Content uploaded by Michael Schultz
Author content
All content in this area was uploaded by Michael Schultz on Sep 06, 2018
Content may be subject to copyright.
PREPRINT
Dynamic change of aircraft seat condition for fast boarding
December 2017. Transportation Research Part C: Emerging Technologies 85:131-147
DOI: 10.1016/j.trc.2017.09.014
Dynamic change of aircraft seat condition for fast boarding
Michael Schultz
Institute of Flight Guidance, German Aerospace Center (DLRe.V.),
Lilienthalplatz 7, Braunschweig, Germany
michael.schultz@dlr.de
ABSTRACT
Aircraft boarding is a process mainly impacted by the boarding sequence, individual passenger
behavior and the amount of hand luggage. Whereas these aspects are widely addressed in scientific
research and considered in operational improvements, the influence of infrastructural changes is only
focused upon in the context of future aircraft design. The paper provides a comprehensive analysis of
the innovative approach of a Side-Slip Seat, which allows passengers to pass each other during
boarding. The seat holds the potential to reduce the boarding time by approx. 20%, even considering
operational constraints, such as passenger conformance to the proposed boarding strategy. A validated
stochastic boarding model is extended to analyze the impact of the Side-Slip Seat. The
implementation of such fundamental change inside the aircraft cabin demands for adapted boarding
strategies, in order to cover all the benefits that accompany this new dynamic seating approach. To
reasonably identify efficient strategies, an evolutionary algorithm is used to systematically optimize
boarding sequences. As a result, the evolutionary algorithm depicts that operationally relevant
boarding strategies implementing the Side-Slip Seat should differentiate between the left and the right
side of the aisle, instead of the current operationally preferred boarding from the back to the front.
1 INTRODUCTION
Aircraft boarding holds the potential to significantly influence the entire aircraft trajectory over the
day of operations, since it is the last process of the turnaround (critical path) and determines the
estimated off block time of the aircraft (SESAR 2014, Eurocontrol/IATA/ACI 2014, IATA 2016).
Deviation in aircraft boarding (extension or reductions) could directly result in additional delays or
compensation of inbound delays. In particular, short-range flights require a reliable turnaround and
boarding to prevent the accumulation of delays during the aircraft rotation over the day. The
following analysis of infrastructural changes provides a fundamental background for evaluating the
boarding process in detail (Schultz et al. 2008, 2013) and provides an evaluation of the innovative
Side-Slip Seat concept (Molon Labe Seating 2017), which is expected to shorten the aircraft boarding
time significantly.
1.1 Previous Related Work
In the following section, a short overview concerning scientific research on aircraft boarding
problems is given. Relevant studies concerning aircraft boarding strategies include, but are not limited
to, the following examples. More comprehensive overviews are provided by Jaehn and Neumann
(2015) for boarding and by Schmidt (2017) for the aircraft turnaround. Comparisons and reviews of
different boarding approaches mainly focus on fast and reliable operational progress, but also address
the economic impact (Nyquist and McFadden 2008, Mirza 2008).
A common goal of simulation-based approaches is to minimize the time that is required for
passengers to board the aircraft. Taking into account specific boarding patterns, a study by Van
Landeghem and Beuselinck (2002) investigates the efficiency of different boarding strategies. A
similar approach is used by Ferrari and Nagel (2005), particularly focused on disturbances to the
PREPRINT
Dynamic change of aircraft seat condition for fast boarding
December 2017. Transportation Research Part C: Emerging Technologies 85:131-147
DOI: 10.1016/j.trc.2017.09.014
boarding sequence caused by early or late arrivals of passengers. The results show faster boarding
times for the commonly used back-to-front boarding in the case of passengers not boarding in their
previously assigned boarding block. This fact indicates that a back-to-front policy is not an optimal
solution for the boarding problem. Picking up the idea of block boarding, a study based on an
analytical model by van den Briel et al. (2005) shows significantly improved boarding times for block
policies compared to the back-to-front policy. In contrast, Bachmat and Elkin (2008) support the
back-to-front policy in comparison to the random boarding strategy. Schultz et al. (2008) demonstrate
with a stochastic cellular automaton model that back-to-front boarding is most efficient if two
boarding blocks are used, which is confirmed by Bachmat et al. (2013) using a 1+1 polynuclear
growth model with concave boundary conditions.
The interference of passengers during the seating process when boarding an aircraft forms the
focus of a study by Bazargan (2006). The mathematical model’s output aims to minimize the
interferences by using a mixed integer linear program for optimization. A stochastic approach to
covering both the individual passenger behavior (e.g. passenger conformance to the proposed
boarding strategy, individual hand luggage amount and distribution) and the aircraft/airline
operational constraints of aircraft/airlines (e.g. seat load factor, arrival rates) is in the focus of the
research of Schultz et al. (2008). Using a Markov Chain Monte Carlo optimization algorithm, Steffen
(2008a) develops a boarding strategy assuming that the handling of the hand luggage is a major
impact factor for the boarding time and provides a model based on fundamental statistical mechanics
(Steffen 2008b). Frette and Hemmer (2012) identify a power law rule, where the boarding time scales
with the number of passengers to board, which allows the prediction of the results of the back-to-front
boarding strategy, and Bernstein (2012) extends this approach to large numbers of passengers.
Tang et al. (2012) develop a boarding model considering passengers individual physique
(maximum speed), quantity of hand luggage, and individually preferred distance. Based on a boarding
strategy from Steffen (2008a), Milne and Kelly (2014) develop a method, which assigns passengers to
seats so that their luggage is distributed evenly throughout the cabin, assuming a less time-consuming
process for finding available storage in the overhead bins. Qiang et al. (2014) propose a boarding
strategy which prioritizes passengers with a high number of hand luggage items to board first. Milne
and Salari (2016) assign passengers to seats according to the number of hand luggage items and
propose that passengers with few pieces should be seated closed to the entry. Kierzkowski and Kisiel
(2017) provide an analysis covering the time needed to place items in the overhead bins depending on
the availability of seats and occupancy of the aircraft.
Bachmat et al. (2009) demonstrate with an analytical approach that the efficiency of boarding
strategies is linked to the aircraft interior design (seat pitch and passengers per row). Chung (2012)
and Schultz et al. (2013) address the aircraft seating layout and indicate that alternative designs could
significantly reduce the boarding time for both single and twin-aisle configuration. Fuchte (2014)
focusses on the aircraft design and, in particular, the impact of aircraft cabin modifications with
regard to the boarding efficiency whilst Schmidt et al. (2015, 2017) evaluate novel aircraft layout
configurations and seating concepts for single and twin-aisle aircraft with 180-300 seats.
In the context of deboarding, the seat interference disappears and only the interference of
passengers in the aisle is important for an efficient process. Wald et al. (2014) provide a study of
deplaning strategies using stochastic optimization methods, Qiang et al. (2016) use a cellular
automaton approach and Miura and Nishinari (2017) employ a model using an ex-Gaussian
distribution.
New topics of boarding research are focusing on dynamic allocation and control aspects.
Notomista et al. (2016) realize an efficient boarding procedure by allocating the seat numbers
adaptively to passengers when they pass the boarding gate. Zeineddine (2017) emphasizes the
importance of groups when traveling by aircraft and proposes a method whereby all group member
should board together, assuming a minimum of individual interference ensured by the group itself.
The methods applied to boarding problems range from analytical approaches (e.g. Frette and
Hemmer 2012), mixed integer linear programs (e.g. Bazargan 2007), polynuclear growth models (e.g.
PREPRINT
Dynamic change of aircraft seat condition for fast boarding
December 2017. Transportation Research Part C: Emerging Technologies 85:131-147
DOI: 10.1016/j.trc.2017.09.014
Bachmat 2013), Markov Chain Monte Carlo model (Steffen 2008a), statistical mechanics (Steffen
2008b), stochastic cellular automaton approaches (e.g. Schultz et al. 2008) up to pedestrian-following
models (Tang et al. 2012). If the research is aimed at finding an optimal solution for the boarding
sequence, evolutionary/genetic algorithms are used to solve the complex problem (e.g. Li et al. 2007,
Wang and Ma 2009, Soolaki et al. 2012).
