FioreSelig-2015-AIAA-2015-3393-AirfoilShapeOpt-for-Erosion
AIAA Aviation 2015-3393 22-26 June 2015, Dallas, TX
33rd AIAA Applied Aerodynamics Conference
Optimization of Wind Turbine Airfoils Subject to Particle
Erosion
Giovanni Fiore?and Michael S.Selig?
University of Illinois at Urbana-Champaign,Department of Aerospace Engineering,Urbana,IL61801
An optimization of wind turbine airfoils subject to particle erosion is investigated.The erosive
damage to the surface was represented by sand grains colliding with the blade leading edge.The
optimization was performed by using a genetic algorithm approach(GA).A newly-written GA code
was coupled with a two-dimensional inviscid?ow?eld solver(XFOIL)and with a particle dynamics
code(BugFoil).The chosen scheme of the GA was based on random tournament selection.Each
geometry was evaluated through a?gure of merit that represented the airfoil?tness to damage and
to aerodynamic performance when the airfoil is damaged.Multiple approaches to the optimization
were taken.The initial step of this study involved optimization through geometry perturbations of
the leading edge by means of Bezier curves.It was found that the leading edge shape modi?cations
of existing airfoils increase the?tness of the airfoils subject to erosion.The second approach in-
volved a crossover of existing airfoil geometries along with leading edge shape mutations,in order
to expand the design space.The?tness of such airfoils was improved with respect to the?rst ap-
proach.However,for both these approaches the airfoil aerodynamic performance was disregarded,
as they were intended to give insight on the optimal airfoil shapes.The?nal step of the study
made use of an inverse airfoil design method(PROFOIL)to widen the design space even further.
In this approach the aerodynamic performance of the airfoil was evaluated,and the GA became
a two-objective optimization process(maximum lift-to-drag ratio,and erosive?gure of merit).It
was observed that bulbous upper leading edges,and slanted,?at lower leading edges allowed for
best erosion performance.Moreover,it was seen that the airfoils with a positive camber required
a smaller angle of attack to operate at the same lift coef?cient,hence reducing the erosive damage
experienced for sand grain impacts.The airfoils obtained from the inverse design GA algorithm
were characterized by a lift-to-drag ratio greater than310for a Reynolds number of5.84×106.
Nomenclature
AK=particle nondimensional mass
c=airfoil chord length
COE=Cost of Energy
C d=airfoil drag coef?cient
C D=particle drag coef?cient
C l=airfoil lift coef?cient
d=particle diameter
D=particle drag force
E=erosion rate
FOM=?gure of merit
GA=genetic algorithm
GAEP=Gross Annual Energy Production
K=erosion rate constant
?Graduate Student(Ph.D.),Department of Aerospace Engineering,104S.Wright St.,AIAA Student Member.
?Associate Professor,Department of Aerospace Engineering,104S.Wright St.,AIAA Associate Fellow.
