ARTICLE IN PRESS

Journal of Air Transport Management 12 (2006) 267273

www.elsevier.com/locate/jairtraman

Risk assessment modeling in aviation safety management

Wen-Kuei Lee
Department of Industry and Business Management, The Open University of Kaohsiung, 436 Daye North Road, Siaogang, Kaohsiung 812, Taiwan

Abstract
Safety risk management is important in aviation. This paper develops a quantitative model for assessing aviation safety risk factors as
a means of increasing the effectiveness of safety risk management system by integrating the fuzzy linguistic scale method, failure mode,
effects and criticality analysis principle, and as low as reasonably practicable approach. The model is developed by evaluating all related
estimation factors based on their importance, how hazardous they are, their detectability, probability, criticality, and frequency. An
empirical study demonstrates the modeling process.
r 2006 Elsevier Ltd. All rights reserved.
Keywords: Risk assessment; Safety management; FMECA principle; ALARP approach

1. Introduction
Safety analysis of accidents is an important but
challenging issue in the civil aviation industry. Air
passenger transportation is growing, with annual increases
exceeding 5% forecast for the next 20 years. From a safety
perspective, this means that continuous improvement is
necessary to maintain high safety levels (Button et al.,
2004). During recent decades, the focus has been on
qualitative analysis or post-event studies of accidents.
Nevertheless, whether considering the qualitative/quantitative analysis or the post-event/pre-event approach, these
methods are generally based on either reactive or proactive
analysis (Lee and Chang, 2005b). The reactive approach is
a method of taking precautions following a loss, and its
efcacy in preventing air accidents is limited by its ex post
facto nature. Consequently, a before-the-fact diagnostic
and predictive method may be more useful for safety risk
management (SRM). In Taiwan, the Civil Aeronautics
Administration (CAA) has promulgated some fundamental
SRM measures for use by airlines and airports. It even
announced a reward system integrated with the allocation
of aviation resources, such as air routes, operational rights,
and ight frequency quotas, to monitor SRM performance.
The airlines and airports, however, lack an adequate

E-mail address: [email protected].

0969-6997/$ - see front matter r 2006 Elsevier Ltd. All rights reserved.
doi:10.1016/j.jairtraman.2006.07.007

method for accurately assessing the signicance of SRM

systems.
From the perspective of prevention, if risks can be
efciently diagnosed before serious failure occurs the
incident may be markedly reduced. The failure mode,
effects and criticality analysis (FMECA) principle is a
useful tool for considering risk. It has been extensively
applied in the national safety defense (US Department of
Defense, 1980). Subsequently, the FMECA was universally
used in high risk industries, for example nuclear energy,
chemical engineering, and petrochemical manufacturing.
However, in it for multiple estimation factors has a fatal
weaknessthe dilution phenomenon. To avoid this, Lee
and Chang (2005b) designed a unique method of formulating the combined measurement scores. To obtain a rapid
and convenient risk analysis model, the graphic analysis
integrated the as low as reasonably practicable (ALARP)
approach. ALARP evolved from the safety case concept
developed in the UK (Health and Safety Executive, 1992).
Through proper application of graphic analysis, the SRM
overseer can monitor individual risk factor without
constraints of time and place and immediately adopt the
preventive measures to avoid imminent accidents.
Finally, because of the intangible nature of judging
measurement scores of aviation risks for certain estimation
factors, such as importance and detectability, and to reect
the inherent subjectivity and imprecision of even expert
judgments, the fuzzy linguistic scale method (Buckley,

ARTICLE IN PRESS
268

W.-K. Lee / Journal of Air Transport Management 12 (2006) 267273

1984) can be used to derive fuzzy judgments. The fuzzy

membership functions can be deduced, and then the a-cut
technique is used to calculate the risk levels of the risk
factor.

unable to rise/fall, attrition, etc., are called failure modes.

Numerous random factors also affect risk factor failure,
for example articial operations, weather effects, mechanical faults, and organizational culture.

