Within this work, an existing model of a Suzlon S111 2.1 MW turbine is used
to estimate potential cost savings when the conventional upwind rotor concept
is changed into a downwind rotor concept. A design framework is used to get
realistic design updates for the upwind configuration, as well as two design
updates for the downwind configuration, including a pure material cost out of
the rotor blades and a new planform design. A full design load basis according
to the standard has been used to evaluate the impact of the redesigns on the
loads. A detailed cost model with load scaling is used to estimate the impact
of the design changes on the turbine costs and the cost of energy. It is shown
that generally lower blade mass of up to 5

Historically, the first wind turbines were dominantly downwind turbines, for which the rotor was placed behind the tower, as seen from the incoming wind. This
turbine configuration was considered safer than the alternative upwind
configuration with the rotor in front of the tower since the rotor blades
would bend away from the tower under turbine operation. Early research, mainly
by NASA and associated partners, compared the downwind rotor configuration with
the upwind configuration.

Many residents living near early downwind wind turbines reported
high noise levels and especially the high unsteadiness, a “thumping” sound
being reported as a nuisance (

Cost-driven industrial designs prefer larger rotor areas to capture more energy. The rotor blades for modern-sized upwind wind turbines are designed under a constraint of maximum blade tip deflection to avoid a collision of the blades with the tower. Aiming to eliminate the tip deflection constraint for modern-sized wind turbines under normal operation, the downwind configuration is currently coming into research focus again.

Advances in wind turbine noise mitigation techniques since the 1980s, as well
as airfoil design, could overcome the previously reported noise issues and bring them to an
acceptable level.

In a system-level design study for large rotors,

A reduced edgewise damping for a downwind configuration compared to an upwind
configuration was identified by

Aligning the blades with the loading direction of aerodynamic
forces, gravity, and centrifugal force is an opportunity for the downwind
configuration to significantly reduce flapwise bending loads, instead loading the
blade in axial tension. Such a load distribution is achieved by
adjusting the cone angle and blade prebend. These downwind rotors with
so-called “load alignment” have been suggested as an option to reduce blade
mass significantly, utilizing the large cone angles and downwind prebend from

Downwind configurations with a passive wind direction
alignment are often proposed. Such yaw systems could be cost efficient as they simplify the
turbine control and reduce operation and maintenance costs as they could
purely be used for cable unwinding. However,

The cost-efficient design of wind turbines has been approached to an increasing extent by the use of optimization frameworks. Over the years, rotors designed for the maximum efficiency result in the most
cost-efficient turbine designs have been questioned. Optimizing a conventional 10

Lower rotor loads could potentially result in the possibility to increase the
rotor length and therefore increase the overall power capture. This could be a
more cost-efficient rotor than a traditional design approach also for upwind
turbines.

This paper shows how rotor design trends for a downwind configuration differ
from design trends for an upwind rotor configuration due to differences in
design loads inherent to the configuration. Full design load bases (DLBs) are
calculated to evaluate the impact of the rotor design trends on the turbine
loads, AEP, and estimated cost of energy. The work in this paper is based on
the specific example of the commercial S111 2.1

This work aims to compare design trends for an upwind configuration of an
existing turbine with a downwind configuration from a cost and mass
perspective. The chosen example turbine is the Suzlon S111 2.1

For this turbine, a new baseline rotor blade is defined, inspired by the
commercial blade, which is adapted to the framework. For the baseline rotor,
an upwind turbine configuration is generated, called S111uw. Additionally, a
downwind baseline turbine configuration is defined with the baseline rotor
called S111dw. The downwind configuration utilizes the same cone and tilt
angle, both increasing blade tip to tower distance. Since the blade prebend of
the rotor is towards the blade pressure side, the prebend decreases the blade
tip to tower distance in the downwind configuration. Three rotor redesigns are
made. For the upwind configuration, a blade planform and internal structural
redesign is made. The design is called S111uw PF. For the downwind
configuration, two scenarios are regarded. Firstly, a pure blade material
reduction is performed, called S111dw STR. This corresponds to a configuration
change from an existing upwind configuration into a downwind configuration while
keeping the blade molds but saving blade material. Secondly, a blade planform
and structure redesign in the same manner as for the upwind redesign is called
S111dw PF. Table

Turbine configurations regarded in design and cost estimation.

