Impact assessment of new generation high-speed agricultural tractor aerodynamics on transportation fuel consumption and related phenomena

New generation agricultural tractors contribute to transportation by increased travel speeds. There is not any available aerodynamic data on the authentic agricultural tractor form. On-road transportation by tractors is between 8 and 30% of their operational time. In this work, two agricultural tractors are modelled via computational fluid dynamics for nine different speeds to determine aerodynamic resistances. Constant speed travel scenarios are analyzed. Corresponding speeds are 5 and 10 to 80 km/h with 10 km/h increments. Reynolds number changes between 1.6 × 105 and 2.98 × 106. The characteristic lengths are taken as the square root of the streamwise projected area of the tractor geometries. Aerodynamic forces exerted on the tractors change between 3 and 746 N. The calculated drag coefficients are found as independent from Reynolds number and are 0.6 and 0.78 for the two different types of driver compartments. The approximated aerodynamic related fuel consumptions for 1-h changes between 0.002 and 8.28 lt/s which correspond to 0.001 to 5.76 kg/s carbon emission. A potential improvement in decreasing aerodynamic resistance about 20% is discussed by spatial data. Since the conducted work is being regarded as the first instance in the literature, it is estimated that several consecutive reports will be triggered.


Introduction
Agricultural tractors are gaining new abilities. Their power, torque, speed, operator comfort offerings, utilities, and features are ever increasing. Today, transportation by agricultural tractors with relatively high speeds is common. In a very recent work of Mattetti et al. (2021), 579 h of realworld operation from an agricultural tractor is processed and 8% of the operation is found to be on-road with travel speeds between 40 and 50 km/h. In addition, the engine operational power values approach to field operation values during onroad transportation. Harland et al. (2014) suggest that 30% of agricultural tractor included traffic incidents occur in urban areas. Authors also indicate that most frequent incidents among mentioned percentage are in 55 mph speed limit roads. They add that seasonal activities contribute to the intensity of incidents. Mederle et al. (2015) justify the reason of agricultural tractor usage in road transportation and also state that agricultural tractor tires, which are essentially for farm operations, are used in on-road transportation at a rate of 30%. Therefore, they conducted an experimental tire examination for transportation speeds of 50 and 60 km/h. Cavallo et al. (2015) focused on expectations of farmers regarding technological developments in agricultural tractors. One of the three major expectations is on increasing usage of agricultural tractors and their travel speeds for on-road transportation. The authors emphasize motivating parameters for high-speed agricultural tractors and also state that agricultural tractors are being used in on-road transportation. Finally, a domestic study determined that 50% of operational time of agricultural tractors are dedicated to on-road transportation (Altintas 2015). Although above references stress the usage of agricultural tractors in on-road transportation with relatively high speeds, their impact on energy consumption and environment has not been addressed. Aerodynamic component of such transportation seems to be missing totally. Even if the necessary statistics of agricultural tractor usage in on-road transportation are obtained to evaluate its impacts, aerodynamic behavior is still absent, and it is needed to be addressed for decomposing total impact into source components. As a general rule of thumb, 40 km/h is regarded as a threshold for aerodynamic resistance. However, this threshold is generated for vehicles having drag coefficient (C D ) about 0.2 or 0.3. If C D about 0.5 is considered, the limit reduces to 30 km/h. Adding increased surface areas of high C D vehicles, aerodynamic investigation of agricultural tractors deserves attention as commercial tractors do.
The literature survey that is conducted for the present paper did not reveal necessary amount of works in the literature. Only two papers, dealing with a type of meshing technique in computational fluid dynamics (CFD), use tractor geometry since this geometry has the needed complexity by the authors to test their approach Xu et al. 2016). However, the two papers do not deal with physics of the agricultural tractor aerodynamics for on-road transportation. Instead, they focus on the effect of the mesh types. Only provided information related to the present work is the C D of the tractor model, i.e., about 0.8 for 11 m/s inlet velocity and Re about 4.5 × 10 6 . As it will be shown in the present work, drag coefficients are not consistent with the abovementioned two papers. However, drag coefficient definition based on characteristic length and drag components can cause differences between results. Unfortunately, authors did not report the force values. On the other hand, wake structures downstream of the tractors have qualitative resemblances with the present paper.
