V02162017
Virtual Risk-informed Agile Maneuver Sustainment (VRAMS) - Fabrics of Artificial Intelligence-informed Technology for
Holistic Sustainment (FAITHS)
Institute for Materials, Manufacturing, and Sustainment (IMMS) Texas Tech University (TTU)
Dy D. Le, Director
Presented to: Defense Advanced Research
Project Agency (DARPA) Arlington, VA
March 24, 2017
V02162017
•!“Leap-ahead” capabilities to sustain Next-Generation Aircraft with “little” or without maintenance, “Zero-Maintenance (ZM)” “Leap-ahead” capabilities to sustain Next-Generation Aircraft with “Leap-ahead” capabilities to sustain Next-Generation Aircraft with
•!Evolutionary discoveries to enable the development of Next-Generation Air Platforms with “Fatigue-Free (FF)” characteristics
!"##$%&'!"(&)*+,-+&'.-/0+$1$23'4'5)#)6*1*&3'7$%'89#-:*;$+)%3'<$%/-'=>=?'4'@-3$+:'A*&0'B-:"/-:'C$2*(;/'<$$&#%*+&'
2 Defying “Impossibilities”: U.S. Army Aviation Sustainment Vision
V02162017
“Digital Nanomaterial Architecture”- Additive Manufacturing
Self-Healing/Material State Awareness
Material Damage Precursor - Failure Correlation
Virtual Risk-informed Agile Maneuver Sustainment
(VRAMS) Capability
!"#$%&'($
•!")*)+,($,-$.-&/0)1(2$$!!!!"!#$$%&'()%!*%+',-%!+%(-)+.,(/(0'-')1!2)()%!!!!!!!!!!"!34*'56!0%145&!745&'845"0(2%&!!!!!!!!!!!!!!!3('5)%5(5,%!97:3;!)4!$'22'45"!!!!!!!!!!!!!!!!!'5<4=$%&!$()%='(-!2)()%"0(2%&!!!!!!!!!!!!!!!!!!!!!(>(=%5%22!!!!!!!!!!!!!!!!!!!!"!?50@=&%5!+(=&>(=%"-%*%-!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!,45,%=5!<4,@2!45!+'6+%="!!!!!!!!!!!!!!!!!!!!!!!!!!!!-%*%-.$'22'45"=%-%*(5)!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!4/%=(8452!
•!345-6)1$7148$>')+!A0=%(B)+=4@6+C!)%,+54-46'%2!D!!!!!!,(/(0'-'8%2!)4!,45&@,)!%E/%&'845(=1!$(5%@*%=2!>')+!!!!!!!!!"#!$%&'%()*+,"-'.&)/&).0+*%'.&)1)!"!$%/&2+&$!,42)2!
9:;<$
=>.$!-
(,($
3 Envisioning Discoveries: A System Level Approach Perspective
V02162017
Risk Assessment
Quantitative Total Risk
Health State
4
Health Treatment Longevity Sustainment
Genetics Life Style/Stress-Fatigue
Accident/Transmission
Human-Machine Longevity Sustainment
Autonomous & Assisted Environment
Preventive
Autonomous & Assisted
Design Usage/Stress & Fatigue
Environment Accident/Threats
NOT Autonomous BUT Assisted
Preventive
cs.stanford.edu Courtesy of photosearch
Adaptive
NOT Adaptive NOT Autonomous BUT Assisted
cs.stanford.edu
Platform Design Multifunctional Structures Intelligence Next-Gen Platform Intelligence Intelligence Intelligence Next-Gen Platform Intelligence Next-Gen Platform
Bio-inspired Living Aerial Platform (BiLAP)
Living Skin
Science for Autonomous Prognosis and Healing for Longevity Sustainment
V02162017 V02162017
FEEL
Multifunctional
Risk Assessment
Precursor ID
Data Fusion
Diagnostics Info Format Precursor-based Indication
Physics Models
Diagnostics
Prognostics
High