Flow chart of the work flow for design and cost estimation.

The rotor design procedure uses a low-fidelity optimization tool to create a blade planform and stiffness distribution. The planform and stiffness distribution are afterwards matched within the HAWTOpt2 framework

The baseline blade is set up in BECAS (a 2D cross-sectional analysis tool;

The baseline blade is described according to the parameterization adopted in
HAWTOpt2

For the baseline blade structural properties, total mass, static mass moment, and blade eigenfrequencies are compared to the commercial blade to assure the baseline is reasonable and fairly close to the commercial blade. The same has been done with turbine eigenfrequencies and damping, as well as the design-driving loads for blades, main bearing, and tower.

Full design load bases are simulated with HAWC2 (version 12.7) according to
the IEC standard 61400-1 Edition 3 (

The annual energy production (AEP) is calculated for all designs. It is calculated from the normal operation load case with six turbulence seeds for all configurations without inclination or yaw angle. The turbulence intensity follows the class A IEC standard.

For all load calculations, the controller setup from DTU
(

Start-up and shutdown pitch speed in the implemented routines of the DTU controller need different values for downwind configurations than comparable upwind configurations. The moment due to both thrust force and the gravity overhanging moment of the rotor nacelle assembly increases the tower bottom bending moment. Start-up routines, especially at high wind speeds, need to have a lower pitch speed in downwind configurations than the comparable upwind configurations. Shutdown routines, especially during gusts, have to be of faster pitch speed in the downwind configuration. Both adjustments have to be made to unload the tower bottom as the moment due to the thrust force is aligned with the overhanging rotor moment due to gravity. A faster pitch decrease in gust situations reduces overshoot in the thrust force due to the gust and therefore the tower base loads. During start-up, a slower pitch increase avoids a thrust overshoot and related high tower loads.

For a control routine that reflects an industrial controller, three failure
scenarios are adapted. Firstly, the failure scenario of one blade getting
stuck at a current pitch angle, which means that the pitch angle of one blade is kept constant
at the current pitch angle at the time of failure. The deviation of the pitch
angle from the set point initiates a stop routine of the turbine. Secondly, the
pitch run away (design load case, dlc, 2.2p) is not included since the failure mode is prevented
by the type of pitch actuators used. Thirdly, for the scenario of a parked
turbine with high yaw errors, the wind field is interpreted as a wind direction
change of 360

To eliminate fault cases from the design-driving loads and to stay similar to an industrial controller, two additional control features are implemented as separate dynamic-link libraries manipulating the output or input from the controller to HAWC2 for practical reasons. The first addition is a thrust control aiming to reduce fluctuations of the thrust. The second addition is a conditional stop routine avoiding operations at high yaw errors and high wind speeds. The following explains the two additions in more detail.

Flow chart of the thrust control controller addition.

Figure

The conditional stopping routine triggers the turbine stop as soon as the filtered wind speed and the filtered wind direction are above a certain threshold. For practical reasons of implementation, the emergency stop is triggered.

The redesign of the rotor blades is performed using the in-house code STORM
(Suzlon Turbine Optimization fRaMework). The code is aimed at fast conceptual
rotor design optimization studies and couples steady aerodynamic AEP
considerations with a simplified blade structural estimation. In the present
study, it minimizes the blade mass under AEP constraints. The code, written in
MATLAB, is organized as a nested optimization problem. The outer optimization
loop controls the blade geometrical planform and minimizes the blade mass
subject to linear constraints on the geometrical design variables, nonlinear
constraints on minimum AEP, and feasibility of all the inner optimization
problems (Eq.

In this study, the blade geometry design variables are limited to four spline
control points that set the thickness-over-chord (ToC) ratios in fixed points
along the blade span. The geometry at the blade root is fixed up to the point
of maximum chord for all configurations. For each iteration of the outer
optimization loop, six steps are taken; they are described in the following
sections and briefly consist of the following.

The blade ToC spline is defined from the control points (the four design variables).

The blade geometrical planform is outlined in terms of chord, twist, and thickness distribution. An inner optimization returns the chord distribution that minimizes the squared difference from a target axial induction distribution.

