Probabilistic Risk Analysis Procedure for Aircraft Overruns · Probabilistic Risk Analysis...
Transcript of Probabilistic Risk Analysis Procedure for Aircraft Overruns · Probabilistic Risk Analysis...
Probabilistic Risk Analysis Procedure for Aircraft Overruns
P. Trucco, M. De Ambroggi, M.G. Gratton Politecnico di Milano – Italy
M.C. Leva Trinity College Dublin - Ireland
3rd Symposium on Games and Decisions in Reliability and Risk (GDRR) Kinsale, Ireland - 8th–10th July 2013
Three possible scenarios:
• Landing Overrun,
• Takeoff Overrun
• Landing Undershoot
Runway Excursion: when the aircraft departs from the runway limits during the operation.
2 INTRODUCTION
GDRR 2013 - Kinsale, Ireland - 8th–10th July 2013
Runway-related accidents represent a relevant fraction of the total number of recorded accidents in airport operations
3
Other 70%
Excursions 29%
Runway incursions
1%
Runway-related
accidents 30%
[Flight Safety Foundation, 2009]
INTRODUCTION
GDRR 2013 - Kinsale, Ireland - 8th–10th July 2013
Overrun Probability Models: a comparison
Authors Year # of factors in the model
Input/Output variables
Based on international data (accident / NOD)
NOD
Eddowes et al.
2001 1 Deterministic No No
Kirkland et al.
2004 1 Deterministic No No
Wong et al. 2008 17 Deterministic No Yes
Hall et al. 2008 14 Deterministic Yes Yes
4 STATE OF THE ART REVIEW
FAA - Airport Cooperative Research Program (ACRP) - Report 3 “Analysis of Aircraft Overruns and Undershoots for Runway Safety Areas”
GDRR 2013 - Kinsale, Ireland - 8th–10th July 2013
State of the art review and
model selection
Event probability
and location model
Event-Consequence
model
Topological model
“Risk Map” Test case
application
Objective
5 AIM AND SCOPE OF THE STUDY
Development of a Probabilistic Risk Analysis Procedure to design a topological “risk map” of overrun events based on the statistical characterisation of airport operations and context conditions
Methodology
ACRP Model
• Ops statistical characterisation • Monte Carlo
• Kinetic Energy as Severity index
GDRR 2013 - Kinsale, Ireland - 8th–10th July 2013
The ACRP Model (Hall, 2008) • Made of three modules:
• Accident Probability Model: returns the overrun probability value for a given operation
• Longitudinal Location Model: returns the conditional probability that the longitudinal excursion is greater than a certain value starting from the runway end
• Lateral Location Model: returns the conditional probability that the lateral excursion is greater than a certain value starting from the runway centre axis
• Based on international accident data and NOD in countries with accident rates similar to US. • Input data are normalised by means of a deceleration model
developed by Kirkland et al. (2003)
6 THE ACRP MODEL
GDRR 2013 - Kinsale, Ireland - 8th–10th July 2013
ACRP Accident Probability Model
7
b(Landing Overrun)=-15.456+0.551(Heavy aircraft)-2.113(Commuter Aircraft)-1.064(Medium Aircraft)-0.876(Small Aircraft)+0.445(Turboprop Aircraft)-0.856(Foreign Origin)+1.832(Ceiling Height<1000 ft)+1.639(Ceiling Height 1001-1500 ft)+2.428(Visibility<2 SM)+1.186(Visibility 2-4 SM)+1.741(Visibility 4-6 SM)+0.322(Visibility 6-8 SM)-0.532(Crosswind 2-5 kt)+1.566(Crosswind 5-12 kt)+1.518(Crosswind>12 kt)+0.986(Electrical Storm)+1.926(Icing Conditions)+1.499(Snow)-1.009(Temperature<5°C)-0.631(Temperature 5-5°C)+0.265(Temperature>25°C)+1.006(Non Hub Airport)+0.924(Significant Terrain)
INPUT: Categorical variables of relevant factors
P =
OUTPUT: Overrun Probability for a single movement
Limitation: available excess runway is not considered among the relevant factors
THE ACRP MODEL
GDRR 2013 - Kinsale, Ireland - 8th–10th July 2013
ACRP Longitudinal Location Model
8
0 0.2 0.4 0.6 0.8
1
0 500 1000 1500
Pro
babi
lity
(dis
tanc
e >
x)
Distance x from threshold [m]
Raw distances model for Landing Overruns
• Regression model:
• Separate models for Landing and Takeoff Overruns
• Lateral Location Model is very similar (over variable y)
THE ACRP MODEL
GDRR 2013 - Kinsale, Ireland - 8th–10th July 2013
Deceleration Model (Kirkland et al. 2003) • Relates the travelled distance (s) during an excursion to the initial
speed of the aircraft (u)
9
q = constant [ms] s = travelled distance [m] p = coefficient dependent on ground type:
Ground Type p Wet Grass/Dry Grass/Pavement -0.0185 Mud/Gravel -2.8065 Obstacles/Water -8.5365
0 200 400 600 800
0 500 1000 1500
Spe
ed [K
m/h
]
Travelled Distance [m]
THE ACRP MODEL
GDRR 2013 - Kinsale, Ireland - 8th–10th July 2013
10
Probabilistic topology of the event occurrence: Probability distribution of each single area to be involved in an overrun event [event/movement]
EVENT PROBABILITY AND LOCATION MODEL
GDRR 2013 - Kinsale, Ireland - 8th–10th July 2013
11
Probabilistic topology of the event severity: Iso Kinetic Energy Areas
[KJ/movement]
EVENT CONSEQUENCE MODEL
GDRR 2013 - Kinsale, Ireland - 8th–10th July 2013
Expected speed at a certain point beyond the runway end