Hazard Mapping
Degree of Risk
medium Low
Negligible
Quantitative Total Risk
Functional Hazard Assessment Fault Tree Analysis
Decompose SSC&I into Components
Probable Occasional
Remote Improbable
Identified Risk Acceptable Risk
Unacceptable Risk Unidentified Risk
Risk Matrix
Types of Risk
Health State Precursors
Adaptive Controls
Safety Significant Subsystems & Interfaces
Indirect Adaptive Controls
BiLAP Prototype
Multi-Scale Modeling Nano-Macro Interfaces Assessment Method
Smart Materials
Living-Skin & Robots
Integration
1 0 1 0 1 0 1 1 0 1 1 0 10110110101 0 1 0 1 0 1 0 1 0 1 1 0 1 0 101010
Prototype Prototype 1 0 1 0 1 0 1 1 0 1 1 0 10110 1 0 1 0 1 0 1 0 1 1 0 1 0 10101
Prototype Prototype 1 0 1 0 1 0 1 1 0 1 1 0 10110 1 0 1 0 1 0 1 0 1 1 0 1 0 10101
Distributed Sensing
& Transmission
Direct Adaptive Controls
Assess Remaining Capabilities: Structural Integrity and Survivability
Inform Risk on Demanding Capability ASSESS RECOVER & ADAPT
Quantitative Total Risk
Probable Occasional
Remote Improbable
Identified Risk Acceptable Risk
Unacceptable Risk Unacceptable Risk Unidentified Risk
Risk Matrix
Types of Risk
Adaptive Controls
Indirect Adaptive Controls
BiLAP
Living-Skin & Robots
Prototype Prototype 1 0 1 0 1 0 1 1 0 1 1 0 10110110101 0 1 0 1 0 1 0 1 0 1 1 0 1 0 101010
Direct Adaptive Controls
On-Board Nano-Robots
Fly-by-Feel
1 0 1 0 1 0 1 1 0 1 1 0 10110 1 0 1 0 1 0 1 0 1 1 0 1 0 10101
High
Hazard Mapping
Degree of Risk
medium Low
Negligible
Functional Hazard Assessment Fault Tree Analysis Fault Tree Analysis
Decompose SSC&I into Components
Health State
Safety Significant Subsystems & Interfaces Safety Significant Subsystems & Interfaces Safety Significant Subsystems & Interfaces
Nano-Macro Interfaces
5
Quantitative Total Risk Quantitative Total Risk Quantitative Total Risk 1 0 1 0 1 0 1 1 0 1 1 0 10110 1 0 1 0 1 0 1 0 1 1 0 1 0 10101Self-Healing Precursors
Remaining Life Health
Science & Technology for “Bio-Inspired Living Aerial Platform - BiLAP
V02162017
Cognitive Capability for Legacy-Future Aviation Platforms
VRAMS Concept
A Step Ahead of “IBM WATSON”
Real-Time Self-State Awareness
“Fatigue-Free & Zero-Maintenance
AI Machine Learning Algorithm Suites
Conceptualized Need Developed Concept & Tech Demonstrating
Platform Sustainment and Survivability
6
V02162017
F!
?@.%2$A9B9;$C@.%$DA9B9;$
GHGHGHGHGHGH!HGHGHGHGHGHG!GHGHGHGHGHGH!GHGHGHGHGHGH!HGHGHGHGHGHG!GHGHGHGHGHGH!HGHGHGHGHGHG!GHGGHGHGHHGH!GHGHGHGHGHGH!HGGHGHGHHGHG!GHGGHGHGHHGH!HGHGGHGHGHHG!GHGGHGHGHHGH!GHGGHGHGHHGH-B!
GHGHGHGHGHGHGHGHGHGHGHGHGHGHGHHGHGHGHGHGHH!HGHGHGHGHGHGGHGHGH!GHGHGHGHGHGHGHGHGHHGHGHG!GHGHGHGHGHGH!GHGHGHGHGHGHGHGHGHGHGHGHGHGHGHHGHGHGHGHGHH!HGHGHGHGHGHGGHGHGH!GHGHGHGHGHGHGHGHGHHGHGHG!GHGHGHGHGHGH!
GHGHGHGH!HGHGHGH!GHGHGHGHGHGHGHGH!GHGHGHGHGHGGHIGHGHGHGHGHH!HGHGHGHGHGHGGHGHGH!GHGHGHGHGHGHGHGHGHHGHGHG!GHGHGHGHG!
