2.1 Identification of pre-takeoff and post-landing taxi phases

The first step of the data collection process is to identify the pre-takeoff and post-landing taxi phases of the ADS-B data file. The pre-takeoff taxi phase is defined from the first broadcast row to the broadcast row which immediately precedes the first broadcast row within the data file that has a non-zero altitude (i.e. the aircraft has become airborne). The post-landing taxi phase is identified as the first broadcast row with a zero-altitude following the airborne rows of the data file to the last row of the data file.

2.2 Heading and speed change

The pre-takeoff and post-landing broadcast ADS-B data rows are then worked through row-by-row to identify whether the aircraft heading/direction and speed has changed with respect to the previous data row. Heading/direction changes, which are computed by comparing the aircraft heading of the current data row with the previous data row, are used to identify turns. Turns are characterised using the ‘compass’ visualised in Fig. 5, which enables the turn direction to be established using a range of conditional statements based upon the start and end values of the heading/direction change. It is assumed that the aircraft will take the shortest path around the compass, and this is considered an appropriate assumption due to typical airport geometries which result in individual turns usually being no greater than 180° (turns on airport taxiways are typically 45° or 90°). Following the application of the compass during row-by-row characterisation, an Identification (ID) code is applied to each row, either 11 (right turn), 12 (left turn) or 66 (straight taxi). Numeric ID codes, rather than strings or characters, are used to simplify the data processing.

Figure 5.

A visualisation of the compass used to characterise heading/direction changes in the ADS-B data file.

The row-by-row characterisation of the ADS-B data files also identifies the speed change between data rows. A deceleration is marked with the ID 42, whilst any increase in speed, or constant speed, is marked with ID 41. An example of a completed row-by-row characterisation is shown in Fig. 6 for a pre-takeoff taxi phase.

Figure 6.

The process of characterising heading/direction and speed changes in the ADS-B data file sourced from Flightradar24^®. Dataset reproduced with kind permission from Flightradar24^®.

Due to the potential for multiple ADS-B broadcasts every second⁽²³⁾, it is expected that individual turning and braking manoeuvres could be split across several broadcasts and therefore data rows. As a result, the ADS-B data file is again worked through row-by-row and manoeuvres that are adjacent and identical are combined. It should be noted that braking manoeuvres are only combined during straight taxi manoeuvres, as the data within the ADS-B data file cannot identify whether decelerations during turns are as a result of braking or the turning manoeuvre itself. The combination of adjacent manoeuvres results in the manoeuvre sequence shown in Fig. 7.

Figure 7.

The process of combining identical adjacent manoeuvres.

2.3 ADS-B data smoothing

During the development of the algorithms, it was observed that noise within the ADS-B data file could result in the identification of turns which were not performed when comparing the ADS-B data file with the Flightradar24^® visual ground track. It was identified that the incorrect identification of manoeuvres was as a direct result of noise within the ADS-B data files.

The first source of noise was found to be as a result of small fluctuations of heading/direction value from row to row, typically of 1° to 2°. These small deviations in heading would be counted as a turn, whilst comparison with the Flightradar24^® ground track would show that the aircraft was taxiing on a straight taxiway. Therefore, a minimum turn angle of 5° was defined as a threshold for identifying a turn manoeuvre, and this threshold successfully eliminated the false identification of manoeuvres from this source of ADS-B data noise.

Another source of noise was identified in ADS-B data files that showed the aircraft with zero speed but with a constantly changing heading/direction value (i.e. the aircraft was continually pivoting on the spot). Therefore, an additional algorithm was developed that would screen the ADS-B data files prior to row-by-row characterisation and remove elements of the ADS-B data files within which the aircraft heading changed while the aircraft speed was zero.

