Burdun AUVSI-2007

AUVSI's Unmanned Systems Europe 2007 8-9 May 2007, Hilton Cologne, Kln, Germany
INTELONICS
UAV "Built-in" Safety Protection: "Built-in" A Knowledge-Centered Approach Knowledge-Centered

Presentation Slides Version #2 of 30.04.2007 Ivan Y. Burdun, Ph.D Ivan Y. Burdun, Ph.D Chief Scientist Chief Scientist Intelonics Ltd. Intelonics Ltd. Russia Russia ivan.burdun@mail.ru ivan.burdun@mail.ru
INTELONICS
Copyright 2007 Intelonics Ltd.
Presentation Plan
INTELONICS
Problem: UAV flight safety performance prediction and protection in complex (multifactor) situations Solution approach: Knowledge is Power Methodology conceptual framework (introduction):
- micro- and macro-structural knowledge models of flight - flight situation scenario - operational factor - operational hypothesis - situational tree - safety spectrum - flight safety [performance] window - last chance for recovery point, self-preservation decision making - safety chances distribution time-history - dynamic safety window tree
Case study: Notional UAV Low-Altitude Flight in the Presence of Urban Infra-Structure Obstacles Conclusions
2007 Intelonics Ltd. 2
INTELONICS
Natural Tree Analogy of Pilots Situational Knowledge Base

E B
Legend:
A E B A B
Characteristic zone of a pilots situational knowledge base Natural tree analogy Space available for tree growth Trees trunk First-order derivative branch Second-order derivative branch Absent but possible branching Dry or broken branches Excessive, chaotic branching Insufficient, sparse branching
D 1 A C A A E 2
Space of possible complex flight situation scenarios Basic (standard/non-standard) flight situation scenario One-factor non-standard flight situation scenario Two-factor non-standard flight situation scenario Missing knowledge Forgotten or shadowed knowledge Non-systematic , occasionally developed knowledge Fragmentary, incomplete knowledge
0 A D B C
1 2 A B C
B C D D F
A C F
D E F
Optimally dense Systematic, yet economically developed and stored, knowledge branching Physically unattainable flight situation scenarios A sub-domain where branching is impossible
Strengths to model by means of AI in UAVs

A, B,C, D main defect types of a human pilots situational knowledge.
Defects to back up by means of AI in UAVs
Lack of theoretical and practical training (design and testing) especially under complex (multifactor) conditions may result in structural disparity of a human pilots (automatons) internal situational tree of flight.
INTELONICS
Fractal Tree Model Of Pilots Situational Expertise Growth In Long-Term Memory
10
desirable maturity levels of AI knowledge for flight safety protection in UAVs

Legend: Characteristic levels of piloting expertise: k{1, 2, 3} experience of a student pilot, k{8, 9, 10} experience of a professional pilot, ace, or test pilot, k{4, , 7} interim (immature) states of experience. The most valuable asset of an expert pilot (a perfect automaton) is the reliability and comprehensiveness of his/her (its) knowledge of the system behavior under complex (multifactor, non-standard) operational conditions. This expertise is of critical importance for reliable prediction, timely avoidance or/ and safe resolution of chain reaction type emergencies in UAV flight.
INTELONICS
Micro- and Macro- Structural Models Of Complex Flight Situation Domain

Situational tree-network of flight Flight situation scenario ... ... 14 ... 2 E6 9 E2 8 10 5 E4 4 3 E8 ... E1 E5 ... 11 7 15 6 E7 C4 B-1 Macro-structure of flight 13 ... E4 12 ... 1 Elementary situation (Ei, j, Ek) Ei j Ek Event E Process Legend: Ei - flight event; j - flight process; Cm fuzzy constraint of flight; - reference state; - bud type state; - target state (leaf); source state (root); B-1 parent branch; B0 - main branch (trunk) basic flight scenario; Bn n-th order derivative branch (non-standard scenario with n factors, n = 1, 2, )
C1 B0 C2
B1
B2 C3
Micro- and macro- structures of flight Micro-structure are two interconnected components of the of flight
2007 Intelonics Ltd.
developed generalized knowledge model of a complex flight situation domain.
INTELONICS
Basic Flight Situation Scenario (Examples)

Si S1 S2 S3 S4 S5 S6 Content Description Normal takeoff, maintaining commanded flight path and bank angles during initial climb Normal takeoff under crosswind and given runways surface conditions, maintaining commanded flight path and bank angles during initial climb Continued takeoff (left-hand engine out at given VEF), maintaining commanded flight path and bank angles during initial climb Normal takeoff under wind shear conditions, maintaining commanded flight path and bank angles during initial climb Continued takeoff (left-hand engine out at VEF), under crosswind conditions, maintaining commanded flight path and bank angles during initial climb Low-altitude level flight
Scenario #6 will be used in the notional case study
Basic (Baseline) Scenario Si is a plan of some central or reference flight situation be it standard or non-standard one. It represents the situational trees trunk. Variations of the basic scenario derivative cases constitute the situational trees crown. The vehicles flight safety knowledge base is in fact a collection (a forest) of the situational trees, which are constructed for various basic scenarios and exemplify a complex (multi-factor) flight situation domain.
Joint Graph of Basic Scenarios (Example)