GHGHGHGHHGHGHGHGHHGHGHG!GHGHHGH!HGHGHGHGGHIGHGHGHGGHIGHGHGHHGHGH!GHGHGHGHGHGHG!GGHIGHGHGHHGH!GHGHGHGHGHGGHIGHGHGHGHGHGHHHGHG!
.8(,)4$E8*F40G($
!F5,'10*H$EF4FH)$%1)G'1(-1$
.8(,)4$"F()&0*)$
?)F&ID04)$#J.$
#-/)&0*H$
?0(K$7(()((4)*,$
#F*)'L)1$7/F5,FM-*$
GHGHGHGHGHGH!HGHGHGHGHGHG!GHGHGHGHGHGH!GHGHGHGHGHGH!HGHGHGHGHGHG!GHGHGHGHGHGH!HGHGHGHGHGHG!GHGGHGHGHHGH!GHGHGHGHGHGH!HGGHGHGHHGHG!GHGGHGHGHHGH!HGHGGHGHGHHG!GHGGHGHGHHGH!GHGGHGHGHHGH-B!
7 VRAMS Concept
V02162017
*!*!
*!
Loading Threshold
Optimizing Solution for Sustaining “Fatigue-Free”
Delayed Failure
Loading Threshold
.. !.. !.. !.. !.. !.. !..
!.. !.. !.. !.. !.. !..
.. ..
....
..
..
1
5
*!*!*!Fatigue Cracks
Optimized Load
Core Engine: Virtual Risk-informed Agile Maneuver Sustainment (VRAMS)
7
3
6
Extended Usage
.)*()N$E),)G,N$>$$!OF1FG,)10P)$
7(()((0*H$EF4FH)$>$?0(K$
Combat Damage Delayed Failure
7
6
Extended Usage Highly Damaging Load
4
#F0*,)*F*G)IQ1))$=5)1FM*H$%)10-/$2 *! 2 Damage Precursor
3R)G'M*H$7/F5ML)$$#F*)'L)1($
Optimized Load
8
V02162017
FAITH
(a) Rule-based Pattern Recognition
(b) Artificial Intelligence Machine Learning
(a) Rule-based Pattern
(b) Artificial Intelligence (c) C
ogni
tive
Cap
abili
ty
(3) Facilitate system behavior change for sustaining longevity or self healing to disrupt failure cascade $
(2) Enable cognitive cueing and human-machine teaming, interaction, & communication in RT
(1) Identify Model of
Models properties, e.g.,
ingredients of longevity
or precursors of onset
of failure
machine teaming, interaction, & communication in RT
(2) Enable cognitive cueing and human-
Self-Learning
FAITH: Model-based System of Systems (MSoS) Concept
Achieve “Zero-Maintenance” Thru Intelligent Comprehensive Integrated Solution
9
1
3
Goal
2 3 3 (a) Model of Models
(b) System of Systems
V02162017
(2)
DESCRIPTIVE PREDICTIVE PRESCRIPTIVE
What
(2)Quantify Forecast Prognosticate
!! Aggregate data in a way that relationships, meaningful patterns, and/or insights of anomaly might emerge
!! Quantify, forecast, and prognosticate the likelihood of future events
"!Machine Learning
!! Optimize hindsight/insight into the past/future, assess effect of future decisions and projected outcomes, and inform course of actions
When Why
"!Analytical Methods
Probability
1 0 1
0 1 0
1 1 0
1 1 0
1011
0010
1010
1011
1 0 1
0 1 0
1 0 1
0 1 1
0 1 0
1010
1011
1001
01
1 0 1
0 1 0
1 1 1
0110
1101
0101
0101
1110
11
0 1 0
1 0 10
1 1 0
1 0 1
0101
0101
0110
0101
1 0 1
0 1 0
1 1 0
1 1 0
1011
0110
1010
1010
111
0 1 0
1 0 1
0 1 0
1 1 0
1 0 10
1010
1010
1100
101
1 0 1
0 1 0
1 1 0
1 1 0
1011
0110
1010
1010
111
0 1 0
1 0 1
0 1 0
1 1 0
1 0 10
1010
1010
1100
101
1 0 1
0 1 0
1 1 0
1 1 0
1011
0010
1010
101
0 1 0
1 0 1
0 1 0
1 0 1 0
1 0 1
0 1 0
1 1 01 0
1 0 1
1 0 1
001011010
111
1 0 1
0 1 0
1 1 0
1 1 0
1011
0110
1010
0 1 0
1 0 1
0 1 0
1 1 0
1 0 10
1010
1010
1 0 1
0 1 0
1 10
1101
1010
1010
1011
1101
1 1 0 1
1 01 0 1
0 1 1
0 1001
011010
1010
0101
011110000
101011
1 0 1
1 0 1
0110
1101
0101
0101
11 1
0 1 0
1 1 0
1 0 1
0101
0101
0110
0101011
0 1 0
1 1
0 10 1
1 0 1
