GPU Technology conference 2018 in Silicon Valley S8911 Practical Application of Deep Learning in Smart Factory : Visual Inspection System of Semiconductor Laser Hiroyuki Kusaka, Masahiro Kashiwagi, Yuya Sato, Masahiro Iwasaki, Shinichi Nakatori, Kiminori Kurosawa, Taku Taguchi, Masanori Muto*, Yumi Yamada*, and Kenji Nishide Fujikura Ltd. and *Optenergy, Inc. (Fujikura group) Project Project
Today’s Topic: Deep Learning in Manufacturing Difficulty of DL in Manufacturing Few defective images. Advertise Service Financial The requirement is quite different from ment ordinary DL. ex. image size, criteria of classification, etc. Agriculture Medical We have successfully overcome these. Deep Learning (AI) The platform of DL visual inspection system Retail Education was developed to apply the various production line. Logistics Security A visual inspection system has been Manufacturing implemented to actual production line. 1 27
1. Introduction of our company 2. Fujikura’s “ Monodukuri innovation” 3. Fiber Laser 4. Visual inspection using deep learning 5. Other application, future work. Project
Fujikura Ltd. corporate profile Headquarters 4 business Areas Tokyo, JAPAN Founded February 1885 （ 132 Years ） Head office and consolidated companies Fujikura Group ： 29 countries, about 140companyies 3 27
History of our company Toward Forth Industrial Now Revolution (Industrial 4.0) Third Industrial Revolution Second Industrial Revolution 4 Electrification, Mass Production, … Computer, Robot, Automation … 27
Industry 4.0 & Smart factory Manufacturing Execution System Cyber-Physical system Internet of things Connected Industries M2M:Machine to Machine Big Data Interoperability Information transparency Technical assistance Decentralized decisions Highly flexible mass production. Condition awareness of machine and process Intelligent support of workers Product quality 5 Just-in-time maintenance and near-zero downtime 27
1. Introduction of our company 2. Fujikura’s “ Monodukuri I nnovation” 3. Fiber Laser 4. Visual inspection using deep learning 5. Other application, future work. Project
2.1“Monodukuri Innovation” in Fujikura ● Our main business field is Level 1:Q ・ C ・ D Innovation, Level 2A:Creation of new service “ Monodukuri ” means production Level2 （ Create new value ） Construct global 2B ： provision of information platform ecosystem 2A ： Creation of new service using Innovation utilizing the information our own strength Level1(Q ・ C ・ D Innovation ) Solution in real world. Measures by AI 1D ： Application of IoT in Today‘s Talk and Control of robots by AI Real World Progress of IoT 1C ： Analysis using IoT AI analysis, Optimization using simulation 1B ： Aggregation and Manage information Construction of IT infrastructure 1A ： Digitization of information Level0 Digitization using sensors and cameras Business improvements Business improvements 7 Creation of new value using IoT so far utilizing IoT 27
2.2 “ Monodukuri Innovation”( Level 0) Kaizen (business improving) process so far engineer’s insight, intuition, guess etc. ⇒ Total process speed depends on humans processing speed. Grasping Identifying Kaizen Testing Counter the current the root Verification Process hypothesis measure situation cause Intuition Experience Conventional Measurement SO far Insight guess Principle technology by human Little & Slow 8 Designed by D3Images / Freepik Designed by Pressfoto / Freepik 27
2.2 “ Monodukuri Innovation” (Level 1) ● Kaizen process with IoT (without AI) Human centric process cannot deal a large amount of various sensors data Grasping Identifying Kaizen Testing Counter the current the root Verification Process hypothesis measure situation cause Sensors Intuition Experience Measurement Conventional IoT Cameras guess Principle by human technology Large amount of information ● Data mining, BI tools, etc. Internet of Things ⇒ Human centric result and speed is limited. 9 Designed by Jannoon028 / Freepik 27
2.2 “ Monodukuri Innovation” (Level 1) ● Unprecedented speed up, effectivity of countermeasure ⇒ AI & IoT is complementary relationship. Promote both as one entity . Grasping Identifying Kaizen Testing Counter the current the root Verification Process hypothesis measure situation cause Sensors AI AI AI Sensors IoT Cameras Analysis prediction Countermeasure Cameras Artificial Intelligence Internet of Things Internet of Things 9 27