Steady operational loads and the power curve are retrieved from a standard steady blade element momentum (BEM) formulation. An inner optimization sets the pitch angle to maximize aerodynamic power, subject to limitation on maximum power, thrust, aerodynamic flapwise bending moment, and angle of attack (for stall considerations).

The steady BEM loads are scaled to extreme loads to be used in the structural optimization.

The blade structural properties are determined, solving a fast low-fidelity structural optimization problem. The blade structure is simplified to two symmetric glass fiber spar caps joined by an ellipse (Fig.

Finally, the outer loop optimization objective function is evaluated. The estimated blade mass is here taken as an objective function, and minimum AEP output is enforced as a nonlinear constraint.

Figure

Flow chart of the optimization routine with STORM.

Simplified model of the blade structure for each cross section, as applied in STORM. The section height

The outer optimization problem (Eq.

The design variables are here the four thickness-over-chord (ToC) control point ratios. Linear constraints on the design variables are enforced to ensure that they are maintained within reasonable ranges and that monotonically decreasing values are selected from root to tip. The objective function for this problem consists of minimizing the estimated blade mass, subject to nonlinear constraints to reaching a minimum AEP output (as derived from the BEM steady power curves) and ensuring feasibility in all the inner optimization problems.

Once the iteration ToC control points are fixed, the ToC distribution along the blade span is outlined with a piecewise cubic Hermite interpolating polynomial. A wind speed in the below-rated variable speed range is chosen, and the target axial induction distribution for the blade at that wind speed is fixed as an input. Similarly, the angles of attack at which the airfoils are expected to operate at that wind speed point are also fixed. The target axial induction is kept according to the original commercial blade. It has been assumed that this is a typical induction distribution resulting from a commercial aerodynamic design process. It has been kept as the optimization tool is not capable of reflecting the complexity of fully variable induction, also regarding related concerns such as stability or stall margins.

With the given input set (ToC, target induction, target angle of attack) the blade
geometry is then retrieved in terms of chord, twist angle, and thickness for
each section along the blade span. The chord is retrieved by solving a set of
independent minimization problems (Eq.

In the current iteration, the axial induction is retrieved from a steady BEM
formulation, following Ning's implementation (

Given the blade geometrical definition like from the step above, the steady
power and loads curves are then determined running a standard steady BEM
formulation

The operational pitch angle at each wind speed is retrieved from a simple
optimization loop, in which the objective is to maximize the aerodynamic power
output, subject to constraints on maximum power (the aerodynamic rated
power), maximum thrust force, maximum aerodynamic blade flapwise bending
moment, and minimum “stall distance” (Eq.

The maximum aerodynamic steady flapwise bending moment is retrieved from the
step above and is scaled up to an extreme load using a ratio retrieved from
full DLB HAWC2 simulations of the baseline blade:

In the case of the downwind configuration, a second flapwise design load case
for cut-out wind speed is considered as the minimum tower-blade clearance
arises in different loading conditions. The load distribution for the maximum
deflection towards the tower

The edgewise loads remain unscaled as they are driven by the aerodynamic torque, as well as the gravity load.

To verify the load scaling approach, it has been checked that the tower clearance from dynamic HAWC2 simulations is captured reasonably well. Also the failure indices for each blade section have been checked to assure that also locally on the blade sections the approach captures the loads reasonably well. No direct comparison between the scaled loads and the dynamic loads from HAWC2 has been done.

The simplified blade structural model is based on the work of

The load cases described in the previous section are applied to the finite
beam element model, and the structural optimization aims at minimizing the
blade static-mass moment, subject to constraints on the range of the design
variables, maximum strain levels on caps and ellipses, maximum tip deflection
for the deformed blade, and maximum buckling coefficient for a single spar
cap. The structural optimization problem can be stated as

The buckling coefficient is added to the optimization problem compared to the
references. The buckling coefficient is calculated under the assumption of an
orthotropic plate under compression load. The compression load

The optimization is solved with the Interior Point Optimizer, Ipopt
(

The optimized planform (chord, twist, and thickness distribution) and
the changes in the structural geometry (spar cap width, thickness of the spar,
and trailing edge caps) are applied in HAWTOpt2 according to the planform
calculated by STORM. All thickness distributions are fitted by hand at five control points, and a spline fit is applied in between the control points. The
HAWC2 inputs are extracted from HAWTOpt2, and a DLB is calculated for each
redesign. From the DLB, the maximum load at each blade cross section is
extracted. The failure index is calculated with BECAS for each
cross section. The design is accepted if the failure index

The DLB calculation is further used to calculate the tower wall thickness

The cost model used for the cost evaluation consists of costs that scale with
the mass, such as tower and blade costs. For other components, the costs scale
with a design-driving load or measure called cost driver (CD). The cost driver
is scaled with a factor

Other cost components, e.g., logistics or operation and maintenance costs, are scaled directly with the factor

Cost drivers (CDs) for turbine cost and mass split by main cost components.

