JP2022038020A - Inspection system - Google Patents

Abstract

To provide an inspection system for allowing a worker to recognize a change in inspection precision when the inspection precision is changed by a change in an external factor.SOLUTION: An inspection system 10 includes: a determination part 323 for performing quality determination for determining whether an inspection object is in a desired state following a learning model to which a threshold of quality determination is set such that an error detection ratio for showing a degree of error detection for determining that an article that should be determined to be conforming is determined to be nonconforming becomes a setting value; an error detection determination part 343 for performing quality determination for determining whether defect determination performed by the determination part 323 is error detection or not; an error detection rate calculation part 345 for calculating an error detection rate from the number of articles determined to be detected erroneously in the total number of inspected articles; and a precision change reporting part 349 for reporting a change in inspection precision because of a change in the degree of the error detection rate when the calculated error detection rate is changed from a setting value.SELECTED DRAWING: Figure 2

Description

The present disclosure relates to an inspection system in an article assembly process.

In the process of assembling articles, the use of inspection devices utilizing AI (artificial intelligence) is being promoted. For example, Patent Document 1 discloses an example of an image inspection device using a learning model. In this image inspection apparatus, the quality of the second image is determined based on the first evaluation standard learned and processed using the first evaluation parameter of the first image. Then, when the second image is determined to be defective, the quality of the second image is determined based on the second evaluation criterion learned by using the second evaluation parameter.

Japanese Unexamined Patent Publication No. 2019-39727

In the above-mentioned image inspection apparatus, the inspection accuracy may change due to changes in external factors such as secular variation and changes in the inspection environment. In order to improve this lowered inspection accuracy, it is required to perform the learning process again to update the first evaluation standard and the second evaluation standard. However, since there is no function of detecting a change in inspection accuracy, there is a problem that the administrator of the image inspection apparatus cannot recognize the change in inspection accuracy. It should be noted that this problem is not limited to the image inspection device utilizing AI, but is a problem common to various inspection devices utilizing AI.

The present disclosure can be realized in the following forms.

(1) According to one embodiment of the present disclosure, an inspection system in an article assembly process is provided. In this inspection system, in the pass / fail judgment for determining whether or not the inspection target is in a desired state, the over-detection rate indicating the degree of over-detection for determining an article to be judged as a non-defective product as a defective product is set as a set value. A determination unit that determines the quality of the inspection target, an overdetection determination unit that determines whether the defect determination made by the determination unit is over-detection, and an execution unit according to a learning model in which the threshold value for the quality determination is set. The over-detection rate calculation unit that calculates the over-detection rate from the total number of inspections performed, the number of the total number of inspections determined to be over-detected, and the over-detection rate when the calculated over-detection rate changes from the set value. It is provided with an accuracy change notification unit for notifying that the degree of over-detection has changed and the inspection accuracy has changed.
According to the inspection system of the above form, when the inspection accuracy changes due to the change of external factors such as the secular variation and the change of the inspection environment explained in the problem, the worker is notified that the inspection accuracy has changed. Can recognize changes in inspection accuracy. As a result, the administrator of the inspection system can investigate the state of the inspection system, perform the learning process, update the learning model, and improve the inspection accuracy.

The present disclosure can be realized not only in the above-mentioned form of the inspection system but also in various forms such as an inspection accuracy monitoring device and an inspection accuracy monitoring method.

Explanatory drawing which shows the outline of one Embodiment of the inspection system in an assembly process. A block diagram functionally showing each configuration of the inspection system. A flowchart illustrating the data acquisition process by the data acquisition control unit. A flowchart illustrating the rework history management process by the management department. Explanatory drawing which shows an example of the input screen of the rework history input part. A flowchart illustrating the inspection process of the inspection unit. Explanatory drawing which shows an example of the state to be inspected. An explanatory diagram showing the pass / fail judgment based on the discrimination rate obtained according to the learning model. An explanatory diagram showing a problem of pass / fail judgment. A flowchart illustrating the accuracy change detection process of the inspection unit.

