WO2024009467A1

WO2024009467A1 - Conveyor control device, production system, and conveyor control program

Info

Publication number: WO2024009467A1
Application number: PCT/JP2022/027006
Authority: WO
Inventors: 太朗紫尾; 英松林; 順二西原; 慎一大江
Original assignee: 三菱電機株式会社
Priority date: 2022-07-07
Filing date: 2022-07-07
Publication date: 2024-01-11
Also published as: JPWO2024009467A1; JP7224561B1

Abstract

In a conveyor control device (300), an observation data acquisition unit (331a) acquires observation data representing a time-series of a processing operation by a worker performing predetermined processing on a workpiece being conveyed by a conveyor. An identification unit (331b) uses the observation data to identify an initial operation period, which is a period in which the worker performs a series of initial operations at the start of the processing, and a final operation period, which is a period in which the worker performs a series of final operations at the end of the processing. A required processing time calculation unit (331c) calculates a required processing time, which is the time span from a representative start time belonging to the initial operation period to a representative end time belonging to the final operation period. A feeding speed control unit (331d) controls the feeding speed, which is the speed at which the conveyor coneys the workpiece, on the basis of the calculation result of the required processing time calculation unit (331c).

Description

Conveyor control device, production system, and conveyor control program

The present disclosure relates to a conveyor control device, a production system, and a conveyor control program.

Production systems equipped with conveyors that transport workpieces are known as equipment for manufacturing products. In this specification, the term "work" is a concept that includes semi-finished products, which are products in the process of being manufactured, and parts that constitute products.

A worker is placed next to the conveyor. An operator performs predetermined processing on the workpiece being conveyed by the conveyor. In this specification, "processing" refers to an operation performed on a workpiece in order to make the workpiece resemble a finished product. Typically, "processing" refers to the operation of assembling the rest of the parts that make up the product onto a workpiece that is a basic part or semi-finished product.

As disclosed in Patent Document 1, a conveyor control device is known that controls the speed at which a workpiece is conveyed by a conveyor (hereinafter referred to as conveyor feeding speed) according to the processing efficiency of an operator. . Specifically, the conveyor control device according to Patent Document 1 measures the required processing time, which is the time required for a worker to process a workpiece, and controls the conveyor feeding speed according to the measurement result. do.

Japanese Patent Application Publication No. 2011-248482

The conveyor control device according to Patent Document 1 includes a first receiving section and a second receiving section fixed to the side of the conveyor. At the moment when the transmitting section worn on the worker's arm approaches the first receiving section, a start time, which is the starting point of the required processing time, is detected. Furthermore, the end time, which is the end point of the required processing time, is detected at the moment the transmitter approaches the second receiver. The processing time is measured as the time span from the start time to the end time.

However, with this method of detecting only two points on the time axis, the start time and the end time, depending on the operator's actions, the start time or end time may not be detected, or the first or end time may not be detected. 2. There is a possibility that the input of noise to the receiving section may be mistakenly detected as the arrival of the start time or end time. That is, the method according to Patent Document 1 cannot be said to have sufficient accuracy in measuring the time required for processing.

An object of the present disclosure is to provide a conveyor control device, a production system, and a conveyor control program that can improve the accuracy of measuring the processing time required compared to conventional methods.

The conveyor control device according to the present disclosure includes:
an observation data acquisition unit that acquires observation data representing a time series of processing operations of a worker who performs predetermined processing on a workpiece being transported by a conveyor;
Using the observed data, the worker performs an initial action period, which is a period in which the worker performs a series of initial actions at the start of the process, and a series of final actions at the end of the process. a specific part that specifies a final operating period, which is a period in which the
a processing time calculation unit that calculates a processing time that is a time width from a representative start time belonging to the initial operation period to a representative end time belonging to the final operation period;
a feed rate control unit that controls a feed rate that is a speed at which the workpiece is conveyed by the conveyor based on a calculation result of the processing time calculation unit;
Equipped with

According to the above configuration, the initial operation period and the final operation period each having a time width are specified, and the required processing time is calculated using the specified results. Therefore, the accuracy of measuring the time required for processing is improved compared to the conventional method of directly detecting the times at only two points on the time axis.

Conceptual diagram showing the configuration of main parts of the production system according to Embodiment 1 Conceptual diagram showing the configuration of a conveyor control device according to Embodiment 1 Conceptual diagram showing functions of the conveyor control device according to Embodiment 1 Conceptual diagram for explaining the time required for processing according to Embodiment 1 Flowchart of conveyor control according to Embodiment 1 Conceptual diagram showing functions of a conveyor control device according to Embodiment 2 Conceptual diagram showing functions of a conveyor control device according to Embodiment 3 Conceptual diagram for explaining intermediate operation period according to Embodiment 4

The production system according to Embodiments 1-4 will be described below with reference to the drawings. In the figures, the same or corresponding parts are denoted by the same reference numerals.