https://www.wendangku.net/doc/415462556.html,/m-selig
LE=leading edge
m=particle mass
n=erosion rate velocity exponent
r/R=blade section radial location
Re=freestream Reynolds number
t=time
t/c=airfoil thickness-to-chord ratio
U=chordwise?ow?eld velocity component
V=chord-normal?ow?eld velocity component
V∞=freestream velocity
V imp=particle impact velocity
V s=particle slip velocity
x=chordwise coordinate
y=chordwise-normal coordinate
α=angle of attack
αr=relative angle between?ow?eld and particle velocity
θ=impact angle
τ=nondimensional time
ω=weight used to compute FOM
Subscripts and superscripts
0=initial state
l=lower limit
P=particle
tr=transition
u=upper limit
I.Introduction
Wind turbines used for eletrical power generation are subject to fouling and damage by airborne particles typical of the environment where the wind turbine operates.Throughout the20-year lifespan of a wind turbine,particles such as rain,sand,hail,insects,and ice crystals are major contributors to a deterioration in turbine performance through local airfoil surface alterations.1–6Wind turbine blades accumulate dirt especially in the surroundings of the leading edge. In addition,temperature jumps and freeze-thaw cycles may cause smaller coating cracks to propagate,promoting coat-ing removal and eventually delamination and corrosion damage due to exposure of the internal composite structure. The originally smooth surface of the blades may change considerably,and the increased roughness will cause a drop in gross annual energy production(GAEP)and an increase in cost of energy(COE).7–13
Modern trends in the wind turbine market have shown the bene?ts of offshore megawatt-scale wind turbine instal-lations14,15in order to maximize GAEP while reducing COE.However,offshore locations are subject to more intense sand erosion than the majority of land installations.13,16,17Airborne sand particles collide with the blade and cause micro-cutting and plowing in the coating material18,19resulting in surface abrasion.20,21Such damage is particularly prominent at the outboard sections of the blade where the local relative velocity is larger compared with inboard sec-tions.16,22
The weather conditions of a given wind farm site may vary substantially throughout the seasons.Throughout the world,another common damage scenario is represented by water-based particles,namely raindrops and hail-stones.23–25When intense precipitation occurs,large raindrops may impact with the wind turbine blades and poten-tially promote internal panel delamination.23,26Even smaller raindrops increase the mechanical fatigue in the coating plastic materials27,28which translates into surface micro-cracking and a?nal erosive effect.29 The wind turbine industry has developed several devices to help mitigate the effect of airborne particles striking on the blade surface.In particular,chemical companies have commercialized leading edge tape products which allow for a convenient solution that is integrated into the blade maintenance cycle.11,30,31However,from the standpoint of
an aerodynamic designer such an approach may appear as a later?x.Hence,it is valuable to investigate which blade section geometry features allow for a minimization of the surface erosive damage.
The recent research has shown that the blade damage scenario is complex and multi faceted.22–25,32From an aerodynamic standpoint it must be noted that there exists a correlation between the blade?ow?eld and the particle in-teraction with the blade surface.In particular,small and lightweight particles are more susceptible to the aerodynamic ?ow?eld than bigger and heavier particles.Therefore,a blade shape optimization to minimize the damage of any type of particle may be dif?cult to implement.For example,insects represent a complex scenario which is challenging to undertake due to the large variety of species and knowledge gaps concerning their aerodynamic characteristics.Other heavy particles such as hailstones are characterized by a trajectory that is fairly independent of the blade?ow?eld.23,25 Little may be done from a shape optimization standpoint to minimize the damage to the panel due to hailstone strikes. However,the two particles that are most sensitive to the blade?ow?eld are sand grains and rain drops.22,25Both parti-cles are associated with an erosion rate,hence allowing for an estimate of surface material loss.From an optimization perspective,such a parameter represents a direct feedback of the airfoil?tness with respect to particle erosion.
Large values of erosion rate are responsible for an earlier visible damage on the blade over time when compared with smaller values.Given the air quality characteristics of a chosen wind farm,a blade section geometry that is responsible for higher erosion rates will output less energy than a more damage-resilient blade over time.Such con-siderations motivate the investigation of optimum section geometries,under the novel perspective of airborne particle erosion.The goal of this study is to investigate the blade section geometry optimization for the case of sand erosion because sand represents the type of particle that is mostly susceptible to the blade aerodynamic?ow?eld,and hence the blade section geometry.
This paper is divided into?ve sections:the numerical method used is explained in Section II,the blade operating point,particle aerodynamics and damage models are introduced in Section III,while the results obtained are discussed in Section IV.Finally,conclusions are proposed in Section V.
II.Methodology and Theoretical Development
A.Particle Equations of Motion
In order to predict the erosive effect of sand grains on the blade surface,a lagrangian formulation code was developed in-house and named BugFoil.22,25BugFoil integrates a pre-existing particle trajectory code33and a customized version of XFOIL.34The local?ow?eld velocity components are obtained by querying the built-in potential?ow routine of XFOIL from which the particle trajectory and impact location on the airfoil are computed.
