On the effect of popular additive manufacturing technologies on the performance and noise emission of UAV propellers

On the effect of popular additive manufacturing technologies on the performance and noise emission of UAV propellers J. García-Tíscara,∗, P. Quinteroa, P. Varelaa, F. N. Ramíreza, A. Cremadesb aCMT – Clean Mobility & Thermofluids, Universitat Politècnica de València, Camino de Vera, 46022 Valencia, Spain bFLOW, Engineering Mechanics, KTH Royal Institute of Technology, Stockholm, Sweden Abstract Noise remains a major concern for the widespread employment of Unmanned Air Vehicles (UAVs) operations, especially in urban environments. Moreover, not only must overall noise levels be considered when analyzing the acoustic impact of UAVs, but noise quality based on the spectral content of the acoustic signal must also be considered due to the psycho-acoustic effect on the listeners. Additionally, as UAVs become widespread, the ability to perform field replacement of propellers using additive manufacturing (AM) is of increased interest to many operators. However, as different AM techniques become popular, their impact on performance and noise must be assessed. In this investigation, the effect of manufacturing characteristics, such as surface roughness and flexibility, on the propeller thrust, torque, and acoustic signature is explored through an experimental campaign, revealing differences depending on the selected technique. A numerical experiment is then performed to isolate the effect of flexibility and roughness on these characteristics. Keywords: Propellers; UAV; Acoustics; FDM; SLS; SLA; 1. Introduction1 Unmanned Air Vehicles’ (UAVs) popularity as an aerial 2 platform for a wide range of civil and military applica3 tions continues to increase. From recreational purposes 4 to surveillance and fire safety, passing through filmmaking, 5 infrastructure inspection, road traffic control, cell network 6 enhancement, to name a few, these aircraft are becoming 7 widespread in both the countryside and urban spaces [1].8 However, even though the technology in terms of flight 9 performance is mature, its full potential, given all the de10 scribed applications, is not yet fully realized, partly due to 11 a combination of regulatory, safety, and acceptability con12 cerns. On the first two fronts, several initiatives in the US, 13 EU, UK, etc., aim to achieve a harmonized regulation that 14 allows the safe operation of UAVs [ 2 ]. However, there are 15 still some concerns about the public acceptability of UAV 16 operations. Different public opinion surveys conducted by 17 aviation authorities have found that UAV noise is one of the 18 leading factors among these concerns [3].19 When studying UAV noise, two aspects should be con20 sidered. On the one hand, there is the issue of overall noise 21 level, which is the added global level that continued UAV 22 operation could bring into homes, schools, hospitals, etc. 23 According to the findings of the World Health Organization 24 (WHO), noise levels are the second most harmful form of 25 environmental cause of health problems, just after chemical 26 pollutants [ 4 ]. Thus, care should be taken that the new 27 ∗Corresponding author. [email protected] services provided by using UAVs do not increase these noise 28 levels and their health consequences for the population. On 29 the other hand, the psycho-acoustic impact of UAV noise 30 signatures has been attested. Due to the particular flight 31 dynamics of UAVs, based on multi-propeller speed control, 32 the frequency content of the noise presents a semi-random 33 variance around an average, often perceived as especially 34 annoying [ 3 , 5 ]. Therefore, both overall noise level and fre35 quency content must be studied to address these concerns 36 properly. 37 Consequently, the amount of research in propeller aeroa38 coustics has increased considerably in latter years. Recently, 39 a benchmark for low Reynolds number propellers has been 40 proposed [ 6 ]. Other works have focused on evaluating the 41 accuracy of different numerical methods comparing differ42 ent blade resolved methods with BEMT solutions [ 7 , 8 ], 43 or with the actuator line method [ 9 ]. The influence of 44 non-axial inflow conditions on the acoustic performance 45 has been also explored both experimentally [ 10 ] and nu46 merically [ 11 ]. In addition, aerodynamic and aeroacoustic 47 propeller-propeller and propeller-fuselage interactions have 48 also been studied for both multicopters [ 12 ] and fixed-wing 49 UAVs [13,14]. 