[Embodiment 1]
As shown in FIG. 1, a production system 400 according to the present embodiment includes a conveyor 100 that constitutes a part of a production line that manufactures products in an assembly line. The conveyor 100 conveys the workpiece 500 at a low speed that the worker 600 can follow by walking.

The work 500 is a semi-finished product or a part. The worker 600 performs predetermined processing on the workpiece 500 being transported by the conveyor 100, such as assembling other parts, machining, marking, inspection, and packaging. The worker 600 performs the above processing on the workpiece 500 while walking along the conveyor 100 following the workpiece 500, or in some cases, while standing still.

The works 500 are conveyed one after another toward the worker 600 by the conveyor 100. The operator 600 performs the above processing on the workpiece 500 every time the workpiece 500 is transported. In this way, the above process is repeatedly performed by the worker 600.

If the efficiency of the above-mentioned processing performed by the worker 600 is high relative to the speed at which the workpiece 500 is transported by the conveyor 100, that is, the feed speed of the conveyor 100, the worker 600 has a waiting time for waiting for the arrival of the next workpiece 500. It will happen. On the other hand, if the feed speed of the conveyor 100 is too fast for the efficiency of the processing performed by the worker 600, the burden on the worker 600 may increase or the processing may not be completed.

Therefore, the production system 400 according to the present embodiment further includes an imaging device 200 and a conveyor control device 300 in order to control the feed speed of the conveyor 100 according to the processing efficiency of the worker 600.

The imaging device 200 images the worker 600 beside the conveyor 100. The imaging area IA captured by the imaging device 200 includes a movable area in which the operator 600 can move for one process. The imaging device 200 generates observation data DT representing a time series of the above processing operations performed by the worker 600 by imaging the worker 600. That is, the imaging device 200 is an example of an observation data generation device according to the present disclosure.

The observation data DT is time-series data in which each observation result is associated with the time at which the observation result was obtained. Specifically, the observation data DT is a moving image in which each still image (hereinafter also referred to as a frame) taken in the imaging area IA is associated with the time at which the still image was taken, as an observation result. It is data.

First, the conveyor control device 300 uses the observation data DT generated by the imaging device 200 to determine the processing time required, which is an index representing the efficiency of the above processing by the worker 600. Then, the conveyor control device 300 controls the feed speed of the conveyor 100 according to the determined processing time. The configuration of the conveyor control device 300 will be specifically described below.

As shown in FIG. 2, the conveyor control device 300 includes a communication device 310, which is hardware necessary for communication. The communication device 310 plays the role of taking in observation data DT from the imaging device 200 and the role of outputting a control command to the conveyor 100 for controlling the feed speed of the conveyor 100.

Additionally, the conveyor control device 300 includes a storage device 320 that stores a conveyor control program 321 that defines a procedure for controlling the feed speed of the conveyor 100. The storage device 320 also stores a learned model 322 used to control the feed rate of the conveyor 100 and learning data 323 necessary to generate the learned model 322.

Additionally, the conveyor control device 300 includes a processor 330 that executes a conveyor control program 321. The functions realized by the processor 330 executing the conveyor control program 321 will be described below.

As shown in FIG. 3, the conveyor control device 300 has the function of a control section 331 that controls the feed speed of the conveyor 100, and the function of a learning section 332 that generates a learned model 322 necessary for the control.

First, the control section 331 will be explained. The control unit 331 uses the observation data acquisition unit 331a that acquires the observation data DT from the imaging device 200 and the observation data DT to specify the time necessary for calculating the index representing the efficiency of the above processing of the worker 600. It has a specific part 331b.

With reference to FIG. 4, the function of the specifying unit 331b will be specifically explained. FIG. 4 shows a time axis along the time series represented by the observation data DT. The observation data DT has a data structure in which frames representing one still image are arranged along the time axis.

The specifying unit 331b uses the observation data DT to specify an operation period that is a period in which the worker 600 actually operated for the above processing. The operation period represents the period of one cycle of the above-mentioned processing that is periodically repeated by the worker 600. The presence or absence of motion can be determined by analyzing each frame arranged in time series. Therefore, the operation period can be specified by the observation data DT.

Furthermore, the specifying unit 331b uses the observation data DT to specify the initial operating period TA that constitutes the beginning of the operating period. The initial operation period TA is a period during which the operator 600 performed a series of initial operations at the start of the process. The series of initial operations are operations that follow a predetermined procedure. Therefore, the initial operation period TA can be specified based on the time series of frames represented by the observation data DT.