2. Methodology

2.2. Constructing risk assessment model

2.1. Nouns definition

The level of risk (LoR) index, an integrated concept

(Fig. 1), is deduced from the risk gradient (RG) and risk
magnitude (RM) indexes. The RG index comprises
endogenous estimation factors, and the RM index comprises exogenous estimation factors. To avoid the dilution
effect, this study proposes a coordinate-combination
method for constructing the risk-space diagram (RSD),
and then employs the Euclidian distance formulae to
calculate the RG and the RM indexes. To reect the
subjectivity and imprecision of the questionnaire survey
and routine inspection, the judgments made by experts and
inspectors regarding the scores of each fuzzy estimation
factor, excluding probability and frequency, are represented using a fuzzy linguistic scale.
The triangle fuzzy questionnaire surveys are used to
obtain the fuzzy measurement scores owing to their ease of
comprehension and operation. A ve-point linguistic scale
{very high, high, middle, low, and very low} is used for
designing the fuzzy questionnaires (Buckley, 1984). The
triangle fuzzy measurement scores of importance (I~mkj ) are
provided by Eq. (1), and the sub-total triangle fuzzy
measurement scores (Amk, Bmk, Cmk) are then determined
by Eq. (2), where Bmk is calculated using geometric means.
Given



(1)
I~mkj Amkj ; Bmkj ; C mkj ; 8m; k; j,

To develop the risk assessment model, all considered

estimation factors are assessed in terms of importance,
hazardousness, detectability, probability, criticality, and
frequency. These are divided into two groups. The
endogenous group, comprising importance, hazardousness,
detectability based on the absolute judgments of experts,
and probability, the four of which must generally remain
constant during a risk assessment period. The exogenous
group, containing criticality based on inspectors absolute
judgments, and frequency, both of which must change
according to each inspection result. To reect the inherent
imprecision of the survey and inspection process, these
expert assessment and inspector examination results are
represented by triangular fuzzy numbers, while probability
and frequency are crisp values.
~ represents the safety signiThe factor importance (I)
cance of a risk factor; a high degree of inuence of the
signicance is positively associated with high importance.
~ indicates the possible
The failure hazardousness (H)
severity of the disaster resulting from the failure of a risk
factor. Degree of hazardousness may vary among risk
factors. Some risk factors may cause only light injury or
light property damage or loss, while others may cause
heavy casualties or considerable property damage or loss.
The greater the level of death or injury and property
damage or loss resulting from the failure, the higher the
~ of a risk factor
hazardousness. The failure detectability (D)
indicates whether the failure can easily be detected. If the
failure can easily be detected, then it has low detectability.
The probability (P) of air accidents represents the number
of global air accidents occurring during a risk assessment
~ represents the extent of
period. The failure criticality (C)
risk factor failure. The inspector must clearly detect and
record the failure, the failure mode, and its criticality
during the routine examination. The failure frequency (f) of
a risk factor refers to the number of failures per unit during
a risk assessment period. The measurement unit can be the
number of take-offs and landings, the ight-hours, the
number of departures, and the duration of air trafc
control. The failure frequency is calculated for each risk
factor based on different failure criticalities.
Failure follows a malfunction, fault, breakdown, bad
reaction, or loss of normal function of a risk factor. Most
failures are discovered during routine maintenance, repair,
and inspection. Numerous types of failure exist, including
the fuselage fractures, the mechanical breakdowns, failure
of landing gear to rise/fall normally, attrition of tire
integrity, and so on. These types of fractures, breakdowns,

then


Amk min Amkj ; k 1;


; K; j 2 N mk ,
!1=N mk
N
mk
Y
Bmk
Bmkj
,
j1


C mk max C mkj ;

k 1;


; K; j 2 N mk ,

Risk gradient (RG) index

a. Factor Importance of a riskfactor
b. Failure hazardousness of a risk factor
c. Failure detectability of a risk factor
d. Probability of air accidents
Level of risk (LoR) index

a. Failure criticality of a risk factor

b. Failure frequency of a risk factor
Risk magnitude (RM) index
Fig. 1. The integrated concept of level of risk (LoR) index.

