David Controle, Analytics Accelerator Airbus. Why Invest in AI and Deep Learning NVIDIA GTC

Transcription

1 David Controle, Analytics Accelerator Airbus Why Invest in AI and Deep Learning NVIDIA GTC

2

3 Punctuality issue 20 % flights delayed more than 15 min * 50 min average delay * 57 % airlines experience frequent delays ** 3 * Global airlines OTP, FlightStats Sept 2018 ** Major airlines with less than 80% OTP, FlightStats Sept 2018

4 Low punctuality Impact Low passenger experience $31.2 billions annual costs in US in 2007 * * Nextor report

5 Main causes of delay in US Late arriving aircraft Air carriers National Aviation System Extreme Weather 39 % 31 % 25 % 4 % Maintenance and crew issues Aircraft cleaning Baggage loading Fueling Non-extreme weather Airport operations Heavy traffic volume Air traffic control Tornados Blizzards Hurricanes 5 United States Department of Transportation (Bureau of transportation statistics)

6 Main causes of delay in US Late arriving aircraft Air carriers National Aviation System Extreme Weather 39 % 31 % 25 % 4 % Maintenance and crew issues Aircraft cleaning Baggage loading Fueling Non-extreme weather Airport operations Heavy traffic volume Air traffic control Tornados Blizzards Hurricanes 6

7 Main causes of delay in US Late arriving aircraft Airports and airlines operations Air carriers Air Traffic! National Aviation System Technical Issues Extreme Weather Weather 7

8 Main causes of delay in US! Airports and airlines operations Air Traffic Technical issues Weather Airport operations Aircraft cleaning Baggage loading Fueling Crew issues Heavy traffic volume Air traffic control Maintenance issues Non-extreme weather Tornados Blizzards Hurricanes 8

9 Main causes of delay in US! Airports and airlines operations Air Traffic Technical issues Weather Turnaround Optimization Predictive Maintenance 9

10 Aircraft sensors A320 & A330 FOMAX FOMAX Router Router WACS** 3G Wifi 4G 4G Download valuable data: up to 3 GB per day Ground Distribution Bridge Ground IMACS** Airline IT Airline Operational Data & Aircraft Data* An integrated & secured data pathway from the aircraft to * Aircraft data for Aircraft not equiped with FOMAX **WACS: Wireless Airport Communication System **IMACS: Information Management Aircraft-ground Communication System

11 First use cases: Post-flight analysis and Live monitoring Data visualisation Enhance sensors data visualization and enables to correlate turnaround performances with weather information, traffic information and geolocation data Situational awareness Remote turnaround visualization and live monitoring from airlines operating centers Fleet Planning Compare actual turnaround performances with schedules and identify delay recovery patterns or bottlenecks Root-cause analysis Identify root cause of turnaround delays Statistics Monitor global performances, identify seasonal patterns and trends Operations Planning Monitor tasks performed during the turnaround, identify critical paths and gaps 11

12 What is AI? Computer vision Deep learning Predictions Reinforcement learning Recommendations Machine learning Advanced planning optimisation Natural language understanding 12

13 Why invest in AI Will end-users be comfortable with black box recommendations? How much can we save? How to deal with privacy? How mature is AI? How much does it cost? What s the value for the customer? What is the right organization? How to deal with unexplainable models? 13?

14 3 AI based use cases Computer vision Predictions Advanced Planning optimization Recommendation 14

15 Computer Vision Domain specific object detection and tracking: Ground Support Equipment detection and geolocation Actions recognition Cabin cleaning Passengers boarding and deplaning Baggage loading/unloading Cargo loading/unloading 15

16 Deep-learning / Predictions Weather Real-time alerting system: Air Traffic What s the probability of delay in the current context? 16 First Sensors Values Flight schedules Results: 93% accuracy 10% of false positive 5% of missing alerts

17 Advanced planning optimisation (Experimental) Weather Air Traffic Contextual planning recommender: What s the optimal planning in the current context Historical Sensors Values 17 Flight schedules

18 Why invest in AI and deep learning? What s the value for the customer? On-time performances improvement Better passenger experience expected Less fuel burned to recover delay How much does it cost? GPU resources thanks to NVIDIA 5 sprints run in Agile mode Team of 6 Airbus data scientists / data analysts: Florian Caroline David J Ali Pooja David C What s the right organization? Work in Agile mode Work close to the customer How to deal with privacy? Data governance team Secure Skywise platform Facial recognition not required 18

19

20 Appendix: Predictions performances ROC / AUC Confusion Matrix 20

21 Appendix: Annual cost of delay Cost Component Cost ($ billions) Costs to Airlines 8.3 Costs to Passengers 16.7 Costs from Lost Demand 2.2 Total Direct Cost 27.2 Impact on GDP 4.0 Total Cost