In steady?ight,the forces acting on the particle are perfectly balanced and perturbations to such forces are assumed to be additive to the steady-state forces.For these reasons the equations of motion may be expressed by neglecting the steady-state forces and may be written as functions of increments only.35In the current study,both raindrops and hailstones were treated as aerodynamic bodies whose only associated force is the aerodynamic drag D.
By applying Newton’s second law along the particle trajectory in both chordwise x and chord-normal y directions, the following equations are obtained18,22,36–38
m P d2x P
2
=ΣF x(1)
m P d2y P
dt2
=ΣF y(2)
By projecting the drag of the particle D in both chordwise x and chord-normal y directions using the relative angle between particle and?ow?eld velocityαr,the equations may be rewritten as
m P d2x P
dt
=?D cosαr(3)
m P d2y P
dt2
=?D sinαr(4)
Given the particle velocity components U P and V P and given the velocity?ow?eld components U and V at a certain point along the trajectory,the particle slip velocity V s can be expressed as
V s=
(U?U P)2+(V?V P)2(5)
while the trigonometric functions in Eqs.3and4may assume the form
cosαr=U?U P
V s
=
V rx
V s
(6)
sinαr=V?V P
V s
=
V ry
V s
(7)
By expressing the particle aerodynamic drag D as a function of dynamic pressure and by substituting for the trigono-metric functions,the Eqs.3and4may be rewritten as
m P d2x P
dt2
=
1
2
ρV2s A P C D
V rx
V s
(8)
m P d2y P
dt2
=
1
2
ρV2s A P C D
V ry
V s
(9)
To scale this problem in a non-dimensional fashion,non-dimensional time,space,and mass parameters can be intro-duced here
τ=t U
c
(10)
x P=x P
c
(11)
y P=y P
c
(12)
AK=
2m P
ρA P c
(13)
Nondimensionalization of Eqs.8and9by a reference velocity U yield
d2x P
=1
V r C D V rx(14)
d2y P dτ2=
1
AK
V r C D V ry(15)
which together represent a set of second-order,nonlinear differential equations.Once the particle drag coef?cient is evaluated,the trajectory can be computed by numerically solving both x and y equations.
B.Sand Erosion
Upon impact,sand grains promote a mechanism of surface abrasion.Sand erosion is responsible for an increase in blade surface roughness and a decrease in structural stiffness.The parameter erosion rate E,de?ned as the removed mass of the target material divided by the mass of the impacting particle,is a function of particle impact velocity V imp and angle at impactθ,and it is measured in units of(g/g).19The velocity V imp is related to E through a power-law;whereas,the correlation with impact angle strongly depends on the eroded material properties.Most current materials used for wind blade coating are polyurethane derivatives30and show a primarily plastic erosion behavior with maximum erosion rate atθ=30deg.39Following the approach previously implemented by the authors,22the erosion rate for plastic materials is given by the equation20,40–43
E=K V n imp(16) where K and n are constants of the eroded material.The correlation between E andθis implicit in the parameters K and n?tted at various impact angles and impact velocities.Similar to previous studies,the simulations were performed
by using linear-?tted erosion constants of ultrahigh molecular weight polyethylene(UHMWPE)22because it has the best performance of the polyethylene-based coatings,and because normal impacts cause small erosion rates.40 Each simulation is performed by placing a vertical array of sand particles?ve chord lengths upstream of the airfoil. Once the the simulation is initiated,the trajectory of the particles is evaluated and impingement is estimated.Given the angle and velocity at impact,E assumes different values in the region around the leading edge(LE),with a minimum associated with near-normal impact.
III.Aerodynamics of Damaged Airfoils and Analysis of Section Geometries
The design point for wind turbine airfoils is associated with the maximum aerodynamic ef?ciency(C l/C d)max.To illustrate the signi?cance of blade shape optimization it is of particular interest to predict the performance of blade sections in a damaged con?guration(i.e.,when the aerodynamic ef?ciency is reduced).In this section the aerodynamic performance of a damaged DU96-W-180are discussed.