2.4 Characterisation of pushback entry, runway exit and turn onto stand

During ground manoeuvres, there are a series of ‘milestone’ manoeuvres that mark the transition of the aircraft from one ground taxi phase to the next⁽¹⁵⁾. The first of these is the pushback of the aircraft from the gate/stand, which often involves the aircraft performing a turn in reverse. Prior to the takeoff roll, the aircraft typically enters the runway by performing a turn, and this turn marks the end of the pre-takeoff taxi phase. Following the touchdown and landing roll, the aircraft vacates the active runway, and this turn represents the start of the post-landing taxi phase. Finally, the aircraft typically vacates a taxiway to turn onto the arrival stand or gate, and this represents the final ground manoeuvre of the flight. This section of the paper presents the algorithms used to extract and characterise each of these milestone manoeuvres. The purpose of extracting these manoeuvres is that they typically have different loads applied to the landing gear than standard taxiway turns and braking (e.g. a reversed turn for the pushback and higher turn speeds for the runway exit)^{(Reference Ladda and Struck2,Reference Buxbaum3,15)} .

2.4.1 Pushback

To identify the aircraft pushback, the aircraft must be observed to be travelling in reverse. As ADS-B data files only represent the magnitude of the aircraft ground speed, an alternative algorithm was required to identify whether the aircraft was undergoing pushback. To identify whether the aircraft was travelling in reverse, the latitude and longitude positions from the ADS-B data file could be used to compute the bearing in degrees (°) between two data rows. If this bearing was found to be greater than 90° different than the aircraft heading (i.e. the direction that the aircraft nose is pointing), it was assumed that the aircraft was travelling in a direction opposite to the aircraft nose and therefore pushback was occurring. In the current implementation of the algorithms, only the first manoeuvre in the pre-takeoff flight strip was assessed as to whether the pushback is occurring. This represents a current limitation of the pushback algorithm and is discussed in Section 7.1. Pushbacks are classified as ID 31 for a tail-right pushback or ID 32 for a tail-left pushback. The pushback identification algorithm is visualised in Fig. 8.

Figure 8.

A visualisation of the algorithm used to identify aircraft pushback from ADS-B data files. Original ground track reproduced with kind permission from Flightradar24^{® (23)}.

2.5 Generation of ground manoeuvre occurrence and sequence statistics

Following the identification of the milestone manoeuvres, the completed flight strip can then be assembled. An example of a completed flight strip is shown in Fig. 9 and represents the ground manoeuvres that occur for a flight and the sequence in which they occur. The start row and end row shown in Fig. 9 represent the original data rows and ADS-B broadcasts in the original ADS-B data file.

Figure 9.

A flight strip constructed by applying the algorithms to the ADS-B data file.

For the flight shown by the flight strip in Fig. 9, the number of pre-takeoff and post-landing turn and braking manoeuvres can be counted. This process, when repeated for multiple flights, enables the identification of how the number of turn and braking manoeuvres varies on a per-flight basis across a fleet of aircraft. In addition, over the repeated flights, a ‘tally’ for each milestone manoeuvre type and turn direction can also be recorded.

Finally, the manoeuvre sequencing for each flight strip can be identified. For each type of manoeuvre, a tally is kept of the manoeuvre that immediately follows it (except for the runway entry and turn onto stand, for which it is the manoeuvre preceding these manoeuvres that is of interest). From the tally values, the proportional share of the manoeuvres either following or preceding a certain type of manoeuvre can be computed.

3.1 Implementation of methodology

The algorithms described in Section 2 were implemented within a MATLAB^® environment, which read the ADS-B data files and applied the algorithms to produce a corresponding flight strip for each ADS-B data file. An additional MATLAB^® script was then used to store the ‘tally’ values from each flight strip to produce the ground manoeuvre occurrence and sequence statistics.

3.2 Demonstration dataset

To demonstrate the data collection and processing methodology, a wide-body civil aircraft that is currently in production was selected. A wide-body civil aircraft was selected due to the availability of existing data on wide-body aircraft ground manoeuvres (albeit limited with respect to variability) from Ladda and Struck^{(Reference Ladda and Struck2)} and the FAA statistical loads program⁽¹⁵⁾. At the time of writing, the selected wide-body civil aircraft performs approximately 2,500 individual flights/sectors a week⁽²³⁾. The selected wide-body civil aircraft currently operates into and from both small and medium-sized airports, along with large international airports. The selected wide-body civil aircraft also operates globally across multiple operators, as highlighted by the route network shown in Fig. 10(a).