INTELONICS
W3: cross-wind 10 m/s E44: engine out speed F1: left-hand engine out
Legend:
44 6
S3 S2
S5
E88: altitude 200 m
44
E44: engine out speed
event process
88
T5: maintain commanded bank (G) and sideslip (G) angles
S1
P1: set engine levers to commanded power rating
E6: altitude 10.7 m
P3: wheels - up
F1: left-hand engine out
E7: altitude 120 m
T2: maintain commanded bank P4: flaps - up (G) and sideslip (G) angles T2: maintain commanded bank (G) and sideslip (G) angles P5: maintain commanded IAS E12: flaps retracted
1
E1: situation start
T1: maintain ground-roll path along runways centerline
S6
55
E55: in airborne
190
E190: end of situation T4: maintain commanded flight path angle G2 (2nd phase of climb, level flight, or descent)
3
E3: VR achieved
P2: elevator up for rotation
5
E5: pitch 8
T3: maintain commanded flight path angle G1 (1st phase of climb)
12
S4
W2: windshear
W1: cross-wind -10 m/s
basic scenario Si
A flight situation scenario is depicted as a directed graph. It defines logic and content of flight. Scenario graph is clear and concise formal description of a flight situation. Basic scenario examples S1, , S6 are structurally close. They can be modified by adding new events/processes or by modifying existing ones.
Design Field of Operational Hypotheses

INTELONICS
Elevator deflection for rotation Rotation airspeed
Wheels - runway surface adhesion factor

2
3 2
VR
e
10
4
Wyg
4 3
Cross wind velocity Flaps-up start altitude Commanded flight path angle (2nd phase of climb, level flight or descent)
HFL 6
6
Longitudinal C.G.
1 xCG
7 8 9 11
G2 8
14 13
Left-hand engine failure at VEF
13
LHE VEF
G1 7
12
Commanded flight path angle (initial climb)
Engine out indicated airspeed
12
kP
10
G 11
Commanded bank angle
factors in
kW
Intensity of wind shear
Legend:
13
Engines power rating at takeoff

Cross wind velocity
- operational hypothesis
Wyg 5
- operational
factor
independent - link between dependent
Normally, a single operational factor is not critically dangerous. More important and much more difficult to learn the effects of multi-factor combinations on flight safety. These multi-factor combinations are called operational hypotheses.
8
INTELONICS
Composition of Situation Scenario (S) and Operational Hypothesis () is A Situational Tree (S)
Situational trees branches (flight paths) stand for what-if derivative (non-standard) situation scenarios. All branches are color coded using integral safety spectra colors.
Safety Palette. Fuzzy Constraint

INTELONICS
Safety Palette
green (norm), G yellow/ amber (attention), Y red (danger), R black (catastrophe), B grey/white (uncertainty), W
Color is natural and, perhaps, the most effective and economic medium for communicating safety-related information to/ from an operator (a pilot or automaton).
C(VFL.D.) Fuzzy Constraint (Example)

Legend: c, d characteristic points of the carrier of fuzzy set-constraint C, C(x) L.A. Zadeh membership function
C: permitted flaps-down flying IAS
0
green
390 yellow
d
410
470 red black
VFL.D.
[km/h]
Operational constraints under multi-factor flight conditions are not known precisely. They are inherently fuzzy. The notion of fuzzy constraint (by L.A. Zadeh) and the notion of safety palette are employed for approximate measurement of the compatibility of current (i.e. measured at time instants t) system states with operational constraints for key system variables (monitored flight parameters).
Partial and Integral Safety Spectra

INTELONICS
IAS (F = 0, airborne) IAS (F > 0, airborne) Sideslip Load_factor East_rate (groundroll) East (groundroll) North (groundroll) Bank (airborne) Bank (groundroll) Pitch (airborne) Pitch (groundroll) Vert_rate (airborne) AoA (F = 0) AoA (F > 0) Wheels (airborne) Wheels (groundroll) Elevator (airborne) Elevator (groundroll) Aileron Rudder Integral spectrum
Monitored variables/ constraints
1 2 ... k ... 20
time, s Partial flight safety spectra
Legend: k partial safety spectrum for variable xk, k = 1, , p; p total number of monitored constraints/ variables, p = 20. integral safety spectrum; t flight time; i color from safety palette, i {B (black), R (red), Y (yellow), G (green),}; < colder than operation for comparing two safety colors; max operation of selecting the hottest color at time instant t; || - operation of safety colors concatenation in ; [t*; t*] examined flight time interval; spectrum construction time step. - green (norm), G - yellow (attention), Y - red (danger), R - black (catastrophe), B - gray/white (uncertainty), W
Integral Safety Spectrum Calculation Algorithm:

(t) (t[t*;t*]) ((xk(t)) ((xk(t)){W, G, Y, R, B, } (W < G < Y < R < B)) ((t) = max (xk(t)), k = 1, , p) ((t) = (t*) || (t*+) || (t*+2) || || (t*))
For each flight situation from the situational tree, safety levels are measured for all monitored variables xk at all recorded time instants. As a result, for each situation from the tree, a family of Partial Safety Spectra k, k = 1, , p, and an Integral Safety Spectrum are obtained. The integral safety spectrum is a color-coded time-history of violation and restoration of monitored fuzzy constrains during a flight situation.
Safety Classification Categories