Aerodynamic investigations on commercial vehicles can give an idea for a framework of aerodynamic investigations on agricultural tractors. Commercial vehicles and commercial tractors, trucks in other words, have high drag coefficients and high form and friction drag surfaces as agricultural tractors do. Therefore, not only their CFD framework is useful for the current case but also some aerodynamic aspects and results can be used for the current case. It is clear that numerous passive flow control devices are adapted by commercial tractors for improving fuel economy and reducing negative impacts. Khosravi et al. (2015) used CFD in order to investigate effects of several append devices as passive flow controllers on drag coefficients of commercial tractor trailer couples. They managed to obtain 41% reduction in drag coefficient. Kim et al. (2017) used wind tunnel and particle image velocimetry (PIV) in order to investigate cab-roof fairings as an aerodynamic modification for commercial tractor trailer combinations. Authors also used CFD in their investigation. Several geometric modifications on the cab-roof fairings are taken as geometrical parameters by the authors. Fore body is emphasized as being a major contributor to aerodynamic drag. Drag reduction of 19% is obtained by the fairings. A similar work is reported by Malviya et al. (2009). Authors used experimentation and CFD together. Drag contributions of several parts of commercial tractor trailer combination are given. Passive flow control geometries alongside an active flow control device named as moving surface boundary layer control cylinder are examined. The authors claim that the active flow control device can decrease fuel consumption up to 13%. Hariram et al. (2019) report an extensive review on aerodynamics of commercial truck and trailer combinations. Three main conclusions of the authors are the importance of fore body geometry that discriminates USA trucks from Europe trucks; different contributions of CFD, wind tunnel testing, and truck testing; and drag reduction potential up to 10%. Kim et al. (2019) used wind tunnel and PIV in order to provide quantitative spatial resolution of flow around commercial tractor-trailer combination with several passive flow control geometries. Drag reduction of 26% is determined by the authors. Mosiężny et al. (2020) use Reynolds Averaged Navier Stokes (RANS) turbulence modelling for steady solution and delayed detached eddy simulation (DDES) for transient CFD solution of commercial tractortrailer combination. Authors tried an active flow control arrangement at the rear part of the investigated vehicle and obtained 11% drag reduction. Aerodynamics of a vehicle or truck type is also important for platooning, which is highly studied in transportation. The work of Dávila and Nombela (2011) can be given as an example. Authors report 20% drag reduction for some of the vehicles in the platoon, mostly due to the aerodynamic features of the leading vehicle. Xie et al. consider aerodynamic drag for arranging distances between vehicles in platoons, especially commercial tractors or trucks (Xie et al. 2020). It is seen that drag coefficient changes about 50% according to the distance between platooning vehicles. Authors emphasized its energy implications. Drag is very related for energy consumption and environmental impacts and hence, researchers are trying to find innovative geometries to reduce drag by having inspirations from nature (Huang et al. 2021). Obtained experiences are adapted in realworld applications. The necessity of drag data for a vehicle form can be assessed by life cycle assessments (Schäfer et al. 2006). Authors consider drag coefficients originating from vehicle bodies and tires during constructing their models. Similar vehicle modelling works considering aerodynamic drag for a type of vehicle in respect of energy issues exist in the literature (Tian et al. 2019). Almost all reviewed literature works emphasize aerodynamic resistance impact on fuel consumption and related negative impacts on environment due to exhaust emissions. Additionally, it is understood that passive flow controllers are widely studied and also commercially available. The current experience on aerodynamics of commercial tractors justifies a work on aerodynamics for transportation by agricultural tractors. Literature data is also an advantage that can be adapted rapidly and hence give positive results in relatively short times.
Drag force acting on ground vehicles has several components emerging from different geometrical sources. Breaking down drag force into components creating a chance to reduce them by focusing on a certain source. Real vehicle models can be used for this task as well as simplified models that isolate most of other sources. There are instances for both approaches. Wiedemann (1996) stresses importance of wheel turning and ground moving in wind tunnel tests. Author shows that three important factors are not available in conventional wind tunnel tests, i.e., ground relative motion, rotation of the wheels, and air flow through engine cooling system. It is also mentioned that 1% drag reduction corresponds to about 10 to 15 kg weight reduction. Sivaraj et al. (2018) study a base-bleed solution for drag reduction on a simplified model geometry. The simplified car model is modified with a base-bleed geometry in wind tunnel tests and 6% drag reduction is attained. Vignesh et al. (2019) focus on front and rear windscreens and hood angles by using CFD. Authors determine the best arrangement of angles for the lowest drag value. Their car model is also a simplified one and imposes almost only the three geometrical parameters. This simplified geometry approach actually based on pioneering works in 1970s and 1980s (Ahmed et al. 1984;Morel 1978;Templin and Raimondo 1986). Whether it is due to petrol crisis or not, starting from 1970s, drag reduction is a main concern and dealing with it has been done mostly by simplified models that isolate most of the factors and focus on one or two main parameters. Le Good and Garry (2004) lists about 25 reference simplified models. However, it is again evident that there is not any reference geometry that features characteristics of agricultural tractor form. Therefore, it is suggested by the authors of the present work that breakdown of drag relating to the agricultural tractor geometry should be done in the future and reference geometries can be developed in order to investigate drag components. It should be noted that there are works on tractor transportation (Jokiniemi et al. 2016) but they are related to mechanical resistances. Gao et al. (2019) view eco-driving potentials for diesel vehicles and drag reduction significantly takes place in the list. However, the test cycle that is cited by the authors is for light-duty diesel vehicles and it is not clear that transportation by agricultural tractors is subjected to any regulation at all. The aerodynamics of agricultural tractor will also provide for additional aspects such as air conditioning in the tractor cabin since previous works are done for stationary or quasi-stationary conditions (Oh et al. 2020).
Transportation by agricultural tractors with high travel speeds is expected to have higher impact on energy utilization and environmental effects. However, it is not possible to make an assessment without aerodynamic performance perspective. In the current state of the literature, no relevant aerodynamic assessment has been encountered in the literature. Previous studies on agricultural tractor energy consumption focus on tire inflations and other issues considering off-road operations related with traction (Janulevičius and Damanauskas 2015). However, it is apparent that present day capabilities of agricultural tractors, including travel speeds, enable them for high-speed transportation. There are also literature papers that have already been mentioned, dealing with the on-road transportation with agricultural tractors (Altintas 2015;Cavallo et al. 2015;Harland et al. 2014;Mattetti et al. 2021;Mederle et al. 2015).