Figure 10.

(a) The route network for the wide-body civil aircraft included in the demonstration dataset⁽²³⁾. (b) A demonstration of the fidelity of the ADS-B ground tracks, which trace out the key features of an airport geometry.

The route network for each operator of the wide-body civil aircraft was identified using Flightradar24^® and the ADS-B coverage was evaluated at each departure and arrival airport within the route network⁽²³⁾. It was observed that limited reception was available at certain airports (especially at stand/gate locations). As this limited reception resulted in inaccurate ground tracks, these routes were omitted from the dataset. The aircraft utilisation on the remaining routes was then identified. If the wide-body civil aircraft was used on a given route five times in a week, five ADS-B data files for the route were randomly selected from across the period of a single year. Random sampling from across the period of a year was performed with the intention of capturing the aircraft fleet operating in different seasons, to ensure that the effect of environmental conditions and weather (e.g. rain and snow) on aircraft ground manoeuvres would be captured. The random sampling of flights in a manner that is proportional to the actual aircraft utilisation in-service was performed to produce ground manoeuvre statistics that would be representative of the entire wide-body civil aircraft fleet (i.e. all aircraft of a given type in-service, also known as the manufacturer’s fleet). In addition, random sampling across all operators in proportion to the weekly utilisation of the aircraft ensured that various taxiway and runway layouts and surface conditions were captured, whilst also accounting for differing operational practices that may exist across the globe. The random sampling of flights resulted in 1,265 Flightradar24^® ADS-B data files being sourced from Flightradar24^{® (23)} and the resulting routes are shown in Fig. 10(a).

Figure 11.

A histogram of the number of pre-takeoff turns for the wide-body civil aircraft.

Figure 10(b) shows the ground tracks across the wide-body civil aircraft fleet at a single airport; the high fidelity of the Flightradar24^® ADS-B data files can be observed as key elements of airport geometry can be identified. It should be noted that the randomly sampled ADS-B data files were manually checked to ensure high-fidelity ground tracks were available.

4.1 Pre-takeoff taxi phase manoeuvres

Figure 11 shows the total number of pre-takeoff turns per flight and the frequency with which they were observed across the 1,265 flights of the demonstration dataset study. The mode number of pre-takeoff turns was found to be three pre-takeoff turns per flight, which shows good agreement with previous studies into wide-body aircraft ground manoeuvres, which suggest that four pre-takeoff turns per flight is typical^{(Reference Ladda and Struck2,15)} . The minimum number of pre-takeoff turns was zero, representing a flight that entered the runway directly from pushback, which is possible when operating from smaller airport geometries. The maximum number of pre-takeoff turns was 26, and an investigation into this result highlighted that the aircraft had performed a Rejected TakeOff (RTO), which would be expected to result in additional turns. Figure 11 also shows that a significant number of flights have greater than five pre-takeoff turns, and these flights were found to be operating from large international airports with complex airport geometries.

It can be observed from Fig. 11 that the histogram of the number of pre-takeoff turns demonstrates positive or ‘right-tail’ skew. The presence of a skewed distribution results in the assumption made in previous studies of an ‘average’ number of turns per flight^{(Reference Ladda and Struck2)} failing to accurately represent the underlying variability in the manoeuvre occurrence. The use of the average number of turns would require the assumption of a symmetric distribution, where every number of turning occurrences ‘above’ the average value is also represented ‘below’ the average value. As a result, the assumption of an average number of turns would fail to account for flights that will have a large number of turns when constructing landing gear load spectra for fatigue design.

The pre-takeoff turns across the flights of the demonstration dataset can also be decomposed into the number of pre-takeoff left and pre-takeoff right turns for each flight, as shown respectively in Fig. 12(a) and (b). The mode number of occurrences for both turn directions is one while the median is two, highlighting the positive skew in the histograms. Figure 12(c) shows that the share between left and right turns is approximately equal. Within existing landing gear fatigue design practice, it is typically assumed that the number of left and right turns across the load spectrum are equal⁽⁴⁾, thus the occurrence results generated from the ADS-B data shown in Fig. 12(c) provide support and validation for this existing design assumption.