Further, the specifying unit 331b uses the observation data DT to specify the representative start time t1 belonging to the initial operation period TA. In this embodiment, the representative start time t1 is the time at the moment when the worker 600 performs a specific action (hereinafter referred to as representative start action) among the above-mentioned initial actions. Since the representative start motion is a predetermined specific motion, the representative start time t1 can be specified in the time series of frames represented by the observation data DT.

As one specific example, consider a case where the worker 600 uses a tool to perform the above process on the workpiece 500. In this case, the period from when the operator 600 prepares the tool until he starts operating the tool can be defined as the initial operation period TA. Further, the action of the worker 600 applying the tool to the workpiece 500 can be defined as a representative start action, and the moment when the representative start action is performed can be defined as the representative start time t1.

Furthermore, the specifying unit 331b uses the observation data DT to specify the final operating period TB that constitutes the final part of the operating period. The final action period TB is a period during which the worker 600 performed a series of final actions at the end of the process. The series of final operations is an operation that follows a predetermined procedure, and is clearly different from the series of initial operations described above. Therefore, the final operating period TB can be specified based on the time series of frames represented by the observation data DT.

Furthermore, the specifying unit 331b uses the observation data DT to specify the representative end time t2 belonging to the final operation period TB. In this embodiment, the representative end time t2 is the time at the moment when the worker 600 performs a specific action (hereinafter referred to as representative end action) among the final actions. Since the representative end action is a predetermined specific action, the representative end time t2 can be specified based on the time series of frames represented by the observation data DT.

Regarding the specific example described above, the period from when the operator 600 finishes operating the tool until the tool is returned to a predetermined position can be defined as the final operation period TB. Further, the action of the operator 600 separating the tool from the workpiece 500 can be defined as a representative end action, and the moment when the representative end action is performed can be defined as the representative end time t2.

The time difference between the representative start time t1 and the representative end time t2 identified as described above will be referred to as the required processing time. The processing time required represents the time actually required by the operator 600 to perform the above processing once. In other words, the processing time is an index representing the efficiency of the above-mentioned processing by the worker 600.

Here, as a typical example, the processing time required is the time difference between the representative start time t1 and the representative end time t2, but is not limited to this. That is, in the present embodiment, the processing time refers to the period between the initial operation period TA and the final operation period TB, and for example, the time required for processing is the period between the initial operation period TA and the final operation period TB. It may be the difference between In addition, the processing time is the time after a predetermined period has elapsed from the start time of the initial operation period TA (for example, the time after 10% of the period length of the initial operation period TA has elapsed from the start time of the initial operation period TA), It may be a difference from the time after a predetermined period has elapsed from the start time of the final operation period TB (for example, the time after 90% of the period length of the final operation period TB has elapsed from the start time of the final operation period TB). In other words, the processing time is the difference between the time obtained based on the initial operation period TA and the time obtained based on the final operation period TB.

Hereinafter, the reason why not only the representative start time t1 and the representative end time t2, but also the initial operation period TA and final operation period TB are specified in order to obtain the processing time that represents the efficiency of the processing described above will be described.

If we try to directly specify the representative start time t1, which is one point on the time axis, without specifying the initial movement period TA, the representative start time t1, which is one point on the time axis, may be There is a high possibility that the representative start time t1 may be erroneously identified as the time when an action different from the action was performed, or that the representative start time t1 is failed to be identified.

In other words, since the accuracy of specifying the representative start time t1 largely depends on the characteristics of the movement of the worker 600, it is difficult to specify the representative start time t1 with high accuracy. The same applies to the case where the representative end time t2, which is one point on the time axis, is directly specified without specifying the final operation period TB.

In contrast, in this embodiment, the initial operation period TA is first specified. Since the initial movement period TA has a time width, unlike the case of directly specifying one point on the time axis, even if the operator 600 performs an unexpected movement, highly accurate identification is possible. be.

Furthermore, since the representative start time t1 exists within the initial operation period TA, once the initial operation period TA is specified, the range in which the representative start time t1 is searched can be limited to the initial operation period TA. Therefore, it is possible to reduce the possibility of failing to specify the representative start time t1 or erroneously specifying a different representative start time t1. The same applies to specifying the representative end time t2.

That is, in this embodiment, in order to increase the accuracy of specifying the representative start time t1 and the representative end time t2, and ultimately the required processing time, not only the representative start time t1 and the representative end time t2, but also the initial operation period TA and The final operating period TB is also specified.

Returning to FIG. 3, the explanation will continue. The specifying unit 331b uses the learned model 322 to specify the above-described initial operation period TA, representative start time t1, final operation period TB, and representative end time t2.

The trained model 322 detects, from the observation data DT, first partial data representing the series of initial operations in the initial operation period TA, and second partial data representing the series of final operations in the final operation period TB. At the same time, machine learning is performed to detect the representative start time t1 using the detected first partial data and to detect the representative end time t2 using the detected second partial data.