INTELONICS
Flight Safety Category Color Code I Name Safe Conditionally Safe a Conditionally Safe b Potentially Unsafe Dangerous (Prohibited) Catastrophic (Chain Reaction)
Situation Classification Criterion The system state resides mainly inside the 'green' zone. As a maximum, the system state may stay, for a short time, in close proximity to operational constraints, i.e. inside the yellow zone, but must leave it by the end of the flight situation As a maximum, the system state may stay for a medium time in close proximity to operational constraints, i.e. inside the yellow zone As a maximum, the system state may stay for a long time in close proximity to operational constraints, i.e. inside the yellow zone As a maximum, the system state may violate operational constraints, i.e. enter the red zone, for a short or medium time, but must leave it by the end of the situation As a maximum, the system state may stay beyond operational constraints, i.e. inside the red zone, for a long time or till the end of the flight situation There is at least one (for a short time) occurrence of a black violation of any operational constraint
II-a II-b III
IV V
One more level of flight safety knowledge generalization is introduced. The goal is to measure the vehicles safety performance in a flight situation as a whole. With this aim, a generalized safety ruler consisting of five Safety Classification Categories I, , V is employed. Why five? It is because experts cannot reliably recognize and use more than 5-10 gradations of a complex, difficult-to-formalize systemlevel property (e.g.: Cooper-Harper scale). New light green (salad green) and orange colors have been added to the existing Safety Palette in order to denote interim Categories II-a and III, respectively.
INTELONICS
Safety Window for Situational Tree S111: Takeoff. Errors of Selecting Commanded Flight Path nd Bank Angles in Climb
Flight Safety Window (FSW)
A
Safety Chances Distribution Pie Chart

B
Let us map safety classification levels (categories) obtained for all situations for tree S111 onto a two-factor plane. This gives a Flight Safety [Performance] Window (FSW). In FSW above, cell C is located at column A - row B crossing. This cell depicts safety status of one flight pathbranch from the tree. It is a non-standard situation with values of 30o and 14o of factors 7 and 11 in S1. This cell is painted using the situations Flight Safety Category color (orange). The FSW has a dangerous corner (upper-left). Rapid transition (3) from safe (salad green) to dangerous (red) zone is possible (Cat. II-a IV), bypassing interim zones (II-b, III). Control at such corners therefore requires enhanced 2007 Intelonics Ltd. attention.
Category
nj 37 8 29 1 55 0 130
j, % 28 6 22 1 43 0 100
13
I II-a II-b III IV V nj, j | S111
Flight Safety Topology Map

INTELONICS
Operational factor 1
3 5 2 3 4
Transitions 6 must be known and prevented!
2 1
Flight Safety Categories
I II-a II-b
Transitions 3 must be known and controlled!
III IV V
6 1 4
3 2
Operational factor 2
In general, the following characteristic objects can be defined within Flight Safety Topology Map:
1 Abyss (catastrophe) 2 Hill (danger) 3 Slope (reversible state transitions)

4 Valley (standard safety, norm) 5 Lake (maximum safety, optimum) 6 Precipice (abrupt, irreversible state transitions, chain reaction)
14
INTELONICS
S412: Normal Takeoff. Strong Wind Shear. Errors of Selecting Commanded Flight Path and Bank Angles in Climb
Flight Safety Window Safety Chances Distribution Pie Chart
6
j j, % 13 15 15 1 46 10 100
15
This safety topology corresponds to the tree S412 obtained under strong wind shear conditions. At small flight path angles G1 and any bank angles G it reveals a stable catastrophic abyss (a black strip in the bottom) and precipice type transitions (6). It means that attempts of climbing at small commanded flight path angles (1o 2o) will inevitably lead the vehicle to a fatal outcome.
Category
nj 17 19 19 1 61 13 130
I II-a II-b III IV V j n , j | S412
INTELONICS
Real-Time Safety Knowledge Map (Dynamic Safety Window)

t = t0: benign weather forecast
Presented is a time-history of safety windows and safety chances distribution pie charts that correspond to a hypothetical complex flight situation domain - a union of three compositions S4(11+12+13): Normal takeoff. Possible variations of wind-shear intensity, errors/ variations in maintaining commanded flight path and bank angles during initial climb.
The concept of dynamic safety window is based on use of a forest of situational trees. Provided that key operational factors are measurable on board the vehicle in real time, a dynamic safety window can be used as a medium for coherent monitoring of tactical goals and constraints of flight under uncertainty.
t0 t = t1: strong wind-shear warning t1 t = t2: very strong wind-shear warning t2