50 Because of the low Reynolds at which these propellers 51 operate, certain phenomena like laminar separation bubbles 52 become important [ 15 , 16 ]. Moreover, broadband noise be53 comes more relevant, so efforts are being made to improve 54 the accuracy of numerical methods [ 17 , 18 ]. The influence 55 of the Additive Manufacturing (AM) technology used to 56 manufacture the propeller has proven to have an impact on 57 1 this broadband noise [17,19].58 In parallel with UAVs, another technology that is be59 coming widespread is Additive Manufacturing (AM). Since 60 UAVs are often subjected to crashes or forced landings, 61 which may imply a degree of damage, and given that the 62 propellers are often the most easily damaged part, rapid 63 in-field replacement of these components becomes an ex64 tremely desirable capability for many kinds of operators 65 [ 20 ]. For example, UAVs are being used to deliver medical 66 supplies in remote locations or even embarked on ocean67 going vessels. Obtaining spare parts such as propellers from 68 the UAV manufacturer is difficult or impossible in these sit69 uations, and thus, the ability to rapidly manufacture them 70 from a supply of raw material becomes highly desirable.71 However, replacing a commercial propeller with a 3D72 printed one is not straightforward, even if the geometry 73 is exactly the same. As different AM techniques result in 74 pieces with distinct mechanical properties such as flexibil75 ity, surface finish, roughness, anisotropy, etc., there is no 76 certainty that the performance of a recreated propeller, in 77 terms of thrust, torque, and noise emission, is equivalent to 78 that of the original. Previous works [ 17 , 21 , 22 , 23 ] have 79 proven that hover performance and noise emissions are af80 fected by the AM technology used, but further work needs 81 to be done to fully comprehend these effects under different 82 operating conditions and for the entire noise spectrum. To 83 address this issue, in this investigation, we consider one 84 wooden commercial propeller typical of UAVs (XOAR 9x7) 85 and recreate it using the three most widespread AM tech86 niques, expanding the results presented by García-Tíscar 87 et al. [ 24 ]. All the proposed propellers have been tested 88 at CMT’s anechoic chamber and wind tunnel to assess the 89 impact of the different AM methods on thrust, torque, and 90 noise emission, including overall level, directivity, and spec91 tral content. Since each technique introduces its particular 92 combination of mechanical properties, a numerical simu93 lation campaign is performed, in which each mechanical 94 property of the material is modified separately, allowing 95 a better understanding of their impact on the propeller 96 performance.97 The present paper is structured as follows: this section 98 provides a brief introduction to the subject and a contextual99 ization of the current state-of-the-art of the field. Section 2 100 presents the propeller used, the additive manufacturing 101 techniques evaluated, the experimental setup, and the nu102 merical approach. Section 3shows and discusses the ex103 perimental results, and the conclusions obtained from the 104 numerical analysis. Finally, in Section 4, the final conclu105 sions reached are summarized.106 Scanning Processing + Additive manufacturing Mesh detail Figure 1: Workflow of the scanning, meshing and manufacturing of the propellers used in this work. Top image courtesy of UAV Works Group. 2. Methodology 107 2.1. Selected propeller 108 The commercial wooden XOAR PJN 9x7" (22.86 cm) 109 propeller was selected for this research. This model of pro110 peller is typical of fixed-wing UAVs such as the VALAQ120 111 by the UAV Works Group. In order to create the additive112 manufactured (AM) versions of the propeller, the geometry 113 Building platform Extruder head Heater Supports Thermoplastic filament spool Figure 2: Conceptual schematic of the FDM technology. 2 of the propeller was extracted by manually fine-tuning a 114 structured-light 3D scanner digital model. Figure 1depicts 115 the aforementioned process.116 2.2. Additive manufacturing117 The term Additive Manufacturing (AM), also known 118 popularly as 3D printing, encompasses a large variety of 119 very different technologies. Each technology has its own 120 strong points and weaknesses, especially in terms of me121 chanical properties and surface finish of the resulting items. 