In other words, the identification unit 331b inputs the observation data DT acquired by the observation data acquisition unit 331a into the learned model 322. Then, the specifying unit 331b obtains outputs of the detection results of the first partial data, the representative start time t1, the second partial data, and the representative end time t2 from the learned model 322.

Note that since the observation data DT is time-series data, detecting the first partial data from the observation data DT is equivalent to specifying the initial operation period TA. Further, detecting the second partial data from the observation data DT is equivalent to specifying the final operating period TB.

As described above, the specifying unit 331b specifies the initial operation period TA, representative start time t1, final operation period TB, and representative end time t2 based on the output of the learned model 322 into which the observation data DT has been input.

Furthermore, the control unit 331 includes a processing time calculation unit 331c that calculates the processing time required, which is the time span from the representative start time t1 to the representative end time t2 specified by the identification unit 331b. The processing time required is the time difference between the representative end time t2 and the representative start time t1.

The lower the efficiency of the above process by the worker 600, the longer the time required for the process. Furthermore, the higher the efficiency of the above processing by the worker 600, the shorter the time required for the processing. Therefore, the processing time is an index representing the efficiency of the above-mentioned processing by the worker 600.

Therefore, the control unit 331 includes a feed rate control unit 331d that controls the feed rate of the conveyor 100 based on the required processing time that is the calculation result of the required processing time calculation unit 331c.

Although not shown, the conveyor 100 includes a conveyance belt on which the workpiece 500 is placed, a roller on which the conveyance belt is hung, and a motor that rotates the roller. The motor has a configuration that allows its rotational speed to be controlled. Control of the feed speed of the conveyor 100 is realized by controlling the rotation speed of the motor of the conveyor 100, specifically, by controlling an inverter.

Specifically, the feed rate control unit 331d determines that the longer the processing time required by the worker 600, the lower the feed speed of the conveyor 100, and the shorter the processing time required by the worker 600, the higher the feed speed of the conveyor 100. The feed speed of the conveyor 100 is controlled based on the conditions. As a result, in the next process described above, the worker 600 is less likely to have a waiting time, and moreover, the worker 600 is less likely to be overloaded.

Next, the learning section 332 will be explained. The learning section 332 includes a learning data acquisition section 332a that acquires the learning data 323. The learning data 323 includes the actual measured values of the first partial data, representative start time t1, second partial data, and representative end time t2, the first partial data, representative start time t1, second partial data, and The observation data DT for which the representative end time t2 was actually measured is used.

Here, "actual measurement" means correctly specifying by means other than the learned model 322. That is, the "actual measured value of the first partial data" means the first partial data correctly specified by means other than the learned model 322. The same applies to the “actual measurement value of the second partial data”. Furthermore, the “actual measured value of the representative start time t1” means the representative start time t1 correctly specified by means other than the learned model 322. The same applies to the “actual measurement value of representative start time t2”.

Note that "actual measurement" can be performed by a person who creates the learning data 323 (hereinafter referred to as the creator). As a specific example, a video represented by the observation data DT is played back, and the creator visually identifies the initial operation period TA, representative start time t1, final operation period TB, and representative start time t2 in the video. In this way, the actual measured values of the first partial data, the representative start time t1, the second partial data, and the representative end time t2 can be obtained. When specifying by visual recognition, the video may be played back frame by frame or played back in slow motion.

In addition, as another specific example, the creator directly visually confirms the work of the worker 600 at the site, and determines the times corresponding to both end points of the initial operation period TA, the representative start time t1, and the final operation period TB. The times corresponding to both end points of and the representative start time t2 may be recorded respectively. By comparing each recorded time with the time information represented by the observation data DT, it is possible to obtain the actual measured values of the first partial data, representative start time t1, second partial data, and representative end time t2. can.

Furthermore, the learning unit 332 includes a generation unit 332b that generates the learned model 322 using the learning data 323 acquired by the learning data acquisition unit 332a. The generation unit 332b uses the learning data 323 to learn a policy for detecting the first partial data, the representative start time t1, the second partial data, and the representative end time t2 from the observation data DT. Through such machine learning, a trained model 322 is generated.

However, the representative start time t1 is defined as a time that is uniquely determined from the length of the initial operation period TA (for example, the time after 10% of the length of the initial operation period TA has elapsed from the start time of the initial operation period TA). In this case, the learning data 323 does not necessarily need to include the actual measured value of the representative start time t1. In this case, the representative start time t1 can be specified from the length of the initial operation period TA in the specifying unit 331b or the processing time calculation unit 331c without using the learned model 322 to specify the representative start time t1. can.