0 1 1
0 1 1
0 0 1
0 1 0
1 0 1
0 1 1
0
Augmented Reality
Logistic Planning & Execution
Hindsight into the Past-Insight into Current Operation-Foresight into Future
Of Outcomes
Prognosticate
Machine Learning
Insight into Current Operation-Foresight into Future
Of OutcomesOf Outcomes
Real-Time Data Stream 1 0
1 0 1
0 0 1
0 1
0 1 0
1 0 10 1
Augmented Reality Augmented Reality
0 1 0 Augmented Reality
0 1 0
1Augmented Reality 1 1
Augmented Reality 1
0 1
Augmented Reality
0 10 1
Augmented Reality
0 10 1Augmented Reality 0 10 1Augmented Reality 0 1
101
1010
1011111
00001010
1
1010
101011
1
1 0 1 0
1 0 1
0 1
11
1010111 1010
Cognitive Cognitive Cognitive Cognitive Real-time Self State Awarenes
s
"!Deep Learning
"!Artificial Intel Self State Self State Self State Self State
0 1 0
1 1 0 1
0 1
0111
00
101
011Cognitive
Real-time Real-time Self State
Deep Learning Cognitive Deep Learning Cognitive Artificial Intel Cognitive Artificial Intel Cognitive 1 Self State
Formats
Cognitive ""Deep Learning
"Artificial Intel (4)
FAITHS-VRAMS Core Engine Algorithm Architecture Modules
Quantify Forecast Prognosticate Prognosticate
"!Derivative data
"!Fused data
Decision Options
When Why Why
Optimize Inform (3)
Ingest Digest Extract Prepare
Transform
"!Structured/Unstructured
"!Related/Unrelated
(1)
Data Binning
Ingest Digest Extract Prepare
Transform
(1)
Formats
Data Stream
Formats
Data Stream
Sources
Formats
Impacts
Data Storage
Informed Course of Actions
10
V02162017
Information Assurance & Governance
Sensing Network
Data Bus data
Safety Databases
Electronic Logbook
Processed & Relational Database
Standing Queries
Cognitive Ad-Hoc Queries
(1)
(2) (3) (4)
Big Data In-Memory Processing
Processed & Processed &
Data Scoring & Ranking Models
Working Stream
Warehouse RAM
Data Streams
APIs
Data Streams Processed & Processed &
Feature Vectors
(2) (3) (4)(2) (3) (4)(2) (3) (4)(2) (3) (4)(2) (3) (4)(2) (3) (4)(2) (3) (4)(2) (3) (4)(2) (3) (4)(2) (3) (4)(2) (3) (4)(2) (3) (4)
AI-based Analytics
Tran
sien
t t-S
trea
ms
Maintenance Databases
Data Bus data
Performance Data
FAITHS (1)
Sensing Network
Sensing Network Performance Data
DIGITAL TWIN
ALIS
Safety Databases
Safety Databases Digital Modeling
Electronic Logbook
Maintenance Databases
Maintenance Databases
Maintenance Databases Digital Computation Outputs
Electronic Logbook Safety Databases
Safety Databases
Safety Databases
Safety Databases Digital Modeling
Electronic Logbook
Electronic Logbook
Electronic Logbook
Electronic Logbook
Digital Computation Outputs
Digital Computation Outputs
Digital Computation Outputs
Digital Computation Outputs
Digital Computation Outputs
Digital Computation Outputs
Digital Computation Outputs
Digital Computation Outputs
Digital Computation Outputs Digital Analysis Outputs
Data Bus data
Data Bus data
Performance Data
Performance Data
DIGITAL TWIN
DIGITAL TWIN
DIGITAL TWIN SLED/STAMP
Filters
Cogn
itive
Tech
nolo
gy
GHGHGHGHGHGHGHGHGHGHGHGHGHGHGHHGHGHGHGHGHHGHGHGHGHGHGHGHGHGHGHGHGHHGHHGH!HGHGHGHGHGHGGHGHGHGHGHGHGHGHGHGHGHGH!