1.2 Objectives and Document Structure
This paper provides fundamental in-depth analysis of the Side-Slip Seat according its potential to
reduce the aircraft boarding time under operational conditions (individual passenger behavior and
operational constraints). After a brief introduction of the continuously developed stochastic boarding
model (Schultz 2008, 2013, 2017) and a baseline simulation of a set of boarding strategies, the
operational concept Side-Slip Seat (Molon Labe Seating 2017) will be presented. The boarding model
is extended according to the operational requirements and used to evaluate the effect of the Side-Slip
Seat. Then these simulation results are compared against the baseline. Since the introduction of new
infrastructures often demands adjusted operational procedures, in the second part of the paper, the
approach of evolutionary algorithms will be used to derive appropriate boarding sequences. The idea
behind evolutionary algorithms is biologically inspired: start with a set of possible solutions (random
boarding sequences) and allow them to evolve over time using the processes of selection, heredity
(cross-over and mutation) and replacement of least-fit solutions. Instead of the systematic testing of
conceived boarding strategies (it is not possible to check all existing sequences in an appropriate
amount of time), evolutionary algorithms explore the problem space in a more efficient way. The
application of the developed evolutionary algorithms allows the basic evaluation of boarding
sequences, which are used to derive superior boarding strategies that will be analyzed in detail with
the stochastic boarding model. Finally, the appropriately adapted boarding strategies will cover all
operational benefits of the Side-Slip Seat. The conclusion provides an overview of the achieved
results and an outlook on future research activities.
2 STOCHASTIC AIRCRAFT BOARDING MODEL
In the context of aircraft boarding, the passenger (pax) behavior is assumed to be a stochastic,
forward-directed, one-dimensional and discrete (time and space) process. The proposed dynamic
passenger movement model is based on the asymmetric simple exclusion process (ASEP), which has
already been successfully adapted to model the individual passenger behavior in the airport terminal
(Schultz and Fricke 2011, Schultz 2013). To provide both an appropriate set of input data and an
efficient simulation environment, the aircraft seat layout is transferred into a regular grid with aircraft
entries, the aisle(s) and the passenger seats. Fig. 1 (left) clearly demonstrates the regular structure of
the aircraft seats using a typical interior arrangement of a single-class Airbus A320-200 in a high-
density configuration (Airbus 2017). For the reference implementation, an Airbus 320-200 single-
class seat layout is used with 29 rows and 174 passenger seats (economy), which is a commonly used
configuration for evaluating the efficiency of boarding procedures. The regular grid consists of equal
cells with a size of 0.4 x 0.4 m, whereas a cell can either be free or contain exactly one passenger (see
Fig. 1, center).
The boarding progress consists of a simple set of rules for the passenger movement: a) enter the
aircraft at the assigned door (based on the current scenario), b) move forward from cell to cell along
the aisle until reaching the assigned seat row, and c) store the baggage (aisle is blocked for other
passengers) and take the seat. The movement process only depends on the state of the next cell (free
or occupied). The storage of the baggage is also a stochastic process and depends on the individual
amount of hand luggage and the amount of time to store the luggage in the compartment. The seating
process is stochastically modelled as well, whereas the time to take the seat depends on the already
used seats in the corresponding row. The stochastic nature of the boarding process requires a
minimum of simulation runs for each selected scenario in order to derive reliable simulation results.
In this context, a simulation scenario is mainly defined by the underlying seat layout, the number of
PREPRINT
Dynamic change of aircraft seat condition for fast boarding
December 2017. Transportation Research Part C: Emerging Technologies 85:131-147
DOI: 10.1016/j.trc.2017.09.014
passengers to board (seat load factor, default: 85%), the arrival frequency of the passengers at the
aircraft, the number of available doors (default 1 door), the specific boarding strategy (default:
random boarding) and the conformance of passengers in following the current boarding strategy
(default: 85%). Random boarding is defined as a random chronological order of passengers, who
already have dedicated aircraft seats. The conformance rate describes several operational deviations
from the intended boarding strategy, caused by boarding services provided by airlines (e.g. priority
boarding, 1st class seats) or late arrival of passengers. Further details regarding the model and the
simulation environment are available at Schultz (2013). To model different boarding strategies, the
grid-based approach enables both the individual assessment of seats and classification/aggregation
according to the intended boarding strategy of the airline. In Fig. 1 (right), the seats are color-coded
(grey-scale) and aggregated to a superior seat block structure. In this case, the structure contains four
blocks and boarding is from the back to the front (in order of the grey scale value, dark first).
Figure 1: Aircraft model based on an Airbus A320-200 (left, Airbus 2017), transferred to regular grid (center,
Schultz 2013), and a seat block structure to provide different boarding strategies (right)
During field measurements in cooperation with a German airline, the boarding model parameters
are measured and used to calibrate the simulation environment (Schultz 2017). As an example, the
Airbus A320 aircraft characteristics document (Airbus 2017) exhibits a constant arrival rate of 12
pax/min per door. Since initial observations indicate a higher arrival rate, previous research on aircraft
boarding (Schultz 2008, 2013) was based on the assumption of a 14 pax/min arrival rate. The field
measurements, however, exhibit an arrival rate distribution with a linear declining behavior over the
boarding time, starting with a value of 16.6 pax/min and decreasing with a rate of 0.33 pax/min per
minute. This behavior is shown in Fig. 2, where the median of the arrival rate is additionally covered
by the 25% and 75% quantiles.
1
3
5
7
29
27
25
23
......
seat row
front door
rear door
IV
III
II
I
back to front
boarding
seat block
PREPRINT
Dynamic change of aircraft seat condition for fast boarding
December 2017. Transportation Research Part C: Emerging Technologies 85:131-147
DOI: 10.1016/j.trc.2017.09.014
Figure 2: Field measurements show a declining arrival rate of passengers at the aircraft door during the boarding
progress (Schultz 2017)
Besides the arrival rate, two other relevant input parameters are measured in the field: the time to
store the individual hand luggage and the time need to take the seat (including the seat change
process, if seats are occupied in between – seat shuffle, cf. Bazargan 2006, Schultz et al. 2008). To
calculated the time t for storing the hand luggage, a Weibull distribution is calibrated W (1) with
=
1.7,
= 16.0 s, and tmin = 0.
(1)
The seat shuffle is defined by the movements needed for all involved passengers, where each
single movement is defined by a triangular distribution with 1.8s, 2.4s, and 3s for tmin, tmode, and tmax
respectively (Schultz 2013). Finally, the walking speed of the passenger in the aisle is set to 0.8 m/s
(60% of average walking speed (Schultz 2011)). With these calibrated input parameters, the boarding
model achieves a deviation of ±5% compared with the operational measurements of boarding times
(Schultz 2017). Using the calibrated boarding model, several boarding scenarios could be evaluated
against their average value and the standard deviation of the boarding time. Due to the stochastic
nature of the boarding model, each scenario has to be simulated 100,000 times to derive reliable
results.
A fast boarding strategy has to efficiently manage two relevant processes: the storing of hand
luggage in the overhead compartment and the seating process. Both processes result in a blocked aisle
situation which directly affects other passengers. Commonly used boarding strategies are defined by a
specific sequence of passengers (and their seats) which is mainly driven by an airline-specific block
arrangement (defined by number and size of blocks, cf. Fig. 1) and a call-in procedure (such as board
from back to front). The major boarding sequences are based on three basic elements: individual
seating, categorization by seat rows, and categorization by seat position (window, middle, aisle). In
Fig. 3, three representative boarding sequences using these basic elements are shown on the left side:
random, block, and outside-in. In the center of Fig. 3, two derivative strategies are shown.
0
4
8
12
16
20
0 4 8 12 16
arrival rate (pax/min)
boarding time (min)
Q.25%
median
Q.75%
^
^
PREPRINT
Dynamic change of aircraft seat condition for fast boarding
December 2017. Transportation Research Part C: Emerging Technologies 85:131-147
DOI: 10.1016/j.trc.2017.09.014
Figure 3: Representation of different boarding strategies and operational constraints in the boarding model
(boarding in order of the grey scale value, dark first)
The reverse pyramid boarding sequence is composed of the block and the outside-in sequence: in
particular, six blocks are boarded from back-to-front and in each block the seats are additionally
sorted from window to aisle seats. The optimized block strategy is based on the major block strategy,
but the blocks are called in a more efficient way (Schultz et al. 2008, 2013). In the further evaluations,
the optimized block, back-to-front, and reverse pyramid strategies are based on the six-block
arrangement. The grid layout also allows an efficient implementation of the identified operational
constraints, such as pre-/priority boarding, passenger conformance regarding the boarding strategies
(e.g. late arrivals), specific seat load factor, or the consideration of individual group constellations.