A.Parametric Study on the Location of Fixed Transition
The erosion rate E curves due to sand on a DU96-W-180airfoil(α=6.0deg)are shown in Fig.1.The earliest damage to appear on the surface corresponds to the location of maximum erosion rate,when the blade is exposed to a prolonged erosion(as discussed by the authors in previous works22).In fact,at such locations,one can assume the quantity of eroded material to be the largest.Therefore,two?xed transition locations are considered on the upper and lower surfaces:x u tr=3.13%and x l tr=8.39%,respectively.These chordwise locations are obtained with BugFoil for a blade section located at r/R=0.75andα=6deg,since it represents a suitable design point near(C l/C d)max.
For a more general scenario,the effects on E due to the blade angle of attack,particle diameter,and section spanwise location(through local blade velocity)are shown in Fig.2.It can be seen that an increase inα[Fig.2(a)] is translated into a shift of the maximum erosion rate toward the leading edge for the upper surface and toward the trailing edge for the lower surface.On the other hand,an increase in particle diameter[Fig.2(b)]is responsible for an increase in maximum E on both airfoil sides.Also,a shift of the peaks occurs toward the trailing edge on the upper surface and toward the LE on the lower https://www.wendangku.net/doc/415462556.html,stly,an increase in blade velocity[Fig.2(c)]causes an increase in maximum E that is roughly proportional to V3∞.In all cases,however,it can be noticed that the highest values of E are on the blade upper surface.
The current parametric study shows that it is relevant to investigate the effects of the chordwise position of the aerodynamic transition imposed by the erosion on the airfoil aerodynamic performance.Therefore,the aerodynamic performance of the DU96-W-180airfoil was analyzed with XFOIL by parameterizing the chordwise position of upper (U)and lower(L)?xed transitions.The effect on C l/C d due to the chordwise transition location can be seen in Fig.3. Two curves are proposed,each obtained by modeling one surface with chordwise variable transition while imposing a ?xed transition point due to sand erosion on the opposite surface(x u tr=3.13%,and x l tr=8.39%).It can be seen that the location of aerodynamic transition has a detrimental effect on(C l/C d)max as soon as the transition point moves from its natural location toward the LE.The most signi?cant effect on the aerodynamic ef?ciency is due to the transition on the
Figure1.Erosion rate E curves on a DU96-W-180airfoil,α=6.0deg.Red segment:upper maximum erosion rate(E u max);cyan segment: lower maximum erosion rate(E l max);blue circles:particle impingement points.
(a)
(b)(c)
Figure2.Effect of(a)blade angle of attackα,(b)sand grain diameter,and(c)blade velocity V∞on sand erosion rate E on the DU96-W-180 airfoil.
Figure3.Effect on C l/C d of the chordwise location of?xed transition for the DU96-W-180.
upper surface where a decrease of C l/C d≈40%is associated with an early upper transition.This?gure is relevant in showing the magnitude of loss in GAEP of an eroded wind turbine.Similar values were obtained for the experimental investigation of damaged wind turbine airfoil performance carried out in previous studies by Sareen et al.44
B.Polars of a Blade Section Subject to Sand Erosion
The aerodynamic polars of the DU96-W-180airfoil(Fig.1)are computed by using XFOIL for clean surface condi-tions,upper only?xed transition(U)(set by erosion),lower only?xed transition(L)(set by erosion),and both upper and lower?xed transitions(U+L).All computations were carried out at Re=6.88×106as it represents a reasonable Reynolds number for a blade section located at r/R=0.75and c=1.7m.
Figure4shows the effects on the aerodynamic performance due to the various combinations of?xed transition. By assuming that the earliest damage in time would appear on the blade upper surface,an initial large increase in C d is observed with respect to the clean con?guration.Although smaller,a further increase in C d is observed once the blade lower surface damage also appears.Finally,a drag coef?cient approximately double the clean C d for U+L transitions is observed.A larger contribution to C d of the upper?xed transition may be explained by considering the pressure gradients acting on the upper surface and the longer curvilinear path covered by the boundary layer on the blade upper surface when compared with the lower surface.In fact,a thicker boundary layer would be observed on the upper surface when compared with the boundary layer on the lower surface.This would contribute to the overall form drag of the blade section which would increase more for the upper surface contribution than for the lower surface.