In Theater
Transforming BIG DATA through DATA PIPELINE and Insight VISUALIZATION
Data Streams Data Streams
In TheaterIn TheaterIn TheaterIn TheaterIn Theater
ALIS ALIS
Digital Modeling
Digital Modeling
Data Bus data
Data Bus data
Data Bus data
Data Bus data
Pilot Sentiment
Flight Critical
Systems
Digital Modeling
Digital Modeling
Safety Databases
Safety Databases
Safety Databases Digital Modeling
Digital Analysis Outputs
Digital Analysis Outputs
Digital Analysis Outputs Digital Modeling
Safety Databases
Safety Databases
Safety Databases Digital Modeling
Maintainer
Sentiment
Database Database
LSTM AI Network
LSTM: Long-Short-Term Memory
Complex Data Processing – Cognitive, Scalable, & Open Architecture - Descriptive Module 1: Manage Automated Data Pipeline in Real Time -
GHGHGHHGHHGHGHGH
HGHGHGHGGGHHHHGHH
FAITHS (1)Ingest – Digest – Extract –
Prepare – Transform
Descriptive Models
AI Network
Redstone Arsenal IMMC
Redstone ArsenalAMCOM LCMC
IMMC PMs
AMCOM AED PEO
AVN IMMC CAB CAB
AI Network
Artif
icial
Inte
lligen
ce (A
I) Co
mm
unica
tion
Chan
nel
Metadata Warehouse
No Time Restriction
Streaming Engines
AMCOM AMCOM
Prepare Prepare
Stream Stream Stream Warehouse Warehouse
11
V02162017
Cognitive
Output (Y) Input (X)
Quantify Forecast Prognosticate
"!Machine Learning
"!Analytical Methods
Of Outcomes
Optimize Inform
"!Fused data
(2) (3)
Probability Decision Options
Impacts
Regression Models
Regression Analysis
Risk & Probability of Outcomes
Yd = f(Vi) + fe
Ref: Statistical Analytics System (SAS) Institute Knowledge Discovery in Databases
Reinforcement Learning For
Dynamics Events
Machine Learning
Algorithms
FAIT
HS
(2-3
)
(4)
FAITHS (1)
Multi-Scale Modeling
Enabling Autonomous Prognostics Via Machine Leaning
FAITHS (1)
Remaining Useful Life
Signal Detection & Data Fusion Theory
Nano Micro Macro
Cognitive
Output (Y) Output (Y) Input (X)Input (X)
Quantify Forecast Prognosticate Prognosticate
"Machine Learning Machine Learning
"Analytical Methods Analytical Methods
Of Outcomes
Optimize Inform
"Fused data Fused data
Quantify Forecast Prognosticate
(2) Optimize Inform
(3)
Probability Decision OptionsDecision OptionsDecision OptionsDecision OptionsDecision OptionsDecision Options
ImpactsImpactsImpacts
Regression Regression ModelsModelsModels
Regression Regression AnalysisAnalysisAnalysis
Risk & Probability Risk & Probability of Outcomesof Outcomes
Output (Y) Output (Y) Output (Y) Output (Y)
YYdYdY = f(V = f(Vi = f(Vi = f(V ) + ) + = f(V ) + = f(V = f(V ) + = f(Vi) + i = f(Vi = f(V ) + = f(Vi = f(V ffefef
Ref: Statistical Analytics System (SAS) Institute Ref: Statistical Analytics System (SAS) Institute Ref: Statistical Analytics System (SAS) Institute Output (Y)
Ref: Statistical Analytics System (SAS) Institute Output (Y) Input (X)
Ref: Statistical Analytics System (SAS) Institute Input (X)
Ref: Statistical Analytics System (SAS) Institute Output (Y)
Ref: Statistical Analytics System (SAS) Institute Output (Y)
Ref: Statistical Analytics System (SAS) Institute Knowledge Discovery in Databases Knowledge Discovery in Databases