The statistical result of a simulated boarding scenario consists of the average boarding time, the
standard deviation (SD), and quantiles of the boarding time. Fig. 4 shows the probability density
function (PDF) of 100,000 simulation runs using a random boarding strategy (with the default values
for seat load 85% and conformance rate 85%). The simulated values are marked with a triangle and
the corresponding normal distribution with a solid line. Additionally, the 10%, 25%, 50% (median),
75% and 90% quantiles indicate the slight positive skewness of the boarding time. The average
boarding time for the random boarding is defined with 100% and will be the reference boarding time
for the following evaluations, the corresponding standard deviation is 7.3%, and the values for the
boarding time quantiles are 91%, 95%, 100%, 105% and 110% respectively.
PREPRINT
Dynamic change of aircraft seat condition for fast boarding
December 2017. Transportation Research Part C: Emerging Technologies 85:131-147
DOI: 10.1016/j.trc.2017.09.014
Figure 4: Comparison of simulation results and corresponding normal distribution ( = 1, = 7.3%)
In Fig. 5, the result of a sensitivity analysis shows a converging behavior of the average boarding
time with an increase of the passenger arrival rate. At an arrival rate of 16 pax/min, each additional
passenger arriving at the aircraft will decrease the boarding time by less than 0.25%.
Figure 5: Comparison of simulation results and corresponding normal distribution ( = 1, =
7.3%)
To evaluate the introduced boarding strategies of back-to-front, optimized block, outside-in, and
reverse pyramid, a sensitivity analysis with regard to the passenger arrival rate is exemplarily shown
in Fig. 6 (cf. Schultz 2013, Qiang et al. 2014, Zeineddine 2017). All of these strategies point out the
same saturation behavior with an increased arrival rate but at different levels. Since the back-to front
strategy is only favorable for a maximum of three blocks (cf. Schultz et al. 2008, Bazargan 2013), the
implemented six-block arrangement increases both boarding time and standard deviation (SD). The
optimized block strategy results are slightly faster and both outside-in and reverse pyramid lead to
significantly faster aircraft boarding. At an arrival rate of 40 pax/min, the boarding times are 109%,
98%, 92%, 78%, and 72% (SD: 8.1%, 7.4%, 7.0%, 5.8%, and 5.4%) for the back-to-front, random,
optimized block, outside-in, and reverse pyramid strategies respectively. In contrast to the shape of the
average boarding time, the standard deviation exhibits a minimum between arrival rates of 10 and 20
passengers per minute. The minimum SD is located at an arrival rate of 10, 13, 26, 14, and 17 pax/min
for back-to-front, random, optimized block, outside-in, and reverse pyramid.
0
3
6
70 85 100 115 130
PDF (%)
simulation results
normal distribution
Q.10%
Q.25% Q.50%
Q.75%
Q.90%
80
100
120
140
160
180
200
010 20 30 40
boarding time (%)
arrival rate (pax/min)
Q.90%
Q.10%
Q.75%
Q.25%
Q.50% (median)
PREPRINT
Dynamic change of aircraft seat condition for fast boarding
December 2017. Transportation Research Part C: Emerging Technologies 85:131-147
DOI: 10.1016/j.trc.2017.09.014
Figure 6: Evaluation of different boarding strategies with increasing arrival rate (one-door configuration)
All introduced boarding strategies primarily focused on a reduced interaction during the seating
process. The most favorable seating process starts with the window seat, followed by the middle and
aisle seats. If this process runs in reverse, the maximum quantity of interactions in the seat row is
reached and results in a significant interference with the passengers waiting in the aisle. There are four
different ways to further improve the boarding process (Schultz 2017): implementation of more
complex boarding procedures (from 6 blocks to the arrangement of 174 individual seats (Steffen
2008), luggage constellations (Milne and Kelly 2014) or dynamic approaches (Zeineddine 2017)), use
of the second aircraft door (Marelli et al. 1998, Schultz et al. 2008, Nyquist and McFadden 2008), and
the efficient consideration of hand luggage (Qiang et al. 2014), and the introduction of infrastructural
changes (Schultz et al. 2013, Fuchte 2014). Since individual seating results in a 35% faster boarding
time, the use of the second aircraft door holds the potential of 30% faster boarding and is more or less
standard procedure for apron positions (bus shuttle, walk boarding). However, the use of the rear
aircraft door at apron positions will eliminate the prior ensured sequence of passengers. To ensure a
nearly unimpaired boarding sequence for A320/B737 types, a gate position with an additional aircraft
entry is necessary, such as the over-the-wing bridge (FMT 2017). Depending on the specific airline
boarding procedure, the reduction to strictly one piece of hand luggage per passenger results in 5%-
15% savings and the avoidance of suitcases (e.g. only allow few shoulder bags) results in 20%-25%
savings in the boarding time. The infrastructure (e.g. size of aircraft doors, width of cabin aisle) today
still limits the boarding time in the case of high arrival rates (>20 passengers per minute, cf. Fig. 6).
As an example, if the arrival rate for the optimized block strategy increases by 25%, 50%, and 100%,
the savings only increase by 2%, 3%, and 4% respectively. Furthermore, the analysis of data from
field measurements demonstrates that the arrival rate substantially depends on the passenger behavior
(e.g. late arrivals, abilities to scan the boarding ticket, cf. Fig. 2). Changes in the aircraft design
through the provision of a double-sized door in the center of the aircraft could result in savings for the
boarding process (two-door approach), but are not favorable in the context of structural integrity of
the aircraft. A highly promising infrastructural approach to improving the aircraft boarding is the
Side-Slip Seat (Molon Labe Seating 2017), which will now be analyzed in detail.
510 15 20 25 30 35 40
50
75
100
125
150
175
200 arrival rate (pax/min)
boarding time (%)
back-to-front
random
optimized block
outside-in
reverse pyramid
5
6
7
8
9
10
510 15 20 25 30 35 40
SD (%)
arrival rate (pax/min)
PREPRINT
Dynamic change of aircraft seat condition for fast boarding
December 2017. Transportation Research Part C: Emerging Technologies 85:131-147
DOI: 10.1016/j.trc.2017.09.014
3 INFRASTRUCTURAL CHANGES – SIDE SLIP SEAT
Standard approaches to accelerating the boarding process mainly address the management of
passenger behavior by generating boarding sequences or reducing the amount of baggage. In the prior
evaluation, only the use of a second door to board the passengers could be understood as a significant
change in the infrastructure of the day of operations. The most prominent effect on the boarding time
is accompanied with a blocked aisle due to passengers storing baggage or entering their seat row.
With the innovative technology of the Side-Slip Seat, the available infrastructure could be
dynamically changed to support the boarding process by providing a wider aisle which allows two
passengers to pass each other in a convenient way. Two additional benefits come with this new
technology: the wider aisle allows airlines to offer full-size wheelchair access down the aisle (meeting
the needs of disabled passengers (Chang and Chen 2012, Reinhardt et al. 2013, Holloway et al. 2015,
Davies and Christie 2017)) and the middle seat is 2 inches wider than the aisle and window seats
(aisle and window seats retain their standard width). In Fig. 7, the design of the Side-Slip Seat is
shown. In the “boarding-ready” position, the aisle seat is hand-pushed half over the middle seat,
which is in a staggered position. To unfold the seat to the “aircraft-ready” position, the passenger
simply grabs the seat at the side and pulls the seat out into a locked position. All seats have to be in
this position before the aircraft starts. To bring the seat back to the initial position, the passenger/crew
has to press a mechanical button on the side, which releases the locked position and allows pushing of
the aisle seat over the middle seat.
