Noise mitigation of UAV operations through a Complex Networks approach

Harun Siljak 1

Karina Einicke 2

John Kennedy 3

Trinity College Dublin, the University of Dublin College Green Dublin 2 D02 PN40 Ireland

Stephen Byrne 4

Transport Infrastructure Ireland Parkgate Business Centre Dublin 8 D08 DK10 Ireland

ABSTRACT This research combines complex systems science, geographical information systems, and environmental noise modelling to analyse e ff ects of future air mobility in urban settings and plan e ffi cient routes for vehicles. The research used the environmental noise maps of an urban agglomeration produced under the END as input to inform the UAV operations. These maps reveal potential routes for the UAV operations where the noise impact of the vehicle can be embedded within a high background noise due to the existing sources modelled under the END. When an agent based model is superimposed on a real-world map simple strategies of the diverse agents in interaction with the environment reveal patterns, such as dominant paths, points of congestion, and suggest positioning of terrestrial infrastructure. We investigate how agents can overcome the conflicts and find trade-o ff solutions by interacting only with their immediate neighbours—therefore enabling autonomy, decentralization, and putting to use emergent self-organising behaviour. The potential impact of increased UAV operations on urban and peri urban regions is significant. Route optimisation which does not consider the noise is likely to impact on quite areas within our residential spaces and should be considered as part of noise action planning.

1 harun.siljak@tcd.ie

2 einickek@tcd.ie

3 jkenned5@tcd.ie

4 stephen.byrne@tii.ie

a slaty. inter.noise 21-24 AUGUST SCOTTISH EVENT CAMPUS O ¥, ? GLASGOW

1. INTRODUCTION

The European Union has implemented a new regulatory framework for the operation of UAVs or Unmanned Aerial Vehicles (UAV) [1]. These regulations were also retained in UK law [2]. A key consideration for the operation of many UAVs in European cities will be the noise emission. This new and poorly understood noise source has serious potential to disrupt the lives of many European citizens. These regulations set the method for determining the A-weighted sound power level for di ff erent classes of UAVs and testing follows the ISO 3744:2010 standards. This procedure requires the UAV to be tested under hovering conditions at its Maximum Take O ff Mass (MTOM), above one reflecting surface and su ffi ciently far away from any other reflecting surface. The testing is carried out in a hemispherical measurement surface around the source and above the reflecting plane. While it is clear to an acoustician that this test procedure may change the UAV noise source from what is experienced in flight, this procedure does represent a pragmatic choice that enables mass acoustic testing of UAVs. A recent study has shown that the sound power measurements from this procedure are suitable as an input to environmental noise modelling [3]. The result of this testing will be a requirement for the guaranteed sound power level to be represented in a pictogram on the UAV. They also set the maximum allowable sound power level of the UAV depending on its class. This maximum allowable sound power will reduce 2 years after the document came into force, and from 4 years after the document came into force. The legally required pictogram only requires a A-weighted total sound power level but the ISO 3744 procedure easily enables frequency dependent data to be captured for subsequent modelling steps. At present there are limited o ff the shelf noise reduction technologies available for implementation on commercial UAVs to help manufacturers meet these targets. Therefore one of the most attractive options to UAV operators will be changes to flight paths to minimise noise impact on the ground. In addition to higher flying altitudes there is also the possibility to mask the UAV noise emission near existing noise sources. It has been noted that public perception of UAV noise varies significantly in di ff erent soundscapes [4]. The research demonstrated that the perceived annoyance of UAV noise was 6.4 times higher than in soundscapes with low road tra ffi c than those with high road noise. A key recommendation of the research is that concentrating UAV flight paths along busy roads could mitigate the impact of UAV noise annoyance. This finding was echoed in the work by Palmer et al. [5], in this case “sporadic complaints” become “widespread” when there is a 5-6dB increase in noise emission due to UAVs. A cause for further concern is that UAV noise emission has been reported as more annoying than noise from traditional road vehicles and aircraft of the same ‘loudness’ level [6, 7]. With the advent of the Environmental Noise Directive (END) noise maps for many European cities are available and updated every five years. At present urban agglomerations having a population in excess of 100,000 persons are mapped. These noise maps are intended to be used to develop noise action plans which will eventually include UAV noise. Additionally researchers are beginning to make road network planning decisions on the basis of noise emission [8]. These factors have inspired this research where the END noise maps are used as input data to an agent based model of the UAV operations.