Similarly, the representative end time t2 is defined as a time that is uniquely determined from the length of the final operation period TB (for example, the time after 90% of the length of the final operation period TB has elapsed from the start time of the final operation period TB). In this case, it is not necessarily necessary to include the actual measurement value of the representative end time t2 in the learning data 323. In this case, without using the learned model 322 to specify the representative start time t1, the representative end time t2 can be specified from the length of the final operation period TB in the specifying unit 331b or the processing time calculation unit 331c. can.

Hereinafter, with reference to FIG. 5, conveyor control performed in the production system 400 according to the present embodiment will be described.

As shown in FIG. 5, when an unprocessed workpiece 500 arrives at the worker 600 on the conveyor 100 (step S1), the imaging device 200 starts generating observation data DT (step S2). The generation of observation data DT by the imaging device 200 continues in parallel with the processing of the workpiece 500 by the worker 600 (step S3). After the processing of the workpiece 500 by the worker 600 is completed, the generation of observation data DT by the imaging device 200 is completed (step S4).

Next, the identification unit 331b of the conveyor control device 300 uses the observation data DT to identify a representative start time t1 belonging to the initial operation period TA and a representative end time t2 belonging to the final operation period TB. Further, the processing time calculation unit 331c of the conveyor control device 300 calculates the processing time, which is the time difference between the representative start time t1 and the representative end time t2 (step S5).

As described above, the processing time is an index representing the processing efficiency of the worker 600. The feed speed control unit 331d of the conveyor control device 300 performs processing under the conditions that the longer the processing time required, the lower the feed speed of the conveyor 100, and the shorter the processing time of the worker 600, the higher the feed speed of the conveyor 100. The feed speed of the conveyor 100 is controlled based on the required time (step S6).

Thereafter, the process returns to step S1 again, and the unprocessed workpiece 500 arrives at the operator 600 at the controlled feed speed. In this way, steps S1 to S6 are repeated.

According to this embodiment, in particular, the following effects can be obtained.

In step S5 described above, in order to specify the representative start time t1, an initial operation period TA with a time width is once specified, and in order to specify the representative end time t2, a final operation period TB with a time width is once specified. be done. Thereby, the range in which the representative start time t1 is searched can be limited to the initial operation period TA, and the range in which the representative end time t2 is searched can be limited to the final operation period TB.

Therefore, compared to directly specifying the representative start time t1 or the representative end time t2 without specifying the initial operation period TA and the final operation period TB, it is easier to specify the representative start time t1 or the representative end time t2. It is possible to reduce the possibility of erroneously specifying the representative start time t1 or the representative end time t2. As a result, the accuracy of measuring the processing time is less likely to be influenced by the characteristics of the operator's 600 movements. In other words, the accuracy of measuring the time required for processing is improved compared to the conventional method.

In step S6 described above, the feed speed of the conveyor 100 is controlled based on the processing time required, so that the worker 600 is less likely to have to wait in the next processing of the unprocessed workpiece 500. 600 is less likely to be overloaded.

[Embodiment 2]
In the first embodiment described above, supervised learning using the learning data 323 as teacher data is illustrated as a learning algorithm for generating the trained model 322. The learning unit 332 may have a function of updating the learned model 322 by reinforcement learning. A specific example will be described below.

As shown in FIG. 6, the learning unit 332 according to the present embodiment uses observation data DT acquired by the observation data acquisition unit 331a to calculate a reward, which is a parameter used for reinforcement learning. A portion 332c is provided.

The reward represents the validity of the feed rate control performed after the previous processing of the workpiece 500. In the following, when there is time remaining after the worker 600 completes processing of the workpiece 500 until the next workpiece 500 arrives, the remaining time length will be referred to as waiting time. Furthermore, when the processing of the workpiece 500 by the worker 600 is incomplete, the length of time that is insufficient to complete the processing is referred to as insufficient time.

The remuneration calculation unit 332c uses the observation data DT to calculate the waiting time or insufficient time. When waiting time or insufficient time occurs, the feed rate control performed last time is not necessarily appropriate. Therefore, the remuneration calculation unit 332c calculates the remuneration under the conditions that the longer the waiting time or the shortage time is, the more the remuneration is reduced, and the closer the waiting time or the shortage time is to zero, the more the reward is increased.

Furthermore, the learning unit 332 according to the present embodiment includes an updating unit 332d that updates the learned model 322 according to the reward calculated by the reward calculating unit 332c. Specifically, the update unit 332d sets a policy for detecting the first partial data, representative start time t1, second partial data, and representative end time t2 from the observed data DT by the learned model 322, as follows: Update with the conditions that will give you the most of the above rewards.

According to the present embodiment, the validity of controlling the feed speed of the conveyor 100 can be increased each time the worker 600 repeats the process on the workpiece 500. Other configurations and effects are similar to those in the first embodiment.

