 500 will be partially described as the function of the present embodiment. Therefore, in the explanation of thefactor estimation system 100 of the present embodiment to be described later, differences from thefactor estimation system 500 will be mainly explained.

First, thefactor screening unit 102 selects a portion of a factor of a failure based on information of thefailure pattern 107. As an example, when the defective pattern is "PICK NG" of therobot 203, the learning completionstatistical model unit 106 described later is searched, and "the gripping parameter of the hand at the robot tip does not match the size of the workpiece" is extracted as the defective factor.

The factor display/countermeasureinput function unit 103 is a part for displaying the factors of the failure screened out by thefactor screening unit 102 and inputting the result of the countermeasure in the field. This is the content displayed on the display andinput terminal 202 shown in fig. 2, and an example thereof is shown in fig. 4.

The display includes afactor display area 403 indicating the factor of the failure screened by thefactor screening unit 102. In fig. 4, as an example of thefactor display area 403, "the gripping parameter of the hand at the robot tip does not match the size of the workpiece" is displayed as a failure factor.

The on-sitecountermeasure recording area 404 receives contents input by theworker 201 via the input interface, the contents being contents of countermeasures actually taken by theworker 201 on the site. In the example of fig. 4, "readjustment of gripping parameters of the hand at the robot tip" is input.

Next, returning to fig. 5, themodel relearning unit 105 will be described. Themodel relearning unit 105 relearns the learning-completedstatistical model 501 stored in the learning-completedstatistical model unit 106 on the basis of the content input via the fieldcountermeasure recording area 404 by thefailure pattern 107.

Here, a method of relearning in themodel relearning unit 105 will be described with reference to fig. 7. Fig. 7 is an example of a flowchart showing a processing flow of themodel relearning unit 105.

First, when the present process is started in step S701, themodel relearning unit 105 reads the learnedstatistical model 501 from the learnedstatistical model unit 106 in step S702. Fig. 8 shows an example of the learning completionstatistical model 501. As shown in fig. 8, the learning completionstatistical model 501 is data in a table format in which failure patterns, countermeasures, and weights corresponding to the failure patterns and the countermeasures are stored.

Returning to fig. 7, the description will be continued on the contents of the processing. In step S703, themodel relearning unit 105 searches the learning completionstatistical model 501 for a defective pattern sequence and searches for a PICK NG of the robot. That is, information corresponding to the error content shown in fig. 6 described later is searched for.

Next, in step S704, themodel relearning unit 105 determines whether or not the searched defective patterns match. That is, themodel relearning unit 105 determines whether or not the learning completionstatistical model 501 includes the PICK NG. In the present example, yes is included, and therefore, the process proceeds to step S706.

When themodel relearning unit 105 determines that the PICK NG is not included in the learning completion statistical model 501 (no), the process proceeds to step S705. In step S705, themodel relearning unit 105 adds the PICK NG of the robot to the failure sequence of the learning completionstatistical model 501.

In step S706, themodel relearning unit 105 searches the strategy sequence of the learnedstatistical model 501, and searches for readjustment of the gripping parameters of the hand at the tip of the robot. This means that a search is made for a readjustment of the holding parameter in connection with the bad pattern PICK NG.

Next, in step S707, themodel relearning unit 105 determines whether or not the countermeasure sequence matches. In the above example, it is consistent (yes), and therefore, the process proceeds to step S709. If the matching is not satisfied (no), the process proceeds to step S708, and themodel relearning unit 105 readjusts the gripping parameters of the hand at the tip of the robot to be added to the strategy row.

Next, in step S709, themodel relearning unit 105 adjusts the gripping parameter of the hand at the distal end of the robot by the weight + 1. Then, in step S710, themodel relearning unit 105 stores the learning completionstatistical model 501. This completes the process (S711).

As described above, the weight of #13 for which the statistical model is learned is given by +1 by the processing of themodel relearning unit 105.

Next, thefactor screening unit 102 uses the learning completionstatistical model 501 whose weight has been updated as reference information in the factor screening. Specifically, when thedefective pattern 107 is input to thefactor screening unit 102, if there are a plurality of corresponding factors in the learning completedstatistical model 501, the factors of the rows having a higher weight are selected and screened.