time
Safety chances distribution pie charts are expedient to use in UAV safety indicators to monitor current state and predict the system safety chances dynamics under anticipated operational conditions during flight. Note that in this particular example, the share of red and black scenario options increases at the expense of reducing the share of safer outcomes.
16
INTELONICS
Legend: to current flight time, t* prediction start time, t* prediction stop time, = (t* to) decisionmaking delay, t = (t* t*) prediction time range (depth of tree-based multi-factor domain exploration)
Situational Trees for Short-Term Prediction of Flight Safety
safety prediction sub-tree
t*
multi-factor situation domain exploration cone (future-looking knowledge radar)
t t0 t*
Situational tree construction and tree-based safety prediction (a what-if analysis) methodology accounts for both physics and logic of multi-factor flight situation domain.
INTELONICS
S614: Low-Altitude Level Flight. Errors Of Selecting Commanded Flight Path And Bank Angles
Flight Safety Window Safety Chances Distribution Pie Chart
Category
nj 77 12 35 4 82 37 247
j, % 33 5 14 2 33 15 100
18
This Safety Window has two catastrophically dangerous corners (6) corresponding to (G1, G) (-10o-12o, |37.5o45o|). Sharp transition (3) of states from safe (salad green) to dangerous (red) zone is also possible in the left upper corner (Cat. II-aIV), bypassing interim zones (Cat. II-b, III).
I II-a II-b III IV V j n , j | S614
INTELONICS
S614: Low-Altitude Level Flight In the Presence of Urban Obstacles

S0 S
tower building type obstacle (top view) S614 | t13
t14 t15 t16 t17 t18 t19
30 o 45 o
t0 t1 t2 t3 t4 t5 t6 t7 t8 t9 t10 t11 t12 t13
S614 | t19
-15 o -30 o -45 o 30 o 0 15 o
Scenario time lines
S
15 o 0
Legend: S0, S, S - scenario segments, S0 obstacle approach, S - imminent collision, S collision avoidance, S614 | ti - tree location at ti S0 S - terrorist-/ fool-type control, S0 S AI-based self-preservation control. yellow and red zones of obstacles fuzzy constraint S614 | t1 IAS 320360 km/h
24 20 16 12 8 4 0 -4 -8 -12
19 18 13
-30 o
45 o
-15 o
17 12 11 16 15 9 14 8 7
-15o 6 -30o
24 20 16 12 8 4 0 -4 -8 -12
10
S614 | t1
0 15o
-45 o
5 4 3
-45
o
30o
2 1 0 -1
45o
S614 | t13 H 200 400 m tower building type obstacle (side view)
4 5 6 7 8 9 10 11 12 13
-1
Note: not to scale
19
S0: Obstacle Approach (t0)

INTELONICS
Safety Chances Distribution Safety Window

100 80 j, % 60 40 20 0 -1 0 1
-15
The vehicle is approaching the obstacle a tower building at G2 = 0 (commanded flight path angle) and G= -15o (commanded bank angle). No threat is observed in the safety window at t0.
2 3 4 5 6 7 8 9 10 11 12 13
0
13 12 11
10
current tactical goal-cell

24 20 16 12 8 4 0 -4 -8 -12
7 6
-15o 5 -30o 0 15o
4 3
30o
-45o
2 1
45o
Note: not to scale

-1 0 1 2 3 4 5 6 7 8 9 10 11 12 13
0 -1
i
20

INTELONICS

j, % 100 80 60 40 20 0 -1 0 1 2
A fuzzified safety window state at t1 is shown. The white rectangular in the window is a current tactical goal-cell (G2/ G) = (0/ -15o). Still no threat is observed in the safety window.
3 4 5 6 7 8 9 10 11 12 13
13
12
11
10

24 20 16 12 8 4 0 -4 -8 -12
7
-15o 6
15o
5
-30o
30o
4 3
45o
-45o
2 1
Note: not to scale

-1 0 1 2 3 4 5 6 7 8 9 10 11 12 13
0 -1
i
21

INTELONICS

j, % 100 80 60 40 20 0 -1 0 1 2 3 4 5 6 7 8 9 10 11 12 13
Some branches of the prediction sub-tree hit the yellow zone of the obstacles fuzzy constraint.
13
12
11
10
15o 30o
8 -15 7 6
o -30
24 20 16 12 8 4 0 -4 -8 -12
5 4
45o
-45
3 2 1 0 -1
-1
10 11 12 13
i
22

INTELONICS

j, % 100 80 60 40 20 0 -1 0 1 2 3 4 5 6 7 8 9 10 11 12 13
The yellow zone in the safety window is expanding at the expense of the green zone, which is shrinking respectively
13
12
11
10
15
9o -30
15
7 6 5
30
24 20 16 12 8 4 0 -4 -8 -12
45
-45
4 3 2 1 0 -1
-1
10 11 12 13
i
23

INTELONICS

j, % 100 80 60 40 20 0 -1 0 1 2 3 4 5 6 7 8 9 10 11 12 13
Now the yellow zone occupies almost all the lefthand half of the safety window.
13
12
15
11
10
-15
30
o -30
7 6 5
45
24 20 16 12 8 4 0 -4 -8 -12
-45
4 3 2 1 0 -1
-1
10 11 12 13
i
24