[Embodiment 3]
In the first embodiment described above, the feed speed of the conveyor 100 is controlled according to the efficiency of the processing by the operator 600. The feed speed of the conveyor 100 may be controlled by further taking into consideration the tendency of change in the efficiency of the above-mentioned processing by the worker 600. A specific example will be described below.

As shown in FIG. 7, the control unit 331 according to the present embodiment further includes a change amount calculation unit 331e that calculates a difference in the processing required time calculated by the processing required time calculation unit 331c.

Specifically, the change amount calculation unit 331e calculates the processing time calculated by the processing time calculation unit 331c for processing the workpiece 500 in the current turn and the processing time required for processing the workpiece 500 in the previous turn. The difference between the required processing time calculated by the calculation unit 331c (hereinafter referred to as the amount of change in the required processing time) is calculated.

The amount of change in processing time is a physical quantity that also includes positive and negative signs. A positive value of the amount of change in processing time indicates a tendency that the processing time is increasing. One of the factors for this is a decrease in efficiency due to fatigue of the worker 600. On the other hand, a negative value of the amount of change in processing time indicates a tendency for processing time to be shortened. One of the factors for this is improvement in efficiency as the worker 600 becomes accustomed to the work.

The feed rate control unit 331d calculates not only the calculation results of the processing time calculation unit 331c, that is, the processing required time calculated for processing the workpiece 500 in the current turn, but also the processing that is the calculation result of the change amount calculation unit 331e. The feed speed of the conveyor 100 is also controlled based on the amount of change in required time.

Specifically, in step S6 of FIG. 5, the feed speed control unit 331d determines that if the amount of change in the required processing time is positive, the feed speed of the conveyor 100 becomes smaller as the absolute value of the amount of change in the required processing time becomes larger. , if the amount of change in the required processing time is negative, the feed speed of the conveyor 100 is controlled based on the amount of change in the required processing time, with the condition that the larger the absolute value of the change in the required processing time is, the greater the feed speed of the conveyor 100 is. do.

According to the present embodiment, the feed speed of the conveyor 100 is controlled taking into account not only the processing time but also the amount of change in the processing time, so the feed speed of the conveyor 100 can be controlled to a more appropriate value. . Other configurations and effects are similar to those in the first embodiment.

[Embodiment 4]
In the first embodiment described above, the feed speed of the conveyor 100 is controlled according to the processing time required. Furthermore, the feed speed of the conveyor 100 may be controlled in consideration of the length of the intermediate operation period between the initial operation period TA and the final operation period TB. A specific example will be described below.

FIG. 8 shows the position of the intermediate operation period TC on the time axis. The identifying unit 331b according to the present embodiment uses observation data DT to identify not only the initial operating period TA and the final operating period TB, but also the intermediate operating period TC between the initial operating period TA and the final operating period TB. do.

The intermediate motion period TC is a period in which the worker 600 performs a specific series of intermediate motions that constitute a part of the motion of the worker 600 from the end of the initial motion period TA to the start of the final motion period TB. be.

The series of intermediate operations are operations that follow a predetermined procedure, and are different from the series of initial operations and the series of final operations described above. Therefore, the intermediate operation period TC can be specified based on the time series of frames represented by the observation data DT.

It is preferable that the series of intermediate motions is a motion that is more likely to fluctuate depending on the degree of fatigue of the worker 600 and the degree of familiarity with the work than the series of initial motions and the series of final motions. As a specific example, consider a case where the above process is an operation of assembling a plurality of parts onto the workpiece 500. In this case, among the plurality of parts, it is preferable to determine as an intermediate operation the assembly of a part that requires relatively more effort to assemble to the workpiece 500.

The length of the intermediate operation period TC varies depending on the degree of fatigue of the worker 600 and the degree of familiarity with the work. On the other hand, a change in the length of the intermediate operation period TC may be difficult to manifest as a change in the required processing time. This is because even if the period length of the intermediate operation period TC changes, the operator 600 may unconsciously adjust the efficiency of the remaining operation.

Therefore, by measuring the period length of the intermediate operation period TC, it is possible to grasp the degree of fatigue and the degree of work familiarity of the worker 600, which is difficult to grasp only by measuring the required processing time. Particularly in this respect, there is significance in measuring the period length of the intermediate operation period TC.

The feed rate control unit 331d according to the present embodiment controls the feed rate of the conveyor 100 based not only on the required processing time but also on the length of the intermediate operation period TC. That is, the feed speed control unit 331d is configured such that the longer the intermediate operation period TC is, the lower the feed speed of the conveyor 100 is, and the shorter the intermediate operation period TC is, the higher the feed speed of the conveyor 100 is. The feed speed of the conveyor 100 is controlled.

According to the present embodiment, the feed speed of the conveyor 100 is controlled taking into consideration not only the processing time but also the length of the intermediate operation period TC, so the feed speed of the conveyor 100 can be controlled to a more appropriate value. Can be done.