Figure 12.

Pre-takeoff turn direction statistics for the wide-body civil aircraft: (a) histogram of number of pre-takeoff left turns, (b) histogram of number of pre-takeoff right turns and (c) proportional share between turn direction for pre-takeoff turns.

Figure 13 shows the total number of pre-takeoff braking applications per flight and the frequency with which they were observed across the 1,265 flights of the demonstration dataset study. The mode and median number of pre-takeoff braking applications were three and four, respectively, showing consistency with the previous studies into wide-body aircraft ground manoeuvres^{(Reference Ladda and Struck2,15)} . The presence of the positive skew of the histogram in Fig. 13 also suggests that it is inappropriate to consider ‘average’ number of braking applications per flight, based upon the reasoning above, during the pre-takeoff taxi phase. As for the pre-takeoff turning manoeuvre occurrences, previous studies have assumed an average number of braking occurrences^{(Reference Ladda and Struck2)}, so it is possible that landing gear load spectra based upon this assumption would fail to account for flights in the upper tail of the pre-takeoff braking occurrences histogram.

Figure 13.

Histogram of the number of pre-takeoff braking occurrences for the wide-body civil aircraft.

4.2 Post-landing taxi phase ground manoeuvres

Figure 14 shows the total number of post-landing turns per flight and the frequency with which they were observed across the 1,265 flights of the demonstration dataset study. The mode and median number of turns post-landing were found to be two and three, respectively. Therefore, the mode number of turns post-landing was found to be smaller than the mode number of turns pre-takeoff. This is to be expected as the pre-takeoff taxi out to the runway will require the aircraft to taxi to the threshold at the end of the runway, rather than routing directly from where the aircraft vacates following landing, as aircraft do not use the full runway length on landing for every flight. The more direct taxi routing could potentially result in fewer turns being performed by the aircraft during the post-landing taxi phase, prior to turning onto the arrival stand.

Figure 14.

Histogram of the number of post-landing turns for the wide-body civil aircraft.

Note that the mode number of post-landing turns is significantly lower than the six turns for a wide-body civil aircraft suggested by Ladda and Struck^{(Reference Ladda and Struck2)}. It is currently unclear why the wide-body aircraft within the ADS-B data files demonstrated lower post-landing turn occurrences. The minimum number of turns was found to be zero, corresponding to flights arriving at smaller airports where the aircraft taxies directly onto stand after exiting the runway. The maximum number of turns was found to be 15 as a result of the aircraft being towed from the arrival gate. Figure 14 also demonstrates positive skew in the number of turns post-landing, again contradicting the use of average ground manoeuvre occurrences within landing gear load spectra.

The post-landing turns can also be decomposed into the number of post-landing left and post-landing right turns for each flight, as shown respectively in Fig. 15(a) and (b). The mode and median number of occurrences for both turn directions are both one. Figure 15(c) shows that the share between post-landing left and right turn directions is approximately equal. The approximately equal share between post-landing left and right turn directions provides further validation of the assumption of equal share of turn direction currently used when constructing landing gear load spectra⁽⁴⁾.

Figure 15.

Post-landing turn direction statistics for the wide-body civil aircraft: (a) histogram of number of post-landing left turns, (b) histogram of number of post-landing right turns and (c) proportional share between turn direction for post-landing turns.

Figure 16 shows the variability in post-landing braking occurrence per flight across the 1,265 flights of the demonstration ADS-B dataset. Figure 16 demonstrates strong agreement with the pre-takeoff braking applications shown in Fig. 13, as for both pre-takeoff and post-landing phases, the mode and median values were three and four braking occurrences, respectively. These values also show good agreement with previous studies on wide-body aircraft ground manoeuvres, as shown in Fig. 16 ^{(Reference Ladda and Struck2,15)} . Once again, the histogram for the post-landing braking occurrences demonstrates positive skew, further suggesting that the assumption of average manoeuvre occurrences per flight is unsuitable for the wide-body civil aircraft.