In addition, an example of the abnormal HITL machine learning is to update the weighting according to the actual results of the scene based on the factor countermeasure table, but the implementation of HITL is not limited to this. For example, there is an example realized by an idea of a neural network that optimizes migration parameters (weights) of neurons between them with a bad pattern as an input and a countermeasure as an output. In addition, there are the following methods: the bad pattern is input, a condition is searched for in a binary tree (binary tree), and in the binary tree which finally reaches a node indicating a countermeasure, a threshold value branching from the vertex is optimized. Thus, HITL can be realized by a general multivariate analysis method, a statistical method, or a machine learning method. This is also the same in the present embodiment described later.

In addition, problems in the conventionalfactor estimation system 500 described so far will be described below.

Fig. 9 is an example of a production line having asupply apparatus 901. Theproduction line 900 is a production line in which asupply device 901 for supplyingworkpieces 206 is added to theproduction line 200 shown in fig. 2.

Thefeeding device 901 conveys theworkpiece 206 at a constant speed, for example, and places theworkpiece 206 at a determined position on thebelt conveyor 205. As an example, theworkpiece 206 is placed at the center of thebelt conveyor 205. Here, the speed at which thesupply device 901 conveys theworkpiece 206 is assumed to vary depending on some factor. For example, deterioration of the motor of thesupply device 901. Noworkpiece 206 is disposed at the determined position on thebelt conveyor 205.

Therefore, a deviation occurs between theworkpiece 206 and the hand at the tip of therobot 203, and theworkpiece 206 is not gripped by the hand at the tip of therobot 203 and falls. Here, in the conventionalfactor estimation system 500, as shown in fig. 2, the selection is performed from the bad pattern "robot PICK NG" due to the discrepancy between the gripping parameters of the hand at the robot tip and the size of the workpiece. Therefore, theoperator 201 takes measures to readjust the gripping parameters of the hand at the tip of the robot at the display andinput terminal 202. Alternatively, wear of the distal end portion of the robot hand is estimated as a factor according to the situation of weighting, and a countermeasure for replacement of the robot hand distal end member is taken on site.

In short, in theproduction line 900 of fig. 9, although a change in the conveying speed of theworkpiece 206 due to deterioration of the motor of thesupply device 901 is a factor, theoperator 201 takes measures against the factor that is suspected of the robot hand at the tip of the robot or the relationship between the robot hand and theworkpiece 206. As described above, at the manufacturing site where theworker 201 performs the work, the countermeasures against the factors may not be in a one-to-one relationship (disagreement), and the countermeasures at the site may not be accurate. For example, a machine-dependent failure occurs for a long time, but a component (Material) is suspected and countermeasures are continuously taken. In addition, as another example, a portion where a defective pattern appears may be separated from a portion to be countermeasures. Therefore, since learning is continued with an erroneous countermeasure, system support may have an adverse effect, and these problems are not solved in the conventionalfactor estimation system 500.

As described above, by using thefactor estimation system 500, which is an example of which is shown in fig. 5 and which flexibly uses the conventional HITL machine learning, at the manufacturing site, when a plurality of factors are considered for a defective pattern, there are the following problems. Conventionally, theoperator 201 estimates the cause based on the experience or the like so far and takes measures, and therefore, it is personalized and requires experience until the failure measures can be completed with a short TAT. In addition, there is a problem that it takes time until the failure countermeasure is completed. In contrast, in the present embodiment, by accumulating information on factors and countermeasures at the site, countermeasures can be taken from past results, and countermeasures against a failure at a short TAT can be realized regardless of the experience of the operator.

Next, the present embodiment for solving these problems will be described. Fig. 1 shows an example of a functional configuration of afactor estimation system 100 according to the present embodiment.

Thefactor estimation system 100 of the present embodiment receives various information such as adefective pattern 107, human operations, logs, and sensor data from sensors. As described above, thefactor estimation system 100 includes the 4M powerflow determination unit 104 and the 4M powerflow knowledge unit 108 with respect to thefactor estimation system 500.

Here, a hardware configuration of thefactor estimation system 100 will be described. Fig. 15 shows an example of the hardware configuration of thefactor estimation system 100. As described above, thefactor estimation system 100 can be realized by a computer system such as a server device. Therefore, thefactor estimation system 100 includes: a processingunit 11, astorage unit 12, and an input/output unit 13. Thefactor estimation system 100 is connected to theinternal network 14 via acommunication path 15 connected to the input/output unit 13. Theintranet 14 is used in the

production lines

200 and 900, and a wide area network such as the internet may be used when thefactor estimation system 100 is installed on the cloud.