In this work, two new commercial agricultural tractor models, originally emerging from a single model, are modified for and adapted to CFD investigation. On-road highspeed scenarios that may be due to transportation are used for investigating flow around the tractor. Aerodynamic drag forces acting on the geometry are determined. Fuel consumption and related carbon emission are approximated. Experience related to the CFD modelling is shared since this work is one of the few examples in the literature. A framework for future studies is presented for reducing drag force during transportation operations by agricultural tractors. It is projected that scientific community will respond rapidly and diversify findings to give a complete resolution on the topic.

Method
The 3D solid models for agricultural tractors that can do the transportation tasks on road with relatively high speeds were obtained from Erkunt Traktör Sanayii A.Ş, an agricultural tractor manufacturer in Turkey. The model number is "Nimet 75." The main difference between the two models is based on driver compartment. One model is isolated from ambient by a "cabin" and the other one has direct contact with the ambient. However, commercial models are too complex for CFD modelling. Therefore, surfaces were simplified prior to the CFD analyses. The final models are shown in Fig. 1. Some dimensions are given in Fig. 2.
Before meshing of the calculation domain, flow field is constructed with two main zones. The first zone is the zone that surrounds the tractor at proximity. This zone is called internal enclosure. The second zone surrounds the internal enclosure and called external enclosure. Together with the tractor geometry, three solid volumes exist. However, for CFD purposes, only fluid volumes around the solid surfaces are enough since the surfaces can act as fluid-solid interfaces by manual selection. Therefore, meshing cost of the solid parts is avoided. Tractor models are only for creating enclosures. The CFD investigation in this work is a steady analysis. Therefore, flow around the tractor is symmetrical according to middle plane of the tractor in travel direction. Accordingly, only half of the enclosure volumes are transferred into meshing software. Figure 3 shows the 3D volume that contains one side of the computational domain based on the symmetry plane. This figure also marks dimensions and dimensional ratios of the enclosures. As a general practice, preliminary CFD runs were performed and accordingly, boundaries are placed distant enough in order to reduce their effect on the solution. Actually, this is done since known values are used in boundary conditions and those boundaries sufficiently distant from the tractor ensure that the boundary conditions are valid. Especially, outlet boundary has the most distant location due to the spatial length necessity of the wake flow of the tractor model to be diminished at the outlet. All domain sides except the road and the tractor surfaces (no slip walls) are set to symmetry boundary condition. Inlet is arranged as having one directional velocity inlet with 5% turbulence intensity. The intensity value is about 1% in controlled wind tunnel tests (Ahmed et al. 1984). In reality, it can be higher due to atmospheric conditions. Turbulence intensity of 5% is regarded as a steady real-world air condition. Also, CFD method approaches to necessary level right after the inlet during iterations (Canli et al. 2021). Turbulence intensity at the inlet is just for estimating values necessary for conducting numerical iterations. Tractor model surfaces and road are modelled as no slip walls. One may question boundary conditions on tractor surfaces and the road since some of the tractor surfaces are porous, and some are rotating in reality. Also, relative motion between the tractor and the road is generally simulated by means of moving wall in CFD. Nevertheless, this work targets comparative results and approximations to evaluate the topic whether further details are necessary to be investigated. The main trend and approximate results reported in this work will further approach to reality and accuracy by means of further studies that will focus different parameters. The assumptions and simplifications in this work isolate parameters that are examined in the present work and enable an easier comparison.
Meshing fluid volumes for a successful CFD analysis is essential. However, tractor geometry, even with its simplified form, is very complex for a structured mesh. Inflation layers cannot be generated by the software. Therefore, internal enclosure is meshed with finer element sizes and external enclosure is meshed with coarser elements. Element types are unstructured tetrahedral. By doing so, 2 layers of mesh elements sizes are obtained. Internal enclosure has constant element size. External enclosure grows to a maximum elements size value starting from the internal enclosure element sizes. Three meshes were constructed for mesh independency. They are labelled as coarse, medium, and fine. Internal mesh element size doubled for each mesh transitions, i.e., coarse to medium and medium to fine. For nodes, linear distribution is dropped, and program-controlled quadratic interpolation is used for increasing number of nodes. Therefore, all node numbers are higher than the element numbers.
Turbulence model benchmarking was done with two different Reynolds Averaged Navier Stokes (RANS) turbulence models. The selected turbulence models are k-ω SST and Realizable k-ε turbulence model with standard wall functions. k-ω SST turbulence model is the default turbulence model that Ansys Fluent version 20 proposes. This turbulence model uses k-ω in the wall proximity and transits to k-ε far from wall. Therefore, mesh resolution near tractor walls requires dimensionless wall distance y + to be below 1. However, due to the complex geometry of the solid model, it is not possible to meet this criterion with the best level of confidence. Therefore, Realizable k-ε turbulence model with standard wall functions was also tried since standard wall functions work between y + ≈11 and 500. Tractor geometry is too complex to have a structured mesh having inflation layers on solid boundaries. The scale of edge sizes on the geometry varies greatly. Accordingly, an unstructured mesh having smallest possible elements was aimed. On the other hand, standard wall functions tolerating the high y + values enable the simulation. The default turbulence model k-ω SST then provides a comparison tool where no wall functions were utilized with relatively high y + values. The difference between the two models solely arises from the proximity of the wall and boundary layer dealing method. The comparison, which is also used for mesh structure evaluation, gives the effect of wall functions. Also, using the two approaches together and obtaining a band of results may compensate for some of the undetermined uncertainty arising from using one of the two turbulence modelling and wall treatment approaches solely.
