Aeronautical engineering - Aerodynamics of transport vehicles

Final project of the course

Projects

Numerical and Experimental Analysis of Aerodynamic Tail Devices for Drag Reduction of a Semi-Trailer Final project of the Aerodynamics of Transport Vehicles course Group: Francesco Lorenzo Alice, 10678140 Alessandro Crotta, 10621786 Matteo Dall’Ora, 10687723Lecturers: Prof. Luigi Vigevano Prof. Paolo Schito Ing. Francesco Fabio Semeraro 1.Introduction In the last decades, aerodynamic performance and drag reduction have played a central role in the design process of production vehicles, often clashing with overall dimensions and aesthetic goals, as today’s demand of reducing fuel consumption is one of the most challenging issues within the automotive industry. When long-haul commercial vehicles are concerned, the issue acquires an even greater importance as single percentage points of fuel saving can make a significant difference. In most cases, yet, semi-trailers external shape is nowadays still driven by the need of optimizing the internal storage capability while keeping overall dimensions limited in compliance to local regulations, limiting the room of improvement to external add-ons or minor shape changes. Multiple drag reduction devices have already been conceived and tested, showing promising results both in literature studies and practical applications, with some of them being commonly adopted by the latest production vehicles; this category includes roof spoilers and fairings, side skirts and more rounded and well-designed A-pillars and side mirrors. In spite of the effectiveness of these features, none of them directly targets the main source of trucks aerodynamic drag, being the large, low pressure separation bubble present at the back of the trailer. While boat tails and frame extensions do exist and have had a decent amount of research put into their development [3][5][7], they have yet to see widespread adoption. The ob jective of this study is thus to focus on trailing edge aerodynamic devices, performing a parametric analysis aimed at quantifying the importance of length and slant angle on the effectiveness of the tail plates in reducing the drag coefficient. Alternatively, more invasive inflatable devices targeting the trailing edge separation were also tested and compared to the simpler tail plates. The CFD analysis, performed on Ansys Fluent, was based on the use of incompressible RANS equations. Numerical results were later compared and validated using experimental wind tunnel data obtained on a scaled, 3D printed model of the same semi-trailer geometry.Figure 1:Stagnation region and separated wake 2.Case Study 2.1.CAD The baseline geometry was created from scratch using the CAD software SolidWorks, taking the Mercedes Actros tractor and aerodynamic trailer as a reference; the modeling process involved a reasoned geometry simplification with respect to the real semi-truck, aimed at providing a significant model for the CFD analysis, in compliance with the available computational resources. The main aspect of simplification was the removal of the vast ma jority of details having practical tasks in real world application, but negligible effects on the overall aerodynamics of the vehicle. Specifically, the intricate geometry characterizing the underbody of both the tractor and the trailer was strongly modified towards an easier, streamlined geometry; the underbody was extended to include both the wheels axis and the side-skirts, and the existing diffuser was heavily simplified. Also, the narrow spaces in correspondence of the connection between tractor and trailer were closed; the CFD simulation of those narrow spaces, practically allowing the real trailer to turn, is possible but computationally very expensive, as it would require several small elements to properly resolve the flow field in the area. As shown in Fig. [2], the small dimensions of the gap between cab and trailer of the Mercedes Actros semi-truck used as reference justified the choice to actually neglect its presence, realistically having little to no effect on the flow at the rear end of the trailer on which the following analysis is mainly focused. For the reasons listed above, a fairing was modeled by the group, and considered part of the baseline vehicle. The same reasoning was applied to further simplifications such as the absence of the side-mirrors, as well as the smoothing of surfaces like wheels, front grill and any external panel in general. The original frame extensions present on the Mercedes aerodynamic trailer were removed in order to perform the parametric analysis proposed in the following sections.Figure 2:CAD geometry - Mercedes Actros tractor and aerodynamic trailer 1 2.2.Mesh The mesh was created using Fluent Meshing. Fluid volumes surrounding each CAD geometry were obtained through Ansys Design Modeler, which also allowed to name every boundary surface and introduce appropriate refinement