Thefactor estimation system 100 is connected to thesensors 17 of the production line via theinternal network 14, and can acquire sensor data via the input/output unit 13. Thefactor estimation system 100 is connected to thedatabase 16 via acommunication path 15 connected to the input/output unit 13. Therefore, thefactor estimation system 100 can access the learning completionstatistical model 501, the log, and the 4M power flow knowledge via the input/output unit 13. Thefactor estimation system 100 can be connected to the factor display and countermeasureinput function unit 103, i.e., the display andinput terminal 202, via wireless communication. The display andinput terminal 202 is configured to communicate wirelessly, but may be configured to be connected via thecommunication path 15 or theinternal network 14. In addition, the display andinput terminal 202 can be implemented as a terminal device such as a smartphone, a tablet computer, a personal computer, or the like.

Here, theprocessing unit 11 can be realized by an arithmetic unit such as a CPU, and includes afactor screening unit 102, a 4M powerflow determination unit 104, and amodel relearning unit 105. These operations can be realized by operations according to a program stored in thestorage unit 12 such as a memory. That is, thefactor screening unit 102, the 4M powerflow determination unit 104, and themodel relearning unit 105 can be installed as computer programs.

Next, the contents will be described with reference to fig. 1. Note that, in the following, the contents overlapping with those of thefactor estimation system 500 described with reference to fig. 5 are omitted.

Thedefective pattern 107 is information indicating a production state when production is not performed in accordance with a predetermined setting or design in a production line of a factory, for example. Fig. 3 shows an example of thedefective pattern 107 when the workpiece is dropped. Fig. 3 shows an example of thedefective pattern 107 in the production line 200 (or theproduction line 900, the same applies hereinafter), which is, for example, operation log information output by therobot 203. The state of whether the robot hand at the tip of the robot grips or does not grip is generated as text data together with the date and time. Thefailure log 301 in the

production line

200 or 900 when a failure occurs is an example in which therobot 203 senses that theworkpiece 206 is not gripped by the hand at the tip of therobot 203 and the workpiece is dropped, and records a failure state in a string such as "PICK NG". Thefailure pattern 107 is information indicating a production state when production is not performed according to a predetermined setting or design, and may be represented by a character string such as OK/NG as in thefailure log 301. The above is an example of thedefective pattern 107 when the workpiece is dropped. In addition, a camera image, a moving image, and sensor information in which some physical quantity is captured by a sensor, in which a failure is reflected, are also exemplified as a failure pattern.

In theproduction line 200 of fig. 2, when a workpiece is dropped, the production is stopped in theproduction line 200, and theworker 201 is notified of the state by a notification means such as an electronic mail or the like or the lighting of a colored lamp (not shown in fig. 2).

Theworker 201 confirms the information of theproduction line 200 and thefailure log 301, recognizes that the worker is not caught by the hand at the tip of therobot 203, and drops the workpiece. Here, theworker 201 estimates a cause of a failure, that is, a drop, as described below, from thefailure log 301 among the logs stored in thedatabase 16. For example, there is an error in the input information of cad (computer Aided design) software for which theproduction line 200 is designed, and the size of theworkpiece 206 is subtly small, and therefore, the workpiece is dropped. In this case, theoperator 201 again adjusts the gripping parameters of the hand at the tip of therobot 203. After the readjustment is finished, theproduction line 200 starts producing again.

The display/input terminal 202 may display the state of the manufacturing failure in theproduction line 200, or record the cause of the failure estimated by theoperator 201 or the contents of the countermeasures on the spot. Here, the display contents of the display andinput terminal 202 are the same as those of fig. 4 described above. The details thereof will be described below. The failure occurrenceportion display area 401 receives thefailure log 301 via another computer (not shown), labels the failure of therobot 203, and displays the result transmitted to the display andinput terminal 202. The failure occurrencetime display area 402 receives thefailure log 301 through a different computer (not shown) and displays the result of transmitting the time information and the error content to the display andinput terminal 202, as in the failure occurrencepart display unit 401.