The following section presents the resulting design configurations regarding the planforms and resulting blade masses. Further, the design-driving loads and the resulting changes in turbine costs and COE are presented. All results are shown relative to the S111uw design configuration as the data are confidential.

Comparison of planforms for different designs. Thickness over chord ratio and the range of the thickness constraints, chord and twist are normalized with the maximum chord.

Figure

While the S111uw PF design is constrained in blade deflection, in none of the downwind designs is the blade deflection constraint active. All the resulting redesigns are generally utilizing the maximum strain of the material over a larger blade span than the S111uw and S111dw design configurations. All downwind redesigns are fully strain constrained in the spar caps. However, the tower clearance for the S111dw PF design is only marginal. In the structural module of the optimization, the buckling constraint is active along the full blade span. The downwind configurations generally show greater shell thickness than the upwind configuration.

Generally, the difference in active design constraints between the S111uw PF and S111dw PF design are that the S111uw PF design is strain constrained only in a small part of the mid-span section, and the tip deflection constraint is active. The S111dw PF design, on the other hand, is fully strain constrained over the full blade span, and the tip deflection constraint is not active.

For all redesigns of the rotor blade, significant mass savings of at least 12

The planform redesigns utilize greater stiffness with less material by using thicker airfoils in the inboard part, resulting in an overall reduction in mass. In the outboard part, thinner, more efficient airfoils compensate for a production loss of the inboard part of the blade. This effect is amplified as a small AEP penalty was allowed in the design procedure. From the S111dw STR, it can be seen that the downwind configuration benefits from lower flapwise loads and a release of the tower clearance constraint resulting in a reduced blade mass. A greater shell thickness is required to carry the higher edgewise loads in the downwind configurations. Comparing the S111dw PF design to the S111uw PF design, a further effect of the edgewise load increase can be seen. To carry the increased edgewise loads, there are two options. The first one is to increase the shell thickness like for the S111dw STR design. The second option is to increase the stiffness by using airfoils with higher relative thickness. The solution found in the optimization routine for the S111dw PF is a combination of both, showing slightly thicker airfoils on the inboard part for the S111dw PF than for the S111uw PF. Another solution to carrying the increased edgewise loads is an increased chord, but since the variation in chord is limited due to a fixed induction and tip speed ratio, this design freedom is not utilized. The lower flapwise loads in the S111uw PF design allow us, on the other hand, to compensate for a power loss with slightly thinner airfoils in the outboard part. The chord distribution is hardly changing as the AEP is constrained to not deviate from the baseline AEP. As this results in a similar lift level along the blade for all designs and the induction distribution is frozen, the chord length does not change. The twist is simply adjusting the given operational point of the airfoils at the given spanwise position.

Turbine loads for mass and cost drivers. Blade root moment (BRM), tower bottom bending moment (TBM), and tower top moment (TTM).

The following section shows the loads driving either the cost components in
Table

The table also shows that the downwind designs generally benefit on the flapwise mean, flapwise extreme blade root moment, and the related tower top yaw moment. This is mainly due to the alignment of the rotor cone and the rotor forces (“load alignment”). The tower top tilt moment is increased in the downwind designs compared to the upwind designs. Here, the influence of the tower shadow, as well as the alignment of the rotor overhanging gravity moment with the moment due to thrust force, is observed. Due to the latter, an increase in the extreme tower bottom bending moment is also seen compared to the S111uw design. The gravity-related loads, e.g., tower top tilt moment and longitudinal tower bottom bending moment are reduced for each configuration by the reduction of mass due to the redesign (e.g., S111uw vs. S111uw PF and S111dw vs. S111dw PF). With the reduced flapwise stiffness of the S111dw PF design, the tower shadow effect is overcome. As a result, the fatigue load of the S111dw PF is reduced to the level of the S111uw. A relative reduction of the flapwise stiffness compared to the edgewise stiffness increases the edgewise damping. Therefore, a load decrease for edgewise extreme and fatigue loads of the S111dw STR and S111dw PF compared to the S111dw is observed.

