(This is a sample cover image for this issue. The actual cover is not yet available at this time.)

(This is a sample cover image for this issue. The actual cover is not yet available at this time.) This article appeared in a journal published by El...
Author: Cory Holt
2 downloads 0 Views 421KB Size
(This is a sample cover image for this issue. The actual cover is not yet available at this time.)

This article appeared in a journal published by Elsevier. The attached copy is furnished to the author for internal non-commercial research and education use, including for instruction at the authors institution and sharing with colleagues. Other uses, including reproduction and distribution, or selling or licensing copies, or posting to personal, institutional or third party websites are prohibited. In most cases authors are permitted to post their version of the article (e.g. in Word or Tex form) to their personal website or institutional repository. Authors requiring further information regarding Elsevier’s archiving and manuscript policies are encouraged to visit: http://www.elsevier.com/copyright

Author's personal copy

Journal of Air Transport Management 27 (2013) 34e38

Contents lists available at SciVerse ScienceDirect

Journal of Air Transport Management journal homepage: www.elsevier.com/locate/jairtraman

Analysis of an aircraft accident model in Taiwan Pei-Chi Shao a, Yu-Hern Chang a, *, Hubert J. Chen b, c a

Department of Transportation and Communication Management Science, National Cheng Kung University, 1 University Road, Tainan 70101, Taiwan Department of Statistics, The University of Georgia, Athens, GA, USA c Department of Accountancy & Graduate Institute of Finance, National Cheng Kung University, Tainan, Taiwan b

a b s t r a c t Keywords: Aircraft incident occurrences Aircraft accidents Accident fatalities Aviation safety

This paper examines factors that have influenced the average accident rate per million departures in Taiwan from 1985 to 2011 involving turbojet aircraft hull loss. Our analysis is based on the nature of rare events, used to find the importance of the International Civil Aviation Organization occurrence categories. The most significant occurrences were in order of importance are: takeoff, landing, and ground operations; aircraft; miscellaneous; weather; and airborne. The subcategory of runway incursiondvehicle, aircraft, or person was the most significant effect for accidents; runway excursion for serious incidents; system/ component failure or malfunction in non-power plant; turbulence encounter for occurrences; and controlled flight into or toward terrain for fatal accidents. Ó 2012 Elsevier Ltd. All rights reserved.

1. Introduction Between 2002 and 2011, the average accident rate per million departures in Taiwan involving turbojet aircraft hull loss was 2.57 times the world’s average (Civil Aeronautics Administration (CAA) (2012)); the average rate of hull loss occurrences on commercial jets was 1.75 per million departures and on turboprop aircraft was 1.31. The International Civil Aviation Organization’s (ICAO) (2009) safety targets include reduction in fatal airlines accidents, serious incidents, runway excursion events and ground collision events and here we look at the causes of these events and how they relate to the safety targets set out in Taiwan’s Civil Aviation Administration’s (2011) Sate Safety Program (SSP) using the fixed-wing aircraft investigation reports of aviation accidents in which the aircraft were registered in Taiwan. These accidents are classified according to the International Civil Aviation Organization (2008) Aviation Occurrence Categories (AOC). There were 55 aviation occurrences by fixed-wing aircraft in Taiwan, where 32 were accidents, including 14 fatal, and 23 serious incidents.

2. Methodology We use both CAA and Aviation Safety Council (ASC) (2009) investigation reports to classify the aviation occurrence categories

* Corresponding author. Tel.: þ886 958669951. E-mail addresses: [email protected] mail.ncku.edu.tw (Y.-H. Chang).

(P.-C.

Shao),

yhchang@

0969-6997/$ e see front matter Ó 2012 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.jairtraman.2012.11.004

in accordance with the ICAO’s classification standard. Referring to the definition of occurrences by the ICAO, the ASC investigates aviation occurrences that involve fatality, injury, or substantial damage to aircraft. Because there are few records of aviation occurrences or accidents per ten thousand departures, we used the CAA reports from 1985 to 1998 as well as the ASC reports from 1999 to 2011.1 The common taxonomy and definitions established by the ICAO include six groups of incidents (Table 1). According to Aviation Safety Council (2012), between 2002 and 2011 runway excursions caused one accident and eight serious incidents; turbulence encounter (TURB), four accidents, other two accidents and two serious incidents; and fire/smoke (Non-Impact) (F-NI), one accident and three serious incidents (Fig. 1). We focus on the accident, serious incident, and fatal accident rates, respectively, as measures of safety performance (Chang and Yeh, 2004) to build relationships between aviation occurrence rates and AOC.