122 In this investigation, three of the most popular technolo123 gies have been selected, which are briefly described in this 124 subsection.125 2.2.1. Fused Deposition Modeling (FDM)126 Fused Deposition Modeling (FDM), also known as Fused 127 Filament Fabrication (FFF), is probably the most well-known 128 AM technology and can be considered as the one kickstart129 ing its recent popularity. In its essence, a continuous fila130 ment of thermoplastic material is extruded through a heated 131 printing head, which causes the material to melt or soften 132 enough to fuse with the previously deposited material. The 133 extruder head is continuously moved as required to form the 134 geometry of the desired item. Figure 2shows this process.135 The movements of the extruded heads are pre-computed 136 through software called slicers because the typical approach 137 is to “slice” the item geometry into a series of horizontal 138 planes. Commonly, the head will complete all the required 139 XY movements in each Z plane before moving to the next 140 one, which is typically accomplished by lowering the build141 ing platform itself. This process results in a manufactured 142 item that clearly shows the different Z layers in its surface 143 finish, in addition to possible “Z seams” resulting from the 144 filament extrusion’s starting and finishing points at each 145 layer. Moreover, the overhanging parts of the geometry 146 require supporting material that needs to be removed in 147 postprocessing steps.148 In addition, the process of building the object through 149 subsequent Z layers of fused or deposited cordons of thermo150 plastic filament results in anisotropic mechanical properties, 151 as the behavior in the Z direction is usually quite different152 from that of the X and Y directions. Furthermore, objects 153 are not fully filled even if 100% infill is selected, as the 3D 154 space cannot be filled by the cylindrical geometry of the 155 filaments.156 2.2.2. Selective Laser Sintering (SLS)157 SLS technology solves many of the common issues of 158 FDM. In this approach, a high-power, orientable laser melts 159 small particles of the building material (typically a polyamide 160 such as PA11 or PA12), fusing them together to form a 161 solid object. Again, the process involves slicing the object 162 into Z planes, as the laser can only traverse the XY plane. 163 Powdered material from a reservoir is thinly spread over a 164 movable bed through a roller, where the laser sinters the 165 required slice. Then, the bed is lowered, a fresh and un166 sintered powder layer is spread on top, and the process 167 repeats until the object has been fully built. The process is 168 shown in Fig. 3.169 This technique presents several advantages. First, the 170 resulting object has practically isotropic mechanical proper171 ties, as the powder particles have fused together. Also, the 172 surface resolution is better than that of the FDM process, as 173 it is not limited by the geometry of the deposited filaments 174 but by the effective size of the laser. Finally, as the sintered 175 material is supported by the unsintered powder, no support 176 material is required. In fact, very complex and interlocking 177 geometries can be produced by taking advantage of this 178 fact. 179 However, this method is not without disadvantages. 180 First and foremost, the requirement of a high-powered laser 181 increases the complexity and cost of the machine, restricting 182 its use to more professional settings rather than hobby or 183 household environments. Working with powdered material 184 can also be a nuisance, and require procedures and/or a 185 dedicated workshop. The surface finish is usually rough, 186 even if it is free of visible Z layers or seams. 187 2.2.3. Stereolithography (SLA) 188 Finally, the SLA technique relies on the photopolymer189 ization of resin. When a UV light source is focused on the 190 bottom of a vat containing photopolymer resin, the resin 191 solidifies through a photochemical process. The laser is 192 steered to solidify the Z slice as required, and then, the liq193 uid vat is lowered and the laser draws the next slice, which 194 is fused with the previous one, as shown in Fig. 4.195 Through this process, layering or seam effects are greatly 196 reduced. Also, the resulting pieces exhibit isotropic me197 chanical properties. Nowadays, SLA machines are more 198 accessible than SLS ones. Furthermore, a wide array of 199 resins with different properties is available, although the 200 material choice is not as varied as with FDM. 201 A variant of this technique uses a projector to shine 202 the complete image of each slice into the resin vat, which 203 lowers the cost of the machine even more. These are known 204 as Digital Light Processing (DPL) machines. However, the 205 projects work by projecting a raster image (made up of 206 Unsintered powder Roller Laser head Building platform Figure 3: Conceptual schematic of the SLS technology. 