This section shows the estimated costs resulting from the load and mass
difference of the design configurations. Figure

Blade mass, tower mass, CAPEX, AEP, and COE differences for the regarded turbine configurations relative to the S111uw design.

Table

Turbine CAPEX cost split by main cost components normalized by the sum of the S111uw configuration with an indication of constant costs not affected by the redesign process.

Within this study, the COE reduction potential for the Suzlon S111
2.1

New planforms were optimized for upwind and downwind configurations for
minimum blade mass under the constraint of a minimum AEP. The new planforms
were shown to have higher thickness over chord ratios inboard, utilizing
higher stiffness with less material. This design trend agrees well with
findings by

The downwind designs were generally subject to lower flapwise blade root
moments than the comparable upwind designs due to the coning direction, as
also proposed by, for example,

The load saving on the blade in the downwind configuration is offset by an
increase in the tower bottom bending moment as the gravity overhanging moment
of the rotor nacelle assembly is aligned with the thrust force, as also shown
by

The downwind configurations are subject to a lower AEP production due to the
coning direction. This effect has also been observed by, for example,

Lower rotor and nacelle costs can be achieved by the downwind designs. However, the downwind designs also come with higher tower and foundation costs. Overall, the downwind configurations of comparable rotor size achieve a lower total turbine cost than the upwind design configuration. The difference in cost is due to the lower OPEX cost and does heavily depend on the cost model. Overall, the lower turbine cost does not compensate for the loss in AEP. The lowest COE level is achieved by the S111uw PF design configuration which achieves a significant mass and load reduction for a small sacrifice in AEP compared to the baseline.

This study has shown, for the example of the Suzlon S111 2.1

These results depend on the very baseline-specific cost model. Scaling the OPEX with the AEP has been the only cost driver for the OPEX which results in the lower turbine costs for the downwind configuration. It could be expected that the higher fatigue load of the downwind configuration would increase the material wear, but this does not enter the OPEX model.

It should also be highlighted that the costs are effected by the chosen optimization approach, namely a mass minimization under AEP constraint. This does not give the true optimal solution in the sense of cost of energy. However, it does show the influence of the observed design trends on the turbine cost and the cost of energy.

The cost model generally depends on the loads simulated. This comes with uncertainty due to the seed number, the seeds themselves, and the assumptions of the wind field inclination angle. In the case of the downwind configuration, additionally, the dynamic effect of the tower shadow is not captured correctly within the HAWC2 simulations. Within HAWC2 the tower shadow model for downwind configurations is a pure deficit model and the increased vorticity behind the tower is not reflected. It can be expected that especially flapwise blade root and tilt-related fatigue loads are underpredicted. Further research would need to be done to quantify the impact of this effect.

Generally, fatigue loads should be part of the design process in future work. In the chosen approach, fatigue loads are not regarded in the design process and hardly reflected in the cost model. This might be a valid assumption in the upwind configurations, but for downwind configurations, this approach needs to be proven. Due to the tower shadow effect, as well as a possible decrease in edgewise damping, it might be possible that rotors of downwind configurations are driven by edgewise fatigue loads rather than flapwise extreme loads.

Prescribing the induction distribution in the optimization is a major restriction of the chosen design approach. The resulting chord and twist distributions are therefore very similar. The induction should be a design variable in future work as unloading the tip might allow for increases in rotor diameter and therefore AEP increase. The latter does not just hold for the downwind configuration but also for the upwind configuration.