ARTICLE IN PRESS
W.-K. Lee / Journal of Air Transport Management 12 (2006) 267273

where I~mkj represents the fuzzy importance of the jth

expert, the kth linguistic scale of risk factor m; Nmk
represents the number of the kth linguistic scale for
assessing the importance of risk factor m.
Subsequently, the fuzzy membership functions are used
to calculate the aggregate triangle fuzzy measurement
scores (I~m ) by
(
)
K
X
N mk Amk ; Bmk ; C mk  =nI
I~m
k1

I mL ; I mM ; I mR ; 8m,
3
PK
where nI k1 N mk .
The aggregately triangle fuzzy measurement scores of
hazardousness H~ m H mL ; H mM ; H mR  and detectability
D~ m DmL ; DmM ; DmR  can be obtained in the same way.
Finally, the a-cut technique (defuzzied method) is
employed to calculate the left-hand values (I amL , H amL ,
and DamL ) and right-hand values (I amR , H amR , and DamR )
of the triangle fuzzy measurement scores of estimation
factors by

H amL
H amR

for each risk factor. In RSD, the sufx L on each axis

denotes the left-hand value of each estimation factor, while
the sufx R denotes the right-hand value of each estimation
factor; the superscript L in quadrant space denotes the leftbound RG Lm index, while the superscript R represents the
right-bound RG R
m index. Each risk factor has its RSD. The
RG Lm index comprises I amL , H amL , DamL , and P. Meanwhile,
a
a
a
the RG R
m index comprises I mR , H mR , DmR , and P.
L a
a
a
a
Consequently, RG m (I mL , H mL , DmL , P ) and RG R
m (I mR ,
a
a
H mR , DmR , P) represent respectively the coordinatecombination functions of the left-bound and right-bound
RG indexes of risk factor m under a-cut. Here, the
probability (P) is included in these functions and its
left-hand and right-hand values are equivalent for all
risk factors.
The probability can be based on the global air accidents.
Janic (2000) designed the probability distribution function
based on global air accidents during 19651998, and
adopted regressive analysis to deduce the global probability of an accident occurring by
PTpt 1  exp0:020t ;

I amL I mL aI mM  I mL ;
I amR I mR  aI mR  I mM ;

8m,

8m,

DamL DmL aDmM  DmL ;

DamR DmR  aDmR  DmM ;

tX0,

(7)

H mL aH mM  H mL ;
H mR  aH mR  H mM ;

269

8m,

where 0.0oao1.0, and is considered a risk control

variable.
The coordinate-combination method is used for avoiding dilution. Fig. 2 shows the concept upon which it is
based. Each coordinate axis represents a different estimation factor, and all axes construct a RSD of an RG index

where t is a time variable (expressed in days). Consequently, the probability is 0.0198 per day t 1, 0.0392
per 2 days t 2, and so on.
Finally, the Euclidean distance formulae is used to
calculate the distances between xamL and xamin , xamR and xamin ,
xamax and xamin by
v,v
u X
u X
X

uX

2 u
2
L
t
t
a
a
RG m
xmL  xmin
xamax  xamin ;
x1

x1

x1

x1

v,v
u X
u X
uX

X

2 u
2
R
t
t
a
a
RG m
xmR  xmin
xamax  xamin ;

8m,

8m.
8

(0.0,0.0,1.0)

Dm

(0.0,1.0,1.0)

G:1.0(1.0,1.0,1.0)

(1.0,0.0,1.0)
D
mR
RGmR

D
mL
O:0(0.0,0.0,0.0)
I
mR

RG mL

H
mR

(0.0,1.0,0.0)
Hm

H
mL

I
mL
Im

(1.0,0.0,0.0)

(1.0,1.0,0.0)

Fig. 2. The RSD possessing three estimate factors.

where x denotes endogenous estimation factors; xamax are

the maximum values of x under a-cut, which generally
equal 1.0 (standardized); xamin are the minimum values,
which generally equal 0.0. The denominators are the
standardized formulae.
The RSD of RM index resembles Fig. 2, but it is really a
two-dimensional diagram. The RM index comprises C~ m
and fm. The criticalities (C~ m ) are the aggregate triangular
fuzzy measurement scores given by
C~ m C mL ; C mM ; C mR ;

(9)

8m.