A. Embodiment:
FIG. 1 is an explanatory diagram showing an outline of an embodiment of an inspection system in an article assembly process. The inspection system 10 is an image inspection system used in an inspection process executed in an inspection process area after an assembly process area of an assembly line LA of a vehicle (corresponding to an article). The image inspection system takes an image of the part assembled in the previous assembly process area as an inspection target, analyzes the state of the inspection target included in the captured image by AI, and determines whether or not the inspection target is in the desired state. Make a pass / fail judgment.

The inspection system 10 manages a data acquisition unit 20 for acquiring inspection data, an inspection unit 30 for inspecting using the acquired data, and a rework process executed in the rework process area after the inspection process area. It includes a management unit 40 and a management terminal 50 for an administrator.

The data acquisition unit 20 includes a data acquisition unit 22 and a data acquisition control unit 24 that controls the data acquisition unit 22 to acquire data. In the present embodiment, since the inspection system 10 is an image inspection system as an example, for example, an image pickup camera is used for the data acquisition unit 22, and the data acquisition control unit 24 controls the image pickup camera to acquire the data of the image pickup image. Control device is used.

The inspection unit 30 includes an inspection unit 32 and an inspection result display unit 34. As will be described later, the inspection unit 32 inspects the inspection target using the inspection data (captured image data) acquired by the data acquisition unit 20, and also detects the change in the executed inspection accuracy. The inspection result display unit 34 displays the inspection result by the inspection unit 32. The inspection unit 32 is composed of a computer having an AI for inspection. The inspection result display unit 34 is composed of a display device such as a liquid crystal display.

The repair management unit 40 includes a management unit 42, a repair history input unit 44, and a repair completion button 46. When the repair completion button 46 is pressed by the operator, the management unit 42 is notified of the completion of the repair work. The rework history input unit 44 is composed of various terminals such as a tablet terminal, and is an input device for the worker to input information on the rework work performed by the worker as a history. The management unit 42 is composed of various computers such as a server, and is a management device that actually manages the repair process and manages the repair history input by the repair history input unit 44, as will be described later.

The control unit 60 is a control device for controlling the flow of each process of the assembly line LA.

FIG. 2 is a block diagram functionally showing each configuration of the inspection system 10. The data acquisition control unit 24 of the data acquisition unit 20 includes an acquisition instruction receiving unit 241, a control processing unit 243, and a data transfer unit 245. Each of these units is realized by software by at least one processor included in the data acquisition control unit 24. However, the present invention is not limited to this, and may be realized by hardware such as a dedicated circuit. It may also be realized by a combination of software and hardware.

FIG. 3 is a flowchart illustrating the data acquisition process by the data acquisition control unit 24. This data acquisition process is started according to a start instruction from the control unit 60 accompanying the operation of the assembly line LA, and is terminated according to the end instruction.

The control processing unit 243 (see FIG. 2) of the data acquisition control unit 24 first waits in step S112 until the data acquisition instruction from the control unit 60 is received by the acquisition instruction receiving unit 241. Then, when the data acquisition instruction is received by the acquisition instruction receiving unit 241, the control processing unit 243 controls the operation of the data acquisition unit 22 in step S114 to capture an image including the inspection target, and the captured image data. Is acquired as inspection data, and in step S116, the acquired data is transferred to the inspection unit 32 by the data transfer unit 245. Each process of steps S112 to S116 is repeatedly executed until an end instruction is received from the control unit 60 (step S118: NO).

The management unit 42 of the rework management unit 40 shown in FIG. 2 includes a rework history management unit 421, a rework history transmission unit 423, and a stop trigger output unit 431. Each of these parts is realized by software by at least one processor included in the management unit 42. However, the present invention is not limited to this, and may be realized by hardware such as a dedicated circuit. It may also be realized by a combination of software and hardware.