Thefactor display area 403 shows an example in which theoperator 201 confirms the error content "PICK NG" displayed in the failure occurrencetime display area 402 and inputs "the disagreement between the gripping parameter of the hand at the tip of the robot and the size of the workpiece". A modification of thefactor display region 403 will be described with reference to fig. 6.

Fig. 6 is a diagram showing a modification of the display area of the defective factor in the display andinput terminal 202. The display andinput terminal 202 uses the defective factorlist display area 1501 instead of thefactor display area 403 of fig. 4. The display and input terminal 202 searches thedatabase 16 based on the information in the failure occurrenceregion display area 401 and the failure occurrencetime display area 402, and causes the difference between the gripping parameter of the hand estimated as the robot tip and the size of the workpiece. The present estimation may be executed by theprocessing unit 11 of thefactor estimation system 100.

Then, the display/input terminal 202 displays "the holding parameter of the hand at the robot tip does not match the size of the workpiece" in the failure factorlist display area 1501. As a result, theoperator 201 can check the display content of the defective factorlist display area 1501 and input an inspection in a check box or the like.

Returning to fig. 4, the display contents thereof will be described. The fieldcountermeasure recording area 404 displays, as an example, the contents recorded by theworker 201 via the display andinput terminal 202, which are actually taking countermeasures on the field. As another example of the on-sitecountermeasure recording area 404, theprocessing section 11 or the display andinput terminal 202 of thefactor estimation system 100 stores countermeasure contents in a list form in advance. Further, an example is also conceivable in which theprocessing unit 11 or the display andinput terminal 202 displays the list in the fieldcountermeasure recording area 404 to allow theoperator 201 to select the list. In the display andinput terminal 202, after theoperator 201 completes inputting and selecting the on-sitecountermeasure recording area 404, the information is stored in the display andinput terminal 202 or thedatabase 16. Accordingly, the display andinput terminal 202 receives the selection of the OK button from theoperator 201.

Using the information input on the spot, thefactor estimation system 100 of the present embodiment estimates the factor of the occurrence of thedefective pattern 107. In addition, thefactor estimation system 100 displays the estimation result of the factor on the display andinput terminal 202 as a method for facilitating the countermeasure. As a result, theworker 201 refers to the contents of the display andinput terminal 202 and inputs the actual countermeasure result to the display andinput terminal 202. The following description will be made of thefactor estimation system 100 using an example flexibly applied to HITL, returning to fig. 1.

The4M information 101 shown in fig. 1 is, for example, an element of aproduction line 900, and is 4M information (Man, Machine, Material, Method) indicating 4M. The4M information 101 can also be realized by a combination of sensor data from sensors in the production line, operations on the production line by theoperator 201, and logs. In the present embodiment, in theproduction line 900 of fig. 9, a sensor for acquiring the4M information 101 is added to the robot arm at the tip of therobot 203.

Fig. 10 shows an example of mounting thecamera 1000 to the robot hand at the robot tip. The positional relationship between the position of the robot and theworkpiece 206 is acquired in sequence by thecamera 1000, for example. Here, fig. 11 shows an example of an output of thecamera 1000 of the robot hand attached to the robot tip. The good in the figure represents the positional relationship between the hand and the workpiece sensed by thecamera 1000 for the gripping by therobot 203. In the case where the good quality exists in the portion written as the centering, the positional relationship between the robot and the workpiece is just caught in the middle, and in the case where the good quality deviates from the middle, the good quality moves to the next upper side. As can be seen from fig. 11, as the number of times of gripping increases, the position of good gripping also starts from centering, the deviation increases, the deviation amount remains constant from a certain position, and this tendency occurs before the timing of occurrence of poor gripping.

Therefore, in the present embodiment, the 4M powerflow determination unit 104 generates a weight by estimating a portion that may become a cause of a failure by using the 4M power flow knowledge flexibly based on the failure pattern and the 4M trend of the4M information 101. For example, when receiving the output of thecamera 1000, the 4M powerflow determination unit 104 determines that the deviation between the hand at the tip of therobot 203 and theworkpiece 206 is kept at a constant deviation amount from the occurrence of the gripping failure. Thus, the 4M powerflow determination unit 104 estimates that the cause of the failure is not a cause of wear of the distal end portion of the robot hand.