Subsequently, the left-hand values (C amL ) and right-hand

values (C amR ) under a-cut are calculated by
C amL C mL aC mM  C mL ;
C amR C mR  aC mR  C mM ;

8m.

10

ARTICLE IN PRESS
W.-K. Lee / Journal of Air Transport Management 12 (2006) 267273

270

Furthermore, the Euclidean distance formulae are

provided by
RM Lm

s,s
Y

Y

2
2
P
P

yamL  yamin
yamax  yamin ;
y1

y1

y1

y1

s,s
Y

Y

2
2
P
P
yamR  yamin
yamax  yamin ;
RM R
m

8m;

8m;
(11)

where y denotes the exogenous estimation factors; RM Lm

represents the left-bound RM index of risk factor m under
a-cut, and RM R
m represents the right-bound RM index of
risk factor m under a-cut. The denominators are also the
standardized formulae.
Following the calculation of RG and RM indexes, the
LoR indexes are constructed by
q

2

 2
LoRLm
RM Lm  RG Lm 2 ; 8m;
q

(12)


R  2
R 2
2

RM

RG
LoRU
;
8m;
m
m
m
where LoRLm represents the lower-bound LoR index of risk
factor m under a -cut, and LoRU
m represents the upperbound LoR index of risk factor m under a-cut.
2.3. Constructing risk-monitoring diagrams and analyzing
risks
The ALARP approach is an effective tools for analyzing
safety risks in personal insurance or environmental monitoring systems (Tam et al., 1996). Lee and Chang (2005b) have
applied this approach to develop a risk-monitoring diagram
(RMD) in aviation SRM, which integrates the RG and RM
indexes into the LoR index for each risk factor. In RMD, the
two boundaries that divide the three zones are the LoRLm
index and the LoRU
m index given by Eq. (12). The two
oblique lines represent the RM index. Notably, the baseline
of the inversed triangle increases with the RG index. Upon
checking the status of a risk factor, the inspector must
transform the inspected result into the LoR index (the new
LoRU
m index is then called RS-line).
Generally, the inversed triangle can be separated into
zones:
1. Top zoneintolerable region (InTo-region ): If this
region contains the RS-line, the inspected result (the
U
new LoRU
m index) exceeds the set LoRm index. Consequently, the SRM overseer must immediately adopt
appropriate measures to eliminate the impact of failure.
2. Middle zoneas low as reasonably practicable region
(ALARP-region): If this region contains RS-line, the
SRM overseer only needs to monitor the new risk status.
However, more rigorous measures must be adopted
when the RS-line approaches the upper section of this
region.

3. Bottom zonebroadly acceptable region (BA-region): If

this region contains the RS-line, no measures need to be
taken. Meanings vary among regions, and the related
measures that need to be adopted also differ.
Anyhow, if airlines and airports apply this model, the
checklist of FMECA for each risk factor needs to be
established rst. To monitor the RS-line, it is necessary to
implement routine inspection work to collect and completely record the failure data. Particularly, the SRM
overseer would need to continually analyze the risk status
and adopt appropriate measures to prevent worsening
failure.
3. Empirical study
3.1. Screening risk factors and deducing RG indexes
While the proactive function emphasizes all aviation
safety factors are treated as risk factors. Previous studies
were unable to determine which classications were correct
with respect to risk factors (Lee and Chang, 2005b). The
Boeing company separated aviation safety factors into
seven groups crew, airline ight operations, airplane design
and performance, airplane maintenance, air trafc control,
airport management, and weather information. Heinrich
(1959) sorted them into ve types: human, machine,
mission, management, and environment, and Edwards
(1988) categorized them into four types: livewire, hardware, software, and environment. Meanwhile, the International Air Transport Association (IATA) classied them
into ve categories human, organization, machine, environment, and insufciency (HOMEI). Among these categorizations, the categorization of the IATA is the most
widely applied and used here (Civil Aeronautics Administration, 1999; Civil Aeronautics Administration, 2001) to
derive these screened risk factors. However, only 14 risk
factors (mechanical category) are selected.
Table 1 lists the defuzzied measurement scores of
importance, hazardousness, and detectability with respect
to expert questionnaire surveys the risk factors, with
a 0:5. The experts included 21 airline safety supervisors,
10 academics, 13 research department and Aviation Safety
Council (ASC) experts, 11 directors of safety management
departments in the Taiwanese CAA and two international
airports. The survey process from March to June 2004
contained two stages of interviews and mailings; the rst
aimed to screen the estimation factors (Lee and Chang,
2005a), and the second was to obtain the measurement
scores of considered risk factors. In the second stage, a
total of 55 copies were issued, and 45 were returned. The
effective response rate was around 81.82%. Since routine
inspection work is executed everyday, the probability is
0.0198 per day. In Table 1, the average scores of
importance and hazardousness all exceed 0.70, meaning
that these two estimation factors are highly represented.
The average scores of detectability are almost all below