Figure 7: Fundamental design approach of the Side-Slip Seat (Molon Labe Seating, 2017)
If a passenger wants to sit on the window seat, the Side-Slip Seat could stay folded, but when
taking a seat on the middle or aisle seat, it is necessary (recommended) to unfold the seat. In any case,
the Side-Slip Seat finally has to be unfolded when boarding ends and the aircraft is ready to leave the
gate/apron position. In Fig. 8, two configurations of the Side-Slip Seat are shown, which highlight the
potential benefit of a wider aisle.
Figure 8: Side-Slip Seat provides a wider aisle for boarding: seat in initial condition (left) and unfolded
operational condition (Molon Labe Seating 2017)
The developed boarding model is adapted to allow the parallel movement of two passengers along
the aisle. Furthermore, the dynamic status of the seat row (folded/unfolded) is implemented to
PREPRINT
Dynamic change of aircraft seat condition for fast boarding
December 2017. Transportation Research Part C: Emerging Technologies 85:131-147
DOI: 10.1016/j.trc.2017.09.014
enable/disable the parallel movement. If both sides of the aisle are in the initial folded condition, a
second passenger can pass without reducing the walking speed. If only one side is folded, the walking
speed is reduced to 50%. If both sides are used by passengers and the Side-Slip Seats are unfolded,
only one passenger is allowed to move in the aisle. The different seat conditions are shown in Fig. 9 in
the simulation environment: in the first row, both sides of the seat rows are folded, in the second row,
both sides are unfolded, and in the third row, the sides of the seat row possess different conditions. At
this row, the orange-marked passenger wants to take the window seat on the right but the middle seat
is blocked. This situation requires a seat shuffle which would fully block the aisle with the standard
seat, but only partially blocks the aisle with the Side-Slip Seat installed.
Figure 9: Implementation of the Side-Slip Seat in the boarding model and simulation environment. A color code
indicates the level of inference, caused by necessary seat changings (green = no interference, red = highest level
of interference)
To cover the potential operational behaviors of the passengers, two usage scenarios are taken into
consideration: if a passenger wants to sit on the middle/aisle seat, he stores the hand luggage first and
then unfolds the seat or the other way around. Due to the lack of operational data, the scenarios are
assumed to be equiprobable. As expected, the effect of passing passengers in the aircraft aisle allows
for significant savings in the average boarding time. The standard deviation exhibits a more complex
behavior. In comparison to the prior analysis (see Fig. 6), the minimum of SD using the Side-Slip Seat
exhibits a more prominent characteristic (Fig. 10) for the back-to-front, random, and optimized block
strategies in particular. The higher standard deviations are caused by the additional seat row states.
The minimum SD is now located at arrival rates of 12, 14, 15, 22, and 27 pax/min for back-to-front,
random, optimized block, outside-in, and reverse pyramid. These prominent minima are caused by the
fact that even in a wider aisle, blocking situations occur, where each specific boarding procedure
exhibits an optimal operational point. At an arrival rate of 40 pax/min, the boarding times are 86%,
78%, 74%, 53%, and 42% (SD: 8.5%, 7.8%, 7.1%, 5.5%, and 4.4%) for the back-to-front, random,
optimized block, outside-in, and reverse pyramid strategies respectively. Even the fastest boarding
sequences will significantly benefit from the Side-Slip Seat. Finally, considering an operationally
relevant arrival rate of 14 pax/min, the application of the Side-Slip Side will result in additional time
savings for the aircraft boarding of 19%, 11%, 12%, 8%, and 11% for the back-to-front, random,
optimized block, outside-in, and reverse pyramid boarding strategies, without significant changes of
the standard deviation (± 0.25%).
PREPRINT
Dynamic change of aircraft seat condition for fast boarding
December 2017. Transportation Research Part C: Emerging Technologies 85:131-147
DOI: 10.1016/j.trc.2017.09.014
Figure 10: Side-Slip Seat - Evaluation of different boarding strategies with increasing arrival rate (one-door
configuration)
These first simulation results are very promising as regards achieving significant improvement of
the aircraft boarding time due to the dynamic change of the cabin infrastructure. But neither the
outside-in nor the reverse pyramid strategy is operationally considerable, since the pre-sorting effort
at the boarding gate is too high (e.g. specific sorting areas, or call-in of too many groups) and
inconvenient from the passenger perspective. The introduction of new infrastructure approaches often
demands adjusted operational procedures. It is expected that a customized boarding strategy may
greatly benefit from the Side-Slip Seat. Instead of the systematic testing of conceived strategies,
evolutionary algorithms will be used to identify an appropriate boarding strategy.
4 EVOLUTIONARY ALGORITHM
Since the prior analysis focuses on specific boarding sequences, the question arises as to whether
there is an optimal solution for achieving a minimum boarding time with the given requirements of
the Side-Slip Seat. It has already been shown that specific strategies such as outside-in or reverse
pyramid hold the potential of further improvements of the commonly used block boarding. A
systematic evaluation of the block strategy exhibits that an alternation of seat blocks (cf. optimized
block, Fig. 3) reduces the boarding time (Schultz 2013). If one passenger stores his baggage,
passengers with a smaller seat row number are able to start their storing and seating process in
parallel. A gap between these passengers ensures a minimum of potential interferences in the aisle.
Consequently, a combination of these effects in one approach - board from outside to aisle, from back
to front and use alternating sequence - should result in an optimal boarding sequence and a minimal
boarding time. From a combinatorics point of view, three different blocks can be arranged in six
different sequences (123, 132, 213, 231, 312, 321), six blocks in 720 sequences and 174 individual
passengers result in 174! = 6.42x10315 different sequences. It is not possible to check all existing
sequences in an appropriate amount of time to prove the optimality of a given sequence. One
approach towards finding the optimal sequence is to apply evolutionary algorithms to explore the
problem space efficiently. The idea behind this approach is to start with a set of possible solutions
510 15 20 25 30 35 40
25
50
75
100
125
150
175 arrival rate (pax/min)
boarding time (%)
random
outside-in
back-to-front
optimized block
reverse pyramid
4
5
6
7
8
9
510 15 20 25 30 35 40
SD (%)
arrival rate (pax/min)
PREPRINT
Dynamic change of aircraft seat condition for fast boarding
December 2017. Transportation Research Part C: Emerging Technologies 85:131-147
DOI: 10.1016/j.trc.2017.09.014
(population) and allow them to evolve over time using biologically-inspired processes of selection,
heredity (cross-over and mutation) and replacement of least-fit population.
4.1 Definition of Fitness Function
In a first step, a set of valid sequences have to be provided: in the case of boarding of the reference
A320, each sequence has to contain all aircraft seats (no double entries). Then, the sequence has to be
simulated to define the average boarding time (
bt) and standard deviation (
bt). In the case of
identifying an efficient boarding strategy, these two values are summed up to a value of fitness F (2),
because a boarding strategy with a smaller standard deviation is preferable to a sequence with a same
average boarding time but higher standard deviation.
(2)
Hence, the fundamental approach for passenger behavior is based on a stochastic movement
model, hand luggage amount and storage time as well as seat interactions; a minimum of simulation
runs is needed to calculate the average boarding time and standard deviation. In Fig. 11, the effect of a
higher number of calculations is shown. If only a small number of simulation runs are aggregated to
one exercise (one boarding scenario), the result for the expected value of the boarding time
exercise,bt
(3) and accompanied standard deviation
exercise,bt (4) exhibit significant differences when running the
same exercise again.
(3)
(4)
In Fig. 11, the distribution of the boarding time (random strategy) in 100 exercises is shown using
50, 200, 800, 3200, 6400 and 10000 simulation runs per exercise. Even at 10000 simulation runs per
exercise, the results differ by ±0.1% from the average boarding time (not every exercise returns the
same value for average boarding time). That implies a poorer boarding sequence could still perform
better than a superior sequence and could be rated as a least-fit sequence according to F.
Figure 11: Distribution of boarding times of 100 exercises using different numbers of simulation runs per
exercise (from 50 to 10,000)
0
20
40
60
rate (%)
10000
6400
3200
0
10
20
-0.4 -0.2 100.0 +0.2 +0.4
rate (%)
boarding time (%)
800
200
50
PREPRINT
Dynamic change of aircraft seat condition for fast boarding
December 2017. Transportation Research Part C: Emerging Technologies 85:131-147
DOI: 10.1016/j.trc.2017.09.014
It is assumed that a sufficient amount of simulation runs should lead to
excercise,bt = 0, so that
excercise,bt =
bt and each exercise with the same input parameters shows the same result. To find an
appropriate number of calculations needed, both the required precision and the calculation time have
to be taken into consideration. At 10,000 simulation runs per exercise, the calculation time reaches a
level of 18 s per exercise to determine the value of fitness for one specific boarding sequence (Fig.