boxes in the computational domain. As steady 3D CFD simulations were performed, on a semi-trailer endowed with a longitudinal plane of symmetry, only half of the actual fluid volume around the vehicle was considered and meshed, in order to halve the computational domain and, hence, significantly reduce time and resources required by each simulation. A poly- hexcore mesh was generated, consisting of a boundary layer of prismatic elements with hexagonal base and an octree volume mesh, with a polyhedric transition layer between the two. Particular attention was given to the contact points between the tyres and the ground, where the cusp shape of the fluid cavity between the two could potentially lead to significant troubles amidst the meshing process. To avoid this problem, the lower portion of wheels was sliced off by the ground and filleted, creating a contact patch between the two surfaces. This is a widely used and realistic expedient, as real wheels are also compliant and do deform in the lower contact region under the weight of the vehicle. Refinement boxes were created to enclose both the vehicle and its wake, avoiding sharp transitions to the larger elements used on the external boundary of the computational domain; smaller refinement zones were also inserted to capture the main rear recirculation region and to guarantee appropriate element sizes in the underfloor region of the trailer. Surface refinement was also implemented on high curvature zones, as well as on the entire vehicle to avoid an excessive aspect ratio in the boundary layer region.Figure 3:Computational domain and refinement boxes Figure 4:Poly-Hexcore Mesh around semi-trailer geometry The boundary layer thickness was specifically set in order to provide acceptable first celly+ values (50-100), with the aim to employ wall functions. A first guess of the first cell height was obtained through the turbulent flat plate theory; the value was then adjusted using a dummy simulation. The dimensions of the computational domain were chosen basing on a literature round-up and shown in [6] More on this choice is presented in Sec. [3.1]. Cell dimensions and grid refinement were also ob ject of a study briefly presented in Sec. [3.2]. 2.3.Simulation Setup Simulations were run using the commercial software Ansys Fluent, exploiting the pressure-based incompressible RANS pseudo-transient solver and the coupled algorithm; turbulence was modeled usingk−ω−S S T. A freestream velocity of 25m/swas imposed at the inlet as well as on the ground, while the outlet was modeled with a constant pressure; it is important to note how all shown values of pressure are relative toP atm. Simple no-shear slip wall boundary conditions were chosen for the upper and lateral boundaries as they were far enough from the vehicle as to not interfere with the flow; the same treatment was used to impose the slip condition on the volume’s symmetry plane. The reference area was chosen equal to6m2 , which corresponds to frontal area of the halved truck. To facilitate convergence an initial flow field was computed using1st order upwind methods for both momentum and turbulence equations, later used as initial guess for a second simulation performed using2nd order methods. It is important to notice that while steady RANS were used, the problem, involving large separated regions and vortices, is in reality intrinsically unsteady. Second order simulations were considered at convergence when the behaviour of the force coefficient started showing a neat periodicity or, at least, confined oscillations around a constant mean value. To better interpret the solution, all results presented later were obtained through averaging among the last 500 or 1000 iterations. An example of the drag coefficient evolution is shown in Fig. [5]. First order initialization can be noticed as well from the firstC Dplateau.Figure 5:Drag coefficient convergence behaviour 2 3.Domain and Grid Independence 3.1.Domain Independence The purpose of the domain independence study is to verify the independence of the numerical results from the extension of the chosen computational domain. In this analysis, three grids with different farfield dimensions have been generated (approximately considering finite multiples of the truck length in each direction), while keeping constant the height of the cell closest to the boundary and the overall refinement of the grid. Secondly, a simulation for each mesh has been performed, keeping fixed all the boundary conditions and the flow properties. Finally, the results in terms of drag coefficient have been compared, providing the results presented in Tab. [1]. Farfield dimensions are given in terms of truck’s length, height and semi-width. Looking at the results it was chosen to proceed with the medium farfield option, as an error of5%was deemed excessive, and the savings in terms of computational cost are not significant enough to warrant it. 3.2.Grid Independence Grid