Edge-Cloud Collaboration Architecture for AI Transformation of SME Manufacturing Enterprises

2020 IEEE / ITU International Conference on Artificial Intelligence for Good (AI4G)
Date of Conference: 21-25 September 2020
Conference Location: Geneva, Switzerland

Transcript

1. 1 Edge-Cloud Collaboration Architecture for AI Transformation of SME Manufacturing Enterprises Jeffrey Ying, Jackie Hsieh, Dennis Hou, Janpu Hou Caloudi Corporation Tuo Liu, Xiaobin Zhang, Yuxi Wang , Yen-Ting Pan Yuanjie Semiconductor September 23, 2020
2. 2 Abstract: A new edge-cloud collaboration architecture for small and medium sized enterprises (SME) manufacturers is introduced in this work. Lager manufacturers with sufficient resources already invested heavily in smart manufacturing system. There are rapidly emerging needs to help small and medium sized enterprises manufacturers with limited resources to implement smart and highly adaptable manufacturing systems to compete and sustain in global economy. We present an illustrative case study of how to implement and manage AI projects in practice for SME manufacturers. We demonstrated how our proposed architecture can help accelerate one of the United Nations Sustainable Development Goals, i.e. Goal 9: Industry, Innovation and Infrastructure, by exhibiting the practicality and scalability of our proposed solution. In particular, we elaborate on the key manufacturing issues concerning company-wide resource distribution, problem solving and decision making. It will be demonstrated that more advanced AI systems such as deep learning and deep reinforcement learning emerge naturally with one's quality management system which already in place and come with a well-defined semantics of their process functions in the context of collaborative edge-cloud architecture. Furthermore, equipment used in the smart factory includes manufacturing equipment, functional testing equipment and defect detection equipment. In this work, we will present the management and implementation of on-device AI defect detection and classification to show the feasibility and effectiveness of the edge-cloud collaboration architecture approach.
3. 3 Outline: 1. Introduction 2. Proposed Framework 3. Results and Discussion 4. Summary
4. 4 Global Manufacturer Large Manufacturer Small and Medium Manufacturer Industry 2.0 Industry 3.0 Industry 4.0 On-premise Cloud-Based Edge-Based Challenges Facing Small and Medium-sized Enterprises With limited resources how can we upgrade to distributed control systems – Industry 4.0 ? Manufacturing Operations Management (MOM)
5. 5 IoT Is Demanding AI at the Edge Edge as primary with cloud in a supporting role Interpretable AI Model Software Platform Hardware platform for AI on the Edge Successive Subspace Learning Scattering Convolution Network Raspberry Pi Intel Up Board Google Coral NVIDIA Jetson Nano Intelligent Edge
6. 6 Manufacturing Execution System (MES) Manufacture workflow management Quality Management System (QMS) Defining, improving, and controlling processes to meet customer and regulatory requirements Cloud-Based Edge-Cloud Collaboration Based Current Approach: Proposed Approach: Implementation of Industry 4.0
7. 7 Process Control (IoT + Edge AI) Maintenance Inventory Production Definition Scheduling Resource management Execution Data Collection Performance Evaluation Tracking Quality System ISO/IEC 62264 Part 3 Activity model of manufacturing operations ISO/IEC 62264 Part 1 and 2 Object model Information Product definition Product capability Product schedule Product response Enterprise Application Integration PLM ERP SCM CRM Intelligent Cloud Intelligent Edge IEC 62264 Manufacturing Functional Model
8. 8 Edge-Cloud Collaboration for Optimizing Process and Decision
9. 9 Business Model : Edge-Cloud Collaboration
10. 10 Decision Making Model: Edge-Cloud Collaboration
11. 11 Implementation Steps for SMEs
12. 12 Production Equipment 𝒅𝒙𝟐 𝒅𝒕 + 𝒅𝒙 𝒅𝒕 = 𝒇(𝒙, 𝒖) Controller u=K(s) Sensors Actuation Disturbances Cost Function x(t) Human Decision Transform to AI-assisted Decision Making Controller Input • Sensors Feedback • Human Policy Actuator Input u S-RNN Adjustment Operational Decision Making On the Edge s u Kuhn, H. W.; Tucker, A. W. (1951). "Nonlinear programming" Optimization with Constrains
13. 13 Deep Learning Model Physical metrology result Recommended Actions Processed Wafers Sampled Wafers Process Equipment Metrology equipment Physical Factory Virtual Factory Intelligent Edge Measure Intelligent Cloud Cyber-Physical System = Cloud-Edge Transition Intelligent Edge Control Latent Representations 1. Delivery Schedule Forecast 2. Accounts receivable (AR) Forecast 3. Revenue Forecast 4. ….
14. 14 Incoming inspection Substrate Epitaxy Device Process Facets Cleaving Chip Dicing Die/Wire Bond Probing Test Burn-In Final Test Visual Inspection Intelligent Edge: Process in Physical System
15. 15 Process 1 Process 2 Process 3 Process 4 Process 5 Material Product QA QA QA QA QA QA Process 2 Process 3 Process 5 Prediction 79% Process 3 Process 4 Prediction 65% Process 4 Process 5 Prediction 75% Process 5 Prediction 69% Intelligent Edge/Cloud: Process in Cyber System
16. 16 Customer Group by Segment Time Product Indicators Customer Segment Intelligent Edge Example of Operation Management Cube Intelligent Cloud • Manufacture Resource Planning • Supply Chain Management • Customer Relationship Management • Total Quality Management • Lean Manufacture • Six Sigma
17. 17 CNN Customer Multi-factor Customer m Embedding Attention Model Feature Map Operational Representation For Customer m Customer m Status Prediction Operation Management Cube Week n Week n+1 Fully Connected Layer Customer-Attention Recurrent Neural Network R N N Intelligent Cloud Learned Policy Decisions Feedback Ran Zhao et. al. “Visual Attention Model for Cross-sectional Stock Return Prediction and End-to-End Multimodal Market Representation Learning”
18. 18 Machine Failure Prediction Process Improvement Product Improvement Maximize Equipment utilization Minimize process time Minimize material consumption 60% reduction on maintenance costs 20% longer in hash environment 80% manufacturing cost reduction Highest Reliability Highest Yield Lowest Downtime 18% material/labor cost reduction Goal Constrains 20% defect reduction 12% Less machine downtime Intelligent Edge for 120 days Six Sigma Improvement
19. 19 Physical Space Cyber Space Virtual Factory Actual Factory Deep Reinforced Learning Control Latent Representations Recommend Monitor Intelligent Cloud Intelligent Cyber-Physical Manufacturing System Intelligent Edge Policy
20. 20 Intelligent Manufacturing ITRI and Microsoft collaborate on Intelligent Manufacturing Pilot Plant in Taiwan Fan-Tien Cheng: NCKU, Intelligent Manufacturing Research Center Interpretable (Explainable) Machine Learning Jay C. C. Kuo: USC, “Successive Subspace Learning” Stéphane Mallat: Collège de France, “Scattering Convolutional Network” Cynthia Rudin: Duke University, “Optimal Sparse Decision Trees” Large-scale Machine Learning Systems Li Deng, Citadel Investment, “Large-scale deep learning techniques in speech recognition”, (former Chief Scientist of AI , Founder and Head of Deep Learning Technology Center at Microsoft) Inderjit Dhillon, The University of Texas, Austin, “Large-scale Multi-Output Prediction: Theory and Practice”, (Amazon Fellow and Head, Amazon Research Lab) Acknowledgement