12).
Figure 12: Comparison of required precision of calculation and calculation effort
Using the value of 500 runs per exercise, a fair trade-off between precision (by means of standard
deviation) and simulation time is reached: standard deviation of
excercise,bt = 0.19% and calculation
time per exercise is 0.9 s. The fitness value F of the boarding sequences can now be evaluated and
ranked accordingly in an ascending order (smaller values represent a higher fitness regarding the task
of searching for a boarding sequence with a minimum boarding time).
4.2 Recombination of Sequences
After compiling a set of possible solutions (population) and defining a selection approach (fitness
function), evolutionary algorithms require the development of a procedure for heredity (algorithm for
replication and mutation), in order to derive a new generation of population. In this context, only the
part of the population with the best fitness is used for replication and the least-fit population will be
replaced by the next generation. The underlying idea of biological replication uses specific sequences
of each parental gene and recombines them to form new gene sequences. Finally, the new generation
of children contains partial sequences of the involved parents. In the context of aircraft boarding, this
simple approach of recombination will fail using the uniform crossover principle. The result of the
uniform crossover is shown in Fig. 13, where the stochastic recombination of the elements of two
parents (grey, white) results in two children. Applying this recombination to an underlying sequence,
the created sequence will not necessarily contain all basic elements of the sequence, as Fig. 13
exhibits with an exemplarily used sequence from one to six.
Figure 13: Recombination using uniform crossover fails if applied to sequences
0
10
20
0
0.5
1
1.5
10 100 1000 10000
run time (s/100)
standard deviation
bt (%)
runs per exercise
variation of boarding time
calculation effort
uniform crossover
1 2 3456
12
3
4
56
1
23
4
5
6
1
2
3
4
5
6
uniform crossover - sequence
invalid sequencevalid sequence
PREPRINT
Dynamic change of aircraft seat condition for fast boarding
December 2017. Transportation Research Part C: Emerging Technologies 85:131-147
DOI: 10.1016/j.trc.2017.09.014
The uniform order-based crossover (Davis 1991) uses the same initial approach of the stochastic
recombination by introducing a bitmask to assign elements of the parents to the two children. If the
bitmask value is zero, the element from the first parent is transferred to the first child; if the value is
one, the element from the second parent is transferred to the second child. In the following
intermediate step (see Fig. 14, lower right), the missing elements are filled up in the order in which
they appear at the other parent. As an example, the elements 2-3-5 are missing at the first child (grey)
and appear at the second parent in the order 5-3-2, which results in the final sequence of 1-5-3-4-2-6
for the first child.
Figure 14: Recombination using order-based crossover approach for considering sequences
In the context of the travelling salesman problem, the edge recombination (Whitley 1989) is
another approach towards solving the identified combinatorial task. Both approaches are implemented
and in the following analyses the uniform order-based crossover approach is used, since it requires a
smaller calculation effort. After building up a population, defining the fitness function and modelling
the replication process, the final step will be to set up an algorithm to allow for a mutation (minor
changes in the sequence caused by random processes) to continuously influence the evolution of the
population. Two typically used approaches for mutation are the swapping and the insert approach.
The swapping algorithm exchanges the position of two elements of the sequence and the insert
algorithm removes one element of the sequence and inserts this element in a different position (Fig.
15).
Figure 15: Mutation algorithms to swap and insert elements in the existing sequence
4.3 Test Implementation of the Evolutionary Algorithm
To test the evolutionary algorithm for optimizing the boarding sequence, an initial population is
created containing randomly chosen boarding sequences. As a starting point, these boarding
sequences follow no specific strategy. To apply the evolutionary algorithm, rules have to be defined
for selection and heredity of the fittest sequences and replacement of the least-fit part of the boarding
sequence. To start the algorithm, additional parameters (e.g. mutation rate) and the specific progress
of the evolutionary algorithm have to be defined. The computation procedure for one evaluation run
(tournament) in the test implementation is defined by five repeating steps as follows.
1. The start population consists of 100 randomly chosen boarding sequences.
2. The fitness for each sequence is calculated with 500 simulation runs per sequence
3. The 25 fittest sequences are used to create the child population with 50 children
uniform order-based crossover
1 2 3456
12
3
4
56
146
24
6
1 1 1
0 0 0
stochastic bitmask
crosswise sequencing
valid sequence
53
235
1135
2
3
5
5
3
2
3
5
1
1 4 6
24
65
3
2
3
5
1
1 4 6
2
3
51 4 6
2351 4 6
2
3
51 4 6
2
35
mutation - swapping mutation - insert
PREPRINT
Dynamic change of aircraft seat condition for fast boarding
December 2017. Transportation Research Part C: Emerging Technologies 85:131-147
DOI: 10.1016/j.trc.2017.09.014
4. The new generation consists of the 25 fittest and 50 child sequences and each sequence is
mutated with a probability of 0.2% (mutation insert).
5. 25 random sequences are added to the new population.
In Fig. 16, the converging progress of the evolutionary algorithm is shown. After 40,000
tournaments, the boarding time of the best sequence is 81% with a standard deviation of 6.1%, which
is still slower than the outside-in boarding strategy. The steps in the fitness function progress show
several aspects: stagnation, improvements, and degradation (caused by mutation).
Figure 16: Progress of the boarding sequence development and evaluation. The solid line depicts the values for
the fittest boarding sequences
The basic principles of the evolutionary algorithm are now described and tested in a first
implementation. These tests with the simulation environment demonstrate that after 6.5 hours of
calculation time with more than 25,000 tournaments (12.5 million simulated boarding sequences), the
evolutionary algorithm still identifies fitter sequences. Since the simulation environment was initially
developed as a comprehensive tool to cover all operational aspects of the aircraft boarding process
(e.g. data analysis, visualization, testing), it is expected that changes in the implementation of the
boarding model will not result in an absolutely essential increase of the calculation speed by at least
one order of magnitude. To benefit from the evolutionary algorithm, a faster model of the boarding
process is mandatory.
4.4 Development of a Fast Sequence Evaluation Model for Boarding
To enable a faster calculation, some assumptions of the comprehensive, stochastic boarding approach
have to be ignored, if these assumptions will not affect the efficiency of the boarding sequence
(fitness). The following elements will not be included, since they are randomly distributed and result
in average values at high numbers of calculations: hand luggage distribution, walking speed, arrival
rate, conformance rate, seat load factor. Primary elements of the fast boarding model are the seat
shuffle and the influence of the blocked aisle due to the seating process. Therefore, the aircraft is
parted into a left and a right side; for each side, a status array consisting of all rows is implemented.
Each row status Srow aggregates the status of three accompanied individual seats Sseat, n (window,
middle, aisle), with 0 if seat is free and 1 if seat is occupied, to an integer value (5).
(5)
The status demands a specific amount of movement of the involved passengers in order to solve
the seating process (see Schultz 2013). Each status directly results in an amount of movements; only
the case of an occupied middle seat additionally depends on the arriving passengers. If the passenger
takes the aisle seat, one movement is necessary; if the window seat is used, five movements are
0
10
75
100
125
150
0246810
standard deviation (%)
time (%)
tournaments (x 10,000)
fitness
boarding time
standard deviation
PREPRINT
Dynamic change of aircraft seat condition for fast boarding
December 2017. Transportation Research Part C: Emerging Technologies 85:131-147
DOI: 10.1016/j.trc.2017.09.014
necessary to unblock the aisle. If all seats are occupied, the seat row status reaches a value of seven
(22 x 1 + 21 x 1 + 20 x 1 = 7); if only the middle seat is free, the status value is five (22 x 1 + 21 x 0 +
20 x 1 = 5).
To cover the impact of a blocked aisle, the number of influenced passengers has to be defined. In
the boarding sequence, a passenger is marked as influenced if he/she is positioned later in the
sequence and has a higher seat row number. This influence is expected to have a decreasing behavior
(decreasing with a higher distance dx in the sequence between passenger i and j) and will be modelled
with an exponential function (6).
