Improving drone flight safety using machine IVHM COE learning · Improving drone flight safety...
Transcript of Improving drone flight safety using machine IVHM COE learning · Improving drone flight safety...
IVHM COE
http://www.cranfield.ac.uk/ivhmIVHM©2012 Cranfield University All rights ReservedIVHM©2015 Cranfield University All rights Reserved
Improving drone flight safety using machine
learningJonathan Pelham6th October 2015
http://www.cranfield.ac.uk/ivhmIVHM©2015 Cranfield University All rights Reserved
IVHM©2015 Cranfield University All rights Reserved http://www.cranfield.ac.uk/ivhmIVHM©2015 Cranfield University All rights Reserved
“Data does not become information until it has been understood in the mind of the pilot.”
“Data does not become information until it has been understood in the mind of the pilot.”
Introduction
IVHM©2015 Cranfield University All rights Reserved http://www.cranfield.ac.uk/ivhmIVHM©2015 Cranfield University All rights Reserved
• Production, certification & testing
• Total ownership costs• System & life cycle • Requirements• FMECAs• Design models• Failure modes/models• System test data
Design EngineeringManufacturing
• Maintenance Scheduling• Spares Supply• Asset Tracking• Maintenance Execution
Vehicle Maturation/ New Product
• Operational Demand• Fleet Availability• MR & O leading
• Operational Schedule• Operational Effectiveness
Health Status
Act Acquire
Transfer
Sense
Health Status• Current• Predicted
Analyse
Asset
Data Repository& Ground Processing
Maintenance & Logistics
Operational Control
• Production, certification & testing
• Total ownership costs• System & life cycle • Requirements• FMECAs• Design models• Failure modes/models• System test data
Design EngineeringManufacturing
• Maintenance Scheduling• Spares Supply• Asset Tracking• Maintenance Execution
Vehicle Maturation/ New Product
• Operational Demand• Fleet Availability• MR & O leading
• Operational Schedule• Operational Effectiveness
Health Status
Act Acquire
Transfer
Sense
Health Status• Current• Predicted
Analyse
Asset
Data Repository& Ground Processing
Maintenance & Logistics
Operational Control
Integrated Vehicle Health Management
IVHM©2015 Cranfield University All rights Reserved http://www.cranfield.ac.uk/ivhmIVHM©2015 Cranfield University All rights Reserved
Flight!
IVHM©2015 Cranfield University All rights Reserved http://www.cranfield.ac.uk/ivhmIVHM©2015 Cranfield University All rights Reserved
RPAS – a refresher
Operator Tasks• Aviate• Navigate• Communicate• Manage systems
OODA LoopObserve -Gather informationOrient -Put Information in contextDecide-Decide correct actionAct -Perform Action
IVHM©2013 Cranfield University All rights Reserved
UAS Architecture changes with increasing size
6
Single BoardPicopter
Federated Avionics
Global HawkMulti-Board
Demon
IVHM©2014 Cranfield University All rights Reserved
IVHM©2015 Cranfield University All rights Reserved
USAF RPAS Class AMishap Rate
General Aviation Rate 1.2/100,000 fh
Boeing 737 Rate 0.13/100,000 fh
IVHM©2015 Cranfield University All rights Reserved
RPAS Information flow timescales
IVHM©2015 Cranfield University All rights Reserved
A BC
D E FG
Flight I
Flight Data Problems
10Hz 16bit ADC ~
1.2Mb per Minute
30+ Channels ~ 36Mb per minute
Flight 7hrs+ ~
2.1Gb per hour ~ 14.4Gb
Multiple aircraft – Different versions
Multiple Tb per day
IVHM©2015 Cranfield University All rights Reserved http://www.cranfield.ac.uk/ivhmIVHM©2015 Cranfield University All rights Reserved
The Immune system
B-Cells
T-Cells
IVHM©2015 Cranfield University All rights Reserved http://www.cranfield.ac.uk/ivhmIVHM©2015 Cranfield University All rights Reserved
Feature space
Example 2-dimensional feature space• aircraft velocity• load factor
12
34
CaseData
X
IVHM©2015 Cranfield University All rights Reserved http://www.cranfield.ac.uk/ivhmIVHM©2015 Cranfield University All rights Reserved
Feature spacewith AIS
AIS LearningSelf/Non-Self discriminationLearning operational rules from flight data
1. Flight mishap cases within feature space
2. Non mishap Self within feature space
3. Simplified rule for aircraft on-board use
IVHM©2015 Cranfield University All rights Reserved http://www.cranfield.ac.uk/ivhmIVHM©2015 Cranfield University All rights Reserved
Results
IVHM©2015 Cranfield University All rights Reserved http://www.cranfield.ac.uk/ivhmIVHM©2015 Cranfield University All rights Reserved
• Flight History & maintenance data
• better power & processing
• Local humans• Limited bandwidth
• “Fresh” data• Better bandwidth
• Limited power & processing
RPAS AIS structure discussion
IVHM©2015 Cranfield University All rights Reserved
• RPAS need to improve their mishap rate and the lessons learned integrated into future designs
• Machine Learning may be a solution • But not suitable for FMS deployment
• Integrated Vehicle Health management has many opportunities in RPAS
• Linux and Open Source have a major role to play
Conclusions
IVHM©2015 Cranfield University All rights Reserved http://www.cranfield.ac.uk/ivhmIVHM©2015 Cranfield University All rights Reserved