In order to examine convergence or mesh independency, three different unstructured meshes were tried, namely coarse, medium, and fine. Table 1 presents element numbers and node numbers of the three meshes. Mesh quality statistics are given in Table 2.
The difference between element numbers and node numbers for cabin and platform versions of the tractor models is due to the fact that fluid fills driver compartment while cabin leads to a void in that region. Also, a special measure was taken for the steering wheel during the meshing of platform version since steering wheel has a very small volume, resulting in mesh element merging. A sphere of influence function surrounded the steering wheel and kept the element size at 40 mm at that small volume. Considering only Tables 1 and 2, based on self-experiences, medium and fine meshes seem adequate. Mesh instances are given in Fig. 4 for cabin version. The internal enclosure mesh elements cover a big portion of the wake flow of the tractor models.
In order to evaluate results depending on mesh structure, C D and aerodynamic drag force (F D ) acting on the tractor models were calculated. Drag force acting on the tractor surface is calculated by the Fluent software considering streamwise projected areas of mesh elements and then using integration. Normal pressure acting on a mesh surface coinciding tractor surface is multiplied with surface area, yielding the force, and its streamwise component is calculated for summing them covering all tractor surfaces. After summing all forces on a specific direction, which is the streamwise direction in the present case, the total force is calculated. In real-world conditions, the two tractor models distinguish from each other by their projection area in travel direction. At the beginning of the investigation, an aerodynamical difference between them due to the projection area differences were anticipated. On the other hand, CFD work uses finite volume grid elements and the software uses reference projected area that is calculated or drawn from grid elements by the software. Obtaining projected areas from the software were preferred. However, it is realized that the grid resolution and projection calculation parameters effect the projected area results. After having finer grids and trying some parameter values in Ansys Fluent, we managed to have the accurately projected area results. Providing those results in Table 3 enables one to calculate drag force by using projected area and the drag coefficients. Also, it enables one to make a qualitative comparison between tractor models. The projected area values were calculated using Fluent by decreasing the minimum feature size value till the change in the  projected area is below 0.1%. Actually, projected areas in Fluent are half of the values in Table 3 due to the symmetry boundary condition. However, mesh structure changes streamwise projected area of the tractors slightly. Nevertheless, this phenomenon is compensated by the calculation definition of. C D is calculated by below Eq. (1).  In Eq. (1), F D is the drag force, A is the streamwise projected area on the tractor, V is the inlet velocity, and ρ is the air density at room temperature. Although viscous and pressure components of the drag force exist, their total is used in this study. Figure 5 presents changes in C D and F D that are acting on the tractor models according to the mesh types for 40 km/h forward travel speed. Increasing element number by decreasing internal enclosure element size changes C D and F D acting on tractor models. C D and F D decrease as mesh gets finer for cabin version. Nevertheless, k-ω SST turbulence model shows an asymptotic behavior for cabin version. The change rate gets smaller as element number increases. On the other hand, making mesh finer for platform version marks a minimum and then increases the results for C D and F D . Nevertheless, medium and fine meshes for both models give close results. Table 4 is given for percentage changes of C D and F D according to mesh transitions. Both Fig. 5 and Table 4 show that medium meshes are sufficient for mesh independent results. In the present work, to obtain more certain results, fine meshes are used instead. The four graphics in Fig. 5 actually indicate same changes. Decreasing element size means increasing element number. Therefore, trend of lines change direction according to vertical axis between element size and element number graphics. The magnitude of changes in x axis of the graphics is not at same rate for element size and element number graphics. Therefore, lines are stretched for the element number graphics. On the other hand, y axis interval of C D graphics is wide while they are small for F D , making F D changes seem more drastic. However, quantitative results in Table 4 avoid intuitive errors. One last thing about converging F D and diverging C D results is that the reason is the different projected areas of the tractor models.
Since F D results converges about 185 N, difference between projected areas leads to about 30% difference between the C D results of the two models.
It should be noted that one should be careful while calculating C D from the F D value that is taken from Fluent code since the code gives a F D for half tractor. F D value of the whole tractor can be obtained by doubling the software value. Another way is to use software F D and use half of the Table 3 value.  Boundary layer evaluation in respect of mesh structure is also done according to y + values. Figures 6 and 7 show the comparison of y + values by means of surface contours, x-y plots and histogram plots, respectively, for cabin and platform versions. Two dimensional "x-y" plot named visuals contain all y + values from all wall meshes. Contour and histogram visuals were plotted for the y + interval between 0 and 1000. Figures are for 40 km/h travel speed.
After evaluating all mesh dependent results together, fine mesh setup for the two-tractor model was found adequate in order to conduct further analyses. Standard wall functions can handle y + values below 500 by means of the fine mesh setups. Also, as shown by the histograms, most of the y + values are below 300. The change between medium and fine meshes is about 2% for C D and F D . The qualitative extrapolation of the trends according to mesh comparison results implies that the rate of change also decreases. As a result, fine mesh setup was used in the analyses. Also, mesh metrics such as skewness, aspect ratio, and orthogonal quality of the mesh elements would degrade greatly with smaller internal enclosure mesh element sizes. This would create undesired numerical error. On the other hand, using a personal computer, CFD calculation with coarse mesh results approximately in 20 min. Medium mesh and fine mesh CFD calculation times are 1 h and 7 h respectively. Total computation time is then approximately 70 h for the used fine mesh setup. Postprocessing and evaluation extend the work period to 100 h. Increasing mesh element number further would lead to unfeasible calculation times. One major problem with the complex tractor geometry, as stated by the literature Xu et al. 2016), is that meshing is a very hard task. Therefore, a dedicated future work on new meshing strategies would be recommended. Numerical approach is explained in this part. Steady solution was used while gravity was ignored. Constant property air was selected as fluid, evaluated at room temperature. Inlet velocities were selected as 1. 388, 2.77, 5.55, 8.33, 11.11, 13.88, 16.66, 19.44, and 22.22 m/s in normal direction to the inlet boundary. These velocities correspond to 5, 10, 20, 30, 40, 50, 60, 70, and 80 km/h traveling speeds. Reynolds number (Re) in this work was calculated by Eq. (2).