1. Introduction of our company 2. Fujikura’s “ Monodukuri innovation” 3. Fiber Laser 4. Visual inspection using deep learning 5. Other application, future work. Project
3.1 Fiber Laser ● A fiber laser has excellent beam quality, high efficiency and high reliability. ● Laser diodes are key components of a fiber laser Fiber laser products Fiber laser Application kW Hi-power Pulse Single mode cutting welding Surface processing Fiber Laser Basic Configuration Combiner Output 6 kW Oscillator Laser beam High-power Pump light (λ1.1 um ） semiconductor 11 Laser 27
3.2 Pumping LD for Fiber Laser λ :900 nm Output power:10~20 W 0.1 mm Laser Beam 2 ～ 6 mm 12 27
3.3 Manufacturing process of LD for fiber laser Pre process Electrode Die cutting Crystal process growth LD-Chip LD-Bar Visual Inspection Visual inspection by humans AI Post process Assembly Measurement & Evaluation ＋ － Au wire LD-Chip Sub-mount Characteristic 13 inspection 27
3.4 Visual inspection Criteria of LD chips 1.LD chips are classified into 5 categories (A-E) depending on their defects. 2.”Others (defects) mode” needs to be treated and they are classified into different categories depending on their size and position. 3.For multiple defects in an LD chip, it is classified according to a priority of category classification. Defect modes “Others mode” category classification Others mode mode1 mode2 Structural Large small Border Category Category Area 1 Category D Area1 B C Occurrence Area 2 Area 2 Category E Certain distance position - - Category Area 3 Area 3 A -For multiple defects in one chip, categorized accordingly to the prioritized category. Category C > Category D > Category E > Category B> Category A A LD chip with no defect is categorized into A 14 This classification was done by skilled workers. AI 27
1. Introduction of our company 2. Fujikura’s “ Monodukuri innovation” 3. Fiber Laser 4. Visual inspection using deep learning 5. Other application, future work. Project
4.1 Overview of visual inspection ● Purposes 1.Promotion of automatic inspection 3.Productivity and quality improvement 2. Accumulation of knowledge and skills of AI of semiconductor lasers Improve competitiveness of manufacturing Improve competitiveness company (Monodukuri Innovation) of Fiber Laser products ● Organization of Project Ai 1. User (Image preparation) Optenergy, Inc. (Fujikura group) 2. AI technology development Fujikura Ltd. R&D dep. 3. System development Fujikura Ltd. Production facilities dep. ● process Setting a Taking an Cutting out Inspection Output overview Wafer image chip images with DL results … … … … 16 27
4.2 System development process ● Construction of AI preparation learning validation Beginning of use in production line 4.5 Model tuning 4.8 Inference 4.3 Requirement accuracy definition Parameter tuning 4.4 Computer Integration 1 Integration 2 resources 4.6 Data tuning Data preparation 4.7 Verification of learning ● Construction of 4.9 System Presentation scope reliability Interface ● Production facilities (Hardware) ● Production facilities (Software) 17 27
4.3 Requirement in visual inspection system ● The special requirement of our DL system different from ordinary DL is shown. No Special requirements Action, issue Slide The large ratio of chip size to defects Pixel size 4.4 size （ 1:500000 ） Computer resources 1 Size is 30times larger Deep learning model for large image 4.5 There is “ Others ” mode “Others” modes is classified into sub 2 modes. Defects is classified into different Create data base to manage image 4.6 3 class depending on the size and data position There are few images in some failure 4 data augmentation categories. 5 Explanation for AI classification heat map 4.7 6 Implementation in production line High reliability 4.9 18 27
4.4 Pixel size and computer resources ● Pixel size ⇒ minimum defects ： 2x2 pixels, the whole chip ： 2million pixels 2million pixels Minimum defects 2x2pixels The whole ⇒ 30 times large size ⇒ Minimum size chip Half a million （ typical case: 256x256= 65 thousands ） times large 6000 Number of training images MAX 9 images 5000 images 9 Size of mini-batch 5000 8 4000 7 6 3000 DGX-1 with Tesla V100 4 3 2000 Performance 1 peta 2 (GPU FP16) FLOPS 1000 1 System memory 512GB 0 0 0 32 64 96 128 160 0 4 8 12 16 GPU memory 16GB/GPU Capacity of GPU memory (GB) 19 Capacity of Main Memory(GB) 27
Recommend
More recommend