(6)
The factor A is defined as a switch with the value of 1 if the passenger is influenced and 0 if not.
Factor B is a scaling factor and set to the value of 0.4, which results in a suitable decreasing behavior
(influence on the next passenger in the sequence is 33% lower). The seat row status and the influence
calculation are summarized to a new fitness function (7), where Si is defined by the seat row of
passenger i and side of the seat location (left, right).
(7)
4.5 New Implementation of the Evolutionary Algorithm
Finally, the following eight steps are implemented to run one tournament using the developed fast
sequence evaluation. This procedure was developed through a continuous improvement and the
testing of different approaches in order to overcome major problems of convergence, stability, and
computational effort, e.g. the population has to receive a balanced number of new random sequences
and double entries (twins) should be excluded. Furthermore, the mutation rate should have a dynamic
behavior. A high mutation rate supports the search for fitter boarding sequences, but prevents a
convergence to a specific boarding sequence. On the other hand, if one appropriate sequence (or set of
sequences) is identified, the algorithm tends to stay in local minima of the problem space. The
approach of the dynamic, variable mutation rate is comparable to the annealing approach of the
simulated annealing (cf. Youssef et al. 2001).
1. The start population consists of 100 randomly chosen boarding sequences.
2. The fitness F of all sequences is calculated.
3. The best 25 sequences are mutated with a probability of pm and stored (mutation insert).
4. These 25 sequences are used to create the child population (50), where the second parent is
randomly chosen from the least fit sequences.
5. Each of the 50 child sequences is mutated with a probability of 0.1% (mutation insert).
6. 15 of the fittest sequences are copied and mutated (one random mutation insert process).
7. 10 random sequences are added to the new population.
8. Double entries in the list of sequences are deleted and replaced with random sequences.
The new population consists of the 25 fittest survivors, 50 children, 15 mutants, and 10 new
sequences. The fitness of the tournament is defined by the best fitness value of the 100 sequences (in
the case of the fitness function (7): smaller values point out a better fitness). The mutation probability
pm starts with a value of 0.2% and decreases with every tournament by 0.01% if the fitness of the
current tournament is better than the average fitness over the last 100 tournaments (stabilizing the
calculation). If the average fitness over the last 100 tournaments is equal to the fitness of the current
tournament, pm increases by 0.02% (‘heating’ to create disorders). In Fig. 17, the progress of the
evolutionary algorithm is shown with clear local minima as well as the effect of the dynamic mutation
rate.
PREPRINT
Dynamic change of aircraft seat condition for fast boarding
December 2017. Transportation Research Part C: Emerging Technologies 85:131-147
DOI: 10.1016/j.trc.2017.09.014
Fig. 17: Progress of the evolutionary algorithm with fitness function (above) and dynamic, variable mutation
rate (below)
The fast evaluation model for aircraft boarding now reaches a calculation speed of 325 s for 12.5
million boarding sequences, which is approx. 72 times faster than the comprehensive boarding model.
4.6 Application of the Evolutionary Algorithm
Several minor modifications of the algorithm were developed during the simulation runs
considering the expertise gained from the simulation results. The first adaptation was to distinguish
between different levels of the mutation rate. If the mutation rate is higher than 0.1%, the change rate
is set to one order higher. Furthermore, a different change rate for increasing/decreasing the mutation
rate exhibits no beneficial behavior (e.g. faster convergence or higher stability). If the fitness values
start to converge, the value of 15 mutated parents is increased stepwise to 25 and the insertion of
random sequences is finally only needed if double entries occur (see steps 6, 7 and 8 of the
algorithm). In Fig 18, the progress of the evolutionary algorithm is shown using the arisen boarding
sequences for visualization. The disorder at the beginning (random boarding) changes to a regular
boarding sequence (comparable to individual boarding (Steffen 2008)). After 200,000 tournaments,
the pattern of the local minima indicates a nearly optimal sequence which prefers the window seats
first (followed by middle and aisle seats) and inner back-to-front order.
Fig. 18: Progress of the evolutionary algorithm with 200,000 tournaments using standard seats: faster boarding
(increasing fitness) from the left (random) to the right boarding sequence (individual). The order of boarding is
emphasized for visualization by three groups: dark gray to mid-gray followed by mid-green to light green
followed by light gray to white.
1.E-05
1.E-04
1.E-03
1.E-02
0 5 10 15 20
mutation
rate
tournaments (x 10,000)
fitness (a.u.)
random individual
40,000 80,000 120,000 160,000 200,000
tournaments
order of boarding
PREPRINT
Dynamic change of aircraft seat condition for fast boarding
December 2017. Transportation Research Part C: Emerging Technologies 85:131-147
DOI: 10.1016/j.trc.2017.09.014
As an outcome of different simulation trials, new individual boarding strategies could be derived
from the application of the evolutionary algorithm. In Fig. 19, three resulting individual boarding
strategies are shown, which are significantly faster than the prior evaluated boarding sequences
outside-in and reverse pyramid. The first individual strategy is also an outcome of the evaluation of
Steffen (2008). The second individual alternation strategy could also arise from a combination of the
outside-in and back-to-front strategies in combination with a separation of the sides of the aisle (left,
right). The third individual stack strategy represents a block boarding with a half-row structure
combined with an outside-in approach.
Fig. 19: Evolutionary algorithm - generation of individual boarding strategies
Even though the individual boarding sequence proposed by Steffen (2008) is a fast boarding
strategy (boarding time 65.8%), the identified individual alternation is faster (boarding time 63.7%,
see Tab. 1). The implementation and application of the evolutionary algorithm prove the capabilities
of this approach in efficiently solving the common boarding sequence problem by finding already
proposed boarding sequences, providing sequences to identify new boarding strategies, and finding
new sequences in a reasonable amount of time.
In a next step, the evolutionary algorithm will be applied to the Side-Slip Seat in order to identify
an appropriate boarding sequence which will benefit most from this technology approach. The fitness
function F is therefore adjusted by adding a value Ci which considers the condition of the aisle (8) due
to the folded/unfolded condition of the Side-Slip Seat: Ci = 0 if both sides of the seat row are folded
after the passenger takes the seat (passing the row without interactions), Ci = 1 if the Side-Slip Seat at
one side of the row is unfolded (minor interaction) and Ci = 2 if the seats of both sides are unfolded
(major/standard interactions, waiting).
(8)
The implementation of this fundamental behavior of the Side-Slip Seat leads to an efficient
boarding pattern, if the left or right side of the aisle is boarded first. Thus, the other side of the aisle
will stay folded for at least half of the time of the boarding progress and enable passengers to pass
each other. In Fig. 20, the evolutionary progress of the fittest boarding sequence is shown. The fastest
boarding is characterized by both the seat position (window and middle/aisle) and aisle position
(left/right), which results in four boarding zones. Due to the fact that passengers can pass each other,
there is no need for the inner back-to-front approach, as there was for the standard boarding (see Fig.
18).
To visually underline the left/right effect, Ci is slightly modified for the last boarding sequence
(Fig. 20, right) to enforce a differentiation between the left and the right side of the cabin aisle, so
Ci = 1.1 if unfolded left + folded right and Ci = 0.9 unfolded right + folded left.
individual (Steffen 2008)
order of boardingfront door rear door
individual alternation
individual stack
PREPRINT
Dynamic change of aircraft seat condition for fast boarding
December 2017. Transportation Research Part C: Emerging Technologies 85:131-147
DOI: 10.1016/j.trc.2017.09.014
Fig. 20: Progress of the evolutionary algorithm with 200,000 tournaments using Side-Slip Seats: faster boarding
(increasing fitness) from the left (random) to the right boarding sequence (separation of aisle and
window/middle seats as well as left/right side of the aisle). The order of boarding is emphasized for
visualization by three groups: dark gray to mid-gray followed by mid-green to light green followed by light gray
to white)
5 EVALUATION OF THE SIDE-SLIP SEAT
The evolutionary algorithm demonstrates that a boarding sequence which differentiates between
the left and the right side of the aircraft will benefit most from the Side-Slip Seat technology. In
Fig. 3, major and derivative boarding strategies are shown to illustrate the common process of
designing a boarding strategy. If the newly identified left/right (LR) pattern is applied to the random
and the block strategies, the random-LR and the block-ZIP strategies are derived (Fig. 21) and
implemented into the simulation environment.
