Nomenclature

ADS-B

automatic dependent surveillance-broadcast

FAA

Federal Aviation Administration

FSC

full service carrier

GPS

Global Positioning Systems

LCC

low-cost carrier

${N_T}$

total number of taxi turns

${R_{avg}}$

average estimated turn radius

ULCC

ultra-low-cost carrier

Greek symbol

$\alpha $

bearing between two ADS-B position reports

${\rm{\Delta }}\alpha $

change in bearing between three consecutive ADS-B position reports

${\rm{\Delta }}d$

distance between two ADS-B position reports

${\rm{\Delta }}t$

time between two ADS-B position reports

${\rm{\Delta }}{\theta _{s,max}}$

maximum spatial turn rate observed during turn

2.0 Landing gear ground manoeuvre occurrence identification methodology

The exploitation of ‘real-time’ and historic aircraft trajectories derived from crowd-sourced air traffic data has become a significant tool across the research areas of airborne and ground air traffic management [Reference Corrado, Puranik and Mavris19, Reference Schultz, Olive, Rosenow, Fricke and Alam20], aircraft emissions [Reference Sun, Basora, Olive, Strohmeier, Schäfer, Martinovic and Lenders21] and aerospace system design [Reference Hoole, Sartor, Booker, Cooper, Gogouvitis and Schmidt6]. The most common source of air traffic data is automatic dependent surveillance-broadcast (ADS-B) transmissions, now mandated in many civil aircraft [Reference Sun, Olive, Strohmeier, Schäfer, Martinovic and Lenders22], which provides aircraft position reports (latitude, longitude and altitude) at a maximum frequency of two broadcasts a second [Reference Sun23]. When captured by ground-based receivers and collated into repositories such as Flightradar24 [24] and the OpenSky Network [Reference Schäfer, Strohmeir, Lenders, Martinovic and Wilhelm25], historic ADS-B trajectories for individual aircraft and flights are then available. Due to the reliance of ground-based receivers, each repository has differing levels of coverage for airport ground operations. In addition, the temporal resolution of datasets from different ADS-B repositories can vary due to differing data processing and storage approaches. ADS-B data can also contain positional errors as discussed by Ali et al. [Reference Ali, Schuster, Ochieng and Majumdar26].

Within the focus of aerospace fatigue design and structural health monitoring, ADS-B has previously been employed to characterise the variability in training helicopter manoeuvres [Reference Hoole, Booker and Cooper27], provide health monitoring of helicopter gearboxes [Reference Hünemohr, Litzba and Rahmimi28] and to characterise the variability in wide-body aircraft landing gear ground manoeuvres [Reference Hoole, Sartor, Booker, Cooper, Gogouvitis and Schmidt6]. There is also the future ambition to explore the use of remote sensing of airframe accelerations and loads via ADS-B to support structural health monitoring of aircraft structures [Reference Meyer, Zimdahl, Kamtsiuris, Meissner, Raddatz, Haufe and Bäßler29, Reference Hoole, Booker, Cooper, Richardson and Hallam30].

An example of the taxi routes derived from an ADS-B trajectory is shown in Fig. 1 for the pre-takeoff taxi at the departure airport and the post-landing taxi at the arrival airport. It can be observed from Fig. 1 that there are sufficient ADS-B transmission data points to visually identify the taxi routes, from pushback to runway entry and runway exit to the final turn onto stand.

Figure 1. Example of (a) pre-takeoff and (b) post-landing taxi routes from ADS-B trajectories. ADS-B data from Flightradar24 [24]. Map data from OpenSteetMap (https://www.openstreetmap.org/copyright).

As shown in Fig. 1, ADS-B trajectories can provide sufficient resolution of the aircraft taxi route performed during a flight to manually and visually identify the ground manoeuvres performed. Prior work has detailed a framework and rule-based methodology to automatically identify aircraft ground manoeuvres from ADS-B trajectories [Reference Hoole, Sartor, Booker, Cooper, Gogouvitis and Schmidt6]. The existing methodology identifies the changes in aircraft track angle and speed between consecutive ADS-B transmissions and through a series of conditional statements and grouping of consecutive identical changes in the aircraft trajectory extracts, turning and deceleration manoeuvres [Reference Hoole, Sartor, Booker, Cooper, Gogouvitis and Schmidt6].