3 Table 1: Mechanical properties of the additive manufacturing materials used in this study [25,26,27,28]. Property Unit FDM: ABS-M30 SLS: Nylon 12 SLA: Grey v4 (PC) SLA: Rigid (PC) Tensile modulus XZ GPa 2.4 1.85 2.8 4.1 Tensile modulus ZX GPa 2.3 Flexural modulus XZ GPa 2.22 1.6 2.2 3.4 Flexural modulus ZX GPa 1.96 Ultimate tensile strength XZ MPa 28.1 50 65 69 Ultimate tensile strength ZX MPa 26.8 Flexural strength XZ MPa - 66 - - Flexural strength ZX MPa 47.7 Tens. elongation at break X/Y % 8.1 11 6 5.3 Tens. elongation at break Z % 1.8 6 Heat Deflection Temp 0.45 Mpa ºC 103.8 171 73 77 Heat Deflection Temp 1.8 Mpa ºC 99.9 87 58 60 pixels) instead of the continuous movement of the laser spot 207 of SLA machines. Thus, DPL results in pieces exhibiting a 208 certain voxel effect, as if constructed of tiny cubes.209 While SLA presents certain advantages, it also has some 210 drawbacks. Working with the photopolymer can be difficult, 211 and certain precautions must be taken to handle it. The 212 resulting “green” pieces must be cleaned of liquid resin that 213 has failed to solidify. Then, the pieces must be cured with a 214 combination of temperature and UV light in order to attain 215 their best mechanical properties. Also, similarly to FDM, 216 the pieces often require support as they are being built. 217 However, unlike with FDM in which an auxiliary extruder 218 can provide a specialized, soluble support material, the 219 supports in SLA need to be made of the resin available in 220 the vat. When removing the supporting struts, some marks 221 may remain on the surface of the piece, requiring manual 222 sanding.223 2.2.4. Facilities and materials224 For this research, the FDM propeller was manufactured 225 in Acrylonitrile Butadiene Styrene (ABS-M30) thermoplas226 tic using a Stratasys F170 machine. The Z layer resolution 227 was 125 µ m, with the propeller being printed in layers par228 allel to the rotation plane. 100 % infill was selected in the229 slicer software. A limonene-soluble support material was 230 also used to ensure proper attachment, which was removed 231 in postprocessing.232 Supports Liquid photopolymer Building platform Laser head Figure 4: Conceptual schematic of the SLA technology. The SLS propeller was manufactured using a Formlabs 233 Fuse 1 system. It features a 10 W ytterbium laser and is 234 capable of a layer resolution of 110 µ m, with a laser focal 235 point of 200 µ m. The employed material was Nylon 12. 236 This is the only technology that requires no support or 237 postprocessing other than cleaning the unsintered powder 238 off the piece. 239 Finally, the SLA propellers were created with a Formlabs 240 3L printer. This machine features a variant of the SLA 241 method called Low Force Stereolithography (LFS), which 242 uses a roller and a flexible tray floor in order to ensure that 243 only a very thin layer of resin is cured by the laser. Two 244 250 mW laser heads are used to speed up the process. The 245 selected layer height was 50 µ m, and the XY resolution was 246 25 µ m for the standard Formlabs Grey v4 propeller. On the 247 other hand, a layer height of 100 µ m and an XY resolution 248 Figure 5: The propellers tested in this investigation. From left to right: FDM, SLS, SLA-grey and SLA-rigid versions. A zoomed view highlights the differences in surface quality. 4 of 25 µ m was used for the Rigid 4000 resin propeller. The 249 propellers were then cleaned with isopropyl alcohol and 250 cured with the temperature and UV light cycles prescribed 251 by the manufacturer. The support marks were manually 252 sanded off.253 In Fig. 5, the four propellers manufactured by each of 254 the selected technologies are compared against each other. 255 It can be seen how the FDM propeller features very clear 256 layers, along with the paths of the filament (note especially 257 how the shaft wall is created by a circular filament deposi258 tion). The SLS propeller does not show any inhomogeneity, 259 but a rougher surface can be clearly appreciated. Finally, 260 the SLA propeller shows a smooth surface finish, appearing 261 similar to a cast plastic piece.262 Focusing on the material properties, Table 1displays the 263 main mechanical properties of the materials used in this 264 study.265 2.3. Experimental setup266 In order to measure the noise emission of each propeller, 267 an electric motor was installed in the anechoic chamber 268 available at Laboratory 5K of the CMT - Clean Mobility & 269 Thermofluids Institute. This chamber features a frequency270 cut-off of 100 Hz, and an available interior space of 7.5 ×271 6.5 × 6 m. The non-influence of recirculation in this facility 272 for 9-inch propellers measurements has been previously 273 demonstrated [ 29 ]. The engine is an Avenger V3 2812-900 274 kv and is governed by a V-GOOD 2-6S 40A Electronic Speed 275 Controller (ESC). An Arduino UNO microcontroller, linked 276 to a computer in the control room, commands the ESC.277 To measure the acoustic emission, six Brüel & Kjær Type 278 4190 free-field microphones are mounted in an arch, at a 279 Microphones 90º 60º -60º 30º -30º 0º Anechoic walls Motor R = 8.75 Dp ESC RPM PS Cont. PC DAQ Figure 6: Schematic of the experimental setup installed in the anechoic chamber at the CMT Institute. Fairing Propeller 6-axis balance Motor ESC Figure 7: FDM-manufactured propeller installed in the «Prof. Francisco Payri» wind tunnel. distance from the propeller hub of 8.75 propeller diameters 280 (Dp) . The six microphones are mounted in increments of 281 30 º , starting from the vertical (upstream) of the propeller. 