1 The AOC defines “an occurrence” is “an accident or an incident”, and it is focused on powered fixed-wing land and rotorcraft operations; “an accident” is “an aircraft accident associated with the operation of an aircraft which takes place between the time any person boards the aircraft with the intention of flight until such time as all such persons have disembarked, in which such person is fatally or seriously injured or in which the aircraft is substantially damaged or missing”; “a fatal aviation accident” is defined as an accident which has resulted in the death of one or more passengers during the flight; and finally, “a serious incident” is “an occurrence of an incident associated with the operation of an aircraft which takes place between the time any person boards the aircraft with the intention of flight until such time as all such persons have disembarked, which may cause aviation accidents”.

Author's personal copy

P.-C. Shao et al. / Journal of Air Transport Management 27 (2013) 34e38

35

Table 1 ICAO operation grouping categories. Occurrences category

Acronym

Description

Takeoff, Landing, and Ground Operation e TLGO Ground Handling Ground Collision Loss of Control e Ground Runway Excursion Runway Incursion e Vehicle, Aircraft, or Person

RAMP GCOL LOC-G RE RI-VAP

Runway Incursion e Animal

RI-A

Undershoot or Overshoot Abnormal Runway Contact Fire/Smoke (Post-Impact) Evacuation

USOS ARC F-POST EVAC

Occurrences during (or as a result of) ground handling operations. Collision while taxiing to or from a runway in use. Loss of aircraft control while the aircraft is on the ground. A veer off or overrun off the runway surface. Any occurrence at an aerodrome involving the incorrect presence of an aircraft, vehicle, or person on the protected area of a surface designated for the landing and takeoff of aircraft. Collision with, risk of collision, or evasive action taken by an aircraft to avoid an animal on a runway or on a helipad or helideck in use. A touchdown off the runway or helipad or helideck surface. Any landing or takeoff involving abnormal runway or landing surface contact. Fire or Smoke resulting from impact. Occurrence where either; (a) person(s) are injured during an evacuation, (b) an unnecessary evacuation was performed, (c) evacuation equipment failed to perform as required, or (d) the evacuation contributed to the severity of the occurrence happened.

Airborne e AIRBN Airprox/TCAS Alert/Loss of Separation/Near Midair Collisions/Midair Collisions Controlled Flight into/toward Terrain

CFIT

Loss of Control - In flight Fuel Related

LOC-I FUEL

Low Attitude Operations

LALT

Abrupt Maneuver Weather e WTHR Windshear or Thunderstorm Turbulence Encounter Icing

AMAN

Aircraft e ARCFT System/Component Failure or Malfunction (Power plant)

MAC

Airprox, ACAS alerts, loss of separation as well as near collisions or collisions between aircraft in flight. In-flight collision or near collision with terrain, water, or obstacle without indication of loss of control. Loss of aircraft control while or deviation from intended flight path in-flight. One or more power-plants experienced reduced or no power output due to fuel exhaustion, fuel starvation/mismanagement, fuel contamination/wrong fuel, or carburetor and/or induction icing. Collision or near collision with obstacles/objects/terrain while intentionally operating near the surface (excludes takeoff or landing phases). The intentional abrupt maneuvering of the aircraft by the flight crew.

WSTRW TURB ICE

Flight into wind shear or thunderstorm. In-flight turbulence encounter. Accumulation of snow, ice, freezing rain, or frost on aircraft surfaces that adversely affects aircraft control or performance.

SCF-PP

Failure or malfunction of an aircraft system or component-related to the power-plant. Failure or malfunction of an aircraft system or component other than the power-plant. Fire or smoke in or on the aircraft, in flight or on the ground, which is not the result of impact.

System/Component Failure or Malfunction (Non-Power plant) Fire/Smoke (Non-Impact)

SCF-NP F-NI

Miscellaneous e MISCN Security Related Cabin Safety Events

SEC CABIN

Other Unknown or Undetermined Non-aircraft-related e NARCFT ATM/CNS

OTHR UNK

Aerodrome

ADRM

ATM

Criminal/Security acts which result in accidents or incidents Miscellaneous occurrences in the passenger cabin of transport category aircraft. Any occurrence not covered under another category. Insufficient information exists to categorize the occurrence. Occurrences involving Air traffic management (ATM) or communications, navigation, or surveillance (CNS) service issues. Occurrences involving aerodrome design, service, or functionality issues.