However, a number of limitations of the algorithm were identified previously, along with their proposed mitigation [Reference Hoole, Sartor, Booker, Cooper, Gogouvitis and Schmidt6]. A significant limitation highlighted was the reliance on aircraft track for identifying turn direction, especially where ADS-B data was intermittent, missing or noisy and the use of latitude and longitude information was proposed as an alternative [Reference Hoole, Sartor, Booker, Cooper, Gogouvitis and Schmidt6]. Characterisation of ground turning manoeuvres is of vital importance within landing gear fatigue design, as the lateral and torsional loads applied during turns are considered fatigue-critical design conditions [Reference Hoole8]. Therefore, inclusion of ‘tight’ or pivoting turns into the fatigue spectra is especially important, along with turn ‘reversals’ (e.g. a left turn immediately followed by a right turn), due to their increased cyclic loading magnitudes [Reference Buxbaum1, Reference Ladda and Struck4, Reference Hoole, Sartor, Booker, Cooper, Gogouvitis and Schmidt6]. Consequently, exploration of how ground turn characterisation from ADS-B trajectories can be enhanced, both in terms in robustness and extracted information was considered.

2.1 Enhanced turn manoeuvre characterisation

Within the ADS-B trajectory for a given flight, the aircraft’s latitude and longitude position will be reported during the taxi phase, as shown previously in Fig. 1. Between two consecutive ADS-B transmissions, the bearing and distance between the two aircraft positions, as defined using GPS latitude and longitude, can be computed. The definition of the bearing between two latitude and longitude positions is shown in Fig. 2.

Figure 2. Definition of bearing between ADS-B trajectory points and turn direction identification. ADS-B data from Flightradar24 [24]. Map data from OpenSteetMap (https://www.openstreetmap.org/copyright).

Rather than relying on the reported aircraft track, the change in the bearing between three sets of consecutive latitude and longitude positions are used to identify if a left or right turn is being performed, as shown in Fig. 2. The ADS-B ground trajectory is then worked through, point-by-point, and consecutive positions with the same turning direction are grouped together. Any identified turns $\lt{10^ \circ }$ were assumed to be positions of straight taxiing, as prior work demonstrated that deviations of up to ${2^ \circ }$ were observed during straight taxiing [Reference Hoole, Sartor, Booker, Cooper, Gogouvitis and Schmidt6]. In addition, airport geometries typically employ taxiway turns of ${30^ \circ }$ , ${45^ \circ }$ , ${60^ \circ }$ , ${90^ \circ }$ , ${120^ \circ }$ , ${135^ \circ }$ and ${150^ \circ }$ [31], resulting in the selection of a $\gt{10^ \circ }$ turn threshold providing discrimination between straight taxiways and typical taxiway turn angles. ADS-B trajectories have been shown to demonstrate significant noise when the aircraft is stationary [Reference Hoole, Sartor, Booker, Cooper, Gogouvitis and Schmidt6] and consequently, zero-speed trajectory points are also removed from the dataset.

The use of latitude, longitude and bearing information, rather than the reported aircraft heading to identify turn direction also permits two limitations of the prior algorithm to be overcome. Firstly, the original track-based algorithm could identify ‘false’ turns due to noisy ADS-B data, where the aircraft would rapidly jump to another taxiway location, prior to returning to the actual ground trajectory (see Fig. 3) [Reference Hoole, Sartor, Booker, Cooper, Gogouvitis and Schmidt6]. In order to filter out such occurrences, the distance between consecutive latitude and longitude positions were compared to the expected distance travelled by the aircraft based on its current ground speed. Any ADS-B position reports with a disagreement greater than a factor of 2 between the observed and estimated distances was removed from the trajectory.

Figure 3. Example of an erroneous ADS-B trajectory position. ADS-B data from Flightradar24 [24]. Map data from OpenSteetMap (https://www.openstreetmap.org/copyright).