INTELONICS

j, % 100 80 60 40 20 0 -1 0 1 2 3 4 5
Some branches of the prediction sub-tree hit the red zone of the obstacles fuzzy constraint. The yellow and now red zones in the safety window are expanding.
6 7 8 9 10 11 12 13
15
13
12
11
o -15
30
10
o
-30
24 20 16 12 8 4 0 -4 -8 -12
45
7 6 5 4 3 2 1 0 -1
-45
-1
10 11 12 13
i
25

INTELONICS

j, % 100 80 60 40 20 0 -1 0 1 2 3 4 5 6 7 8 9 10 11 12 13
The red zone is spreading through the left-hand part of the safety window.
15
30
13
15 12 11 0 -3
o
10
45
7 6 5 4 3 2 1 0 -1
24 20 16 12 8 4 0 -4 -8 -12
5 -4
-1
10 11 12 13
i
26

INTELONICS

100 80 j, % 60 40 20 0 -1 0 1 2 3 4 5 6
-15
A black (imminent collision) zone appears in the safety window. NB: It overlaps with the current tactical goalcell.
7 8 9 10 11 12 13
0
1 13 12 0 -3
o
0 5
o
15
30
45
11
10

24 20 16 12 8 4 0 -4 -8 -12
7 6 5 4 3 2 1 0 -1
5 -4
-1
10 11 12 13
i
27
S: Imminent Collision (t8)

INTELONICS

j, % 100 80 60 40 20 0 -1 0 1 2 3 4 5 6
o
The black zone is expanding now at the expense of shrinking both the green and yellow zones.
7 8 9 10 11 12 13
15
30
-1
13
o
5o
12
-3 0
11
10
5o
45
7 6 5 4 3 2 1 0 -1
The fact that the goal-cell still remains in the black zone says that the aircraft is on a collision course.
-1 0 1 2 3 4 5 6 7 8 9 10 11 12 13
24 20 16 12 8 4 0 -4 -8 -12
i
28
-4

INTELONICS

j, % 100 80 60 40 20 0 -1 0 1 2 3 4 5
15 o
7 8
30 o
10 11 12 13
-1
5o
13
0o -3
12
11
10
5o
45 o
7 6 5 4 3 2 1 0 -1
No safe (green) flightpath-branch alternatives are available. The share of black scenarios increases. The share of red scenarios remains the same.
-1 0 1 2 3 4 5 6 7 8 9 10 11 12 13
24 20 16 12 8 4 0 -4 -8 -12
i
29
-4

INTELONICS

j, % 100 80 60 40 20 0 -1 0 1 2 3
15 o
5
30 o
7 8
45 o
10 11 12 13
-15 o
13
-30 o
12
11
10
-45 o
7 6 5 4 3 2 1 0 -1
Almost no yellow (conditionally safe) branch options are left in the safety window to use for recovery. A catastrophic trend in the situation continues to build-up steadily.
-1 0 1 2 3 4 5 6 7 8 9 10 11 12 13
24 20 16 12 8 4 0 -4 -8 -12
i
30

INTELONICS

j, % 100 80 60 40 20 0 -1 0 1 2
15 o
4
30 o
6
45 o
7 8
10 11 12 13
-1 5 o
13
-3 0 o
12
11
10
-45 o
7 6 5 4 3 2 1 0 -1
The black zone covers more than 60% of the safety window area, and the rest represents red (dangerous) scenarios, i.e. the flight paths in a close vicinity of the obstacle
24 20 16 12 8 4 0 -4 -8 -12
-1
10 11 12 13
i
31

INTELONICS

j, % 100 80 60 40 20 0 -1 0 1
15 o
3
30 o
5
45 o
7 8
10 11 12 13
-15 o
13
-30 o
12
11
10
-45 o
7 6 5 4 3 2 1 0 -1
The black zone occupies more than 80% of the safety window area. A catastrophic trend accelerates, and the chances of collision are very high.
24 20 16 12 8 4 0 -4 -8 -12
-1
10 11 12 13
i
32

INTELONICS

j, % 100 80 60 40 20 0 -1 0
15 o
1
30 o
4
45 o
7 8
10 11 12 13
-15 o
13
-30 o
12
11
10
-45 o
7 6 5 4 3 2 1 0 -1
The black zone now occupies the entire safety windows area (100%). This means that the collision is unavoidable
24 20 16 12 8 4 0 -4 -8 -12
-1
10 11 12 13
i
33
INTELONICS
Kazimir Malevichs The Black Square Painting and 9/11
The safety window state just before collision point (S | t13), perhaps, helps better understand the meaning of Kazimir Malevichs painting The Black Square - The fatal end is imminent. And there is no chance left to remedy the situation
K. Malevich. The Black Square (1913)

Last Chance for Recovery Point (t t7)

INTELONICS
Safety Window
Safety Chances Distribution

100 80 j, % 60 40 20 0 -1 0 1 2 3 4 5 6 7 8 9 10 11 12 13
-15
0
0 -1 5
o
15
30
13
12
30
o
45
11
10
However, the last chance for recovery point (t) does exist, and it must be assigned to t7. This is marked by the system state when the new black zone (induced by the obstacle) in the safety window first time overlaps with the current tactical goal-cell of the operators flight control.
7 6 5 4 3 2 1
5 -4
Note: not to scale

