- Smart Manufacturing HOME
-
Choice of approach
- Solution
- Products/Services
- Event/Seminar
-
Column/Case Study
- Inquiry
\ I recommend this hotel /
- There is a discrepancy between the production plan and actual results...I want to improve work efficiency!
- I want to correctly grasp the bottleneck of work!
- I want to know how to improve the quality variation and loss caused by the assembly process!
Approach to improve QCD of assembly process
◆Those who want to know the outline first... Please read the Grareco as it is!
◆For those who want to know everything from the beginning... Please see the lecture video below.
Hello Samba!
In this column, the online webinar held on June 30, 2021
"Are you measuring the cycle time correctly? ～Approach for improving QCD by identifying bottlenecks and preventing work mistakes～”
In the webinar, we introduced approach methods for improving bottlenecks, optimizing production plans, and improving quality variations and losses, which are clues for improving QCD in the assembly process.
I think that “shortening the lead time” is an eternal challenge in manufacturing. In particular, many people are working on shortening the cycle time in the assembly process.
However, when I tried to measure it using a stopwatch,
・I don't know if the improvement points are correct because I take it irregularly
・It is difficult to judge the improvement results quantitatively.
・I can't make an optimal production plan
I often hear your voice.
In this webinar, we delivered a method to solve such troubles!
In order to make it easier for people who want to catch up with cutting-edge technology, I would like to pick up each point and tell them through graphic recording.
Let's go♪
Common assembly process site issues
◆◇One comment◇◆
If you've tried using a stopwatch but aren't measuring the cycle time correctly, it can lead to all sorts of problems. It is uneasy not to be able to judge whether the improvement was correct or effective.
Evolution of technology and approach to challenges
◆◇ひとことコメント◇◆
講演では、実際の作業でサイクルタイムの計測や作業ミスの警告が出されるデモ動画をご覧いただきました。ご興味のある方は、ぜひ講演動画もチェックしてみてください！
Answer all your questions in real time!
In the last session, we answered all the questions we received during the webinar! I would like to deliver here the contents that I could not answer due to lack of time ♪
* Please scroll within the gray area to see
|
Q |
In the example you introduced, only one work video was shown. The actual site is crowded with people, but is it possible to determine the time when one person is not working even in a site where multiple people are moving? |
|---|---|
|
A |
If other moving people are reflected in the background, we believe that you can absorb them by shooting a training video with a background pattern of multiple people moving in order to make the AI model learn to ignore them. I'm here. |
|
Q |
How accurate can you detect work order mistakes and omissions? |
|---|---|
|
A |
If the camera can detect that the work procedure before and after is different, it can basically be correctly identified. |
|
Q |
What kind of equipment, environment, etc. are necessary when introducing it? |
|---|---|
|
A |
The following environment is required.
|
|
Q |
There is considerable variation in fieldwork. Is it necessary to learn work in advance for automatic measurement of non-work time? |
|---|---|
|
A |
Each task must be learned in advance. For example, if the work procedure differs for each product to be manufactured, it is necessary to study the work for product A and the work for product B separately. |
|
Q |
How does AI learn? |
|---|---|
|
A |
By uploading training videos taken in advance to the cloud and annotating each work cycle (labeling of work procedures), AI model learning can be performed on the cloud with one click. You can (it takes time to learn and generate a model). |
|
Q |
How long do AI learners need samples? |
|---|---|
|
A |
A video of 100 cycles is recommended for each task of interest. |
|
Q |
How much knowledge and skills in AI and programming are required? |
|---|---|
|
A |
No AI or programming knowledge or skills are required. We believe that a certain level of IT literacy is necessary because you will be able to log in to the cloud environment with a web browser and view the dashboard. |
|
Q |
Is this service available for a short period of time? |
|---|---|
|
A |
Since it is an annual license contract, you can use it in units of one year. We offer individual consultations regarding trials for verification and evaluation. |
|
Q |
Does AI judge which work it is doing by looking at the posture? |
|---|---|
|
A |
It is not posture estimation. |
|
Q |
It is said that it is necessary to teach the work in advance, but is the master necessary for each worker? Or is it enough just to record the work of one person as a master? |
|---|---|
|
A |
Learning does not have to be done for each worker, but it is a mechanism that absorbs differences due to changes in workers by shooting learning videos with multiple different workers. Usually, 100 cycles of learning videos are recommended, with 10 people doing 10 cycles each. |
|
Q |
What happens when there is work that cannot be detected? Will I get an error because I didn't follow that procedure? |
|---|---|
|
A |
If you perform an action (work) that has not been registered in advance, the work and time can be measured as "other" time. If you then perform the next steps you have planned, you should be able to continue working without error. |
|
Q |
For example, if the operator changes the tightening order of the screws, which does not cause any problem even if the procedure is changed, can you tell me whether this software will treat it as an error or not? |
|---|---|
|
A |
Tasks that can be performed in any order, such as screwing, can be handled in any order. |
|
Q |
What kind of logic is the work mistake detected by AI? For example, is it possible to detect incomplete fastening or incorrect bolt size? |
|---|---|
|
A |
It is possible to detect whether or not screws have been tightened, but we believe it is difficult to judge whether or not there has been any loosening. Regarding the difference in the size of the bolt, it is possible to distinguish it from the scene where the bolt is taken out from different places. |
|
Q |
Workers often leave the site, but if they leave and come back, will they be able to record the work of the day correctly? |
|---|---|
|
A |
Even if you stop working and leave the workshop, you can continue to work after returning and you can measure correctly. |
|
Q |
Do you support when working with items that are carried on a belt conveyor? |
|---|---|
|
A |
I think it's possible in principle. If there is variation in the working position, you can absorb the difference by using the video with the variation as a learning video. |
|
Q |
Please let me know the approximate cost of maintenance in the future. |
|---|---|
|
A |
It will be an annual license contract, not a one-time purchase product. The price is set according to the number of contracts, so please consult with us individually. |
Customer feedback after the event
"Concrete examples using videos were very easy to understand and good. It is easy to consider internal introduction images."
"Work measurement using AI cameras, abnormality confirmation, etc., including technological progress, were very helpful."
“I was able to learn about the latest technological trends, which was helpful.”
in conclusion
How was it?
I would be happy if everyone who aims to improve the QCD of the assembly process will get an opportunity to "concretely consider the use of AI"!
However, AI is just one tool.
Do you have data? Can you make AI? Before that, the most important thing is how to set the original purpose and task.
If you are stuck in organizing your assignments, you want to talk to someone about your ideas! If you are interested, please consult our professional ★