Figure 16.

Histogram of the number of post-landing braking occurrences for the wide-body civil aircraft.

4.3 Correlation between turning and braking manoeuvres

Within the existing practice for constructing landing gear load spectra, it is typically assumed that, for every left turn that an aircraft performs, a right turn will also be performed within the same taxi phase, and vice versa^{(Reference Ladda and Struck2)}. This assumption therefore introduces a correlation which represents that, as the number of turns of a given direction increases for a taxi phase, the number of turns in the opposite direction should also increase.

To investigate whether the ADS-B data files for the wide-body civil aircraft support the assumption of a correlation between the occurrence of turn directions within a single taxi phase, the correlation between the occurrence of left and right turns for both pre-takeoff and post-landing was identified for the wide-body civil airliner.

Figure 17(a) shows a contour plot representing how the number of pre-takeoff right turns varies with the number of pre-takeoff left turns, where a lighter shade in the contour plot represents a more commonly observed combination of left and right turns from the ADS-B data files. From Fig. 17(a) it can be observed that the most common combination of turns was one left turn and one right turn during the pre-takeoff taxi phase. Figure 17(a) also highlights that there is a slight ‘stretch’ in the contour plot representing that, as the number of pre-takeoff left turns increases, so does the number of pre-takeoff right turns. This can be supported using the Pearson correlation coefficient^{(Reference Montgomery and Runger45)}, which for Fig. 17(a) is $\rho = 0.660$ , suggesting a moderate correlation (as correlations above $\rho = |0.8|$ are considered ‘strong’ and correlations below $\rho = |0.5|$ are considered ‘weak’^{(Reference Montgomery and Runger45)}). Figure 17(b) shows the contour plot representing the correlation between the number of left and right turns during the post-landing taxi phase. Figure 17(b) demonstrates a similar relationship to the pre-takeoff taxi phase in Fig. 17(a), and this is supported by the correlation coefficient of $\rho = 0.619$ . As a result, it can be summarised that, regardless of taxi phase, the number of left and right turns are moderately correlated. This result supports the assumption currently made in existing design practice, of increasing occurrence of left turns for an increasing occurrence of right turns⁽⁴⁾.

Figure 17.

Contour plots highlighting the correlation between turning and braking manoeuvres: (a) correlation between pre-takeoff left and right turns, (b) correlation between post-landing left and right turns, (c) correlation between pre-takeoff turns and pre-takeoff braking occurrences and (d) correlation between post-landing turns and post-landing braking occurrences.

During the analysis of the results of the ADS-B data files, it was hypothesised that a correlation may also exist between the total number of turns and the number of braking occurrences within a taxi phase, as it would be expected that aircraft would decelerate prior to performing a turn. Therefore, the correlation between turning and braking occurrences for the pre-takeoff and post-landing taxi phases was investigated. Figure 17(c) shows a contour plot of the number of pre-takeoff braking occurrences compared with the total number of pre-takeoff turns. The contour plots highlight that the most commonly observed combination was two pre-takeoff turns with two pre-takeoff braking occurrences. Whilst the contour plot demonstrates a potential relationship of increasing number of braking occurrences for increasing number of turns, the correlation coefficient value of $\rho = 0.452$ suggests that only a weak correlation is present. Figure 17(d) shows the correlation between the number of post-landing braking occurrences and total number of post-landing turns. It can be observed from Fig. 17(d) that the most commonly observed combination was two turns and four braking occurrences during the post-landing phase. The correlation coefficient value of $\rho = 0.503$ demonstrates again that only a weak correlation is to be expected between the number of braking manoeuvres and total number of turns. Therefore, while a moderate correlation is expected between left and right turn occurrences, only a weak correlation is observed for turning and braking occurrences, regardless of taxi phase.