12 EVENT CONSEQUENCE MODEL
GDRR 2013 - Kinsale, Ireland - 8th–10th July 2013
MONTE CARLO SIMULATION Expected kinetic energy at a certain point beyond the runway end
Probabilistic topology of the event severity: Iso Kinetic Energy Areas
[KJ/movement]
13
Probabilistic topology of the event severity: Iso Kinetic Energy Areas
[KJ/movement]
EVENT CONSEQUENCE MODEL
GDRR 2013 - Kinsale, Ireland - 8th–10th July 2013
Meteorological Data Traffic Data
• Source: ENAV • Format: METAR • Large sample
(sampling frequency of 30 min)
• Strong seasonality • Information on Icing
conditions is lacking • Large uncertainties due
to unstable factors (e.g. wind direction)
• Source: Airport Operator
• Format: records from database Businnes Objects ▫ about 40.000
movements/year
• Landing and takeoff weight information is lacking; ▫ substituted with
Maximum Takeoff Weight of Aircraft (MTOW)
14
Linate Airport (LIN) – Milan, Italy
TEST APPLICATION
GDRR 2013 - Kinsale, Ireland - 8th–10th July 2013
• Seasonality of meteorological data ▫ Chi-square and Friedman tests on single
parameters ▫ Four separated models have been developed
for each season
• Sensitivity Analysis ▫ To determine the most relevant factors under
different seasons (tornado plot)
• Correlations ▫ Linear approximation - Pearson’s coefficient
15 TEST APPLICATION
Linate Airport (LIN) – Milan, Italy
GDRR 2013 - Kinsale, Ireland - 8th–10th July 2013
• Major influencing factors: Icing, Snow, Crosswind
• Major differences between landing and takeoff critical conditions
• Relevant traffic conditions: Equipment Class and User Class
Sensitivity Analysis
16
1.00
E-0
6
2.00
E-0
6
3.00
E-0
6
4.00
E-0
6
5.00
E-0
6
6.00
E-0
6
7.00
E-0
6
8.00
E-0
6
Jet/Turbo / Factor D Local/Foreign Origin
Visibility / Factor Equipment Class / Fa
Electrical Storm / F Temperature / Factor
Ceiling / Factor Dis Snow / Factor Distri
Crosswind / Factor D Icing / Factor Distr
Mean of Accident Probability for WINTER LANDINGS
Winter Landings, Sensitivity Tornado
0.00
E+0
0
1.00
E-0
7
2.00
E-0
7
3.00
E-0
7
4.00
E-0
7
5.00
E-0
7
6.00
E-0
7
7.00
E-0
7
8.00
E-0
7
9.00
E-0
7
1.00
E-0
6
Visibility / Factor Temperature / Factor
Ceiling / Factor Dis User Class / Factor Fog / Factor Distrib
Equipment Class / Fa Crosswind / Factor D
Icing / Factor Distr Snow / Factor Distri
Mean of Accident Probability for WINTER TAKEOFFS
Winter Takeoffs, Sensitivity Tornado
DISCUSSION OF RESULTS
GDRR 2013 - Kinsale, Ireland - 8th–10th July 2013
Expected Probability for each single area to be involved in an overrun event [event/movement]
17
1,00E-‐06
1,00E-‐07
1,00E-‐08
1,00E-‐09
<1,00E-‐10
DISCUSSION OF RESULTS
GDRR 2013 - Kinsale, Ireland - 8th–10th July 2013
Event probability - Expected return time [years/event]
18
0-‐50 Years
50-‐100 Years
>100 Years
DISCUSSION OF RESULTS
GDRR 2013 - Kinsale, Ireland - 8th–10th July 2013
IKEA (Iso Kinetic Energy Areas): expected kinetic energy distribution [kJ/movement] beyond runway end
19
1,00E-‐02 1,00E-‐03 1,00E-‐04 1,00E-‐05 <1,00E-‐06
DISCUSSION OF RESULTS
GDRR 2013 - Kinsale, Ireland - 8th–10th July 2013
Advantages of the proposed approach
• First attempt to characterise an airport overrun risk − Probabilistic approach − Topological representation − Based on NOD on traffic and meteorological information
• Easy way of reporting results, understandable by non technical people (e.g. policy makers and population)
• Can be easily combined with structural vulnerability analyses of buildings and infrastructures to arrive at a full risk assessment of the airport area
• IKEA information can be directly used to design protection measures and physical barriers
20 CONCLUSIONS
GDRR 2013 - Kinsale, Ireland - 8th–10th July 2013
Opportunities for future developments • On the side of the Probability and Location Models:
▫ to include Available Excess Runway among the factors included in the ACRP model
• On the side of the Deceleration Model
▫ to enhance the model proposed by Kirkland et al. (2003), to take into consideration a larger variety of ground types and also the contribution of specific protective installations
• On the side of the Consequence Model
▫ Standardisation of basic structural analysis and characterisation of different airport facilities
▫ Modelling more complex scenarios (e.g. overrun + fire/explosion)
21 CONCLUSIONS
GDRR 2013 - Kinsale, Ireland - 8th–10th July 2013
Thank you !
Prof. Paolo Trucco, [email protected] @TruccoPaolo www.ssrm.polimi.it
3rd Symposium on Games and Decisions in Reliability and Risk (GDRR) Kinsale, Ireland - 8th–10th July 2013
Overrun Event Probability
23
• Fall and Winter are the most critical seasons: − Due to adverse
meteorological conditions
• Differences between Landing and Takeoff: − Means differ of about an
order of magnitude: - PLanding = 1.19x10-6 - PTakeoff = 2.81x10-7
− Variances differ of about one order of magnitude
DISCUSSION OF RESULTS
GDRR 2013 - Kinsale, Ireland - 8th–10th July 2013