Fig. 21: New boarding strategies considering the Side-Slip Seat: random-LR and block-ZIP (boarding in order of
the gray scale value, dark first)
In the following Tab. 1, the results of the evaluation of different boarding strategies according to
both average value and standard deviation of boarding time considering a standard seat and the Side-
Slip Seat are shown. Whereas the random-LR is not a favorable solution for the standard seat
environment (boarding time 99.1%), the Side-Slip Seat implementation results in a boarding time of
80.5%, which is nearly as fast as the outside-in strategy using the standard seat. Even the fast
boarding strategies block-ZIP, outside-in, and reverse pyramid significantly benefit from the Side-Slip
Seat, with a reduction of the boarding time by 15.1%, 11.9%, and 14.7% (standard deviation is
reduced as well). All individual boarding strategies possess a higher or similar standard deviation
which is, however, also accompanied with faster boarding. For the standard seat, the individual
alternation is the fastest boarding strategy and for the Side-Slip Seat, the application of the reverse
pyramid strategy leads to the fastest boarding time considering the operational constraints (e.g. seat
load or passenger conformance to the proposed strategy).
random left/right
40,000 80,000 120,000 160,000 200,000
tournaments
order of boarding
block-ZIP
random-LR
front door rear door
PREPRINT
Dynamic change of aircraft seat condition for fast boarding
December 2017. Transportation Research Part C: Emerging Technologies 85:131-147
DOI: 10.1016/j.trc.2017.09.014
Table 1: Evaluation of different boarding strategies considering a standard seat and the Side-Slip Seat
boarding
sequences
boarding time and standard deviation (%)
standard seat
Side-Slip Seat
boarding time
SD
boarding time
SD
back-to-front
109.2
8.1
88.5
8.1
random
(reference) 100
7.3
86.0
7.3
optimized block
95.3
6.9
81.7
7.2
outside-in
79.5
5.7
67.6
5.3
reverse pyramid
75.3
5.3
60.6
4.8
random-LR
99.1
7.1
80.8
6.6
block-ZIP
84.0
6.4
68.9
5.6
individual
65.8
4.8
61.7
5.0
individual stack
66.9
4.9
61.5
4.9
individual alternation
63.7
5.0
62.2
5.2
From an operational perspective, the random-LR boarding strategy appears to be a very promising
candidate with a low level of complexity (just a left and a right boarding block), no separation of
passengers sitting next to each other (e.g. families), and nearly 20% shorter boarding time. If the Side-
Slip Seat is applied to the commonly used three block boarding procedure, boarded from back-to-
front, the boarding time could be reduced by 18% with the Side-Slip Seat and the developed random-
LR strategy additionally provides a more stable boarding process (smaller standard deviation) at the
same efficiency level (Fig. 22).
Fig. 22: Probability density distribution of the commonly used back-to-front strategy (three blocks)
6 SUMMARY AND OUTLOOK
The paper provides a comprehensive analysis of the innovative approach of a Side-Slip Seat from
Molon Labe Seating (2017), which results in a dynamic change of the cabin seating layout during the
boarding progress. Since simplified evaluations already indicate a significant operational impact (less
passenger interference and short aircraft boarding), a reliable and calibrated stochastic boarding model
(Schultz 2008, 2013, 2017) is used for an in-depth analysis of the Side-Slip Seat. To cover the
-40 -20 0 20 40
reduction of boarding time (%)
back-to-front (three blocks)
Side-Slip Seat
random-LR and
Side-Slip Seat
PREPRINT
Dynamic change of aircraft seat condition for fast boarding
December 2017. Transportation Research Part C: Emerging Technologies 85:131-147
DOI: 10.1016/j.trc.2017.09.014
operational concept of the Side-Slip Seat, which allows passengers to pass each other, the boarding
model and the simulation environment were extended. The implementation of such fundamental
change inside the aircraft cabin demands an adapted boarding strategy in order to cover all the
benefits which accompany this new dynamic approach. To reasonably identify an efficient strategy,
the approach of using evolutionary algorithms was used to systematically search for an appropriate
sequence of passengers to board, instead of randomly combining basic elements of standard boarding
strategies (e.g. outside-in, back-to-front).
The stochastic boarding model already considers relevant aspects of the individual passenger
behavior and operational constraints, but has to be extended to cover the dynamic operational concept
of the Side-Slip Seat. At this stage, the simulation results demonstrate the potential of this innovative
approach by additional boarding time reductions of 10%-20% depending on the specific boarding
strategy, even though they show that the calculation time is too high to search for an appropriately
adapted boarding strategy with the stochastic boarding model. Therefore, a fast approach is developed
to identify boarding sequences, which holds the potential to significantly reduce the boarding time
when using the Side-Slip Seat. This approach only covers the passenger interactions during the
seating in a deterministic way and is approx. 72% faster than the comprehensive boarding model. In a
next step, this approach is included in a simulation environment using an evolutionary algorithm to
continuously improve an initial set of boarding sequences. The application of the evolutionary
algorithm identifies two fast boarding sequences, which results in individual boarding strategies with
an alternating and a stack pattern. The individual alternation and individual stack strategies are
evaluated with the stochastic boarding model and the individual alternation strategy is shown to be
faster than the individual boarding strategy proposed by Steffen (2008).
In the case of the Side-Slip Seat, the evolutionary algorithm depicts that operationally relevant
boarding strategies should differentiate between the left and the right side of the aisle (current
operational approaches prefer a boarding from the back to the front). Even under realistic, operational
boundary conditions (e.g. seat load, passenger conformance regarding the boarding sequence), the
average boarding time and standard deviation exhibit significant savings, by means of faster boarding
and smaller deviations of the boarding times (random-LR is 19% faster than random boarding). The
savings are comparable with the use of a second door (rear door) for passenger boarding (Schultz
2013) or the application of the outside-in strategy. The fastest boarding strategy using the Side-Slip
Seat is the reverse pyramid strategy with a boarding time of 60.6% (standard deviation 4.8%), which
is even faster than the individual stack strategy with a boarding time of 62.2% (standard deviation
4.9%).
After introducing a stochastic approach for aircraft boarding considering both individual
passenger behavior and operational constraints (Schultz 2008, 2013), a measurement and validation
campaign (Schultz 2017) and this investigation into infrastructural changes, two new topics in the
context of boarding will be focused upon: the online prediction of the boarding time using sensor
information from the connected aircraft cabin (e.g. seat occupation and queueing in the aisle) and the
SeatNow concept, which addresses operational improvements by efficiently replacing the current
standard call-in boarding procedure.
REFERENCES
Airbus. 2017. “Airbus A320 Aircraft Characteristics - Airport and Maintenance Planning”
Bachmat, E., and M. Elkin. 2008. “Bounds on the Performance of Back-to-Front Aircraft Boarding
Policies”. Operations Research Letters 35: 597–601.
Bachmat, E., D. Berend, L. Sapir, S. Skiena, and N. Stolyarov. 2009. “Analysis of Aircraft Boarding
Times”. Operations Research 57:499–513.
Bachmat, E., V. Khachaturov, and R. Kuperman. 2013. “Optimal Back-to-Front Aircraft Boarding”.
Physical Review E 87(6): 062805.
Bazargan, M. 2007. “A Linear Programming Approach for Aircraft Boarding Strategy”. European
Journal of Operational Research 183(1):394–411.
PREPRINT
Dynamic change of aircraft seat condition for fast boarding
December 2017. Transportation Research Part C: Emerging Technologies 85:131-147
DOI: 10.1016/j.trc.2017.09.014
Bernstein, N. 2012. “Comment on "Time Needed to Board an Airplane: a Power Law and the
Structure Behind it"”. Physical review. E, Statistical, nonlinear, and soft matter physics
86(2):023101.
van den Briel, M.H.L., J.R. Villalobos, G.L. Hogg, T. Lindemann, and A.V. Mule. 2005. “America
West Airlines Develops Efficient Boarding Strategies”. Interfaces 35:191–201.