In Eq. (2), L is the characteristic length and was calculated by square root of the projected area. Dimensions of tractor geometry necessitate a hydraulic diameter like characteristic length. Therefore, square root of the projected area was used as the characteristic length value in Re calculation. In Eq. (2), μ is the dynamic viscosity of air. Air (2) Re = V L thermophysical properties were read from thermodynamic tables for 1 atm pressure and 20 °C temperature. The Re interval of the investigation is 186,291-2,982,270 for cabin version and 162,897-2,607,772 for platform version. Long commercial trucks/tractors and aerofoils use length as the characteristic length while short obstacles use height, width, or hydraulic diameter for the characteristic length value. Agricultural tractor geometry resembles to a low aspect ratio rectangular prism and therefore, characteristic length was calculated by taking square root of the streamwise projected area. Characteristic length L was calculated as 2.02 m for the cabin version and 1.77 m for the platform version. Outlet boundary condition was set to pressure outlet at atmospheric gauge pressure. Spatial discretization scheme was selected as second-order upwind scheme for all governing equations. Pressure velocity coupling was done according to coupled scheme. High-order term Fig. 9 Comparing turbulence models considering − 10 Pa iso surfaces relaxation and pseudo transient options were enabled for a stable solution. This stabling effect slightly increases convergence time. General iteration cycle was limited with 500 iterations while C D , mass flow balance, and scaled residuals were monitored. It was observed that 300 iterations would be sufficient for C D and mass flow rate balance converge to a fixed value. The fluctuation of C D is between 0.60045 and 0.6005 for 40 km/h cabin version simulation while the standard deviation is 0.00002 at 300th iteration. Scaled residuals of velocities and turbulence indicators were also below 10 −8 value while scaled residuals of continuity approaches to 2 × 10 −4 .
One may wonder the effect of y + on the results. This can be answered considering utilization of two different strategies in CFD solution, i.e., using and not using standard wall functions. Since k-ω SST is the default turbulence model in the software and not using wall functions, a benchmarking was done between k-ω SST and Realizable k-ε with standard wall functions. Turbulence model benchmarking between k-ω SST and k-ε is done by tractor model wakes and pressure iso surfaces at − 10 Pa gauge vacuum values in Figs. 8 and 9 respectively. When the two figures are evaluated together, significant differences are hard to detect qualitatively. Viscous to total drag force percentages are 2.2 and 3% for k-ω SST and k-ε respectively, in case of cabin version. For platform version, the percentages are 2.6 and 3.6% for k-ω SST and k-ε, respectively. These results indicate that form drag is the main reason of the total drag. Also, the difference between the two approaches is not significant. By the detailed mesh analyses and the benchmarking work, it was concluded that Realizable k-ε with standard wall functions is sufficient and accurate for the investigated case, also considering the y + values. The reason of selecting Realizable k-ε is also due to the fact that the turbulence model does not use a constant in calculating turbulent viscosity and instead, it calculates a coefficient by an additional equation. Also, it includes molecular viscosity in k and ε equations. The ε equation is also different from the standard model. This type of modification is recommended for separated flows and recirculation regions where form drag is important. Nevertheless, 2D plots for derived results contain results from bot turbulence models in order to provide a band in which the results may have an uncertainty.
Power requirement due to drag force during transportation operation is approximately calculated by Eq. (3). In Eq. (3), P denotes power that is consumed to overcome aerodynamic drag force F D . In order to calculate fuel amount, 0.2 total energy conversion efficiency is assumed in Eq. (4). In Eq. (4), Ė is the energy source per unit time and η is the total energy conversion efficiency. By using lower heat value of the diesel fuel in Eq. (5), necessary fuel mass per second can be calculated.
Emitted amount of carbon is calculated by multiplying fuel mass with 0.849, which is the approximate ratio of carbon mass in diesel fuel.

Results and discussion
Before presenting obtained results, one may wonder about the y + values for the minimum and maximum travel speeds. Figure 10 is given in order to show y + values for the minimum and maximum travel speeds similar to Figs. 6 and 7. Figure 10 shows that approximately 10% of the y + values are below 15 for the minimum travel speeds and 30% of the y + values are above 500 for the maximum travel speeds. In order to evaluate the effect of this distribution, the viscous drag that is very related to the wall functions is evaluated by its percentage in total drag using C D viscous part percentage in total C D , as given in Fig. 11. Also, the Re independency is shown in Fig. 12.