2. INPUT DATA AND MODELLING APPROACH

An environmental noise map containing road tra ffi c noise measurements was used as input data for the agent based modelling. Noise data in the map was produced in 2017 during the Round 3 environmental noise mapping under the END. Strategic noise maps of major roads 5 are generated according to guidelines of the EPA. [9]

5 Major roads are defined as exceeding a threshold of 3 Million vehicles a year.

The tra ffi c noise map shown in Figure 1 was produced by the four local authorities in Dublin (DCC, DLRCC, FCC and SDCC) and reported to the EPA (Environmental Protection Agency) [9] and displayed here in SoundPLAN. The EPA only provides 5 dB(A) intervals. Major streets are in the 70 dB(A) to 85 dB(A) spectrum. In most residential areas in the Dublin region, tra ffi c noise is kept under 70 dB(A). The lowest displayed noise level is 55 dB(A). White areas in the noise map of Figure 1 indicate no given noise data. These areas are assumed to have a road tra ffi c noise under 55 dB(A). Ireland already has one of the first commercial UAV delivery services, Manna Drone Delivery which operate in Balbriggan (Dublin), Ireland. Their UAV operating conditions are used to inform this modelling work. Their UAV’s fly at an altitude of 60 m. The maximum weight of the delivery is currently 2 kg and limited to a package size of 25 cm x 25 cm x 14 cm. Presently the delivery radius is confined to 2 km. [10]

Figure 1: END Noise map of Dublin City produced by Transport Infrastructure Ireland for the Round 3 noise maps, displayed in SoundPLAN

Agent based modelling (ABM) is a of the method from the complex systems science toolbox that has been commonly adopted for large networks in which it is relatively easy to distinguish boundaries between network elements [11]; such is the case in our scenario, as the larger complex behaviour emerges from simple decisions made by UAVs individually reacting to the environment and the mission they have. This is captured by an agent-based model: discrete agents (in our case UAVs) interact with the discretised environment (noise measurements in the area the UAV is flying through), and make decisions on their movement (direction of flying) based on the environment and their goals (travel destination). We set up the agent based model to evolve on the map from Fig. 1 which is discretised into pixels corresponding to 19 m × 19 m area. At every tick (discrete time unit of the agent-based model), every

(a)

(b)

Figure 2: Decision-making for an agent in presence of (a) di ff erent noise values, or (b) established UAV routes

agent (UAV) decides on the direction in which it will step into a pixel adjacent to the one it currently occupies. The rules according to which the agent chooses the direction are illustrated in Fig. 2. In the case of Figure 2a, the agent (blue triangle) is travelling to its destination (blue star) and chooses its orientation based on surveying its environment. Let the agent have a vision radius v ∈ N , which would allow it to observe noise values (map colours) for pixels r steps away from the current position. From all pixels in the surveyed radius that are closer to the destination than the pixel agent is currently on, the agent selects the darkest (noisiest) one and sets course towards it (dotted line), which is then followed by stepping into the pixel this direction meets first. In case there are no coloured pixels that satisfy this criterion (which is a common case in white areas of the map that have not been surveyed for noise), the agent takes the direction of the destination (dashed line). In this model we introduce one exception to this rule, inspired by Paths model in Netlogo [12]. In the case that a certain pixel is popular, i.e. flown through for a certain number of times ( p ∈ N ), we highlight such pixels. In absence of any other coloured pixels, these popular pixels are then chosen by the agent to steer the direction, if they are closer to the destination than the agent currently is. This case is represented in Figure 2b, where the agent is in a dominantly white space, with a frequently flown route marked in black.

3. RESULTS

To observe the global e ff ects such local rules make in the complex network consisting of UAVs and their surroundings, we devised the following scenario. The model was built using NetLogo [13], a common choice for ABM. We placed 15 hubs for UAV delivery services in Dublin area model. The locations were chosen to be realistic based on current purpose and spacing around the city. Beginning with a uniform 5km spacing across the city locations were chosen which were likely to operate UAVs in future. The actual locations include Dublin port, shopping centres, business parks and hospitals. Each UAV leaves one of these defined hubs with a randomly assigned destination within the radius of 3 km from the hub. After successfully reaching the destination through a route determined via local rules, the UAV retreats to the closest hub, and the procedure repeats. We choose to fly 500 UAVs, limit their vision radius to 200 m, and consider pixels to be frequently flown over after 10 flyovers happen. The simulation is ran for 10,000 ticks over a map that has been sampled at 1731 × 1211 pixels. Results of a sample simulation run for this scenario are shown in Figure 3, with the black colour representing frequently flown paths across the map.

Figure 3: Frequently flown paths in our model.