FIG. 4 is a flowchart illustrating the rework history management process by the management unit 42. This rework history management process is started according to a start instruction from the control unit 60 accompanying the operation of the assembly line LA, and is terminated according to the end instruction.

First, in step S152, the rework history management unit 421 of the management unit 42 waits until the rework history is updated by the operator using the rework history input unit 44 (see FIG. 2). Then, when the repair history is updated, the repair history management unit 421 saves the repair history updated in step S154 in a storage area (not shown).

The rework history is updated, for example, as follows. An inspection result for which the state of the inspection target is determined to be defective is displayed as a warning on the inspection result display unit 34, which will be described later, and a worker who confirms the displayed content moves to the repair process area (see FIG. 1). Actually check the condition of the work vehicle to be inspected. As a result of this confirmation, if a rework is required, the worker will rework it. Then, the worker who performed the rework updates the rework history by inputting the information of the rework performed via the input screen displayed on the rework history input unit 44 (see FIGS. 1 and 2), and rework is performed. Complete the process. FIG. 5 is an explanatory diagram showing an example of an input screen of the rework history input unit 44. On the other hand, as a result of the confirmation, if the repair is not necessary, the worker completes the repair process without performing the repair and updating the repair history. The completion of the repair process is notified to the stop trigger output unit 431 (see FIG. 2) by the operator pressing the repair completion button 46 (see FIGS. 1 and 2).

When the stop trigger output unit 431 receives an instruction for monitoring the repair from the inspection unit 32 described later, the stop trigger output unit 431 monitors the completion of the repair by pressing the repair completion button 46 (see FIG. 1). Specifically, if the repair completion button 46 is not pressed by the time the target work vehicle reaches the stop position of the assembly line LA (see FIG. 1), the target work vehicle is stopped at the stop position. The trigger is output to the control unit 60. The control unit 60 that has received the stop trigger stops the target work vehicle at the stop position.

Next, in step S156 of FIG. 4, the repair history management unit 421 monitors whether or not the request for transmission of the repair history from the inspection unit 32 is received by the repair history transmission unit 423. Then, until there is a request for transmission of the repair history (step S156: NO), the repair history management unit 421 repeats the process of waiting for the update of the repair history and saving the updated repair history (steps S152, S154). .. On the other hand, when there is a request for transmission of the repair history (step S156: YES), in step S158, the repair history management unit 421 reads the data of the repair history stored in the storage area, and the repair history transmission unit 423. To the inspection unit 32. Each process of steps S152 to S158 is repeatedly executed until an end instruction is received from the control unit 60 (step S160: NO).

The inspection unit 32 of the inspection unit 30 shown in FIG. 2 includes a data receiving unit 321, a determination unit 323, an inspection result output unit 325, and an inspection result storage unit 327 as inspection execution system blocks. Further, the inspection unit 32, as an accuracy change detection system block, includes an accuracy change determination timing output unit 331, a rework history reception unit 341, an overdetection determination unit 343, an overdetection rate calculation unit 345, and an overdetection rate determination unit. 347 and an accuracy change notification unit 349 are provided. Each of these parts is realized by software by at least one processor included in the inspection unit 32. However, the present invention is not limited to this, and may be realized by hardware such as a dedicated circuit. It may also be realized by a combination of software and hardware.

FIG. 6 is a flowchart illustrating the inspection process by the inspection execution system block of the inspection unit 32. This inspection process is started according to a start instruction from the control unit 60 accompanying the operation of the assembly line LA, and is terminated according to the end instruction.

First, in step S122, when the data reception unit 321 (see FIG. 2) receives the transfer data from the data acquisition control unit 24, the data reception unit 321 stores the transfer data as inspection data in a predetermined storage area. Upon receiving the transfer data in step S122, in step S124, the determination unit 323 analyzes the inspection data according to the learning model by AI and executes the quality determination of the state of the inspection target. Then, the determination result in step S124 is output as an inspection result by the inspection result output unit 325 to the inspection result display unit 34 in step S126, and sequentially stored in the predetermined storage area by the inspection result storage unit 327 in step S128. .. Each of the processes of steps S122 to S128 described above is repeatedly executed until the end instruction is received from the control unit 60 (step S130: NO).