The prior algorithm also failed to identify S-turn reversals when there were insufficient ADS-B trajectory data points to map out the two consecutive turns (see Fig. 4) [Reference Hoole, Sartor, Booker, Cooper, Gogouvitis and Schmidt6]. Therefore, it was defined that if a turn was identified via bearing change between two latitude and longitude positions and that the bearing change between the preceding and following straight taxi elements was $\lt{10^ \circ }$ that an S-turn had been performed, as shown in Fig. 4.

Figure 4. S-turn identification with missing ADS-B trajectory positions. ADS-B data from Flightradar24 [24]. Map data from OpenSteetMap (https://www.openstreetmap.org/copyright).

Finally, the bearing change between latitude and longitude positions was employed to identify the end of the aircraft pushback. Manual identification of pushbacks within trajectories highlighted that the end of the pushback would be marked by a single bearing change between consecutive latitude and longitude positions in excess of ${60^ \circ }$ , as shown in Fig. 5. As a result, the single ADS-B trajectory point in the pre-takeoff taxi phase displaying such a bearing change was marked as the pushback apex, and any turning manoeuvres preceding the pushback apex were relabelled as turns occurring during pushback.

Figure 5. Identification of pushback apex. ADS-B data from Flightradar24 [24]. Map data from OpenSteetMap (https://www.openstreetmap.org/copyright).

2.1.1 Characterisation of tight and pivot turns

As highlighted in Section 2, landing gear ground turning manoeuvres must be characterised into normal taxiway turns and those involving ‘tight’ or ‘pivot’ turns, in which typically the aircraft will turn with one of the main landing gear remaining near stationary [Reference Buxbaum1, Reference Ladda and Struck4]. The transition to identifying turns via latitude and longitude position changes permitted additional parameters relating to each turn to be computed, as visualised in Fig. 6:

• • Mean and maximum temporal turn rate (deg/s)

• • Mean and maximum spatial turn rate (deg/m)

• • Turn radius estimated via turn path distance (m)

• • Turn radius estimated via distance between turn entry and exit (m)

Figure 6. Definition of turn characteristics. ADS-B data from Flightradar24 [24]. Map data from OpenSteetMap (https://www.openstreetmap.org/copyright).

Figure 7. Example of a known pivot turn. ADS-B data from Flightradar24 [24]. Map data from OpenSteetMap (https://www.openstreetmap.org/copyright).

The radius of a turn could be approximated through assuming that all turns took a circular profile, and the computed distance represented the fraction of the circumference in the same ratio as the total turn angle to ${360^ \circ }$ . It was found that estimating the radius via turn path distance was suitable for turns with multiple ADS-B trajectory points and typically of ${90^ \circ }$ , whilst the radius estimation via the entry-exit distance was required for small angle turns or those without a large number of reported positions. Therefore, the average estimated turn radius was also computed using the turn path and entry and exist distance values.

In order to automatically characterise tight and pivot turns within the ADS-B trajectories, 20 ADS-B trajectories containing the following manoeuvres, known to result in tight turns, were selected and the turn parameters computed for each relevant turn:

• • Runway pivot turns after runway backtrack (see Fig. 7)

• • Tight runway entry turns (typically as the final element of a standard taxiway turn)

• • Tight turn onto stand manoeuvres

The variability in each of the turn parameter values was characterised using a Log-Normal distribution due to the observed right-tail skew in the values and potential for zero-thresholds for pivot turn radii and near-zero turn rates for shallow/wide turns. The turn parameters that provided the greatest separation between standard, tight and pivot turns were found to be the maximum spatial turn rate ‘ ${\rm{\Delta }}{\theta _{s,max}}$ ’ and average estimated turn radius ‘ ${R_{avg}}$ ’, as shown by the distributions in Fig. 8.

Figure 8. Variability in (a) maximum spatial turn rate and (b) average estimated turn radius ${R_{avg}}$ for standard, tight and pivot turns.