Since a certain amount of deviation is maintained in fig. 11, the 4M powerflow determination unit 104 performs estimation as follows using the 4M power flow knowledge. The 4M powerflow determination unit 104 estimates whether therobot 203 has a problem in hardware itself or whether a problem has occurred continuously in the preceding stage, for example, thesupply device 901.

Here, theoperator 201 may estimate a cause of an error and take measures against the error. However, the 4M powerflow determination unit 104 can reduce the priority by combining the 4M trend of the4M information 101 and knowledge about theproduction line 900 such as 4M power flow knowledge. In the 4M powerflow determination unit 104, the problem of thesupply device 901 is estimated as a cause of a manufacturing failure other than the target (robot 203) of acquisition of thefailure pattern 107 and the4M information 101.

Through the above processing, the relearning by themodel relearning unit 105 of the present embodiment can be realized. The contents of this relearning are the basis for the description of fig. 7, and only the differences will be described below.

Themodel relearning unit 105 of the present embodiment performs model relearning using the factors, the defective patterns, and the4M information 101 estimated by the 4M powerflow determination unit 104. In the example of the present embodiment, the bad pattern is "PICK NG" of the robot. Further, as a factor thereof, the 4M powerflow determination unit 104 determines that it is not a factor that the distal end portion of the robot hand is worn because a certain amount of deviation is maintained in fig. 11. Therefore, the weight of the factor of wear of the tip end portion is reduced, and themodel relearning unit 105 performs relearning.

In addition, as a factor, it is considered to increase the weight of the factor such as the problem of the hardware itself of therobot 203. However, as an example, when log information from therobot 203 is sequentially collected by the4M information 101, the weight of the factor such as the problem of the hardware of the robot may not be increased when the problem of the hardware of therobot 203 is not reported. Therefore, in step S709, since there is a possibility that some trouble may occur in thesupply device 901, themodel relearning unit 105 relearns the model by setting the weight +1 of the line in which thesupply device 901 exists as the factor. As a result, a learning completion statistical model in which relearning is performed as shown in fig. 12 is generated. Fig. 12 shows an example of a learning completionstatistical model 1201 using the processing result of the 4M powerflow determination unit 104 using the4M information 101.

As described above, according to thefactor estimation system 100 of the present embodiment, in theproduction line 900, it is estimated that a change in the conveyance speed of theworkpiece 206 due to deterioration of the motor of thesupply device 901 is a factor of poor gripping. That is, in thefactor estimation system 100, thesupply device 901 can be accurately estimated as a factor of the failure.

In the present embodiment, even if the countermeasures for the factors existing at the manufacturing site are not necessarily in a one-to-one relationship (inconsistency) and the countermeasures at the site are not necessarily accurate, the countermeasures can be more accurately handled. For example, in the present embodiment, it is possible to suppress the case where a component (Material) is suspected of continuing countermeasures due to a long occurrence of a machine-dependent defect, and a portion where a defective pattern occurs is separated from a portion where countermeasures are required.

[ example 2]

Next, example 2 in which a function of displaying a site of an estimated factor is provided in example 1 will be described. In the present embodiment, theoperator 201 can efficiently estimate the detailed factors by presenting theoperator 201 with the failure factor parts based on the degree (tendency) of fluctuation of peripheral data (for example, 4M (Man, Machine, Material, Method)). In addition, in the present embodiment, it is possible to suppress erroneous factor estimation and countermeasure trial before an event is captured with a short or shallow eye.

Fig. 13 shows an example of a functional configuration of thefactor estimation system 1300 according to the present embodiment. Thefactor estimation system 1300 of embodiment 1 is modified from the factor display and countermeasureinput function unit 103 of thefactor estimation system 100 to the estimated factor part display and factor display andcountermeasure input function 1301. Thefunction 1301 for displaying the estimated factor part and the factor and inputting the measure is connected to the 4M powerflow determination unit 104, and receives the input. Further, the output destination of the estimated factor part display and factor display andcountermeasure input function 1301 is themodel relearning unit 105. The output from the estimated factor part display and factor display andcountermeasure input function 1301 to themodel relearning unit 105 may be output from the 4M powerflow determination unit 104. The estimated cause portion display and cause display andcountermeasure input function 1301 can be realized by a terminal device such as a smartphone, a tablet computer, or a personal computer, as in the cause display and countermeasureinput function unit 103.