Viscous drag accounts about 4% of the total drag. In addition, Re independency is achieved according to the Fig. 12. The distribution of y + values shows that only about 20% of the y + values are not in the desired interval. Therefore, the effect of improper y + values should have a very insignificant effect on total drag results. On the other hand, the difference between the two tried turbulence models is about 1% and the obtained force and coefficient results almost coincide by graphical line plotting. Since the comparison for the two tractor models depends on same numerical setup, relative evaluations are regarded accurate. Absolute results may have some uncertainty; however, they are sufficient for triggering future works. The Re interval of the investigation is 186,291-2,982,270 for cabin version and 162,897-2,607,772 for platform version. As shown in Fig. 12, C D is independent from Re in the investigation interval. Cabin version has about 0.6 C D and the platform version has about 0.78 C D . This phenomenon, the Re independency, can be explained by similar wake and flow structures around the tractor models in a qualitative manner for changing travel speeds in the present work. Boundary layers are very thin, and therefore, viscous forces are too weak to change the flow structures for strong and high magnitude flow inertia at the investigated interval. Since all other parameters in Eq. (1) are constant except the velocity and F D , and the proportionality between the two changing variables as C D is almost constant, travel speed increases F D exponentially as shown in Fig. 13. Figure 13 shows that drag force acting on the tractor models is almost same. The main reason of difference between C D values is then due to the projected area values. Travel speeds and accordingly Re increased 16 times from minimum speed to maximum speed and the proportionality between F D and velocity is not changing. This suggests that cabin utilization has no major drawback in terms of aerodynamic resistance. Considering other comfort advantages of the cabin utilization, it is thought that cabin version is more logical in terms of aerodynamics. Spatial data in three-dimensional space will be given in order to strengthen the physical comprehension.
The reason of aerodynamic resistance acting on the travelling tractor is mostly due to the form drag as previously indicated. Form drag is formed due to the pressure difference Fig. 11 Evaluation of y + according to meshes for cabin version between upstream and downstream surfaces of the tractor. The flow impacts the tractor surfaces and it attaches or detaches/separates to/from surfaces depending on the local boundary layer and pressure gradient. Dynamic pressure is converted to static pressure on the surfaces at different rates. Low-pressure wake occurs downstream of the tractor rear surfaces. Aerodynamic drag force acting on the tractor is a sum of static pressures on the tractor surface. Therefore, static pressure distributions are illustrated in Fig. 14. The static pressure distribution in Fig. 14 is given in such a way that the contours have the same topology but indicated values are changing. This can be also done thanks to the same distribution trends between cases. It is seen that upstream front cover of the engine compartment is exposed to much of the high static pressure values for both tractor models. Therefore, it is evaluated that a porous structure in future works, considering the cooling systems and air suction of the engine, will make some changes.
The static pressure distributions in Fig. 14 for the cabin version also show that front shield of the driver compartment (cabin) also exposed to high static pressure field. The platform model tractor has this pressure distribution on driver seat and inner surfaces of the driver compartment. Since low pressure surfaces at the downstream sides are very similar in terms of static pressure distribution, it can be concluded that the aerodynamic force action on the cabin front shield is almost equal to the aerodynamic force acting on driver compartment in the platform version. The driver comfort is then seriously reduced for the platform version as expected. The higher static pressure values on the cabin suggest a pitching moment acting on the tractor. In transient conditions, this pitching would result in vibrations or oscillations that may degrade driver comfort and induce health issues on long runs. According to above evaluations, it is seen that wind tunnel test is necessary for validation of the CFD approach. Another interesting result in Fig. 14 is that front wheels and partially rear wheels are subjected to high static pressure fields. If wheel rotation will be included in future works, it is predicted that the resulting pressure fields and hence aerodynamic forces due to the wheels will be higher. Maximum travel speeds lead to highest pressure levels up to 300 Pa above the gauge pressure. Also, below gauge pressure values are seen in the rear faces of the tractor models down to − 200 Pa. The lowest pressure values are seen at side surfaces where flow accelerates due to fulfilment of mass conservation. Static pressure levels as low as − 400 Pa are visible at side surfaces.