Employing the ${5^{{\rm{th}}}}$ and ${95^{{\rm{th}}}}$ percentiles of each distribution permitted definition of the turn characterisation space shown in Fig. 9. For example, the bounding box for tight turns shown in Fig. 9 is given by the percentile values derived from the Log-Normal distributions:

• • ${\rm{\Delta }}{\theta _{s,{\rm{max}}}}$ ${5^{{\rm{th}}}}$ Percentile value: 2.24deg/m

• • ${\rm{\Delta }}{\theta _{s,{\rm{max}}}}$ ${95^{{\rm{th}}}}$ Percentile value: 4.43deg/m

• • ${R_{avg}}$ ${5^{{\rm{th}}}}$ Percentile value: 26.5m

• • ${R_{avg}}$ ${95^{{\rm{th}}}}$ Percentile value: 81.9m

Figure 9. Regions for known standard, tight and pivot turns based upon ${5^{{\rm{th}}}}$ and ${95^{{\rm{th}}}}$ ${\rm{\Delta }}{\theta _{s,max}}$ and ${R_{avg}}$ percentiles.

Figure 9 also shows all of the turns extracted from the flights of one of the aircraft described in Section 3. It can be observed that there is overlap in some of the turn characteristic regions, along with undefined regions. Consequently, the turn characterisation space was simplified, assuming the boundaries between overlapping regions were set at the mid-point of the overlap, along with extending any regions to the minimum and maximum axis position. This simplification resulted in the characterisation space shown in Fig. 10 and the turn characterisation conditional statements shown in Equation (1).

(1) \begin{align}{\rm{TurnType}} = \left\{ \begin{array}{l@{\quad}l} {{\rm{NormalTurn}}} & {{\rm{if\Delta }}{\theta _{s,max}} \le 2.33{\rm{deg}}/{\rm{m}}}\\[5pt]{{\rm{TightTurn}}} & {{\rm{if\Delta }}{\theta _{s,max}} \gt 2.33deg/mand{R_{avg}} \gt 27.5m}\\[5pt]{{\rm{PivotTurn}}} & {{\rm{if\Delta }}{\theta _{s,max}} \gt 2.33deg/mand{R_{avg}} \le 27.5{\rm{m}}}\end{array} \right.\end{align}

Figure 10. Assumed ${\rm{\Delta }}{\theta _{s,max}}$ and ${R_{avg}}$ regions for standard, tight and pivot turns.

3.0 Dataset collection

Airline operations can be broadly split into two types, full-service carriers (FSCs) or low-cost carriers (LCCs) [Reference Graham32]. FSC operators are typically based at larger primary airports and operate hub-to-hub routes, with LCC operators focusing on secondary airports, typically operating point-to-point routes [Reference Graham32]. To explore the impact of these two operator characteristics on the ground manoeuvres performed by aircraft, three narrow-body aircraft from a European FSC operator fleet and three narrow-body aircraft from a European LCC fleet were identified. Narrow-body fleets were selected due to the perceived increase in route variability and to enable a comparison with the wide-body global fleet data captured previously [Reference Hoole, Sartor, Booker, Cooper, Gogouvitis and Schmidt6]. The characteristics of each narrow-body aircraft within the data collection timeframe were as follows:

• • FSC 1: aircraft consistently based at a large international airport

• • FSC 2: aircraft consistently based at the same airport as FSC 1

• • FSC 3: aircraft consistently based at the same airport as FSC 1

• • LCC 1: aircraft based at two secondary international airports

• • LCC 2: aircraft rotated around secondary international airports and regional airports

• • LCC 3: aircraft rotated around two regional airports

The FSC fleet aircraft were selected to have the same base airport to permit the variability in ground manoeuvres resulting from solely the operator’s route network to be identified. The results from the FSC fleet could then be compared to the LCC fleet, which were routinely rotated around the operator’s bases and route network, permitting the impact on ground manoeuvre variability to be compared between the two operator types. Multiple aircraft within the LCC fleet were reviewed to ensure the different base characteristics within the LCC route network were captured, whilst the FSC aircraft were selected based on not having significant gaps within their utilisation as a result of being out of service.

For each individual airframe, ADS-B trajectories for 500 consecutive flights in the period of Summer-Autumn 2021 were sourced from Flightradar24 [24]. This period of time was employed to capture the potential increase in seasonal routes flown by the LCC operator and the year of 2021 was required due to the need to be able to select the LCC aircraft based upon their long-term fleet rotation strategy, which can only be apparent from reviewing historic aircraft operations retrospectively. Studies concerning the capture of ADS-B transmissions highlighted that during the latter half of 2021, ADS-B data collection rates had begun to recover after the COVID-19 pandemic [Reference Sun, Basora, Olive, Strohmeier, Schäfer, Martinovic and Lenders21]. The sample of size of 500 flights was selected to represent a typical time between aircraft maintenance checks [Reference Andrade, Silva, Ribeiro and Santos33], assuming that each flight cycle lasts 1.5 hours. Flightradar24 [24] was used as the ADS-B source due to its adequate coverage of ground operations at European airports.