0 -1 35
INTELONICS
Safety Window State at Last Chance For Recovery Point (t t7): S0 S

Commanded bank angle, G
1
Commanded flight path angle, G2
7 (G2, G) = (2o, 7.5o) 8
Legend: 1 zone of -secured non-catastrophic scenarios; 2, 3 zones (islands) of remaining safe/conditionally safe scenarios; 5, 8 C.G. locations for left- and right-hand islands of remaining safe/conditionally safe scenarios; 4, 7 old (catastrophe-prone) and new (safety restoring) cells of the commanded flight path and bank angles, 6 required shift of the tactical flight goal-cell in the safety window.
INTELONICS
Safety Window
S0 S: Self-Preservation Automatic Decision Making at t7 t

-15 30
Based on results of safety topology analysis at t7, a self-preservation decision must be made - the current tactical control goal is shifted from the old (black, collision-prone) cell, (G2/G) = (0/-15o), to a new (green, safe) cell, (G2/G) = (6o/+30o), located in the right-hand safety island of the window.
6 0
0 1 13 12 -3 0
o o
15
30
45
old tactical goal-cell 100 80 j, % 60 40 20 0 -1 0 1 2 3 4
new tactical goal-cell
11
10
7 6 5 4 3 2 1
-4
7 14 15 16 17 18 19
0 -1 37
S: Collision Avoidance (t14)

INTELONICS
Safety Window
The black zone in the safety window is still expanding (due to vehicle dynamics lag). However, the red zone begins to shrink, and the yellow zone size remains unchanged. The commanded (tactical goal) cell is now located outside the danger and catastrophe-prone zones.
0 -15
o
15
3 190
18 13
o -3 0
17 11 16 15 9 14 8 7 6
45
12
o -4 5
10

100 80 j, % 60 40 20 0 -1 0 1 2 3 4 5 6 7 14 15 16 17 18 19
5 4 3 2 1
0 -1 38

INTELONICS
Safety Window
A positive (recovery) safety trend begins to develop. The yellow zone is expanding, and the red zone is shrinking in the safety window.
0 -1 5
o
15
30
19
-3 0
o
18 17 16 15 9 14 8 7
45
13
-4 5
12 o
11
10

100 80 j, % 60 40 20 0 -1 0 1 2 3 4 5 6 7 14 15 16 17 18 19
6 5 4 3 2 1
0 -1 39

INTELONICS
Safety Window
Positive safety trend remains steady. The yellow zone is expanding, and the red zone is shrinking. The black zone begins to shrink as well
0 -15o
15o 30o
-30
19
-45o
18 17 11 16 15 9 14 8 7 6
45o
13
12

100 80 j, % 60 40 20 0 -1 0 1 2 3 4 5 6 7 14 15 16 17 18 19
10
5 4 3 2 1
0 -1 40

INTELONICS
Safety Window
Black zone induced by the obstacle is about to disappear. Positive safety trend is now irreversible. The yellow zone continues to expand.
-15o -30o
15o
30o
-45o
19 18 17
45o
13
12
11

100 80 j, % 60 40 20 0 -1 0 1 2 3 4 5 6 7 14 15 16 17 18 19
10
16 15 9 14 8 7 6 5 4 3 2 1
0 -1 41

INTELONICS
Safety Window
The safety window is about to resume its initial state (vehicles performance only) as the obstacle (a tower type building) has been safely avoided.
-15 o -30 o
0 15 o
-45 o
30 o
19 18 13 17 12 11 16 15 9 14 8 7 6 5
45 o

100 80 j, % 60 40 20 0 -1 0 1 2 3 4 5 6 7 14 15 16 17 18 19
10
4 3 2 1
0 -1 42

INTELONICS
Safety Window
The knowledge-centered process of flight safety restoration is complete by now.
-15 o -3 0 o -45 o
0 15 o
30 o
19 18 13 17 12 11 16 15 9 14 8 7 6 5 4
45 o
10

100 80 j, % 60 40 20 0 -1 0 1 2 3 4 5 6 7 14 15 16 17 18 19
3 2 1
0 -1 43
INTELONICS
Birds Eye View Of Dynamic Safety Window Tree: Catastrophic and Recovery Scenarios
t19 t18 t17 t16 t15 t14
last chance for recovery point (fate switch)
This is a safety window time-history tree. It provides a systematic birds eye view level picture of two alternative scenarios of aircraft flight control in the presence of an urban type obstacle. Such obstacles can be a part of a multi-factor flight situation domain-neighborhood of the current situation.
t13 t12 t11 S t10 t9
S S
t8
Legend:
t7 t6 t5 t4 t3 t2 t1 t0
Scenario segments: S0 obstacle approach S imminent collision S AI based collision avoidance Scenario time lines: {t0, t1, ..., t7} S0 {t8, ..., t13} S {t14, ..., t19} S Key time instants: t7 last chance for recovery t13 just before impact t19 safety restoration complete
S0
S0
44
INTELONICS
Safety Chances Distribution Time-History for Two Control Tactics

(2) AI based self-preservation control
G
100 80
(1) terrorist-/ fool-type control