ARTICLE IN PRESS
W.-K. Lee / Journal of Air Transport Management 12 (2006) 267273

271

Table 1
The defuzzied measurement scores of endogenous estimate factors and P (a-cut 0.5)
Risk factors of machine (Fm)

F1 Airplane structure
F2 Engine system
F3 Landing gear and tire system
F4 Flight control system
F5 Navigation system
F6 Hydraulic pressure system
F7 Fuel system
F8 Automatic driving system
F9 Defending ice, eradicating ice or rain system
F10 Fire and smog warning system
F11 Cabin pressure, lubrication, and electricity system
F12 Ground proximity warning system (GPWS)
F13 Auxiliary approaching system
F14 Early-alarm measures (TCAS, ASDE)

Importance (Im)

Hazardousness (Hm)

Detectability (Dm)

Left-hand

Right-hand

Left-hand

Right-hand

Left-hand

Right-hand

0.735
0.752
0.682
0.737
0.656
0.650
0.667
0.567
0.691
0.771
0.628
0.675
0.621
0.663

0.873
0.887
0.836
0.874
0.804
0.798
0.815
0.730
0.833
0.903
0.780
0.832
0.789
0.823

0.773
0.725
0.652
0.726
0.631
0.629
0.679
0.544
0.717
0.775
0.636
0.705
0.630
0.672

0.892
0.860
0.810
0.864
0.781
0.781
0.826
0.700
0.857
0.889
0.786
0.839
0.774
0.821

0.576
0.471
0.356
0.415
0.394
0.359
0.381
0.366
0.392
0.390
0.374
0.341
0.354
0.350

0.724
0.652
0.575
0.627
0.587
0.561
0.580
0.584
0.566
0.585
0.582
0.542
0.555
0.545

Prob. (P)

0.020
0.020
0.020
0.020
0.020
0.020
0.020
0.020
0.020
0.020
0.020
0.020
0.020
0.020

Table 2
The RGm, RMm, and LoRm indexes (a-cut 0.5)
Fm

F1
F2
F3
F4
F5
F6
F7
F8
F9
F10
F11
F12
F13
F14

Risk gradient (RGm)

Risk magnitude (RMm)

Level of risk (LoRm)

Left-bound

Right-bound

Critical ranking

Left-bound

Right-bound

Lower-bound (C)

Upper-bound (D)

Critical ranking

Interval

0.606
0.573
0.504
0.558
0.496
0.487
0.513
0.434
0.535
0.580
0.484
0.517
0.476
0.503

0.721
0.698
0.649
0.690
0.632
0.625
0.649
0.584
0.661
0.698
0.626
0.650
0.618
0.642

1
2
7
4
10
12
8
14
5
3
11
6
13
9

0.477
0.477
0.477
0.477
0.477
0.477
0.477
0.477
0.477
0.477
0.477
0.477
0.477
0.477

0.566
0.566
0.566
0.566
0.566
0.566
0.566
0.566
0.566
0.566
0.566
0.566
0.566
0.566

0.369
0.382
0.405
0.387
0.408
0.411
0.403
0.425
0.395
0.379
0.411
0.401
0.414
0.406

0.436
0.445
0.463
0.448
0.469
0.472
0.464
0.484
0.459
0.445
0.471
0.463
0.474
0.466

1
2
7
4
10
12
8
14
5
3
11
6
13
9

0.067
0.063
0.058
0.061
0.061
0.061
0.061
0.059
0.064
0.066
0.060
0.062
0.060
0.060

 The aggregately triangle fuzzy measurement scores of criticality and frequency are hypothetical values set as (0.60, 0.75, 0.85).