When the determination of the inspection target in step S124 is a defect determination, the inspection result output unit 325 outputs the inspection result to the inspection result display unit 34 and monitors the completion of the repair by the above-mentioned repair management unit 42. Output instructions.

The determination of the determination unit 323 in step S124 is performed, for example, as follows. First, the determination unit 323 analyzes the captured image data acquired as the inspection data according to the learning model by AI, and determines the state of the inspection target as the identification rate of the probability that the state of the inspection target may be the desired state. Ask. FIG. 7 is an explanatory diagram showing an example of a state to be inspected. In FIG. 7, the presence or absence of a nut (indicated by hatching) attached to the bolt is taken as an example of the state to be inspected, the state in which the nut is attached (the state with the nut) is the desired state, and the state with the nut is present. An example of a possible identification rate is shown.

Then, the determination unit 323 determines whether or not the state of the inspection target is a desired state according to the obtained identification rate. FIG. 8 is an explanatory diagram showing a quality determination based on the discrimination rate obtained according to the learning model. The plurality of symbols “◯” shown in FIG. 8 indicate the identification rate for the inspection target included in each inspection data. Then, among the identification rates for the inspection targets included in the plurality of inspection data, the identification rate on the left side of the alternate long and short dash line indicates the identification rate for the inspection object to be judged as a non-defective product, and the identification rate on the right side of the alternate long and short dash line. The rate indicates the identification rate for the inspection target to be judged as a defective product. As shown in FIG. 8, the determination unit 323 determines that the product is a non-defective product when the identification rate for the obtained inspection target is less than the preset threshold value Rsi, and determines that the product is defective when the discrimination rate is equal to or higher than the threshold value Rsi. The threshold value Rsi is set in advance in the learning model so that the over-detection rate indicating the degree of over-detection for determining the state to be determined as a non-defective product as a defective product is a set value. The learning model is set by machine learning using AI deep learning, a neural network, or the like.

FIG. 9 is an explanatory diagram showing a problem of quality determination based on the discrimination rate obtained according to the learning model. The discrimination rate obtained according to the learning model by AI is the change of external factors such as the secular change of the data acquisition unit 20 and the inspection unit 30 and the change of the inspection environment in which the inspection data is acquired by the data acquisition unit 20. It may change as shown in, and the inspection accuracy may change. The example of FIG. 9 shows a state in which the required discrimination rate has decreased and the overdetection rate has decreased. The changes in the inspection environment of the data acquisition unit 20 include changes in external light incident on the image pickup camera, which is the data acquisition unit 22, changes in ambient temperature, changes in humidity, and accumulation of minute misalignment of the image pickup camera. Changes in various environmental conditions that affect the operating state of the imaging camera, such as the accumulation of fine dirt, are exemplified. However, sudden changes in the environmental conditions of the imaging camera due to large misalignment or dirt are out of scope.

In order to improve the decrease in the overdetection rate that has occurred, it is necessary to perform re-learning by AI to update the learning model. Therefore, as described below, the inspection unit 32 of the inspection unit 30 shown in FIG. 2 executes a process of detecting a change in inspection accuracy, specifically, a change in the overdetection rate.

FIG. 10 is a flowchart illustrating the accuracy change detection process by the accuracy change detection system block of the inspection unit 32. This accuracy change detection process is started according to a start instruction from the control unit 60 accompanying the operation of the assembly line LA, and is terminated according to the end instruction.

First, in step S134, the accuracy change determination timing output unit 331 (see FIG. 2) monitors the accuracy change determination timing. Specifically, the timing at which the preset period T elapses is monitored as the accuracy change determination timing.