3.1 ADS-B trajectory data quality

Prior to identifying the turning manoeuvres performed within the trajectories for each individual aircraft, the ADS-B trajectory data quality was first assessed by manually reviewing the 3,000 assembled flights. Each trajectory was visually checked to identify whether it contained sufficient ADS-B data points to identify the manoeuvres during pushback, pre-takeoff taxi, post-landing taxi and the turn onto stand.

The resulting proportions of flights containing ADS-B trajectories for the four taxi phases are shown in Table 1 and visualised in Fig. 11. When aggregating across all six aircraft, it can observed that approximately 80% of flights will contain pushback, taxi and turn onto stand trajectories, with approximately 60% of flights containing the full taxi route. This latter value represents an $ \approx $ 10% increase in the number of flights with complete taxi trajectories compared to the dataset collected in 2019 for the wide-body aircraft study [Reference Hoole, Sartor, Booker, Cooper, Gogouvitis and Schmidt6].

Table 1. Proportion of dataset containing taxi route elements

Figure 11. Visualisation of proportion of dataset containing taxi route elements.

From Table 1 and Fig. 11, it is also clear to see that the LCC aircraft have lower overall proportions for ADS-B trajectories containing taxi route elements and these proportions are seen to decrease as the airport size reduces from large international airports to regional airports. This especially concerns the availability of turn onto stand information for the LCC aircraft fleet, as 15% fewer flights contain turn onto stand information for the aircraft based exclusively at regional airports, compared to the FSC fleet. It is anticipated that such an observation is due to there being fewer ADS-B ground receivers installed close to smaller airports.

The ADS-B trajectories were also assessed for their temporal and spatial resolution. For each of the 3,000 available flights, the time difference ‘ ${\rm{\Delta }}t$ ’ and distance between the latitude and longitude ‘ ${\rm{\Delta }}d$ ’ for each consecutive ground trajectory point were computed, and the average values for each aircraft and fleet are shown in Table 2.

Table 2. Dataset resolution

It can be observed that the average timestep between ADS-B trajectories for both pre-takeoff and post-landing is significantly higher than the minimum ADS-B specification of 0.5 seconds [Reference Sun23]. This effect is anticipated to be due to the dataset compression typically performed by ADS-B repositories, along with removal of zero-speed trajectories elements as described in Section 2.1. This finding also supports the results in Table 2 which demonstrate that the temporal resolution is doubled for the post-landing taxi phase compared to the pre-takeoff taxi phase. During peak operational times, it has been observed that aircraft queue at runway hold points [Reference Hoole, Sartor, Booker, Cooper, Gogouvitis and Schmidt6], and this will lead to an increased prevalence of zero-speed trajectory elements within the pre-takeoff taxi phase. The increased temporal resolution for regional airports shown in Table 2 also supports the trend of decreased temporal resolution for increased zero-speed trajectory elements, as such airports would be expected to have fewer instances of congestion.

Table 2 also demonstrates the spatial resolution of the ADS-B trajectories, and it can be observed that the average distance between ADS-B position reports is of similar magnitude to the fuselage length and wingspan of a narrow-body aircraft [34, 35]. This observation further supports the adoption of latitude and longitude based manoeuvre identification due to the high average spatial resolution achieved and further supports the adoption the spatial turn parameters ${\rm{\Delta }}{\theta _{s,max}}$ and ${R_{avg}}$ for characterising turn types.