Prescribing the induction distribution did, however, have the advantage that the load scaling approach was possible. Scaling loads from the BEM code loads to the extreme loads has decreased computation time significantly. A drawback of the load scaling approach is that a change in aerodynamic damping is not reflected. For the downwind configuration, the flapwise stiffness could be significantly reduced, while the edgewise stiffness had to be increased; the edgewise whirl modes can, therefore, be expected to increase in damping due to the frequency placement of the edgewise frequency compared to the second yaw frequency. An increase in damping decreases the blade extreme loads. The effect of the loads has been observed in the downwind designs, but there is no feedback within the optimization reflecting the change in damping. In future work, the framework would need to be enhanced with either time consuming load calculations or with a set of transfer functions that can transfer a wind field to extreme loads from a linearized turbine model. Such a linearized turbine model could be extracted, for example, from HawcStab2, which uses these models for eigenvalue analysis. In this case, a representative wind field could be used that represents extreme loads from a simulation set with a much larger seed number and with known uncertainty. This would decrease the computational time drastically while achieving reasonable results.

As designed for fast conceptual rotor design studies, the chosen design approach with STORM is limited in finding truly optimal solutions in the sense of the lowest cost of energy. The reasons are, on the one hand, the simplified structural model. On the other hand, the aerodynamic planform is limited by the assumption of a prescribed target induction, thus, allowing hardly any freedom of chord variations. Further, fixing the rotor diameter limits the investigations of the cost of energy for the different designs as it is fixed to one point of investigation. With the chosen approach and the AEP constraint, it is only possible to find the lowest blade mass for a certain targeted AEP at a specific blade length. With these limitations the resulting designs are not truly optimal in the sense of the lowest cost of energy. However, the chosen approach does allow for the comparison of design trends between the upwind and the downwind configurations.

The COE estimation and therefore success criteria of the downwind concept do also depend on the cost share between the different components. Since in the chosen example turbine the rotor and the tower are similar in the CAPEX share, it is difficult in the downwind configuration to offset the increased tower cost with savings on the rotor. If the baseline had a comparably more expensive rotor and a cheaper tower, the downwind configuration would be more competitive. Possible scenarios could be lower steel prices or higher blade material prices.

Another possibility to increase the competitiveness of the downwind configuration would be a change in the tower configuration, such as a wired tower for which wires are a cheap measure to take the bending loads. Alternatively, a low-labor-cost market could give the options of low tower costs with lattice or hybrid-lattice towers which generate bending stiffness from the increased footprint of the tower rather than large tower wall thicknesses for a tubular tower. These options could make the downwind configuration competitive as the cost share of the tower decreases. However, the cost model with the chosen baseline is not able to reflect such significant design changes.

Compensating the AEP loss in the downwind configuration with a larger rotor
area could be an option to decrease the COE. Nevertheless, this does also
increase the turbine cost not just due to an increased rotor diameter and
therefore rotor mass but also to mass-related loads such as tilt loads and tower
base loads. The rotor diameter has not been part of the rotor design as the
cost model is very specific and does not reflect large differences from the
baseline. Especially for components such as generator or gearbox which are not
available in any possible configuration but are bought as “off-the-shelf”
components, the linear cost scaling is insufficient. A rotor diameter increase
of 4

Future work should also consider a redesign of the nacelle for the better
balancing of the rotor mass on the tower for the downwind configuration, as
suggested by

It should not be forgotten in the discussion of the cost efficiency of
downwind configurations that simple control features such as peak shaving, as
suggested by

Overall, the study shows that a downwind configuration of the chosen example
2.1

The optimization framework would need to be extended to be able to capture the design changes regarding the rotor, but also different tower configurations need to be included. To be able to evaluate such changes, a more comprehensive cost model is required to do a fair comparison of the designs.

It can be concluded from the study that it will be difficult to design a
downwind configuration in the 2

The following Table

List of symbols used in the equations.

The data are not publicly accessible since the research is based on a commercial turbine and the data are not available for disclosure by Suzlon.

GW created the baseline model and implemented the downwind design case in the optimization framework. LB developed and set up the optimization framework. FZ set up the HAWTOpt2 framework for the baseline and design case evaluation. All authors revised the models and results. With revisions of all coauthors, GW and LB prepared Sect. 2.3, and GW prepared the remaining paper. DRV supported the setup of the models and revised the models, results, and the final paper.

This project is an industrial PhD project funded by the Innovation Fund Denmark and Suzlons Blade Science Center. Gesine Wanke is employed at Suzlons Blade Science Center.

This research has been supported by the Danish innovation fund (grant no. 5189-00180B).

This paper was edited by Raimund Rolfes and reviewed by Andrew Ning and Pietro Bortolotti.