Source: International Civil Aviation Organization (2008).

Occurrences 6 5 4 3 2 1

Serious Incidents Accidents

0

Fig. 1. Frequency of occurrences in Taiwan airlines 2000 to 2010 by ICAO occurrence category. Source: http://www.asc.gov.tw/asc_en/index.asp.

Because most aviation accident data are random and accidents are rare, the Poisson probability distribution is generally used for describing their frequency (Raghavan and Rhoades, 2005). Recently, Rose (1990), for example, accepted that the Poisson probability distribution was a natural stochastic model for airline accidents, and Poisson regression was used to explore the relationship between the safety performance and profitability of airlines. Let the random variable Y, the number of occurrences in a given year, have a Poisson probability distribution, P(y: m), where y is a realized value of Y, which may take on values 0, 1, 2, 3,., and m (>0) is the average number of accidents. Based on the occurrence data seen in Table 2, there were 55 occurrences between 1985 and 2011. Given a set of observed accident data, if the actual frequencies of the observed accidents are close to the expected frequencies under

Author's personal copy

36

P.-C. Shao et al. / Journal of Air Transport Management 27 (2013) 34e38

Table 2 Annual occurrences of accidents/serious incidents, departures, and accident/occurrence/fatality rates from 1985 to 2011. Year

1985 1986 1987f 1988f 1989 1990f 1991 1992f 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 Total Mean S.D. a b c d e f

Occurrencesa

Fatale

Accidentsb

Serious incidents

Fatal accidents

Departures (104)

Accidentc rate

Occurrenced rate

Accident rate

1 1 0 0 1 0 1 0 2 1 1 1 1 2 3 1 2 2 1 1 3 1 2 3 0 1 0 32 1.185 0.921

0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 3 1 3 2 1 1 1 1 3 1 3 3 23 0.852 1.167

0 1 0 0 1 0 1 0 0 1 1 1 1 2 2 1 0 2 0 0 0 0 0 0 0 0 0 14 0.519 0.700

7.2101 7.1557 7.9880 8.8747 11.7445 12.5912 15.8741 18.8908 21.0651 22.1306 29.1817 33.4692 36.4678 32.6166 33.2466 30.7042 27.7170 27.6167 26.3311 27.3380 29.2794 26.4551 24.8077 20.8861 19.5601 20.9381 24.2491 604.3893 22.385 8.679

0.139 0.140 0 0 0.085 0 0.063 0 0.095 0.045 0.034 0.030 0.027 0.061 0.090 0.033 0.072 0.072 0.038 0.037 0.102 0.038 0.081 0.144 0 0.048 0 1.474 0.055 0.044

0.139 0.140 0 0 0.085 0 0.063 0 0.095 0.045 0.034 0.030 0.027 0.061 0.090 0.130 0.108 0.181 0.114 0.073 0.137 0.076 0.121 0.287 0.051 0.191 0.124 2.402 0.089 0.067

0.000 0.140 0 0 0.085 0 0.063 0 0 0.045 0.034 0.030 0.027 0.061 0.060 0.033 0 0.072 0 0 0 0 0 0 0 0 0 0.650 0.024 0.036

Fixed-wing aircraft include turbojet and turboprop aircraft. Accidents include fatal accidents. Accidents divided by departures. Occurrences divided by departures. Fatal accidents divided by departures. No occurrences.

the Poisson probability distribution, then the data follow a Poisson probability distribution (Gupta, 1977). Here, based on the occurrences data-set and the frequency information in Table 3 and Fig. 2, the Pearson’s goodness-of-fit Chi-square test statistic can lead to acceptance of the Poisson model at the 0.05 level of significance. Under the Poisson probability model for the number of occurrences, the mean and variance must be equal, and the mean can be expressed by a linear equation; if the variance is larger than the mean, the sample data is overdispersed, and if smaller it is underdispersed. Under such circumstances, adjustments must be made using a rescale in advance so that more accurate standard errors of the estimated Poisson regression parameters and subsequent p-values are obtained. Here, the estimated mean of m is the average number of occurrences, y ¼ 2:037and the estimated variance is 2.019 (Table 3), and the ratio of the variance over the

mean is 0.991, which may indicate some underdispersion. To test for this we consider whether the Poisson variance is smaller than its mean using Equation (1) to check whether a ¼ 0 where g (m) ¼ m2 (Cameron and Trivedi, 1998).