3.2 Verification of manoeuvre identification methodology

The enhanced turn manoeuvre characterisation methodology described in Section 2 also required verification prior to characterising the manoeuvres in the ADS-B trajectories of the six aircraft. For verification, 10% of the dataset (i.e. 50 flights per airframe) were assessed using the methodology, along with the manual characterisation of the taxi route and turns through plotting the ADS-B trajectory in reference to airport geometries (as shown previously in Fig. 1). Table 3 shows the proportion of flights for which the algorithm correctly identified the pre-takeoff taxi, post-landing taxi and complete ground trajectories. It can be observed that in nearly 80% of individual flights the correct ground manoeuvres were identified in the pre-takeoff and post-landing taxi phases. This value represents a significant increase in the $ \approx $ 50% success rate of the heading-based turn identification algorithm [Reference Hoole, Sartor, Booker, Cooper, Gogouvitis and Schmidt6], justifying the adoption of turn characterisation via ADS-B latitude and longitude positions.

Table 3. Verification of manoeuvre identification

Whilst the vast majority of values in Table 3 are consistent across the aircraft and fleets, it is important to note that processing of FSC 1 trajectories showed degraded accuracy for the pre-takeoff taxi phase. As this aircraft was based at the same airport as FSC 2 and FSC 3, it suggested that FSC 1 may have been operating on routes with increased levels of noise within the ADS-B trajectory, leading to increased rates of erroneous manoeuvre identification.

6.0 Limitations, outlook and future work

The current limitations with the proposed study can be decomposed into the accuracy of the manoeuvre characterisation, the quality of the ADS-B trajectory data and constraints within the collected data for the FSC and LCC fleets.

Firstly, whilst the enhanced turn characterisation algorithm showed that it was able to correctly identify the full taxi route in 77% of flights compared to the 50% success rate achieved in prior work [Reference Hoole, Sartor, Booker, Cooper, Gogouvitis and Schmidt6], it is clear that the robustness of the algorithms still needs to be refined. For example, flights demonstrated errors in the pre-takeoff taxi phase propagated from incorrect pushback characterisation, which originates in the noisy ADS-B data observed at stand and gate locations [Reference Hoole, Sartor, Booker, Cooper, Gogouvitis and Schmidt6]. Further investigation into how ground manoeuvre characterisation can be made robust to this lower-quality ADS-B data and propagation of errors needs to be investigated as a priority. Errors during the post-landing phase were typically due to missing or ‘jumps’ in the ADS-B trajectory data, and again, further work is required to increase the robustness of the ground manoeuvre characterisation algorithms, despite the significant improvement achieved over the original methodology presented in prior work [Reference Hoole, Sartor, Booker, Cooper, Gogouvitis and Schmidt6].

It can therefore be seen that the accuracy of ground manoeuvre identification from ADS-B trajectories is inherently linked to the quality of the ADS-B trajectory data. To explore whether data quality had improved since the 2021 data collection timeframe, one aircraft from the FSC and LCC fleets was selected and ADS-B trajectories for one month of operation during Spring 2023 were sourced from Flightradar24 [24]. It was found the LCC aircraft based at a large airport during the study had now entered a fleet rotation phase, highlighting the complexity of operator route networks and fleet rotation strategy. For the 146 FSC and 140 LCC aircraft flights, the ADS-B trajectories were visually assessed for containing taxi route elements as described previously in Section 3.1. The resulting occurrences of taxi route elements in the 2023 data are given in Table 13.

Table 13. Proportion of dataset containing taxi route elements for 2023 data

From comparing Tables 1 and 13, it can be observed for the FSC aircraft that there is an increase in flights containing pre-takeoff and post-landing taxi routes by 5%, leading to a nearly 10% increase in the number of FSC flights showing the complete taxi route. Pushback and turn onto stand elements remained consistent with the 2021 dataset. The LCC fleet on the other hand shows limited change in data completeness between 2021 and 2023 datasets, with less than a 5% increase in the number of flights showing the complete taxi routes. At the narrow-body fleet level, a 4% increase in the number of flights with complete taxi routes has been observed over the two-year window.