independence study aims to ensure that the mesh is sufficiently refined to satisfactorily represent the flow field around the vehicle. Grid refinement strongly affects the number of cells and, hence, the computational cost of the simulation. The number of elements was in fact limited by the available computational resources, to a maximum of around 5-6 millions of cells. This limit, together with the need to achieve valuable results in a reasonable computational time, led to a sensible distribution of the available cells in the areas of greater interest for the present analysis, as described in Sec. [2.2]. Three different refinements of the same, previously chosen computational domain and bodies of influence were tested, yielding to the results, in terms of cell number and drag coefficients, provided in Tab. [2]. The results clearly show a convergent behaviour, thus the final refinement was identified in the 3.4M elements mesh, to guarantee a very respectable accuracy without getting too close to the computational limit of the available machines; in particular it was only possible to efficiently run meshes with less than 5.5M cells without resorting to memory paging, which would increase computational times at least tenfold.DomainC D∆ C D[%]6L ×4H×9W0.41044.69 8L ×5H×11W0.43060.14 10L ×6H×13W0.4312- Table 1:Domain IndependenceNumber of cellsC D∆ C D[%]1 .5×1060.45292.77 2 .1×1060.44072.35 3 .4×1060.43060.89 5 .2×1060.4268- Table 2:Grid Independence 4.Numerical Simulations After a proper validation of the chosen computational domain and the mesh quality, the central part of the study has been carried out, consisting of a parametric analysis aimed at quantifying the effect of tail plates length and slant angle on the vehicle’s drag coefficient reduction with respect to the baseline model. A visual description of the aforementioned tail plates geometry and varying parameters is reported in Fig. [6]. It is important to notice that these drag-reducing devices are meant to be as non-intrusive as possible, and consist of simple straight thin plates attached to the back of the trailer. They are meant to be attached to existing vehicles without any significant modification, and have to be able to fold on the back of the trailer to avoid occupying additional space in parking or maneuvering situations.Figure 6:Tail plates parameters Slant angles of 0°, 12°, 24°and 36°were considered and tested for each of the two lengths chosen, equal to0.4mand 1.2m. The first length was selected in compliance to EU regulations that limit the maximum size of similar devices to half a meter [4], while the second one is meant to allow the lateral plates to fold on the back of the trailer without overlapping, considering a trailer width of around2.5m. The 12°angle was chosen basing on existing devices [6], while larger angles are meant to help understanding whether an increase of the slant is beneficial or detrimental in terms of 3 wake reduction. Tail plates were tested in the 0 °configuration too, acting as simple frame extensions. Inflatable tails, consisting of more intrusive devices characterized by a more rounded, solid shape and reaching the length of around2m, were also analyzed and compared to the planar plates. They are still meant to act as deployable devices, diminishing their size when not required. 4.1.Baseline Results The baseline geometry was simulated first, in order to both perform the grid and domain independence analyses proposed in Sec. [3], and start capturing the main traits of the flow field around the truck in its basic configuration. Velocity magnitude contours are reported in Fig. [7] and Fig. [8], while a pressure contour is reported in Fig. [9]. A wide stagnation region is present ahead of the tractor, as highlighted from the low velocity and high pressure values within the area.Figure 7:Baseline velocity field lateral view - symmetry plane Figure 8:Baseline velocity field top view - trailer mid height plane The rounded lateral edges of the tractor, and the slightly tilted A-pillars, only marginally affect the extent of this high pressure region, whose presence is almost inevitable when considering conventional European tractor design traits, characterized by an almost flat, vertical front face. A significant increase in flow velocity is visible in correspondence of the smoothly curved edges surrounding the frontal area. As briefly described in Sec. [2.1], the baseline geometry considered is not completely bereft of aerodynamic devices: in particular, a fairing was used to connect the tractor to the trailer, leading to a smooth flow deflection towards the top of the trailer, characterized by gradual velocity variations. The air deflector allowed to avoid a secondary stagnation region in the upper part of the trailer which would be otherwise impacted (almost) orthogonally by the incoming airflow resulting in an additional source of drag.Figure 9:Baseline pressure field lateral view - symmetry plane A small recirculation region can be seen, in Fig. [7], in the lower gap between the tractor and the trailer’s underbody. The main recirculation bubble is, yet, the one on the back of the trailer, as mainly highlighted by the velocity contour. The wake region appears to be characterized by a low pressure which is far from recovered with respect to the stagnation 4 value and even the freestream value. This is due to the sudden flow separation occurring at the trailer’s sharp trailing edges. This region will be the main target of the parametric analysis which follows. 