4.4 Share of turns and braking across flight

To identify whether more turning or braking manoeuvres occur during the pre-takeoff or post-landing taxi phases, the total number of turning and braking occurrences were computed for both taxi phases. Figure 18(a) shows the share of pre-takeoff and post-landing turning manoeuvres across the 1,265 flights within the ADS-B dataset, revealing that 53.2% of turns were found to occur during the pre-takeoff phase while the remaining 46.8% of turns were performed post-landing. Figure 18(b) shows the share of braking manoeuvres across the pre-takeoff and post-landing phases. Figure 18(b) demonstrates that the share of braking occurrences is approximately equal across the pre-takeoff and post-landing phases. The approximately equal share of manoeuvre occurrences across the pre-takeoff and post-landing taxi phases contradicts assumptions that are made regarding general aircraft manoeuvres, that is, that different manoeuvre occurrences are observed pre-takeoff and post-landing⁽⁴⁾.

Figure 18.

The proportional share of (a) turning manoeuvres and (b) braking occurrences across pre-takeoff and post-landing taxi phases for the wide-body civil aircraft.

4.5 Occurrence statistics of milestone manoeuvres

Following the characterisation of the variability in turning and braking manoeuvre occurrences, the proportional share of milestone manoeuvre types was identified for pushback, runway entry, runway exit and turn onto stand manoeuvres. Figure 19 shows the proportional share of manoeuvre types for each of the milestone manoeuvres.

Figure 19.

Proportional share of milestone manoeuvres for the wide-body civil aircraft: (a) pushback direction, (b) turn onto stand direction, (c) runway entry type and direction and (d) runway exit type and direction.

Figure 19(a) shows that the direction of aircraft pushback is approximately equal, with 52.1% and 47.9% of pushback manoeuvres being tail-left and tail-right, respectively. The slight discrepancy in the equal share is expected to be as a result of the 283 flights which did not provide pushback data (22.4% of the dataset). Figure 19(b) presents the share of turn onto stand direction, with 54.3% of flights performing a left turn onto stand. In a similar manner to the pushback manoeuvre, it was expected that the share between left and right turn onto stand directions would be approximately equal, and the resulting discrepancy is expected to be as a result of the 16.3% of flights within the dataset which failed to provide information regarding the turn onto stand direction. Therefore, across the pushback and turn onto stand manoeuvres, the turn direction across the wide-body aircraft fleet is approximately equal.

Figure 19(c) shows the runway entry manoeuvre type, revealing that the vast majority (97.2%) of flights within the ADS-B dataset performed a standard runway entry, with the direction of the runway entry being approximately equally shared between left and right runway entries. The 2.8% of flights performing a backtrack runway entry shows good agreement with the wide-body aircraft route network at the time of writing, of which 3.3% of routes (based on their proportion of the weekly fleet utilisation) operate out of airports requiring a runway backtrack prior to takeoff⁽²³⁾. It can also be observed from Fig. 19(c) that the left and right turn directions have an approximately equal share of the runway backtrack entry manoeuvres.

Figure 19(d) shows the proportional share of the runway exit type and direction across the 1,265 flights in the ADS-B demonstration dataset, revealing that 80.8% of runway exit turns were found to be standard runway exits (i.e. low-speed turns), with left and right turn directions having an equal proportion. Runway exits using RETs (i.e. high-speed, <45° turns) were only observed for 18.5% of flights. This result may be as a consequence of the selected wide-body aircraft, which operates into a significant number of airports that do not have RETs⁽²³⁾. In addition, an aircraft may vacate using an RET, but at a slower speed than the runway exit algorithm of 30Kn, which would result in the identification of a standard runway exit. Only 0.6% of flights demonstrated a backtrack exit from the runway. Whilst 3.3% of the wide-body aircraft route network was observed to operate into airports which may require a runway backtrack, the aircraft were typically observed to vacate the runway at an earlier taxiway, removing the need to perform a runway backtrack. It can be observed from Fig. 19(d) that the turn directions for both standard and high-speed runway exits are approximately equal in proportion across the wide-body aircraft fleet.

The results shown in Fig. 19 provide further support to the existing assumption that the direction of ground turning manoeuvres is equally shared between left and right turns⁽⁴⁾. Therefore, it can be concluded that, regardless of taxi phase and turning manoeuvre type, an equal proportion of left and right turns is performed across the wide-body civil aircraft fleet.