From Fig. 14, one may expect that the flow jet through the driver compartment at the platform version is somehow has high velocity values. Also, the wake of the tractor models can be questioned. Accordingly, velocity contours at the symmetry planes are shown in Fig. 15. Velocity contour legends were prepared as between stagnant flow velocity, i.e., 0 m/s and two times the inlet velocity after rounding down the number. General wake topology for cabin version for different travel speeds is almost identical. This is also valid for the platform version. However, the wakes of the cabin version and platform version are not the same. Nevertheless, the major portion of the low velocity wake is mostly downstream of the tractor's lower half body for both cases. This implies that the driver compartment has minor effect on the low velocity wakes. Another thing to be stressed here is the possibility of different wakes for rotating wheels and moving road. These would affect the topology and size of the wake. Cabin version has a longer wake in downstream direction. In the absence of cabin wake, the wake of the platform version moves towards upwards. In the top of both tractor models, a slight velocity increase is apparent for lower travel speeds. The upstream of the front edge of the tractor models has a lower velocity region. Also, values of this region are seen in downstream of the tractors for a long narrow tape through the outlet at above the stagnant zone in the proximity of the road. One major interesting outcome is that the wakes are almost same sizes of the tractors in a mirrored manner. These big wake regions have very low air movement in them.  Also, the boundaries of the wakes create free shear layers with high momentum diffusivities. Wakes suggest a small clockwise vortex on the top region of the wake and a big counter-clockwise vortex on the bottom region of the wake for cabin version. In case of platform version, two similar magnitude vortexes at top and bottom regions of the wake are possible, resembling to Kelvin-Helmholtz instability. The investigated range of Re number is known with intrinsic turbulence. However, by the solid boundaries of the tractors and interacting flow, geometry-induced turbulence is also expected. Flow impact, separation, reattachment, recirculation, shear layers, and possible vortex formations create turbulence. One way of examining turbulence magnitudes is to look for turbulent kinetic energy values. This is done in Fig. 16 with again spatial data on the symmetry planes. Figure 16 marks two major differences between tractor types. The first one is the location of turbulence peaks. The cabin version has its turbulent kinetic energy peak at above the cabin roof upstream edge. On the other hand, the turbulent kinetic energy peak values occur where the driver head would exist in the platform version. One may expect this high turbulent flow would cause further noise and uncomforting conditions. The second major difference between tractor models is that the platform version creates more turbulence. This explains how a lower projected area value can give same amount of aerodynamic resistance force with the high projected area value. The surfaces in the driver compartment of the platform version poses an enhanced surface area for turbulence generation. This turbulence generation draws flow energy and converts it into turbulence. The cost is the momentum diffusion to the driver compartment surfaces of the platform version. The air flow with increased turbulence levels tends to move towards the road surface for the cabin version. On the other hand, the increased turbulence level flow at platform version prevails longer and at about the height of the platform cap. This may lead to disadvantages when a trailer exists. Turbulent diffusion may affect the material that is transported in the trailer. Turbulence magnitudes downstream of the tractor for transportation purposes are actually open for further investigations and studies since vortex shedding, development, dissipation, turbulence diffusion, and relevant phenomena strongly affect downstream of the tractor.
The downstream of the tractor models is also investigated in terms of pressure iso-surfaces, which mark vacuum regions according to gauge pressure in the downstream and proximity of the tractors. Results are given in Fig. 17. Tractor models do not generate significant vacuum regions around the surfaces and in the wake for travel speeds up to 20 km/h. At 20 km/h travel speeds, the cabin version generates 10-Pa vacuum around side surfaces and wheel proximities. This value would be different for rotating wheels. Also, a region of vacuum is generated above the cabin mostly at above the leading edge of the cabin cap. On the other hand, the platform version creates same amount of vacuum mostly in the wake. The cap of the driver compartment does not have comparatively significant vacuum region, probably due to the slower flow rates above the cap since below the cap is also open for flow, and therefore, same acceleration of cabin version is not seen here. The vacuum field in the wake of the platform version compensating the deficit above the cap is due to the same amount of aerodynamic resistance forces acting on tractor models. This also explains the upward tending wake flow of the platform version. This conclusion is justified by the remaining travel speed data. The vacuum fields of the cabin version enclose the cabin mostly. They have a half ring-like shape. However, vacuum fields of the platform version elongate and extend in downstream direction. Transferred energy from upstream to downstream by the jet through driver compartment in the platform version prolongs iso-surface downstream of the platform version. This again is regarded as a drawback for trailer scenarios. On the other hand, moving road surface, rotating wheels, tractor surface porosity, and different means of ventilation would affect vacuum fields of both models. It is also apparent that tractor wheels create and contribute the wake. Increasing travel speeds merge front wheel iso-surfaces with the general iso-surfaces. It is seen that the choice of cabin or platform versions affects upstream iso-surfaces. Side volume of the cabin version iso-surfaces is bigger as expected.
The spatial data emphasizes and explains aerodynamic forces acting on both tractor models for different travel speeds. However, the aerodynamic forces are also affecting the energy consumption of the tractors for transportation purposes. A tractor travelling at constant speed consumes energy for several and constant resistance forces. In order to decompose the resistance components, this work presents the aerodynamic component of the resistances. Figure 18 shows the necessary power to overcome the aerodynamic resistance for different travel speeds. The necessary power to overcome the aerodynamic force increases exponentially with travel speed as expected. It is seen that it can reach to about 16 kW. For more moderate speeds, the necessary power is between 2 and 12 kW. The necessary power for overcoming aerodynamic resistance increases eight times as travel speed grows two times between 40 and 80 km/h. If a 20% decrease in aerodynamic resistance can be achieved by improving tractor aerodynamic performance, 0.4 to 3.2 kW decrease in power requirement can be realized. However, this power amount increases due to energy conversion efficiency values. If 0.2 total energy conversion efficiency value is assumed from combustion to wheels through power delivery equipment, the necessary power amount grows five times. Accordingly, Fig. 19 shows the necessary fuel amount  Figure 19 gives similar trends with Fig. 18 and the essential content is then the diesel fuel amounts per unit time depending on travel speeds in order to overcome the aerodynamic resistance force. This data can approximately be used with the statistical data of tractor usage in transportation in order to determine the aerodynamic-related fuel consumption. Another approximate approach can be using carbon emission data in order to evaluate environmental impact of the transportation by tractors. The aerodynamic resistance-related carbon emission from tractors in transportation duties is approximated in Fig. 20.
Aerodynamic-sourced fuel consumption and carbon emission change between 5.7 × 10 −7 to 2.3 × 10 −3 lt/s and 4 × 10 −7 to 1.6 × 10 −3 kg/s respectively. One-hour transportation with the tractors corresponds to 3.6 lt diesel fuel and 2.52 kg carbon emission at 60 km/h travel speed. These values are 1.44 lt and 0.72 kg for 40 km/h while 7.92 lt and 5.94 kg for 80 km/h. Since any improvement in the aerodynamic performance is directly proportional for decreasing these values, there is a significant potential for working in geometric components of the tractor models. Above numbers would be more prominent when they are multiplied with possible agricultural tractor numbers in transportation. On the other hand, these values will help decomposing impacts of the transportation by tractors on environment.