A
100 80
D E
D H I
K L
j, %
60 40 20 0 -1 0 1 2 3 4 5 6 7 8 9 10 11 12 13
j, %
i
60 40 20 0 -1 0 1 2 3 4 5 6 7 14 15 16 17 18 19
S0
S0
last chance for recovery point
Legend: A, B, , L - characteristic states of the aircraft safety dynamics; j flight safety chances at j level, j{I, II-a, II-b, III, IV, V}; ti time instants, i{-1, 0, 1, , 13} i{-1, 0, 1, , 7, 14, 15, .., 19}. I II-a II-b III IV V safety classification categories and colors Characteristic states {A, B, C, , L} of the vehicles safety dynamics and their recognition criteria are expedient to use in the automatic or manual recovery decision-making process in emergency situations under uncertainty. In accordance with the self-preservation imperative for a civil aircraft, flight control authority in a life-threatening situation must be dynamically assigned/transferred to a most competent agent.
45
Conclusions
INTELONICS
1. Generalized knowledge-centered methodology has been developed for UAV flight safety prediction and protection in multifactor situations near operational constraints. 2. Methods advantages are: use of integrated conceptual framework, simple realtime calculations, open memory-based knowledge system, situation-independent decision-making algorithm, exploration of situation what-if neighborhood tree for short-term flight path probing, use of birds eye view topology maps for flight safety status monitoring and automatic recovery in emergencies. 3. However, prerequisites for successful implementation of developed methodology are: availability of vehicles validated parametric definition database, and onboard integrated sensor suit capable of detecting potentially dangerous physical/ virtual obstacles inside vehicles safety ellipsoid/cone. 4. Potential application areas are as follows: design of affordable, yet expert pilot level AI safety protection systems based on self-preservation imperative for unmanned/ manned air vehicles to prevent key accident/ incident scenarios such as LOC, CFIT, pilot error, hardware failure, mid-air collision, and 9/11 design of adaptive mission control and autonomous collision avoidance systems (integrated with C.Reynolds swarming model, ethology principles, etc.) for heterogeneous multivehicle clusters and free-flight operations. 2007 Intelonics Ltd.
46
Acknowledgements
INTELONICS
The author wishes to thank the following individuals and companies:
Bernd Chudoba, Ph.D.

The University of Texas at Arlington USA
- for sharing innovative ideas and for cooperation in the area of virtual flight testing of a/c for detecting anomalous flight situations in early design. - for multi-aspect support, beginning from 90s during Ph.D studies at Cranfield University, and up to these days. - for offering excellent air travel package (Moscow-Kln-Moscow) to attend the Conference.
Jean-Pierre Cachelet, Ph.D.

AIRBUS France
Till Bunse, Ph.D

Germanwings GmbH Germany
47
Thank You
INTELONICS
Questions, please Questions, please
48
Selected Publications
INTELONICS
Burdun, I.Y., Prediction of Aircraft Safety Performance in Complex Flight Situations (Paper 2003-01-2988), Proc. of 2003 Situations 2003- 01Advances in Aviation Safety Conference, September 8 12, 2003, Montreal, Canada, SAE, 2003, 18 pp. 2003, Montreal, 2. Burdun, I.Y., Studying Physics and Logics of Complex Flight Situation Domains by Means of VATES Modeling and Simulation Tool, Proc. of 2nd Science and Technology Conference Flight Simulation Technologies and Pilot Training: New Tool Approaches and Goals, TsAGI, 24 25 April 2003, Zhoukovsky, Moscow Region, TsAGI, 2003, 11 pp. (in Russian). Goals 3. Burdun, I.Y., Parfentyev, O.M., Fuzzy Situational Tree-Networks for Intelligent Flight Support, Int. Journal of Engineering TreeSupport Applications of Artificial Intelligence (EAAI), 12 (1999), pp. 523 541. EAAI), 4. Schrage, D.P., Calise, A.J., Burdun, I.Y., Pritchett, A., and Rysdyk, R.T., An Integrated Knowledge-Based Approach to KnowledgeImproving Aircraft Safety, White Paper, School of Aerospace Engineering, Georgia Tech, USA, Oct. 1998, 20 pp. Safety Paper, 5. Burdun, I.Y., The Intelligent Situational Awareness And Forecasting Environment (The S.A.F.E. Concept): A Case Study Environment Study (Paper 981223), Proc. of 1998 Advances in Flight Safety Conference and Exhibition, April 6-8, 1998, Daytona Beach, FL, Exhibition, 6USA (P-321), SAE, 1998, pp.131 144. (P6. Burdun, I.Y., and Parfentyev, O.M., Analysis of Aerobatic Flight Safety Using Autonomous Modeling and Simulation and Simulation (Paper 2000-01-2100), Proc. of the 2000 Advances in Aviation Safety Conference, April 11-13, 2000, Daytona Beach, FL, 2000- 0111USA (P-355), SAE, 2000, pp. 75 92. (P- 355), 7. Burdun, I.Y., Parfentyev, O.M., AI Knowledge Model for Self-Organizing Conflict Prevention/Resolution in Close Free-Flight SelfFreeAir Space, Proc. Of IEEE Aerospace Applications Conference, Snowmass, Colorado, March 6-13, vol. 2, 1999, USA, ado, Space Color 6 1999, USA, IEEE, 1999, pp. 409 428. 8. Burdun, I.Y., and Burdun, E.I., VATES Virtual Autonomous Test and Evaluation Simulator (Version 7 Professional), Simulator Users Manual, 2000, 155 pp. User 9. Burdun, I.Y.,Prediction of Aircraft Flight Safety Performance in Complex Situations using Results of Aerodynamics I.Y., Situations Research and Flight Modeling and Simulation, Proc. of the Jubilee Conference 60-th Anniversary of SibNIA Aircraft Simulation 60Aerodynamics and Strength Research Divisions, 15 17June 2004, SibNIA, Novosibirsk, 2004, pp. 45 57 (in Russian). Divisions 2004, Russian). 10. Burdun, I.Y., Theory, Implementation and Proof-of-Concept Study of Flight Safety Topology Knowledge Maps for Accident Proof- ofTopology Prediction and Prevention, Proc. of EWHSFF-2005 Conference, Chinese Aeronautical Establishment and Beihang Prevention EWHSFFUniversity, Beijing, P.R. China, 19-22 October 2005, PRC, pp. 494 502. 19 2005, 11. Burdun, I.Y., C.Reynolds Model of Motion Self-Organization and Some Issues of Application of Highly-Maneuverable SelfHighlyHighly-Autonomous Unmanned Air Vehicles, Proc.of XVI TsAGI Workshop Aerodynamics of Aircraft, 3 4 March 2005, HighlyVehicles Aircraft 2005, Volodarsky, Moscow Region, TsAGI, 2005, pp. 28 29 (in Russian). Volodarsky, Region, pp. Russian).
1.