Fig. 14 shows an example of display contents in the estimated factor part display and factor display and countermeasure input function 1301 (display and input terminal). The estimated factor part display and factor display andcountermeasure input function 1301 has an estimated factorpart display area 1302 that displays the output of the 4M powerflow determination unit 104, that is, the estimated part according to the 4M trend, in the display content of the factor display and countermeasureinput function unit 103. Except for this, the defect factorlist display area 1501 and the on-sitecountermeasure recording area 404 are provided as in example 1, but the sources of acquisition of these pieces of information and the like are different from example 1. These are explained below.

The estimated factorregion display area 1302 displays the factor regions estimated by the 4M powerflow determination unit 104 in a list. Further, as in the example shown in embodiment 1, when the4M information 101 does not include an error relating to the hardware of therobot 203, a failure of thesupply device 901 is suspected as compared with the robot hardware. Therefore, the contents of the estimated causativepart display area 1302 are displayed in the order of high suspicion (priority order).

Then, in the failure factorlist display area 1501, the supply device having the highest possibility of failure factor inspection (high suspicion) is displayed.

The on-sitecountermeasure recording area 404 is a field in which theoperator 201 confirms the estimated factorpart display area 1302 and the failure factorlist display area 1501, and inputs information for taking countermeasures in order of priority, for example.

As described above, thefactor estimation system 1300 according to the present embodiment can cope with a one-to-one relationship (inconsistency) in which countermeasures against factors existing in a manufacturing site are not necessarily one-to-one. In addition, in the present embodiment, it is possible to cope with problems such as inaccurate countermeasures in the field (for example, a case where a component (Material) is suspected of continuing countermeasures due to a long occurrence of a machine-dependent defect, and a portion where a defective pattern occurs is separated from a portion to be countermeasures). That is, by indicating the estimated causative part from the 4M tendency to theoperator 201, efficient countermeasures can be promoted.

Finally, a modification of the mounting of embodiments 1 and 2 will be described. As described above, thefactor estimation system 100 according to embodiment 1 and thefactor estimation system 1300 according to embodiment 2 can be realized by computer systems. Therefore, they can be realized as a server device installed in a production line or an operating enterprise thereof, a so-called cloud system.

First, fig. 16 shows a functional configuration in a case where the estimation system is to be installed as a server apparatus. Note that, although the present configuration is described by taking thefactor estimation system 100 of embodiment 1 as an example, the present configuration can be similarly installed in thefactor estimation system 1300 of embodiment 2. In fig. 16,various devices 1601 such as a robot, aserver device 1602, and a factor display and countermeasureinput function unit 103 are configured to connect theinternal network 14 to each other. Theserver device 1602 is thefactor estimation system 100. Therefore, the present modification can be realized by the hardware configuration shown in fig. 15.

Next, a functional configuration in a case where the estimation system is to be installed as thecloud system 1700 is shown in fig. 17. In this example, thefactor estimation system 100 of embodiment 1 is described as an example, but thefactor estimation system 1300 of embodiment 2 can be similarly installed. In this example, each function is provided to thecloud device 1703. Therefore, theserver device 1602 installed on the production line side may have a function of relaying communication among thevarious devices 1601, the factor display and countermeasureinput function unit 103, and thecloud device 1703. However, thecloud device 1703 may have the 4M powerflow determination unit 104 and themodel relearning unit 105, and theserver device 1602 may have thefactor screening unit 102, the learning completionstatistical model 1201, and the 4M power flow knowledge.

In addition, when the factor estimation system is installed in thecloud system 1700 shown in fig. 17, the relationship with the hardware configuration shown in fig. 15 is as follows. Thefactor estimation system 100 of fig. 15 is installed in thecloud device 1703. Thecloud device 1703 is connected tovarious devices 1601, aserver device 1602, and the factor display and countermeasureinput function unit 103 via a wide area network such as the internet. Thevarious devices 1601, theserver device 1602, and the factor display and countermeasureinput function unit 103 are preferably connected to each other via theinternal network 14.

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

The present invention is not limited to the above-described embodiments, and various modifications are possible. For example, the above-described embodiments are described in detail to explain the present invention easily and understandably, and are not limited to the embodiments having all the configurations described. In addition, a part of the structure of one embodiment may be replaced with the structure of another embodiment, and the structure of another embodiment may be added to the structure of one embodiment. In addition, a part of the configuration of each embodiment can be added, deleted, or replaced with another configuration.