4. DISCUSSION

Figure 3 shows an emergence of complex structure of commonly used routes for UAVs spreading from the hubs in the city. At this point, we can distinguish and separately discuss two families of routes: those in the noise surveyed areas (colour-coded) and those in unsurveyed (white) areas. In surveyed areas, the trajectories follow some roads, pockets of noise around the M50 motorway, and all the motorways present on the map. Some roads in the city remain unused, usually because the alternative roads along the same direction are louder. This is a useful feature of the emergent behaviour as the UAVs are discriminating between roads within the surveyed map and preferring to operate on noisier routes. In the unsurveyed areas, the common routes emerge when UAVs attempt return to the hub: in pursuing noisy (surveyed) ground, they reach a corner of the surveyed area, and from there they take the direct orientation towards the hub to cross the white space. Since this is repeated by all UAVs that end up in a particular region after execution of their mission, such a straight line quickly becomes a popular trajectory. The preferential attachment of UAVs to popular routes in absence of surveyed data causes a formation of rays radiating out of a hub into the white space, if a hub is near a large unsurveyed area (e.g. hubs east of the C-shaped M50 motorway. Without the preferential attachment, UAVs whose destinations are across the white space from these hubs would depart the hub under the angle determined by the destination–a continuum of angles, rather than "runways" that we observe emerging in our case. The consequence of this behaviour is that the UAVs create corridors which can be thought of as additional aerial routes in the city transport network. These routes are in some way optimal based on

the distribution of UAV hubs chosen and potentially useful for defining UAV operations within the city by aviation or local authorities.

5. CONCLUSION AND OUTLOOK

The ABM has shown the potential for UAVs to self determine routes which aim to minimise the noise impact of their operations. Conceptually this is possible if they operate closer to a stronger noise source such as a road or railway line. In this research the input data consisted of a static environmental noise map but the procedure would be equally valid for dynamic, live information about the current noise levels within a city. Environmental noise maps to be prepared for Round 4 under the END in 2022 will be more detailed with the implementation of CNOSSOS-EU and can be supplemented by smart cities technologies thereby providing UAV operators with information that can mitigate the impact of the UAV operations. The next stage of this research will be to estimate the noise impact of the UAV routes and superimpose this on the existing noise map. This will quantify any increase in noise levels due to the operation of UAVs along these roadways. Future investigations could also include more detailed UAV flight profiles including accent and decent of the UAV and hovering above the drop o ff area. At present minimal information is available in literature for the sound power and directivity of delivery UAVs. This information is urgently required if environmental noise mapping is to be used to mitigate the e ff ect of UAV operations in our cities.

REFERENCES

[1] O ffi cial Journal of the European Union. Commission delegated regulation (EU) 2020 / 1058 of 27 april 2020 amending delegated regulation (EU) 2019 / 945 as regards the introduction of two new unmanned aircraft systems classes. (32020R1058), 2020. [2] Commission delegated regulation (EU) 2020 / 1058 of 27 april 2020 amending delegated regulation (EU) 2019 / 945 as regards the introduction of two new unmanned aircraft systems classes. 2020. Publisher: Queen’s Printer of Acts of Parliament. [3] Kai Cussen, Simone Garruccio, and John Kennedy. Uav noise emission - a combined experimental and numerical assessment. Acoustics , 4(2):297–312, 2022. [4] Antonio J. Torija, Zhengguang Li, and Rod H. Self. E ff ects of a hovering unmanned aerial vehicle on urban soundscapes perception. Transportation Research Part D: Transport and Environment , 78:102195, 2020. [5] Nicola Kloet, Simon Watkins, Xu Wang, Sam Prudden, Reece Clothier, and Jennifer Palmer. Drone on: A preliminary investigation of the acoustic impact of unmanned aircraft systems (uas). 23-27 July 2017. 24th International congress on sound and vibration. London, England. [6] Doo Young Gwak, Dongyeon Han, and Soogab Lee. Sound quality factors influencing annoyance from hovering UAV. Journal of Sound and Vibration , 489:115651, 2020. [7] Beat Schä ff er, Reto Pieren, Kurt Heutschi, Jean Marc Wunderli, and Stefan Becker. Drone noise emission characteristics and noise e ff ects on humans—a systematic review. International Journal of Environmental Research and Public Health , 18(11):5940, 2021. [8] Haibo Wang, Bin Sun, and Liang Chen. An optimization model for planning road networks that considers tra ffi c noise impact. Applied Acoustics , 192:108693, 2022. [9] Noise Round 3 Road (Lden) - data.gov.ie. https://data.gov.ie/dataset/ noise-round-3-road-lden . Accessed: April 20th 2022. [10] Manna Drone Delivery. https://www.manna.aero/faq , 2022. Accessed: April 20th 2022.

[11] H Van Dyke Parunak, Robert Savit, and Rick L Riolo. Agent-based modeling vs. equation-based modeling: A case study and users’ guide. In International workshop on multi-agent systems and agent-based simulation , pages 10–25. Springer, 1998. [12] Robert Grider and Uri Vilensky. Netlogo paths model. Center for Connected Learning and Computer-Based Modeling, Northwestern University, Evanston, IL. [13] Seth Tisue and Uri Wilensky. Netlogo: A simple environment for modeling complexity. In International conference on complex systems , volume 21, pages 16–21. Boston, MA, 2004.