Burdun AUVSI-2007

Hochgeladen von

Dokumentinformationen

Originalbeschreibung:

Copyright

Verfügbare Formate

Dieses Dokument teilen

Dokument teilen oder einbetten

Freigabeoptionen

Stufen Sie dieses Dokument als nützlich ein?

Sind diese Inhalte unangemessen?

Copyright:

Verfügbare Formate

Burdun AUVSI-2007

Hochgeladen von

Copyright:

Verfügbare Formate

AUVSI's Unmanned Systems Europe 2007 8-9 May 2007, Hilton Cologne, Kln, Germany

UAV "Built-in" Safety Protection: "Built-in" A Knowledge-Centered Approach Knowledge-Centered

Natural Tree Analogy of Pilots Situational Knowledge Base

Strengths to model by means of AI in UAVs

Defects to back up by means of AI in UAVs

Fractal Tree Model Of Pilots Situational Expertise Growth In Long-Term Memory

desirable maturity levels of AI knowledge for flight safety protection in UAVs

Micro- and Macro- Structural Models Of Complex Flight Situation Domain

developed generalized knowledge model of a complex flight situation domain.

Basic Flight Situation Scenario (Examples)

Joint Graph of Basic Scenarios (Example)

E6: altitude 10.7 m

F1: left-hand engine out

E7: altitude 120 m

T1: maintain ground-roll path along runways centerline

P2: elevator up for rotation

T3: maintain commanded flight path angle G1 (1st phase of climb)

W1: cross-wind -10 m/s

Design Field of Operational Hypotheses

Elevator deflection for rotation Rotation airspeed

Wheels - runway surface adhesion factor

Left-hand engine failure at VEF

Commanded flight path angle (initial climb)

Engine out indicated airspeed

Intensity of wind shear

Engines power rating at takeoff

independent - link between dependent

Safety Palette. Fuzzy Constraint

C(VFL.D.) Fuzzy Constraint (Example)

C: permitted flaps-down flying IAS

470 red black

Partial and Integral Safety Spectra

Monitored variables/ constraints

Integral Safety Spectrum Calculation Algorithm:

Safety Classification Categories

II-a II-b III

Safety Chances Distribution Pie Chart

I II-a II-b III IV V nj, j | S111

Flight Safety Topology Map

Transitions 3 must be known and controlled!

1 Abyss (catastrophe) 2 Hill (danger) 3 Slope (reversible state transitions)

I II-a II-b III IV V j n , j | S412

2007 Intelonics Ltd.

Real-Time Safety Knowledge Map (Dynamic Safety Window)

t0 t = t1: strong wind-shear warning t1 t = t2: very strong wind-shear warning t2

2007 Intelonics Ltd.

Situational Trees for Short-Term Prediction of Flight Safety

safety prediction sub-tree

I II-a II-b III IV V j n , j | S614

S614: Low-Altitude Level Flight In the Presence of Urban Obstacles

t0 t1 t2 t3 t4 t5 t6 t7 t8 t9 t10 t11 t12 t13

Scenario time lines

Note: not to scale

S0: Obstacle Approach (t0)

Safety Chances Distribution Safety Window

current tactical goal-cell

Note: not to scale

2007 Intelonics Ltd.

S0: Obstacle Approach (t1)

Safety Chances Distribution Safety Window

current tactical goal-cell

Note: not to scale

2007 Intelonics Ltd.