Approach to practical application of Deep Learning in manufacturer's - - PowerPoint PPT Presentation

▶

Apr 17, 2023 172 likes •505 views

Approach to practical application of Deep Learning in manufacturer's production line Masahiro Kashiwagi, Hiroyuki Kusaka, Kiminori Kurosawa and Kenji Nishide 1 1 Applications Service Agriculture Advertisement Education Security Deep

SLIDE 1

1 1

Approach to practical application of Deep Learning in manufacturer's production line

Masahiro Kashiwagi, Hiroyuki Kusaka, Kiminori Kurosawa and Kenji Nishide

SLIDE 2

Applications

Deep Learnig (AI)

Service Medical Security Education Agriculture Financial Advertisement Logistics

Manufacture

Retail

SLIDE 3

Productivity improvement

○ Labor cost increase ○ Japan Working-style Reforms

Reduction in long working hours

“ the Official Website of the Prime Minister of Japan and His Cabinet” http://japan.kantei.go.jp/privacy/terms_e.html

Productivity improvement by Deep Learning (AI)

Emerging country Developed country

SLIDE 4

Company using AI

JPN 5.1% US 13.6%

"Ministry of Internal Affairs and Communications" http://www.soumu.go.jp/english/index.html

○ Using AI is still competitive advantage for many companies

SLIDE 5

Fujikura and AI

Image inspection
Data analysis
System Automation

Deep Learning ○ In Fujikura, we have actively studied applications of Deep Learning technology in various products

SLIDE 6

Fujikura and AI

SLIDE 7

Fujikura and AI

Fujikura America NVIDIA HERE

SLIDE 8

Fujikura and AI

Image inspection
Data analysis
System Automation

Deep Learning

SLIDE 9

Products introducing image inspection

Coaxial Cable FPC Laser diode Optical cable

○ High accurate image inspection using Deep Learning is very attractive in many products.

SLIDE 10

Fujikura and AI

Image inspection
Data analysis
System Automation

Deep Learning ○ Data analysis function is important in sensor system for IoT.

SLIDE 11

Energy Harvesting Sensor System

Indoor sensor node

FSN-2001N

Gateway

FSN-2000S communication 920MHz Specified Low Power Radio communication Gateway

FSN-2000S

Browsing Cloud Data Base Server Indoor Sensor Node FSN-2001N

Schematic Diagram

Outdoor Sensor Node FSN-2002N-OD Internet

3G/LTE Wi-Fi etc.

■ Energy-harvester = Dye-sensitized Solar Cell (DSC)

⇒Accelerates implementation of IoT !

○ ○ ○ ○64 sensor nodes ○ ○ ○ ○Temp.,Humidity,Illuminance,Motion,Pressure

SLIDE 12

Fujikura and AI

Image inspection
Data analysis
System Automation

Deep Learning ○ Sensor system with data analysis function is good solution, sinze the size of measurement data is large.

SLIDE 13

Fujikura and AI

Image inspection
Data analysis
System Automation

Deep Learning

SLIDE 14

Fujikura and AI

Image inspection
Data analysis
System Automation

Deep Learning

SLIDE 15

Fiber lasers for material processing

High power CW fiber laser High peak pulse fiber laser Cutting Welding Marking

SLIDE 16

Fiber laser cutting

SLIDE 17

Fiber lasers for material processing

High power CW fiber laser High peak pulse fiber laser Cutting Welding Marking

SLIDE 18

Fiber laser marking

SLIDE 19

Fiber lasers processing sysem using Deep Learning

○ Adjusting processing conditions ○ Adjusting processing position

SLIDE 20

Fiber laser configuration and components

Water-cooled plate Isolator

○ Fiber lasers are made of in-house components.

SLIDE 21

Importance of high accurate image inspection

Image inspection Low accuracy

Fail Pass

Image inspection High accuracy Large cost Process 1 Process 2 Process 3 Process 4

Fail

SLIDE 22

Laser dioede production process

Epitaxy Patterning Cleavage Coating Packaging Testing

Facet inspection Pattern inspection

SLIDE 23

Facet inspection

23 Front facet

Microscopic Image Failure modes

・Particle ・Scratch ・Clack ・・・

SLIDE 24

Training Images

4 Classes Class1 - Pass Class2 - Failure mode 1 Class3 - Failure mode 2 Class4 - Failure mode 3 10,000 images 2,000 images 1,000 images 1,000 images

SLIDE 25

Convolutional Neural Network

Input image Convolutional layers Max pooling Max pooling

...

Max pooling Convolutional layers

...

Convolutional layers

...

Convolutional layers Max pooling Fully-coneted layer

...

Output Convolutional layers

...

SLIDE 26

Training results

○ Image size : 400 x 400 pixcels ○ 15 convolutional layers

GPU : NVIDIA GeForce GTX TITAN X Calculation time : about 20 hours

SLIDE 27

Progress of training model

Input image (Failure mode) 1st epoch 5th epoch Particle Particle 10th epoch Highlighted region Highlighted region

SLIDE 28

Image size 200 x 200 pixcels 300 x 300 pixcels 400 x 400 pixcels 5 convolutional layers 95.7% 98.4% 99.1% 10 convolutional layers 95.6% 98.4% 99.2% 15 convolutional layers 97.1% 98.5% 99.6%

Test results

Skilled worker (2000 x 2000 pixcels) : 97-98%

SLIDE 29

Heatmaps

Particle (Highlighted region) Particle (Highlighted region)

○ Highlighted regions are in good agreement with particles.

SLIDE 30

Laser dioede production process

Epitaxy Patterining Cleavage Coating Packaging Testing

Facet inspection Pattern inspection ○ We have also be developping pattern image inspection using deep learning.

SLIDE 31

Products introducing image inspection

Coaxial Cable FPC Laser diode Optical cable

○ The image inspection method is also being applied to other products

SLIDE 32

Sumary

○We have developed laser diode facet Image inspection method using Deep Learning technology. ○Test accuracies in image size of 400 x 400 pixcels are more than 99%, which are higher than skilled workers ○Large cost reduction and high productivity would be expected. ○We will develop another application of Deep Learning technology. ○We are recruiting new AI research team staff.