282 They are simultaneously acquired by a PULSE Type 3560-D 283 Data Acquisition System from Brüel & Kjær, also connected 284 to the control room computer. A schematic of this setup is 285 shown in Fig. 6.286 After calibrating the microphones through a Brüel & 287 Kjær Type 3541 pistonphone, measurements were made at 288 164.5 rps (9870 rpm), capturing the acoustic signal for 2 289 seconds. 290 The aerodynamic performance characterization of the 291 different propellers was performed in the «Prof. Francisco 292 Payri» wind tunnel at CMT. The test was carried out in the 293 first test section of the wind tunnel reserved for low turbu294 lence aerodynamic tests, through a custom 6-axis balance 295 as shown in Fig. 7. The same motor and ESC as in the 296 anechoic chamber setup have been used. The balance has 297 been designed in-house and built using 6 Tedea-Huntleigh 298 Model 614 load cells at 70 degrees. Readings from the load 299 cells are directly acquired through a National Instruments 300 DAQ system. 301 After calibrating the load cells using reference weights, 302 measurements were made at different rotational speeds, 303 both with the tunnel stopped and at 10 m/s, to obtain data 304 in hover and forward flight conditions. Ten repetitions of 305 the measurements were performed for each propeller to 306 ensure repeatability. 307 2.4. Numerical setup 308 In order to better understand the role of surface rough309 ness and flexing under load on both aerodynamic and acous310 tic performances, a numerical experiment was carried out, 311 where advantage was taken of a previously validated numer312 ical setup for a rigid, smooth propeller previously described 313 by Serrano et al. [ 7 ]. In this work, the simulations were 314 5 performed following the same approach but using the com315 mercial software SimCenter STAR-CCM+ 2210.316 As can be seen in Fig. 8, only the propeller is modeled. 317 It is immersed in a spherical fluid domain with a radius 318 of 7.5Dp . A Moving Reference Frame (MRF) approach is 319 used, with the rotating domain defined as a small cylinder 320 around the propeller. An additional fixed cylinder is used as 321 a refinement volume in order to improve the mesh quality 322 in the propeller wake, bringing the total cell number to 2 323 million. An image of the computational mesh is also shown 324 in Fig. 8.325 The first main modification of the existing model for 326 this investigation was the consideration of Fluid-Structure 327 Interaction (FSI), solving a Finite Element Model (FEM) of 328 the propeller in coupling with the finite-volume fluid solver. 329 Given the nature of the said coupling, both solid and 330 fluid physics are implemented, with a mesh of 400 thou331 sand cells for the solid region. The simulations are treated 332 as pseudo-stationary. To achieve this, an implicit unsteady 333 model is employed with a sufficiently large time step, en334 suring that steady-state conditions are reached at high sim335 ulation times.336 The structural problem is solved using a nonlinear ge337 ometry model, which accounts for large displacements and 338 rotations by satisfying the equilibrium equations in the 339 deformed configuration and iteratively updating the stiff340 ness matrix via Newton’s method. A key implication is the341 treatment of follower forces, where forces, tractions, and 342 pressures adjust their orientation as the structure deforms: 343 force loads retain their vector components, traction loads 344 preserve their direction while their resultant changes and 345 pressure loads remain normal to the deformed surface. This 346 7.5Dp Fluid domain Mesh detail Propeller Refinement volume Rotating volume Figure 8: Numerical domain with detail of the mesh. approach ensures a more accurate representation of slender 347 structures undergoing large rotations with relatively small 348 strains [30,31]. 349 To ensure that the FSI simulation converges to a steady350 state solution, Rayleigh damping is introduced to suppress 351 structural oscillations. The damping matrix C is formulated 352 as a linear combination of the mass and stiffness matrices 353 [32], Mand Krespectively, as can be seen in Eq. 1.354 C=αM+βK,(1) where α is the mass-proportional damping coefficient and 355 β is the stiffness-proportional damping coefficient. In this 356 study, these coefficients are set to α =0.001 Hz and β =0.001 357 s, calibrated based on the timescales of the observed oscil358 lations. This choice ensures numerical stability and accel359 erates convergence, allowing for an efficient evaluation of 360 the steady-state structural response. However, it inherently 361 suppresses dynamic instabilities, which could be relevant 362 in cases of low structural stiffness or high rotational speeds. 