Leveraging AI for Industrial IoT Chetan Gupta, Ph.D. Chief Data Scientist, Big Data Lab, Hitachi America Ltd. Date: Sept. 19 th , 2017
AI
Level Set Machine Learning Data Outcomes & Artificial Intelligence
Data Sensor Data Human Generated Data Enterprise Data Tags, Sensors, Video, etc. Social Media, etc. Transactions, logs, etc. Volume/Speed of data Heterogeneity of Data
Outcome Autonomous Prescriptive/ Recommendation Predictive Descriptive/ Self Driving Car, etc. Movie Recognition Recommendation, etc. Clinical Decision Support, etc. Fraud Detection, etc. Complexity Data Requirement
AI/ML Algorithms Deep Learning “Traditional” Machine Learning CNN, RNN, LSTM, etc. Rule Based AI SVMs, Random Forests, etc. Expert Systems, etc. Compute Accuracy
Example – Fraud Detection Rule Based, Credit Card Flag Credit Card Fraud Anomaly Transactions Detection Transactional Data, Descriptive
Example – Churn Prediction Activity, Classification Predict the probability of Behavioral Data Techniques losing a customer Transaction/Social Media Data, Predictive
Example – Product Recommendation Collaborative Recommend ads, Social Media Filtering products, movies, etc. Data Social Media Data, Prescriptive
Example – Autonomous Vehicles Deep Neural Self Driving Cars Video, Lidar, etc. Nets Sensor Data, Autonomous
Impact Products & Services Personalization of Services, Automation in Products Sales Up Sell, Cross Sell, Customer Retention Marketing Micro campaigns, Targeted Advertising Customer Support Fielding Service/Support Calls Human Resource Talent Acquisition & Retention Operational ….
Industrial IoT
Industrial Analytics Semiconductor Manufacturing • Increase Asset Availability Resources Healthcare Chemicals Financial Railways Mobility Natural BEMS • Increase Asset Utilization • Improve Product Quality Customer • Increase Safety & Reliability of Operations End-to-end Automation & Optimization Operations Optimization Solution • Reduce Operations and Quality Enhancement Cores Maintenance Cost Predictive Maintenance Safety Improvement • Enhanced Operational Control & Planning Lumada
IIoT Problem Taxonomy Analytics Maintenance Operations Quality 1. Equipment Monitoring 1. Operations Monitoring 1. Quality Monitoring 2. Performance Analytics 2. Characterize Process 2. Testing Process 3. Maintenance Analytics 3. Operator Behavior Monitoring & Evaluation Descriptive 4. Equipment Failure Root 4. Operation Failure Root 3. Detect Quality Loss Cause Analysis Cause Analysis 4. Defect Root Cause Analysis 1. Predict Failures 1. Predict Activity Time 1. Early Defect Detection 2. Estimate RUL 2. Predict Production KPI(s) 2. Yield Quality Predict. Predictive 3. Predict Failure Impact 3. Demand Forecasting 4. Supply Chain Disruption 1. Reduce Failure Cost 1. Failure Rate Reduction 1. Process Parameter 2. Reduce Failure Rate 2. Fuel/Energy Reduction Recommendation for 3. Repair Recommendation 3. Equipment Scheduling Quality Improvement Prescriptive 4. Optimize Maintenance and Dynamic Dispatch 2. Improve Testing 4. Operations Recommendation
Example – Maintenance Effectiveness Estimation Determine the effectiveness of each maintenance activity, vendor, practice, etc. to improve maintenance operations Overhaul Overhaul Chemical Cleaning Overhaul Sensor Data, Descriptive Maintenance
Example – Operator Profiling Characterize the efficiency, safety of operator behavior to improve operations Feature Operator Behavior Sensor Mill extraction Machine Profiling Learning Data Video Model Operator KPI Sensor Data/Video Data, Descriptive Operations
Example – Quality Test Failure Prediction Predict failures earlier in process Sensor Data, Predictive Quality
Example – Repair Recommendation Recommend the correct repair to reduce repair mistakes and cost of repairs Symptom (Free text) NLP Recommendations Machine Learning Data Log Model Historical repair data Sensor/Maintenance Data, Prescriptive Maintenance
Example – Mining Operations Improve OEE for mining operations with automated dispatching Operational/Simulation Data, Autonomous Operations
Next Stage of Industrial AI Prescriptive analytics OT×AI × AI Driven Control Value of Insights = Business Impact Total Operation Optimization & Automation Recommendation of best action Operating Prescriptive Maintenance Scheduling Envelope Analytics Recommendation Recommendation Recommendation A view of the future Predictive Failure Activity Time Batch Quality Prediction Prediction Prediction Analytics Insights on the present Descriptive Performance Operations Quality Monitoring Monitoring Monitoring Analytics Predictive Operations Quality Maintenance Optimization Improvement Individual Fleets End-to-end Scope of Control
Next Stage of Industrial AI Recommend actions to achieve multi- AI Driven objective optimization with machine Control learning, AI, and simulation Up to 85% Connected Industries Geographically distributed production systems Material Equipment Process Product Supply Chain & Logistics
Next Steps and Conclusions
Complexity of Automation Control Operations Strategy (secs – mins) (mins – days) (days – months) AI/ML Enterprise Number of sub-components Complexity of Automation
Cost Tradeoffs Performance Degradation Detection: Failure Prediction: Accuracy-Gain tradeoff Accuracy-Latency tradeoff $100,000 $300,000 $ Degradation Cost Failure $250,000 $80,000 Cost $ $200,000 $60,000 Detection Error Cost False $150,000 Cost Alarm $40,000 Cost $100,000 Total Cost $20,000 Total $50,000 Cost $0 $0
In Conclusion “…I am very optimistic about the eventual outcome of the work on machine solution of intellectual problems. Within our lifetime machines may surpass us in general intelligence….” – Marvin Minsky, 1967 It’s difficult to make predictions especially about the future
Thank You
Recommend
More recommend