When the accuracy change detection timing is reached (step S134: YES), in step S136, the repair history receiving unit 341 requests the management unit 42 of the repair management unit 40 to transmit the repair history, and the repair history is transmitted from the management unit 42. Receive the incoming rework history.

Then, in step S138, the over-detection determination unit 343 determines the over-detection state with reference to the inspection results sequentially stored in the storage area by the inspection result storage unit 327 and the received rework history. Specifically, the number of over-detections, that is, the number determined to be defective even though it should have been determined to be non-defective, is actually obtained from the repair history from the number determined to be defective as an inspection result. It can be obtained by subtracting the number modified to.

Next, in step S140, the over-detection rate calculation unit 345 divides the over-detection number Nod by the total number of inspections Nt to calculate the over-detection rate Rod.

Then, in step S142, the over-detection rate determination unit 347 determines whether or not the calculated over-detection rate Rod is within the preset range Rsp ± α centered on the set value Rsp. In the case of Rsp-α≤Rod≤Rsp + α, the change in the overdetection rate, that is, the change in the inspection accuracy is small, so in step S144, the monitoring state of the accuracy change determination timing in the accuracy change determination timing output unit 331 is reset. .. On the other hand, in the case of Rod <Rsp-α or Rod> Rsp-α or Rod> Rsp + α, the change in the overdetection rate, that is, the change in the inspection accuracy is large, and therefore, in step S146, the accuracy change notification unit 349 manages. Notifies the terminal 50 that a change in the overdetection rate, that is, a change in inspection accuracy has occurred. Each of the processes of steps S134 to S146 described above is repeatedly executed until the end instruction is received from the control unit 60 (step S148: NO).

As described above, the inspection system 10 of the embodiment can automatically detect a change in inspection accuracy due to a change in the overdetection rate and notify the administrator via the management terminal 50 for the administrator. .. Then, the administrator who sees the notification of the change in the inspection accuracy displayed on the management terminal 50 immediately or at an appropriate timing, stops the operation of the inspection system 10 and executes re-learning by AI to learn. You can update the model.

B. Other embodiments:
(1) In the above embodiment, the over-detection number is calculated by subtracting the number actually corrected from the number of inspection results determined to be defective, but the present invention is not limited to this. .. For example, the management unit 42 determines over-detection according to the presence or absence of rework for each rework instruction, manages the number of over-detections determined to be over-detection by the management unit 42, and follows the instruction from the inspection unit 32. The number of over-detections may be transmitted to the inspection unit 32. Further, by detecting the use of the tool by detecting the current flowing through the tool required for the repair, the presence or absence of the repair can be detected, and the tool is instructed by the management unit 42 to be repaired. If the use of is not detected, it may be determined as over-detection and the number of over-detections may be managed by the management unit 42. Further, the management unit 42 determines the over-detection by using the detection of the weight change of the defective parts mounting box, the detection of the weight change of the replacement parts mounting box, the pressing detection of the replacement parts removal completion button, and the like. Is also possible. In addition, the management unit 42 monitors the rework of the worker by using a camera that monitors the rework process area, and after the worker arrives at the rework process area, until the completion button 46 is pressed, or the worker evacuates. If the time until this is done is less than or equal to the specified time, it may be determined as over-detection.

(2) In the above embodiment, the timing at which the preset period T elapses is monitored as the accuracy change determination timing, but the total number of executed inspections (total number of inspections) Nt is set to a preset value. The timing may be monitored as the accuracy change determination timing.

(3) In the above embodiment, the case where the attached nut is an inspection target, the case where the nut is present is regarded as a good product, and the case where the nut is not present is regarded as a defective product has been described as an example. However, for example, the case where there is no nut may be regarded as a good product, and the case where there is a nut may be regarded as a defective product. Further, the presence or absence of the inspection target is not limited, and one of the two states of the inspection target may be regarded as a non-defective product and the other state may be regarded as a defective product. That is, when the inspection target is in a desired state, it may be regarded as a non-defective product, and when the inspection target is not in a desired state, it may be regarded as a defective product.