Reinforcement Reinforcement Learning For Learning For Learning For
Dynamics EventsDynamics EventsDynamics EventsDynamics Events
Machine Machine Learning Learning Learning
AlgorithmsAlgorithms
FAIT
HS
(2-3
)FA
ITH
S (2
-3)
Cognitive (4)(4)
Output (Y) Input (X) Output (Y) Output (Y) Output (Y)
Decision OptionsDecision OptionsDecision OptionsDecision OptionsDecision OptionsDecision OptionsDecision Options
Fused data Fused data
ImpactsImpacts
AlgorithmsAlgorithms
Probability
Of Outcomes
FAITHS (1)
Multi-Scale Modeling Remaining Useful Remaining Useful Remaining Useful LifeLife
Signal Detection & Signal Detection & Data Fusion TheoryData Fusion TheoryData Fusion Theory
Y = f(V ) + = f(V ) + = f(V fFAITHS (1)
Nano Micro MacroNano Micro Macro
APIs APIs
Discovered Knowledge
"!Derivative data
Complex Prognostic Processing – Cognitive, Scalable, & Open Architecture Predictive Modules 2-3: Perform AI-based Predictive Analytics in Real Time 12
V02162017
Machine Learning Decision Tree
Uncertainty Entropy
entropy(X) = -!pi log2 pi
Deep Learning Network Training Reduced Uncertainty
Recurring Neural Network
Cognitive Cognitive (4)
Uncertainty
entropy(X) = -!pi log2 piXavier Weight
Initialization
Certain
fc (er , bi)
Training dataset
(e (e ( , b
Network Error
Forward Training
Data & labels
Back-Propagation-Through-Time
Training
Long-Short-Term Memory Units
Forward Forward TrainingTraining
Data & Data & labelslabelslabels Uncertainty Uncertainty Uncertainty
Certain
Back-Propagation-Back-Propagation-Through-Time Through-Time
Long-Short-Term Long-Short-Term Memory UnitsMemory Units
(1) (2) (3) (1) (2) (3)(1) (2) (3)(1) (2) (3)
FAITHS (1-3)
FAIT
HS
(4)
Reaching to The Future of Self-Learning Aviation
entropy(X) = -!
Complex Solution Processing – Cognitive, Scalable, & Open Architecture Prescriptive Module 4: Learn Past, Aware Present, & Predict Future 13
V02162017
Fu
lly A
uton
omou
s Mi
ssio
ns
Automated Logistic Planning
FAITHS: VRAMS Cognitive Capability for Legacy and Future Aviation Platforms
- Self-Sustaining -
#! “Fatigue-free” (FF) structures #! Maintenance-Free Operating Period
(MFOB) – “Zero-Maintenance” (ZM) #! Self-maintenance and optimization of
large data (“Big Data”) #! Expeditionary missions with smaller
logistic footprint - Self-Maneuvering -
#! Reconfigurable controls technologies #! Self-autorotation (rotary-wing) #! Autonomous systems (manned/
unmanned) teaming #! Fully autonomous missions
- Self-Adapting - #! Self-healing and repair #! Self-informed of parts replacement
demands and schedules #! Self-informed of remaining capability to
achieve demanding tasks or maneuvers
FAITHS-VRAMS Technology
Fully Autonomous
Missions
CBM MFOP FF-ZM
Automated Maintenance / Reliability /
Durability
O&S Costs
Past
& P
rese
nt
Futu
re
Chip Detector > HUMS > FAITHS-VRAMS
Fully
Aut
onom
ous
Miss
ions
Fu
lly A
uton
omou
s Mi
ssio
ns
Smaller Logistics Footprint
Fully
Aut
onom
ous
Miss
ions
Integrated Augmented Reality
14
V02162017
QUESTIONS ?