Note that measuring the period length of the intermediate operation period TC is equivalent to specifying the intermediate operation period TC. Those skilled in the art will understand that the intermediate operating period TC can also be specified using the learned model 322 in the same manner as the initial operating period TA and the final operating period TB. As the learning data 323 used to generate the learned model 322, it is preferable to use observation data DT in which the third partial data representing the series of intermediate operations described above is correctly identified in advance. Other configurations and effects are similar to those in the first embodiment.

[Embodiment 5]
In the first embodiment, the learned model 322 is used not only to specify the initial operation period TA and the final operation period TB, but also to specify the representative start time t1 and representative end time t2. The learned model 322 does not necessarily need to be used to specify the representative start time t1 and the representative end time t2. As long as the initial operation period TA is specified in order to specify the representative start time t1, and the final operation period TB is specified in order to specify the representative end time t2, the accuracy of measuring the processing time can be improved more than before. be able to. The representative start time t1 may be a time uniquely determined from the initial operation period TA, and the representative end time t2 may be a time uniquely determined from the final operation period TB.

In the present embodiment, the identifying unit 331b or the processing time calculating unit 331c determines the time at which the time width of the initial operation period TA is internally divided by a predetermined ratio, specifically, the midpoint of the time width. The time corresponding to t is calculated as the representative start time t1. Similarly, the specifying unit 331b or the processing time calculating unit 331c determines the time at which the time width of the final operating period TB is internally divided by a predetermined ratio, specifically, the time at the midpoint of the time width. , is calculated as the representative end time t2.

In this embodiment, the learned model 322 does not estimate the representative start time t1 and the representative end time t2. Therefore, the learning data 323 as teacher data does not include the actual measured values of the representative start time t1 and the representative end time t2. That is, the learning data 323 uses the actual measured values of the first partial data and the second partial data described above, and the observation data DT in which the first partial data and the second partial data were actually measured.

Note that in the present embodiment, the midpoint is exemplified as the point for internally dividing the time width of the initial operation period TA, but the point for internally dividing does not necessarily have to be the midpoint. The same applies to internally dividing the time width of the final operation period TB. Furthermore, in this specification, the concept of "a point that internally divides a time width" includes a start end point and an end end point that are both end points of the time width.

Embodiments 1-5 have been described above. The following variations are also possible.

(1) In the first embodiment described above, video data in which frames as observation results are arranged in chronological order is used as the observation data DT representing the chronological sequence of the actions of the worker 600. The movement of the worker 600 can also be expressed by a change over time in the distance between the reference location and the worker 600, or a change over time in the intensity of the sound that occurs along with the movement of the worker 600. That is, the observation data DT may be time-series data in which observation results such as distance and sound intensity are arranged in chronological order.

(2) As a learning algorithm for generating the trained model 322, supervised learning was exemplified in the first embodiment, and reinforcement learning was exemplified in the second embodiment. As the learning algorithm, known methods other than supervised learning and reinforcement learning may be used. Specifically, the trained model 322 may be generated by unsupervised learning. In this case, the learning data 323 may use past observation data DT in which the first partial data, representative start time t1, second partial data, and representative end time t2 are not correctly specified.

(3) The learned model 322 does not necessarily need to be used to specify the initial operating period TA and final operating period TB. The identification unit 331b searches the time series of observation results included in the observation data DT, and extracts observation results representing a specific action of the worker 600 from the observation data DT by pattern recognition. , an initial operating period TA and a final operating period TB may be specified.

(4) As described in the first embodiment above, the feed speed control section 331d controls the feed speed of the conveyor 100 based on the calculation result of the processing time calculation section 331c. In this specification, "based on the calculation result of the processing time calculation unit 331c" refers not only to controlling the feed speed of the conveyor 100 based on the processing time that is the calculation result of the processing time calculation unit 331c; This term also includes controlling the feed speed of the conveyor 100 using a physical quantity that depends on the processing time. Specifically, the feed speed control unit 331d calculates the target value of the feed speed of the conveyor 100 by dividing the known length by the required processing time as a physical quantity that depends on the required processing time, and adjusts the feed speed of the conveyor 100 by dividing the known length by the required processing time. Control may be performed to bring the speed closer to the target value.

(5) The conveyor control device 300 according to Embodiments 1-4 above can be realized using an existing computer. That is, by installing the conveyor control program 321 shown in FIG. 2 into a computer, the computer can function as the conveyor control device 300. The conveyor control program 321 may be distributed via a communication network, or may be stored in a computer-readable, non-transitory recording medium and then distributed.