Agricultural tractor drag coefficients are determined about 0.6 and 0.78 for cabin and platform versions, respectively. Drag coefficients of the two aforementioned literature works are about 0.8 for cabin holding tractor models Xu et al. 2016). This means about 30% difference between literature data and the present work. Since aerodynamic drag force is not encountered in the literature, it cannot be compared. It is anticipated that additional details such as moving road, porous tractor surfaces, and turning tires will alter and increase the drag coefficient and aerodynamical drag resistance force values.
According to this work, tractor wheels, front face of the engine compartment, top surface above driver compartment, and side surfaces of driver compartment are possible areas and regions for modification in favor of aerodynamic performance increases. However, CFD is not enough by itself. Wind tunnel test should be done in order to assess and validate CFD results. By this way, CFD geometries of agricultural tractors, their mesh structures, numerical models, and schemes can be evaluated and improved. Subsequently, moving road (wall), wheel spinning, transient analyses, vortex shedding frequencies, and other aspects should be determined. Major findings should be studied further systematically by proposing characteristic tractor geometries in a manner that resembles to Ahmed Body approach. After drag breakdown is achieved by systematic and detailed studies, aerodynamic performance improvement measures, by mostly passive flow structures, should be adapted in inspiration from the experience in commercial tractors or trucks. Finally, full-scale agricultural tractors should be tested for various transportation scenarios. In a parallel manner, statistical research relating to utilization of agricultural tractors in transportation Fig. 18 The power amounts that is necessary to overcome the aerodynamic resistance force Fig. 19 The diesel fuel amounts that is necessary to overcome the aerodynamic resistance force Fig. 20 The carbon emission amounts related to the aerodynamic resistance force is necessary in order to assess possible impacts. This research scheme, described in this paragraph, can be extended further to agricultural tractor trailer combinations. According to the literature survey, agricultural tractor aerodynamic performance can be improved up to 20% easily by adapting the experience in commercial tractors. Agricultural tractor front wheels, that are used for steering, have significant patterns and characteristic features. The impact of these wheels to aerodynamic resistance can be mitigated by temporary cases during transportation. Tires used in agricultural tractors have important parameters that affect traction (Ekinci et al. 2015;Taghavifar and Mardani 2014) and some of them such as tire type, pattern, inflation pressure, and position in general geometry can significantly change aerodynamic behavior. Base bleed structures through engine compartment and driver compartment can reduce wake dimensions. Vortex generators and turbulators can be adapted on top and side surfaces of cabin in order to separate and reattach the flow for narrower wakes. Tractor rear surface edges can be modified, as in commercial tractors, for pressure recovery. It is anticipated that the present work will trigger future works of researchers in the field for the abovementioned scenarios.

Conclusion
In this work, agricultural tractors that can be used in on-road transportation tasks are modified for and adapted to a CFD investigation for determining aerodynamic forces acting on them. The obtained aerodynamic forces that are attributed to transportation velocities are used for approximating aerodynamic sourced fuel consumptions and carbon emissions. It is thought that 20% drag force reduction is possible with future investigations on passive flow control devices that are previously tried on commercial tractors by the scientists, to adapt them on agricultural tractors during transportation operations. However, wind tunnel tests, vehicle tests, and improved CFD simulations are needed in order to breakdown drag components and handle them separately in the future. Some key findings in the present work are summarized below.
• Reynolds number independency is realized between 1.6 × 10 5 and 2.98 × 10 6 . Corresponding drag coefficients are 0.6 and 0.78 for cabin and platform versions, respectively. • Aerodynamic force acting on agricultural tractor during transportation changes between 3 to 746 N. This corresponds to 0.02 kW to 82.82 kW energy input to the system if total energy conversion efficiency is assumed as 0.2 considering combustion, power delivery, and other auxiliary steps. • The exponentially increasing power consumption and carbon emission suggest that lower transportation speeds by agricul-tural tractors may be mandated, even if the tractor may have much higher travel speed capabilities. • Cabin version is recommended in respect of aerodynamics since no significant drawbacks is found. • Aerodynamic-sourced fuel consumption and carbon emission changes between 5.7 × 10 −7 to 2.3 × 10 −3 lt/s and 4 × 10 −7 to 1.6 × 10 −3 kg/s respectively. One-hour transportation with the tractors corresponds to 3.6 lt diesel fuel and 2.52 kg carbon emission at 60 km/h travel speed. These values are 1.44 lt and 0.72 kg for 40 km/h while 7.92 lt and 5.94 kg for 80 km/h. • Front and rear wheels pose a resistance to the air flow and should be considered in the future work in order to determine their absolute effect by considering wheel spin. Also, measures should be designed accordingly. • Moving road, engine compartment porosity, and effects of different cooling speeds are other factors to be considered in future CFD work. • Aerodynamic experience on commercial tractors should be adapted for agricultural tractor transportation operations. • Drag breakdown should be necessary for deeper understanding on the topic. A characteristic geometry, representing characteristic aerodynamic features of the agricultural tractor, is needed. • Wind tunnel tests are necessary for CFD validation. • Real-world 1:1 on-road test should be the final step of aerodynamic measures for high-speed travelling agricultural tractors.