0.5000

InTo-region
RS-line
0.721

0.4357
0.4000
ALARP-region
0.3687
Level of risk (LoR)

0.50, indicating that it is easy to examine the failures for

these 14 risk factors.
Using the data in Table 2, all left-bound and right-bound
RG indexes are calculated using Eq. (8). Risk factor F1 has
the highest (RG Lm , RG R
m ) (0.606, 0.721) indexes. Column
4 of the table reveals the critical-ranking order of these risk
factors according to the decreasing order of RG R
m indexes.
Risk factor F1 is the most critical, followed by F2, and F10.
Based on the ranking, the SRM overseer can adopt
different measures relevant for different risk factors.

0.606

LoR mL

0.3000

BA-region
0.2000

3.2. Deducing RM indexes and LoR indexes

Owing to the scarcity of safety inspection data,
hypothetical data is used to obtain the C~ m and fm
measurement scores for deducing the RM indexes. The
aggregately triangle fuzzy measurement score of failure
criticality is assumed to be (0.60, 0.75, 0.85), and the failure
frequency is given by 1/365 (times/days) 0.00274 for each

LoRmU

0.1000

0.0000
Risk gradient (RG)
Fig. 3. The RMD of risk factor F1.

ARTICLE IN PRESS
W.-K. Lee / Journal of Air Transport Management 12 (2006) 267273

272

risk factor. The RM Lm and RM R

m indexes are calculated
using Eq. (11), with a 0:50. The LoRLm and LoRU
m indexes
are thus obtained from Eq. (12), with a 0:50. Table 2 lists
0.5000

InTo-region

the results. While all of the hypothetical RM indexes are

the same, the LoR indexes differ among the risk factors.
The critical-ranking order of these factors ranked in
increasing order of the LoRU
m index shows factor F1 has
the lowest LoRU
index.
Accordingly,
the RS-line (new
m
LoRU
index)
of
risk
factor
F
exceeds
this
lowest threshold
1
m
by more than other risk factors. Comparatively, risk factor
F8 has the highest LoRU
m index, so its RS-line has difculty
exceeding this threshold. A higher RG R
m index will produce
a lower LoRU
index.
This
result
ts
with
the aviation SRM
m
theorem. The SRM overseer must adopt more rigorous
measures to monitor risk factors when LoRU
m index is low.
Furthermore, the nal column of Table 2 displays the
interval between the LoRLm and LoRU
m indexes. When the
hypothetical RM indexes are the same, if the interval is
larger, the deduced ALARP-region of RMD is wider and
its LoRU
m index is lower (Figs. 3 and 4). In this situation, the
RS-line of this kind of risk factor easily falls into the InToregion, and should be considered a more critical risk factor.

0.584

0.4844

Level of risk (LoR)

0.4252
0.4000

ALARP-region

RS-line

0.434

BA-region
0.3000

0.2000

0.1000

3.3. RMDs and risk-monitoring strategies

0.0000

The RMDs of risk factor F1 and F8 are shown in Figs. 3

and 4. Risk factor F1 has the lowest LoRU
m index, while risk
factor F8 has the highest one. Clearly, a rigorous riskmonitoring strategy must be adopted for risk factor F1, but
a slack strategy can be adopted for risk factorc F8.
Matching up the RMD can enable the SRM overseer to
clearly monitor the status of each risk factor. Table 3 lists
the variations of the LoRU
m indexes for the 14 risk factors by
varying the a-cut value, which can be varied between 0.0
and 1.0. When a-cut is increased, the low LoRU
m index is
further reduced. On the other hand, the LoRU
m index is
increased while a-cut is decreased. For example, if a-cut
varies from 0.50 to 0.65 for risk factor F1, the LoRU
m index is
reduced from 0.436 to 0.430. Consequently, more rigorous
risk-monitoring measures are adopted when the larger a-cut
value is used, while otherwise looser measures are adopted.
Numerous results are obtained given different a-cut
values for each risk factor. This study emphasizes critical
U
risk factors with high RG R
m index or low LoRm index
require close monitoring. Fig. 5 displays the relationships
between a-cut values and ( LoRLm , LoRU
m ) indexes of risk
factor F1. For risk-monitoring, the SRM overseer can