Various embodiments and modifications of the present disclosure are possible without departing from the broad spirit and scope of the present disclosure. The embodiments described above are for explaining the present disclosure and do not limit the scope of the present disclosure. The scope of the present disclosure is indicated by the claims rather than the embodiments. Various modifications that come within the scope of the claims and the meaning of equivalent disclosures are deemed to be within the scope of the present disclosure.

100 conveyor, 200 imaging device (observation data generation device), 300 conveyor control device, 310 communication device, 320 storage device, 321 conveyor control program, 322 learned model, 323 learning data, 330 processor, 331 control unit, 331a observation Data acquisition unit, 331b Specification unit, 331c Processing time calculation unit, 331d Feed rate control unit, 331e Change amount calculation unit, 332 Learning unit, 332a Learning data acquisition unit, 332b Generation unit, 332c Reward calculation unit, 332d Update unit , 400 production system, 500 work, 600 worker, DT observation data, TA initial operation period, TB final operation period, TC intermediate operation period, t1 representative start time, t2 representative end time, IA imaging area.

Claims

an observation data acquisition unit that acquires observation data representing a time series of processing operations of a worker who performs predetermined processing on a workpiece being transported by a conveyor;
Using the observation data, an initial action period is a period during which the worker performs a series of initial actions at the start of the process, and a series of final actions is performed by the worker at the end of the process. a specific part that specifies a final operating period, which is a period in which the
a processing time calculation unit that calculates a processing time that is a time width from a representative start time belonging to the initial operation period to a representative end time belonging to the final operation period;
a feed rate control unit that controls a feed rate that is a speed at which the workpiece is conveyed by the conveyor based on the calculation result of the processing time calculation unit;
A conveyor control device comprising:
The identification unit detects first partial data representing the series of initial actions in the initial action period and second partial data representing the series of final actions in the final action period from the observation data. identifying the initial operating period and the final operating period using a trained model that has been subjected to machine learning for
The conveyor control device according to claim 1.
a learning data acquisition unit that acquires, as learning data, actual measured values of each of the first partial data and the second partial data, and the observed data in which the first partial data and the second partial data are actually measured; ,
a generation unit that generates the learned model by the machine learning using the learning data;
The conveyor control device according to claim 2, further comprising:.
The machine learning further includes detecting the representative start time using the first partial data detected from the observation data, and detecting the representative start time using the second partial data detected from the observation data. It is for detecting the representative end time,
The identifying unit uses the learned model to identify not only the initial operation period and the final operation period, but also the representative start time and the representative end time.
The conveyor control device according to claim 2.
Actual values of each of the first partial data, the representative start time, the second partial data, and the representative end time, and the first partial data, the representative start time, the second partial data, and the representative end. a learning data acquisition unit that acquires the observation data whose time is actually measured as learning data;
a generation unit that generates the learned model by the machine learning using the learning data;
The conveyor control device according to claim 4, further comprising:.
The process is repeatedly performed by the worker,
a remuneration calculation unit that calculates a remuneration representing the validity of the feed rate control performed by the feed rate control unit;
an updating unit that updates the learned model according to the reward calculated by the reward calculation unit;
The conveyor control device according to claim 2 or 4, further comprising:
The process is repeatedly performed by the worker,
Amount of change calculation that calculates the difference between the processing time calculated by the processing time calculation unit for the current process and the processing time calculated by the processing time calculation unit for the previous process. Department,
Furthermore,
The feed rate control unit controls the feed rate based not only on the calculation result of the processing time calculation unit but also on the calculation result of the change amount calculation unit.
A conveyor control device according to any one of claims 1 to 6.
The identification unit uses the observation data to identify not only the initial operating period and the final operating period, but also the period between the initial operating period and the final operating period, at the end of the initial operating period. Also specifying an intermediate movement period that is a period in which the worker performed a series of intermediate movements that constitute a part of the worker's movements from the start of the final movement period,
The feed rate control unit controls the feed rate based not only on the calculation result of the required time calculation unit but also on the length of the intermediate operation period.
A conveyor control device according to any one of claims 1 to 7.
A conveyor control device according to any one of claims 1 to 8,
the conveyor, the feed rate of which is controlled by the feed rate controller;
an observation data generation device that generates the observation data;
A production system equipped with
to the computer,
an observation data acquisition unit that acquires observation data representing a time series of processing operations of a worker who performs predetermined processing on a workpiece being transported by a conveyor;
Using the observation data, an initial action period is a period during which the worker performs a series of initial actions at the start of the process, and a series of final actions is performed by the worker at the end of the process. a specific part that specifies a final operating period, which is a period in which the
a processing time calculation unit that calculates a processing time that is a time width from a representative start time belonging to the initial operation period to a representative end time belonging to the final operation period;
a feed rate control unit that controls a feed rate, which is the speed at which the workpiece is conveyed by the conveyor, based on the calculation result of the processing time calculation unit;
A conveyor control program that realizes this function.