363 While such effects may play a role in certain operational 364 regimes, their analysis falls beyond the scope of this study, 365 which focuses on the steady-state behavior of the coupled 366 system rather than transient or instability-driven phenom367 ena, as studied in previous works [33,34,35,36]. 368 The rotation is modeled using a Moving Reference Frame 369 (MRF), which imposes a constant rotational flux in the 370 conservation equations without physically deforming the 371 mesh. This approach significantly reduces computational 372 cost compared to a fully transient rotating mesh simulation, 373 as can be seen in the work of Garofano-Soldado et al. [ 37 ] 374 or Liu et al.[38]. 375 The fluid flow is governed by the Reynolds-Averaged 376 Navier-Stokes (RANS) equations, discretized using second377 order schemes for advection and diffusion terms. To model 378 turbulence, the k−ω Shear Stress Transport (SST) model, 379 proposed by Menter [ 39 ], is employed to compute the 380 Reynolds stress tensor. This two-equation turbulence model 381 is widely used in MRF and propeller simulations [37,38]. 382 Although the flow remains essentially incompressible 383 across all tested advance ratios, the fluid is modeled as an 384 ideal gas to enhance numerical stability and ensure solver 385 convergence. This approach helps avoid potential issues 386 associated with solving the continuity equation in strictly in387 compressible formulations, where small numerical pressure 388 fluctuations can affect computational stability. Additionally, 389 treating the fluid as an ideal gas provides greater flexibil390 ity in solving the conservation equations, reducing the risk 391 of divergence without compromising accuracy within the 392 evaluated range of conditions [40,41]. 393 The second study introduced a variable wall roughness 394 via a modification of the wall treatment to sweep this param395 eter and assess its influence on performance. This allows 396 us to study the effect of both parameters in isolation to 397 improve the understanding of the experimental results. 398 In order to modify the wall roughness, a roughness 399 function is introduced (Eq. 3) [ 42 ]. Said function depends 400 6 on the roughness parameter R+, defined as in Eq. 2.401 R+=rρu∗ µ(2) f=     1if R+≤R+ smooth hBR+−R+ smooth R+ rough−R+ smooth +CR+ia if R+ smooth <R+<R+ rough B+CR+if R+>R+ rough (3) Where B,C,R+ smooth,andR+ rough are model coefficients (de402 fined in Table 2), r is the equivalent sand-grain roughness, 403 ρ the density, u∗ the velocity scale, µ the dynamic viscosity, 404 and ais defined as in Eq. 4.405 a=sin  π 2 logR+/R+ smooth logR+ rough/R+ smooth (4) Finally, the Hanson acoustic model [ 43 ] was used to 406 estimate the tonal noise numerically. This allows us to study 407 the influence of the material and surface roughness on the 408 tonal noise without having to carry out costly transient 409 simulations.410 The Hanson model is a theoretical propeller noise formu411 lation that takes into account both the loading (Eq. 5) and 412 thickness (Eq. 6) noise sources and distributes them through 413 a helicoid. It has been recently used to study propellers’ 414 noise [ 44 ] and compared with other acoustic analogies 415 [ 45 , 46 ]. In this work, the input of the Hanson model is 416 the thrust and torque distributions along the blade span, 417 extracted from the steady CFD simulations.418 PmL=mBMtsin θrex p [imB (ΩSr/c+ (φ0−(π/2)))] 2p2πY rt(1−Mcos θr)× Ztip hub cosθ0 r 1−Mcosθr dT dr−1 r2Mtrt dQ drex p(iφs)JmBΨLdr (5) PmT=−ρc2Bsinθrex p [imB (ΩSr/c+ (φ0−(π/2)))] 4p2π(Y/D)(1−Mcosθr)× Ztip hub M2 s(h/b)ex p(iφs)JmB k2 xΨVdr (6) Loading and thickness components are computed using 419 equations 5and 6, where PmL and PmT are respectively the 420 loading and thickness components of the noise, dT and 421 B C R+ smooth R+ rough 0 0.253 2.25 90 Table 2: Model coefficients for roughness function. dQ are the thrust and torque of the differential element 422 of length dr , m is the harmonic number, B the number of 423 blades, Mt is the tip Mach number, θ the observer angle 424 relative to flight direction, θr the observer angle in retarded 425 reference frame and θ0 r the same angle but relative to the 426 propeller shaft axis, Ω the rotational velocity of the propeller, 427 S the distance of the observer from the propeller hub and 428 Sr the same distance in retarded reference frame, c the 429 speed of sound, φ the tangential angle and φ0 the same 430 angle relative to the propeller shaft axis, Y the observer 431 distance from propeller axis, rt the radius of the propeller, M432 the freestream Mach number, r the non-dimensional radial 433 position, φs the phase shift due to sweep, ρ the density, D434 the diameter of the propeller, Ms the section relative Mach 435 number, h the maximum airfoil thickness, and b the chord 436 of the blade at each section. 