Chang, Y.-C., and C.-F. Chen. 2012. “Meeting The Needs of Disabled Air Passengers: Factors That
Facilitate Help From Airlines and Airports”. Tourism Management 33(3):529–536.
Chung, Ch.-A. 2012. “Simulation Design Approach for the Selection of Alternative Commercial
Passenger Aircraft Seating Configurations”. Journal of Aviation Technology and Engineering
2:100–104.
Davies, A., and N. Christie. 2017. “An Exploratory Study of the Experiences of Wheelchair Users as
Aircraft Passengers - Implications for Policy and Practice”. IATSS Research 41(2):89-93
Davis, L. 1991. Handbook of Genetic Algorithms. New York: Van Nostrand Reinhold.
Eurocontrol/IATA/ACI. 2014. “Airport CDM Implementation Manual”, V.4
FMT. 2017. Over-The-Wing Bridge. www.fmt.se/airport/passenger-boarding-bridge/over-the-bridge
Ferrari., P., and K. Nagel. 2005. “Robustness of Efficient Passenger Boarding Strategies for
Airplanes”. Journal of the Transportation Research Board 1915:44–54.
Frette, V., and P.-C. Hemmer. 2012. “Time Needed to Board an Aircraft: A Power Law and the
Structure Behind It”. Physical Review E 85(1): 011130.
Fuchte, J. 2014. “Enhancement of Aircraft Cabin Design Guidelines with Special Consideration of
Aircraft Turnaround and Short Range Operations”, Ph.D. thesis, Technische Universitt
Hamburg-Harburg.
Holloway, C., R. Thoreau, E. Petit, and N. Tyler. 2015. “Time and Force Required for Attendants
Boarding Wheelchair Users Onto Aircraft”, International Journal of Industrial Ergonomics
48:167-173.
IATA. 2016. “Aircraft Handling Manual, 37th edition”, International Air Transport Association.
Kierzkowski. A., and T. Kisiel. 2017. “The Human Factor in the Passenger Boarding Process at the
Airport”, Procedia Engineering 187:348-355.
van Landeghem, H.V., and A. Beuselinck. 2002. “Reducing Passenger Boarding Time in Airplanes: A
Simulation Based Approach”. European Journal of Operational Research 142:294–308.
Li, Q., A. Mehta, and A. Wise. 2007. “Novel Approaches to Airplane Boarding”. The UMAP Journal
28:3-353-370.
Marelli, S., G. Mattocks, R. Merry. 1998. “The Role of Computer Simulation in Reducing Airplane
Turn Time”. Boeing Aero Magazine 1.
Milne, R.J., and A.R. Kelly. 2014., “A New Method for Boarding Passengers onto an Airplane”.
Journal of Air Transportation Management 34:93–100.
Milne, R.J., and M. Salari. 2016. “Optimization of Assigning Passengers to Seats on Airplanes Based
on Their Carry-on Luggage”, Journal of Air Transport Management 54:104–110.
Mirza, M. 2008. “Economic impact of Airplane Turn-times”. AERO Quarterly 4:14–19.
Miura, M., and K. Nishinari. 2017. “A Passenger Distribution Analysis Model for the Perceived Time
of Airplane Boarding/Deboarding, Utilizing an ex-Gaussian Distribution”, Journal of Air
Transport Management 59:44–49.
Molon Labe Seating. 2017. Side-Slip Seat. www.airlineseats.aero.
Notomista, G., M. Selvaggio, F. Sbrizzi, G. Di Maio, S. Grazioso, and M. Botsch. 2016. “A Fast
Airplane Boarding Strategy Using Online Seat Assignment Based on Passenger Classification”.
Journal of Air Transport Management 53:140–149.
Nyquist, D.C., and K.L. McFadden. 2008. “A Study of the Airline Boarding Problem”. Journal of Air
Transport Management 14:197–204.
Qiang, S.-J., B. Jia, D.-F. Xie, and Z.-Y. Gao. 2014. “Reducing Airplane Boarding Time by
Accounting for Passengers' Individual Properties: A Simulation Based on Cellular Automaton”.
Journal of Air Transport Management 40:42–47.
PREPRINT
Dynamic change of aircraft seat condition for fast boarding
December 2017. Transportation Research Part C: Emerging Technologies 85:131-147
DOI: 10.1016/j.trc.2017.09.014
Qiang, S.-J., B. Jia, R. Jiang, Q.-X. Huang, E. Radwan, Z.-Y. Gao, and Y.-Q. Wang. 2016.
“Symmetrical Design of Strategy-pairs for Enplaning and Deplaning an Airplane”, Journal of Air
Transport Management 54:52–60.
Reinhardt, L. B., T. Clausen, and D. Pisinger. 2013. “Synchronized Dial-a-Ride Transportation of
Disabled Passengers at Airports”. European Journal of Operational Research 225(1):106–117.
SESAR. 2014. “Concept of Operations Step 2 (Ed. 01.01.00)”, Single European Sky ATM Research
Programme.
Schultz, M., C. Schulz, and H. Fricke. 2008. “Efficiency of Aircraft Boarding Procedures”. In
Proceedings of the 3rd International Conference on Research in Air Transportation, 371–377.
Schultz, M., and H. Fricke. 2011. “Managing Passenger Handling at Airport Terminal”. In
Proceedings of the 9th USA/Europe Air Traffic Management Research and Development Seminar,
Berlin.
Schultz, M., T. Kunze, and H. Fricke. 2013. “Boarding on the Critical Path of the Turnaround”. In
Proceedings of the 10th USA/Europe Air Traffic Management Research and Development
Seminar, Chicago.
Schultz, M. 2017. “Aircraft Boarding - Data, Validation, Analysis”. In Proceedings of the 12th
USA/Europe Air Traffic Management Research and Development Seminar, Seattle.
Schmidt, M. 2017. “A Review of Aircraft Turnaround Operations and Simulations”. Progress in
Aerospace Sciences 92: 25–38.
Schmidt, M., P. Nguyen, and M. Hornung. 2015. “Novel Aircraft Ground Operation Concepts Based
on Clustering of Interfaces”. SAE Technical Paper 2015-01-2401.
Schmidt, M., P. Heinemann, and M. Hornung. 2017. “Boarding and Turnaround Process Assessment
of Single- and Twin-Aisle Aircraft”. 55th AIAA Aerospace Sciences Meeting, AIAA SciTech
Forum, (AIAA 2017-1856).
Soolaki, M., I. Mahdavi, N. Mahdavi-Amiri, R. Hassanzadeh, and A. Aghajani. 2012. “A New Linear
Programming Approach and Genetic Algorithm for Solving Airline Boarding Problem”. Applied
Mathematical Modelling 36:4060–4072.
Steffen, J.H. 2008a. “Optimal Boarding Method for Airline Passengers”. Journal of Air
Transportation Management 14:146–150
Steffen, J. H. 2008b. “A Statistical Mechanics Model for Free-for-all Airplane Passenger Boarding”.
American Journal of Physics 76:1114–1119.
Tang, T.-Q., Y.-H. Wu, H.-J. Huang, and L. Caccetta. 2012. “An Aircraft Boarding Model
Accounting for Passengers’ Individual Properties”. Transportation Part C: Emerging
Technologies 22:1–16.
Wald, A., M. Harmon, and D. Klabjan. 2014. “Structured Deplaning via Simulation and
Optimization”. Journal of Air Transportation Management 36:101–109.
Wang, K., and L. Ma. 2009. “Reducing boarding time: Synthesis of improved genetic algorithms”. In
Yu, F., J. Zeng, and G. Yue (Eds.), Proceedings of the 2009 international symposium on web
information systems and applications 359–362.
Whitley, L.D., T. Starkweather, and D. Fuquay. 1989. “Scheduling Problems and the Traveling
Salesman: the Genetic Edge Recombination Operator”. In Proceedings of 3rd International
Conference on Genetic Algorithms 133–140.
Youssef, H., S.M. Sait, H. Adiche. 2001. “Evolutionary Algorithms, Simulated Annealing and Tabu
Search: A Comparative Study”, Engineering Applications of Artificial Intelligence 14(2): 167–
181.
Zeineddine, H. 2017. “A Dynamically Optimized Aircraft Boarding Strategy”. Journal of Air
Transportation Management 58:144–151.