Risk gradient (RG)

Fig. 4. The RMD of risk factor F8.
Table 3
The variations of LoRU
m indexes by varying a-cut

F1
F2
F3
F4
F5
F6
F7
F8
F9
F10
F11
F12
F13
F14

LoRU
m
a 0.8

a 0.65

a 0.5

a 0.35

a 0.2

0.424
0.434
0.454
0.437
0.459
0.461
0.454
0.474
0.448
0.432
0.460
0.453
0.464
0.456

0.430
0.430
0.459
0.443
0.464
0.466
0.459
0.479
0.454
0.439
0.466
0.458
0.469
0.461

0.436
0.445
0.463
0.448
0.469
0.472
0.464
0.484
0.459
0.445
0.471
0.463
0.473
0.465

0.442
0.451
0.467
0.454
0.474
0.477
0.468
0.489
0.464
0.452
0.477
0.468
0.479
0.470

0.447
0.456
0.471
0.459
0.478
0.482
0.472
0.494
0.470
0.458
0.482
0.473
0.483
0.475

upper-bound

lower-bound

0.500
Level of Risk (LoR)

Fm

(0.3, 0.4435) (0.5, 0.4357)

0.450

(0.8, 0.4240)

0.400

RS-line

(0.8, 0.3971)
0.350

(0.5, 0.3687)
(0.3, 0.3496)

0.300
0.0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.0

Fig. 5. The relationship of a and LoR index of risk factor F1.

ARTICLE IN PRESS
W.-K. Lee / Journal of Air Transport Management 12 (2006) 267273

adjust the a-cut value to yield different LoRU

m indexes to
facilitate risk monitoring for each risk factor.
Acknowledgements
The author would like to thank the National Science
Council of the Republic of China, Taiwan, for nancially
supporting this research under Contract no. NSC 95-2416H-408-001. Directors of the Safety Management Department in the Taiwanese CAA and airports, airline safety
supervisors, academic professors and experts in the
research department and the Aviation Safety Council are
thanked for their assistance in problem formulation and
data collection.
References
Buckley, J.J., 1984. The multiple judge, multiple criteria ranking problem:
a fuzzy set approach. Fuzzy Sets and System 13, 2537.
Button, K., Clarke, A., Palubinskas, G., Stough, R., Thibault, M., 2004.
Conforming with ICAO safety oversight standards. Journal of Air
Transport Management 10, 251257.

273

Civil Aeronautics Administration, 1999. Airport Risk Management

Reference Manual. Civil Aeronautics Administration, Taipei.
Civil Aeronautics Administration, 2001. Airport Risk Self-Superintending
System. Civil Aeronautics Administration, Taipei.
Edwards, E., 1988. Introductory overview. In: Winner, E.L., Nagel, D.C.
(Eds.), Academic Press Human Factors in Aviation. Academic Press,
New York.
Heinrich, H.W., 1959. Industrial Accident Prevention. McGraw-Hill,
New York.
Health and Safety Executive, 1992. The Tolerability of Risk from Nuclear
Power Stations. Health and Safety Executive, London.
Janic, M., 2000. An assessment of risk and safety in civil aviation. Journal
of Air Transport Management 6, 4350.
Lee, W.K., Chang, Y.H., 2005a. Applying the fuzzy FMECA to
access the risks of ight-safety. Civil Aviation Journal Quarterly 7,
122.
Lee, W.K., Chang, Y.H., 2005b. Risk assessment model on ight-safety
management. Transportation Planning Journal 34, 145176.
Tam, V., Moros, T., Webb, S., Allinson, J., Lee, R., Bilimoria, E., 1996.
Application of ALARP to the design of the BP Andrew platform
against smoke and gas ingress and gas explosion. Journal of Loss
Prevention in the Process Industries 9, 317322.
US Department of Defense, 1980. Procedures for Performing a Failure
Mode, Effects and Criticality Analysis. MIL-STD-1629A, Washington, DC.