 ðY  mÞ2 Y ¼ a$g m þ 3

(1)

m is obtained using the maximum likelihood method. The negative a found indicates underdispersion and thus adjustments are made using a generalized linear model with a “DSCALE” option in SAS.

Frequency

Table 3 Aviation Accident data in Taiwan (1985e2011). Occurrences

Observed (Oi)

Expected (Ei)

Estimated Poisson probability

0 1 2 3 4 5 6 >6

4 9 4 5 3 1 1 0 27

3.5212 7.1728 7.3056 4.9606 2.5262 1.0292 0.3494 0.1349 27.000

0.1304 0.2657 0.2706 0.1837 0.0936 0.0381 0.0129 0.0050 1.0000

Observed Expected

Fig. 2. Observed versus expected occurrence frequency trend (1985e2010).

Author's personal copy

P.-C. Shao et al. / Journal of Air Transport Management 27 (2013) 34e38

3. Analysis

Table 4 Estimates for the ICAO occurrence categories.

Using the Poisson distribution, we define Yi as the actual number of occurrences, Ni departures in one hundred thousands, and mi the average number of occurrences during the ith year, where i is one of the years from 1985 to 2011. The mean of occurrences mi is treated as a positive function of some regression parameters (b0, b1, ., bk) associated with k exogenous explanatory variables (1, X1, ., Xk) and the average number of occurrences mi is equal to departures multiplied by the true occurrence rate li per departure, i.e., mi ¼ Ni  li where

l ¼ expðb0 þ b1 X1i þ b2 X2i þ . þ bk Xki Þ

37

(2)

The Poisson distribution is a limiting distribution of the binomial and negative binomial distributions because the number of departures (Ni) is very large and the probability of a failure (Pi), the departure accident rate, is relatively small. Because these features in binomial and negative binomial distributions can lead to computational difficulty, a Poisson distribution is used (Ott and Longnecker, 2010). It is also an excellent model for approximating the binomial and negative binomial regressions under their associated distributions.2 Using TLGO, AIRBN, WTHR, ARCFT, and MISCN as exogenous explanatory variables, the maximum likelihood is used to estimated the parameters in Eq. (2). After the underdispersion adjustment of the standard errors, the Poisson regression parameter estimates and associated statistics are given in Table 4. The parameters indicate a positive relationship between occurrences and the five ICAO categories. This means that each grouping category has a positive influence on the occurrences. Based on Table 4, the two grouping categories, TLGO and ARCFT, associated with the largest regression coefficients are the top two most important grouping categories (p-value of 0.0015 or less) for evaluating the occurrence rate. Among the five grouping categories, TLGO has the highest effect on occurrences, ARCFT is the second most significant grouping category which explains a high-risk occurrence during take-off and landing. In the TLGO category, there were 23 accidents and serious incidents, they are consistent with the causes highlighted by Raghavan and Rhoades (2005) who pointed out that accidents of this kind normally occur during takeoffs and landings when pilots’ have the greatest workload, and at the most congested areas and at the lowest altitudes. 3.1. Takeoff, Landing, and Ground Operation (TLGO) During the period, of the TLGO possibilities there were no occurrences in the LOC-G, RI-A, F-POST, and EVAC categories. Table 5, shows the parameters for the other variables to have positive and significant relationship with accidents, with RI-VAP having the greatest impact. A number of specific circumstances lead to runway incursions: confusing airport layouts; visibility limitations; high traffic volume; and communication errors (Young and Vlek, 2009). In looking for causes of GCOL occurrences, the International Air Transportation Association (IATA) (2008) found that factors contributing to runway collisions are deficiencies in regulatory oversight; environmental factors including wildlife and foreign objects; and poor signage, faint markings, and runway or taxiway closure.

2 We use the SAS procedure "PROC GENMOD" to calculate the log binomial regression, the logistic binomial regression, and the log negative binomial regression to check whether the log Poisson regression is an appropriate model to use in the research. All calculations indicated the same results. Thus, the Poisson regression is used to evaluate the importance of the grouping categories and subcategories.

Variable

Parameter

Wald chi-square

p-value

Intercept TLGO AIRBN WTHR ARCFT MISCN

12.641 0.422 0.173 0.264 0.406 0.371

2787.46 18.21 0.32 6.54 10.04 7.89

Suggest Documents