Further, for example, a part or all of each of the above-described structures, functions, processing units, and the like may be realized by hardware by designing an integrated circuit or the like. The above-described structures, functions, and the like may be realized by software by a processor interpreting and executing a program for realizing the functions. Information such as programs, tables, and files for realizing the respective functions can be stored in a memory, a hard disk, a recording device such as an ssd (solid State drive), or a recording medium such as an IC card, an SD card, or a DVD.

The control lines and information lines are illustrated as necessary for the description, and not necessarily all the control lines and information lines in the product. In practice, it is also possible to consider almost all structures connected to one another.

In the present embodiment, although the production line is exemplified as an example of the process, the present invention can be applied to various types of facilities such as a logistics site such as a so-called automated warehouse, a waste collection facility, and a power plant.

Claims

1. A factor estimation system for estimating a factor of a failure occurring in a product in a process using a learning model,

it is characterized in that the preparation method is characterized in that,

the factor estimation system includes:

a storage device that stores the learning model;

a power flow determination unit that receives input of a plurality of pieces of peripheral data indicating conditions of elements of the process, and performs trend analysis for analyzing a fluctuation trend indicating a fluctuation degree of the received peripheral data;

a model relearning unit that periodically performs learning of the learning model based on the executed trend analysis; and

a factor screening unit that specifies peripheral data that affects the factor from a plurality of peripheral data using the learning model learned by the model relearning unit,

the power flow determination unit may analyze the trend of the fluctuation using the peripheral data specified by the factor screening unit.

2. The factor estimation system according to claim 1,

the power flow determination unit also receives a defective pattern representing the defective pattern, and performs trend analysis using the defective pattern.

3. The factor estimation system according to claim 2,

the power flow determination unit estimates a portion that can be a cause of the failure from the received peripheral data, and creates weighted data for learning by the model relearning unit based on the estimation result.

4. 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 that has been specified by the factor screening unit and that has an influence on the factor is displayed, and the result of the countermeasure input by the user is output to the power flow determination unit in accordance with the displayed result.

5. 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 that has been specified by the factor screening unit and that has an influence on the factor and the result of trend analysis performed by the power flow determination unit are displayed, and the result of countermeasures input by the user is output to the model relearning unit in accordance with the displayed result.

6. A factor estimating program is characterized in that,

the factor estimating program causes a factor estimating system, which is a computer device, to estimate a factor of a phenomenon of failure occurring in a process using a learning model and has a storage device for storing the learning model, to function as each of the following parts,

the following parts are included:

7. The factor estimation program according to claim 6,

the power flow determination unit receives a defective pattern indicating the defective pattern, and performs trend analysis using the defective pattern.

8. The factor estimation program according to claim 7,

the power flow determination unit estimates a portion that can be a cause of the failure from the received peripheral data, and generates weighted data for learning by the model relearning unit based on the estimation result.

9. The factor estimation program according to any one of claims 6 to 8,

10. The factor estimation program according to any one of claims 6 to 8,

11. A factor estimating system for estimating a factor of a failure of a phenomenon occurring in a process by using a learning model,

it is characterized in that the preparation method is characterized in that,

the factor estimation system includes:

a power flow determination unit that receives input of a plurality of pieces of peripheral data indicating conditions of elements of the process, and performs trend analysis for analyzing a fluctuation trend indicating a fluctuation degree of the received peripheral data; and

a model relearning unit that periodically performs learning of a learning model stored in a storage device based on the trend analysis performed,

a factor screening unit of a terminal device connected via a network, using the learning model learned by the model relearning unit, for identifying peripheral data affecting the factor and the factor from a plurality of peripheral data,

12. The factor estimation system according to claim 11,

13. The factor estimation system according to claim 12,

14. The factor estimation system according to any one of claims 11 to 13,

in the terminal device, the peripheral data that has been specified by the factor screening unit and that has an influence on the factor is displayed, and the countermeasure result input by the user is output to the power flow determination unit in accordance with the displayed result.

15. The factor estimation system according to any one of claims 11 to 13,

a terminal device connected to the factor estimation system displays the peripheral data that has been specified by the factor screening unit and that has affected the factor and the result of trend analysis performed by the power flow determination unit, and outputs the result of countermeasure input by the user to the model relearning unit in accordance with the displayed result.