4.2.Tail Plates Parametric Analysis The purpose of the tail plates (also referred to asframe extensionsorboat tails) is that of reducing the large separated wake generated downstream of the trailer, aiming to provide a partial pressure recovery and, hence, reduce the aerodynamic drag of the vehicle. The geometry of the plates used is visible in Fig. [6]. The parameters chosen for this analysis were described and motivated above, and led to a total number of 8 combinations to be tested, in addition to the baseline, for which as many simulations were needed. Results were compared to the values obtained for the baseline model in terms of percentage reduction ofC D, and gave rise to the surface plot reported in Fig. [10]. The red dots appearing on the surface plot represent the results of the eight simulations actually performed, in addition to the data coming from the baseline configuration (for which the slant angle loses any practical meaning due to the zero length of the frame extension). For ease of capturing the significant trends, more deeply analyzed in the following sections, the same set of data was unwrapped and plotted in separate parametric curves in Fig. [17].Figure 10:Drag reduction as a function of tail length and angle Figure 11: C Dplotted against angle (top) and length (bottom) 4.2.1.Frame extensions with no slant angle The most basic design of a rear tail is a simple frame extension, with no slant angle with respect to the trailer. Since the tails are flat and thin their direct contribution to the viscous and pressure drag is negligible, but they are in fact able to create a more uniform pressure field on the rear, which is on average higher with respect to the baseline case as visible in Fig. [12], providing an appreciable reduction of the suction effect and, hence, of theC D.Figure 12:Frame extensions at β= 0°withl= 0m,l= 0.4m,l= 1.2m The reason for this pressure increase may be sought in the capability of the flat frame extensions to retard the upper and 5 lateral separation point, hence moving away the upper recirculation vortices as shown by the flow streamlines at the back of the trailer in the comparison between the baseline and the1.2mtail, reported in Fig. [13]. This reduces the velocity induced on the higher part of the rear of the trailer, leading to the greater pressure distributions shown previously.Figure 13:Streamlines at the back of the trailer for l= 0mandl= 1.2m 4.2.2.Effect of the slant angle A clear trend can be seen in the tail plates performance, as their slant angle is progressively increased. The overall behaviour appears to be independent from the tail length, therefore the following results were consistently compared considering a fixed length of1.2m. The role of the tail length variation for each of the angles considered is more deeply examined in the subsequent section. Fig. [15] shows a visual representation of the shape and dimension of the recirculation regions arising at the back of the trailer, obtained with the use of null total pressure iso-surfaces. While Ptotis not inherently linked to flow separation, in practice it nevertheless gives a reasonably good idea of the shape of the wake, as can be seen in Fig. [14].Figure 14: P tot= 0 iso-surface (in red) aroundβ=0°configuration draped by flow streamlines coloured with velocity magnitudeFigure 15: P tot= 0 iso-surfaces - Wake comparison between baseline,β= 12°,β= 24°,β= 36° The second image of Fig. [15], which corresponds to 12°slant, shows how the flow is able to remain attached to the tail and how this deflection creates a slimmer wake, which normally corresponds to a lower drag with respect to the baseline configuration in the first image. The third image shows a similar effect, just more pronounced, while the last one highlights a clear separated flow which completely envelope the tail. From this alone the 24°tail seems to be the best configuration, but theC Dresults discredit this observation. An explanation was provided through the surface pressure in the three cases, as shown in Fig. [16]. 