(4) In the above embodiment and the above other embodiment (3), the case where the quality of the two classifications of whether the inspection target is a desired state is judged as good or bad has been described as an example, but the present invention is not limited to this. It is also applicable when the inspection target determines whether or not the state of multiple classifications is good or bad. For example, it can be applied to the case where the state to be inspected is classified into three categories of A, B, and C, the state of A is regarded as a good product, and the state of B and C is regarded as a defective product.

(5) In the above embodiment and the above other embodiments, an image inspection system using an image pickup camera as an image sensor is described as an example, but the present invention is not limited to this, and vibration, heat, light, and current are not limited thereto. , X-rays and the like can be similarly applied to various inspection systems using sensors and the like. It can be applied to an inspection system that inspects sound and physical vibration, for example, a field inspection system that learns a waveform of normal (or abnormal) vibration and detects the degree of normal (or abnormal) as a discrimination rate. It is also applicable to an inspection system that inspects the operating current of a motor, for example, an inspection system that learns time-series data of normal (or abnormal) operating current and detects the degree of normal (or abnormal) as a discrimination rate. .. Further, it can be applied to an inspection system for inspecting the heat distribution of a press-molded product, for example, an inspection system for learning a normal (or abnormal) heat distribution and detecting the degree of normality (or abnormality) as a discrimination rate.

(6) In the above embodiment, the vehicle assembly process is described as an example, but the present invention is not limited to this, and can be applied to various article assembly processes.

The present disclosure is not limited to the above-described embodiments, examples, and modifications, and can be realized with various configurations within a range not deviating from the gist thereof. For example, the technical features in the embodiments, examples, and modifications corresponding to the technical features in each of the embodiments described in the column of the outline of the invention may be used to solve some or all of the above-mentioned problems. , Can be replaced or combined as appropriate to achieve some or all of the above effects. Further, if the technical feature is not described as essential in the present specification, it can be appropriately deleted.

10 ... Inspection system, 20 ... Data acquisition unit, 22 ... Data acquisition unit, 24 ... Data acquisition control unit, 30 ... Inspection unit, 32 ... Inspection unit, 34 ... Inspection result display unit, 40 ... Management unit, 42 ... Management unit , 44 ... rework history input unit, 46 ... rework complete button, 50 ... management terminal, 60 ... control unit, 241 ... acquisition instruction receiving unit, 243 ... control processing unit, 245 ... data transfer unit, 321 ... data receiving unit, 323 ... ... Judgment unit, 325 ... Inspection result output unit, 327 ... Inspection result storage unit, 331 ... Accuracy change determination timing output unit, 341 ... Rework history reception unit, 343 ... Overdetection judgment unit, 345 ... Overdetection rate calculation unit, 347 ... Over-detection rate judgment unit, 349 ... Accuracy change notification unit, 421 ... Rework history management unit, 423 ... Rework history transmission unit, 431 ... Stop trigger output unit, LA ... Assembly line

Claims (1)

It is an inspection system in the assembly process of articles.
In the pass / fail judgment for determining whether or not the inspection target is in a desired state, the pass / fail judgment is made so that the overdetection rate indicating the degree of overdetection for determining the article to be judged as a non-defective product is a defective product is set as a set value. A determination unit that determines the quality of the inspection target according to the learning model in which the threshold value is set, and the determination unit.
An over-detection determination unit that determines whether or not the defect determination made by the determination unit is over-detection,
An over-detection rate calculation unit that calculates the over-detection rate from the total number of inspections performed and the number of the total number of inspections determined to be over-detection.
When the calculated over-detection rate changes from the set value, the accuracy change notification unit that notifies that the degree of over-detection has changed and the inspection accuracy has changed.
Equipped with an inspection system.

Images