437 Finally, JmB is defined as in Eq. 7, kx is a wave number 438 defined by Equation 8, and ΨV and ΨL are non-dimensional 439 source transforms representing the effect of chord-wise non440 compactness. As suggested in Ref. [ 47 ] and used in Ref. [ 44 , 441 46 ], a parabolic thickness distribution and a uniform lift 442 distribution were applied, defined in Equations 9and 10.443 JmB =JmB mBrMtsinθ0 r 1−Mcosθr(7) kx=2mBbMt Ms(1−Mcosθr)(8) ΨV(kx) = ¨2/3kx=0 8 k2 x [2 kxsin kx 2−cos kx 2]kx6=0(9) ΨL(kx) = ¨1kx=0 2 kxsin kx 2kx6=0(10) 3. Results & discussion 444 In this section, the experimental data will be presented 445 and discussed, and the results of the numerical models will 446 be used to draw further conclusions on the material and 447 surface roughness influence as isolated effects. 448 The performance data will be presented as dimensional 449 forces for the hover results (thrust and torque), and as non450 dimensional coefficients for the forward flight data, using 451 the definitions shown in Equation 11, where ρ∞ represents 452 the undisturbed air density, n is the rotational speed of the 453 blade in rev/s, and Dis the diameter of the propeller. 454 CT=T ρ∞n2D4CP=P ρ∞n3D5(11) For the acoustic data, the spectra at different angular 455 positions will be shown, and the SPL at the BPF and OASPL 456 (Eq. 12) directivities will also be computed. 457 OASP L =20 ·log10 PRMS P0(12) 7 3.1. Experimental campaign458 After manufacturing the different propellers using the 459 AM technologies described in section 2.2, these were in460 2000 4000 6000 8000 10000 12000 Rotational Velocity [rpm] 0 2 4 6 8 10 12 Thrust [N] FDM SLS SLA SLA Rigid 2000 4000 6000 8000 10000 12000 Rotational Velocity [rpm] 0 0.05 0.1 0.15 0.2 Torque [Nm] FDM SLS SLA SLA Rigid Figure 9: Thrust (top) and torque (bottom) vs rotational velocity in hover for each propeller measured in the wind tunnel. 0.2 0.25 0.3 0.35 0.4 0.45 0.5 0.55 Advance Ratio [-] 0.02 0.04 0.06 0.08 0.1 CT [-] FDM SLS SLA SLA Rigid 0.2 0.25 0.3 0.35 0.4 0.45 0.5 0.55 Advance Ratio [-] 0.02 0.025 0.03 0.035 0.04 0.045 CP [-] FDM SLS SLA SLA Rigid Figure 10: CT(top) and CP(bottom) vs Jfor each propeller measured in the wind tunnel at a freestream velocity of V∞= 10 m/s. stalled and tested in the anechoic chamber and wind tunnel 461 test benches, where the raw readings from the microphones 462 and the load cells, respectively, were recorded. As stated 463 before, at least 10 repetitions of the force measurements 464 were performed for each propeller to ensure repeatability. 465 The time evolution of thrust and torque during the ex466 periments was obtained using the load cell calibrations 467 performed before the measurement. Measurements at sev468 eral rotational speeds were carried out with the wind tunnel 469 stopped for the hover measurements, and with the wind 470 tunnel at 10 m/s for the advance operational conditions. For 471 each case, the average of the thrust and torque signals was 472 computed, and the standard deviation of said mean was ob473 tained, with the results being shown in Fig. 9and 10. It can 474 be observed that at hover and low rotational speeds, there 475 is no significant difference in thrust between propellers. 476 In contrast, as the rotational speed increases, there is a 477 substantial increase in the performance of the propellers 478 made by SLA relative to the other two, with the difference 479 reaching up to 9.5% between the standard SLA and the 480 FDM propellers at the highest rotational speed. 481 On the other hand, the propeller manufactured by SLS 482 generates the highest torque, which is appreciable at all 483 rotational speeds. The other three propellers generate a sim484 ilar torque, being the FDM and SLA Rigid propellers slightly 485 higher than the SLA one at high rotational speeds. The 486 difference between the SLA and the SLS propeller torque 487 at the maximum rotational speed is up to 15%. 488 The findings regarding the SLA and SLS propellers are 489 consistent with current literature [ 17 ], where SLA pro490 pellers were found to produce a higher thrust and lower 491 torque at hover compared with SLS ones. 492 Regarding the measurements with mean flow, Fig. 10 493 shows the thrust and power coefficients evolution with re494 spect to the advance ratio. The advance ratio can be defined 495 as J=Un/D , where U is the flow velocity, n is the rotational 496 speed and D is the propeller diameter. It can be observed 497 that the differences in CT are higher than the ones seen in 498 hover. Both SLA propellers produce a higher CT at low val499 ues of J , as expected based on the hover results. However, 500 the slope of the SLA propellers is steeper than the other 501 two. Because of this, the thrust coefficient obtained at high 502 values of Jis higher for the SLS and FDM propellers. 