When compared to 2019 data used in a prior study on wide-body aircraft [Reference Hoole, Sartor, Booker, Cooper, Gogouvitis and Schmidt6], it is suggested that ADS-B coverage for aircraft ground operations increases by approximately 2% of flights a year, although this will be highly sensitive to the global location and airports served by aircraft due to the current reliance on crowd-sourced data from ground receivers. However, with ADS-B set to become a cornerstone technology for air traffic control and air traffic management [Reference Schäfer, Strohmeir, Lenders, Martinovic and Wilhelm25], there is the potential for a rapid increase in the coverage and quality of ADS-B derived ground trajectories in conjunction with such data being sourced from a certified, rather than predominately enthusiast or research-led source. When coupled with ‘map-matching’ algorithms that have been recently developed to enable ADS-B ground trajectories to be linked to known airport geometries [Reference Szymanski, Ghazi and Botez37], ADS-B may continue to evolve into a robust data source for aerospace system design and monitoring. There is also the potential future synthesis of such approaches with aggregated data of aircraft ground trajectories to support the simulation and optimisation of aircraft trajectories at airports as considered by Ma et al. [Reference Ma, Zhou, Liang and Delahaye38].

The collected dataset also contains limitations when attempting to generalise operator characteristics on landing gear ground manoeuvres. Firstly, the data collection window equal to the approximate timeframe between heavy maintenance checks of 500 flights only represented approximately six months of operation. A longer data collection window would be required to capture all aspects of an operator’s route network, and any fleet rotation strategies employed, as observed for the LCC aircraft that during 2021 was based predominately at a large airport being subject to fleet rotation in 2023. The data collection window of Summer-Autumn 2021 may have also captured non-standard route networks for operators recovering from the COVID-19 pandemic [Reference Sun, Olive, Strohmeier, Schäfer, Martinovic and Lenders22].

Another limitation of the collected dataset was that whilst both the FSC and LCC were European-based operators, the aircraft considered were all based at airports within the same country. This limitation could lead to the aircraft across the two fleets operating on similar routes, leading to a reduction in the impact of the operator characteristics on ground manoeuvres. Future work should endeavour to also consider the impact of geographical characteristics of operators, through investigating aircraft not just from European fleets, but global fleets. A wider consideration of other FSC operators would also provide insight as to whether fleet rotation around various large international airport bases impacts ground manoeuvre occurrence, as the FSC fleet within this study were consistently based at a large international airport. In addition, the initial study into ground manoeuvres of an ULCC aircraft in Section 4.4 highlights the need to study ULCC aircraft fleets further. These limitations of the dataset presented in this paper provide the natural follow-on and future work concerning the exploitation of ADS-B trajectories for characterising aircraft ground manoeuvres. Ultimately, such a comprehensive study could support the construction of bespoke fatigue spectra related to operator business and geographical characteristics, leading to more efficient and if required, more reliable, landing gear structures and systems.

Concerning future work, the collected ADS-B trajectories has highlighted significant landing gear focused opportunities. Within on-going work, ADS-B trajectories containing taxi routes related to towing the aircraft from maintenance/remote parking to the departure gate (see Fig. 23) have been observed. The towing portion of the trajectory could be segregated from the pre-takeoff taxi due to the significant duration observed between the towing and pre-takeoff taxi sections due to preparations for flight and boarding. Fatigue loads applied to aircraft landing gear are known to be significantly different under towing operations, especially for the nose landing gear [Reference Buxbaum1, Reference Ladda and Struck4] and consequently, ADS-B trajectories may provide a route to better characterising the manoeuvres performed during aircraft towing. In addition, the increasing number of regional aircraft now fitted with ADS-B transponders [Reference Jux, Foitzik and Doppelbauer39] will permit ground manoeuvre occurrences for such aircraft to also be investigated, permitting comparison across regional, narrow-body and wide-body aircraft. A recent example of the exploitation of ADS-B ground trajectories for regional aircraft concerns the derivation of electric taxiing requirements for regional aircraft presented by Taltaud et al. [Reference Taltaud, Carbonneau-Côté, Bouchard and Rancourt40].

Figure 23. Example of towing trajectory pior to pre-takeoff taxi. ADS-B data from Flightradar24 [24]. Map data from OpenSteetMap (https://www.openstreetmap.org/copyright).

Finally, whilst this paper has considered landing gear specifically, the continual and evolving utilisation of ADS-B data in the area of structural health monitoring highlights future opportunities for aerospace structures and systems in general. Of current interest to the authors is the exploration of inferring airframe loads from ADS-B data [Reference Hoole, Booker, Cooper, Richardson and Hallam30], which could facilitate the remote monitoring of aircraft structures based upon air traffic data.