6 Figure 16:Frame extensions at β= 12°,β= 24°,β= 36°withl= 1.2m Both the 12°and 24°tails guarantee a mainly attached flow, as seen in Fig. [15]. This turns into a smaller wake and a consequent benefit in terms of drag reduction. Fig. [16] clearly highlights that the 24°tail creates a higher average pressure on the back of the trailer, but, at the same time, is also responsible for a low pressure zone on the outer surface of itself. This effect, paired with the higher pressure on the inside surface shown in Fig. [17], results in a pressure drag component on the tail itself, which more than offsets the granted drag reduction with respect to the 12°case. In the last case (β=36°), instead, the flow appears to be completely separated (as shown in Fig. [15], where the tail is completely immersed in theP tot= 0 iso-surface, and in Fig. [18]) therefore the effect of the tail is even weaker. Nevertheless, similarly to the previous configuration, tails still "shield" the truck’s rear from the induced velocity of the shed vortices guaranteeing a small benefit with respect to the baseline geometry.Figure 17:Pressure distribution on both sides of the tail (β= 24°,l= 1.2m)Figure 18:Flow separation over tail plates at β= 36°, detected on symmetry plane 4.2.3.Effect of the length While the slant angle seems to show an ever equal trend in terms of drag reduction, tail length appears to behave differently according to the angle used. For the lower angles (0°and 12°), for which separation doesn’t occur, a longer tail simply improves the performance in terms of drag reduction. In the 24°case, as stated before, the tail itself starts generating a pressure drag which becomes stronger the longer is the tail; this means that an optimal length could be found, after which an increase turns into a loss of performance. Between the two configurations tested, the shorter tail resulted in a higher effectiveness. The same behaviour was found in the minor drag reduction provided by the frame extension atβ=36°. Due to the tail working in a sub-optimal manner, the larger plates tested performed in a worse way with respect to the shorter ones. 7 4.3.Inflatable Tail Analysis One of the limits of the planar tail plates appeared to be the sharp angle between the trailer and the slanted surfaces, leading to early flow separation when the slant angle exceeds a certain value, strongly decreasing the tail effec- tiveness. From a practical point of view, inflatable devices allow for smoother, rounded shapes while keeping their capability to quickly deflate to reduce the occupied space when not in cruising conditions, similarly to what can be done with the foldable planar tail plates. Since an effective parameterization of a curved surface would require an excessive number of simulations, a different approach was chosen to evaluate its performance. Initially, an inflatable tail was modeled and tested, with a total length (of 2m) and geometry inspired by the device presented in [2]. Subsequently a heuristic improvement was de- signed basing on the main flaw detected on the first solution, consisting of the early separation identified in both Fig. [20] and Fig. [21]. While the results of this first geometry were very respectable, it still under-performed compared to the best tail design as seen in Tab. [3], due to the separation preventing a full recovery of the pressure. To avoid this early, undesired separation, a much lower curvature would be required, leading to an impractically long tail. In the second inflatable geometry proposed, the tail curvature was strongly diminished, but the device was still cut off to a length of2min order to con- sistently compare the results with the previous solution; this led to a solid, squared rear surface visible in Fig. [19].Figure 19:Inflatable solutions: baseline (top) and optimized (bottom)Figure 20:Flow separation comparison between the two inflatable solutions The improvement brought by the second geometry was successful, as shown in Tab. [3]. Even this design, though, was unable to outperform the best tail design despite reaching a pretty similar drag reduction percentage; this is most likely due to the inflatable device losing the "cavity effect" provided by the frame extensions, which by itself brings a significant CDreduction, as shown by the flat plates case. Further studies may be performed on hybrid solutions, trying to combine the perks provided by each of the two working principles.Figure 21:Limit streamlines on baseline designResultsC DPercentage saving Baseline0.4306- Best tail plates design0.36714.77% Base inflatable0.375912.70% Improved inflatable0.370114.05% Table 3:Inflatable tail comparison 8 5.Wind Tunnel Test 5.1.Model Geometry and Setup In order to validate the results of the numerical simulations, an experimental wind tunnel test was performed on a scaled model of the semi-truck. The model was an exact replica of the CAD geometry presented in Sec. [2.1], created entirely through 3D printing, along with a detachable tail in the configuration which provided the best results, in terms of drag reduction, in the previous parametric CFD analysis (l=1.2m,β= 12°). The scale of the 3D printed model was chosen as a trade off between a high Reynolds number of the experiment and an acceptable blockage ratio (e.g.