503 In terms of power coefficient, the hover trends seem to 504 continue for all the J range, with the SLS propeller being 505 the one with higher torque and the SLA propeller being the 506 one with a lower CP . Nevertheless, the CP of the SLA Rigid 507 propeller tends to get close to the SLS one at high advance 508 ratios. 509 It can be seen that there are no clear tendencies that can 510 be identified, as the two factors that have the greatest effect 511 on performance (material properties and surface finish) 512 operate simultaneously, hiding the individual effect. For 513 this reason, a numerical investigation will be carried out to 514 try to explain the findings obtained and to acquire a deeper 515 understanding of the underlying phenomena. 516 As for the noise generated by each of the propellers 517 8 manufactured with AM, the SPL at the BPF and OASPL 518 directivity obtained around the propeller at a radius of 519 2 meters are shown in Fig. 12. To expand the analysis, 520 in Fig. 11, the spectrum of the sound signal recorded at 521 −30deg and 30deg is presented.522 It can be observed that at the first blade passing fre523 quency, the sound level achieved is very similar among the 524 four propellers in the −60deg to 30deg range, being slightly 525 higher for the SLS propeller than for the other two. Near 526 the axis of rotation, the difference increases, with the SLA 527 Rigid propeller being the most silent one, with a difference 528 of up to 8.75 dB.529 This sound increase is also observed, with a greater noise 530 increment, in the broadband noise at higher frequencies, 531 where it can be seen how the SLS propeller generates a 532 considerable noise increment, presumably due to the surface 533 finish that causes the generation of a turbulent boundary 534 layer. This phenomenon can be clearly observed in Fig. 11 535 and explains the observed increase in OASPL directivity for 536 the SLS propeller in Fig. 12 (down), with a difference of up 537 to 5 dB depending on the angle.538 In addition, it can be observed in Fig. 11 how at the 539 shaft frequency (half the first blade passing frequency) a 540 peak appears. This peak is higher in the case of the pro541 pellers manufactured in SLA, which would indicate that 542 these are the propellers with the greatest geometrical dif543 ference between blades. This is probably due to the effect 544 of hand sanding on the geometry. In fact, the SLA Rigid 545 propeller shows an increase of up to 10 dB in the noise at 546 0 1000 2000 3000 4000 5000 Frequency [Hz] 20 30 40 50 60 SPL [dB] FDM SLS SLA SLA Rigid 0 1000 2000 3000 4000 5000 Frequency [Hz] 20 30 40 50 60 SPL [dB] FDM SLS SLA SLA Rigid Figure 11: Frequency spectra for the -30º(top) and 30º(bottom) microphone positions hovering at 9870 rpm in the anechoic chamber. 0° 30° 60° 90° 120° 150° 180° 210° 240° 270° 300° 330° 30 40 50 60 70 FDM SLS SLA SLA Rigid 0° 30° 60° 90° 120° 150° 180° 210° 240° 270° 300° 330° 60 70 80 90 FDM SLS SLA SLA Rigid Figure 12: Sound Pressure Level at the Blade Passing Frequency (top) and Overall Sound Pressure Level (bottom) directivity of the propellers at 9870 rpm. the shaft frequency. This is consistent with the needed level 547 of post-processing (higher in the Rigid case compared with 548 the standard SLA one). 549 3.2. Numerical results 550 Regarding the numerical results, as stated before, steady 551 simulations have been carried out, modifying the mate552 rial Young’s modulus (assuming isotropic materials) and 553 modifying the surface roughness of the propeller (using a 554 roughness function and assuming isotropic roughness) 555 3.2.1. Material influence 556 Regarding propeller performance, Figure 13 shows the 557 thrust and power coefficients against advance ratio curves 558 for different values of Young’s moduli. It can be observed 559 that at hover and low values of J there is a small increment 560 in CT for Young’s moduli between 1 and 30 GPa. This 561 tendency increases for lower values of E, with a difference 562 of ∼ 22.5% between the 0.05 GPa case and the rigid (30 GPa) 563 one. Nevertheless, all curves converge to the same value 564 when the advance ratio is increased, reducing the influence 565 of the material. The same behavior can be observed for 566 the power coefficient, with higher increments at hover (up 567 to ∼ 58% for the worst-case scenario). The gain in thrust 568 produced by the influence of the material is found to be 569 smaller than the increase in torque, resulting in reduced 570 9