The aerodynamic ef?ciency C l/C d plot in Fig.4shows a similar effect to Fig.3.It can be seen here that the drop in ef?ciency from clean to?xed upper transition(U)is drastic and brings the blade section to a≈40%penalization in C l/C d atα=6deg.Once the lower blade transition is also?xed(U+L),a further decrease in aerodynamic ef?ciency is observed,bringing the blade section to approximately half the aerodynamic ef?ciency of the initial clean condition. Finally,by observing Fig.5an initial drop in C l can be seen for upper transition,followed by a modest recovery when U+L-transitions are set.Such behavior may be due to the over-prediction in lift coef?cient of XFOIL and should be veri?ed with higher-order methods such as CFD.It should be noted now that the chronological history of damage on the blade section is not crucial to the present analysis.In fact,identical considerations may be drawn when considering an early lower damage and a later upper damage.
C.The Effect of Blade Shape on Sand Erosion
The scenario depicted in Sections III.A and III.B does not include the relationship of the blade section shape to sand erosion rate E.In the current section,the role of different geometry features with respect to sand erosion is investigated.An ef?cient way to compare the results is to?x the particle size,the freestream conditions,and the airfoil operating point while varying the airfoil geometry.At this point,it is signi?cant to choose a design lift coef?cient C l rather than an angle of attack,since it gives a better indication of the blade operating conditions once the airfoil geometry is included in the blade design.
The analysis that has been carried out in previous studies22,25involved the Delft University wind turbine airfoil family applied to the damage problem.12,45,46Even though this choice of airfoils is currently representative of state-of-the-art wind turbines,it is somewhat limited when trying to investigate the geometric features that potentially minimize damage to blade sections.In fact,the NREL airfoil family offers a wider variety of upper and lower surface topology and curvature when compared with the DU family.47–49For such reasons,the NREL S family of airfoils is included here.Finally,because the most prominent effects of particle damage are seen when moving toward the blade tip, interest is aimed at optimizing outboard blade section geometries.A lift coef?cient of1.0is chosen and the results are presented as a comparison between two pairs of airfoils as shown in Figs.6,and7.The erosion peaks on the airfoil upper side are denoted by red marks;whereas,the cyan lines denote peaks of the lower side.It should be noted that more airfoils were analyzed by the authors,but only four are shown in this section for brevity.
A DU96-W-180and a DU96-W-212are compared in Fig.6since they are characterized by very similar upper geometries,whereas the maximum thickness is due to the different curvature of the lower surface.The upper E peaks are virtually identical in value,shape,and chordwise maximum location.On the other hand,the more bulbous front part of the DU96-W-212promotes an erosion peak on the lower side that is noticeably farther downstream,when compared with the thinner DU96-W-180airfoil.In other words,the fuller shape of the airfoil lower side allows for the erosion peak to move downstream while also becoming?atter.No appreciable effect is detected as far as the maximum value of E on the blade lower side.
Figure4.C l,C d,and C l/C d curves of the DU96-W-180for clean and damaged conditions.
Figure5.Drag polar of the DU96-W-180for clean and damaged conditions.
Figure6.Erosion rate comparison for a DU96-W-180and a DU96-W-212at C l=1.0,effect of the lower side geometry.Upper side erosion rate peaks in red;lower side erosion rate peaks in cyan.
Figure7.Erosion rate comparison for a NREL S817and a NREL S831at C l=1.0,effect of the upper side geometry.Upper side erosion rate peaks in red;lower side erosion rate peaks in cyan.
The situation is reversed when considering the airfoils NREL S817and S831in Fig.7.Such airfoils have similar lower side geometries in the forward section,whereas the upper side is more bulbous for the NREL S831.The lower erosion peaks are virtually located at the same chordwise location,although the shape of E appears to be?atter for the NREL S817airfoil.An important difference appears when considering the upper erosion peaks.Again,the bulkier geometry promotes an erosion peak farther downstream of the LE,when compared with the thinner section.Also,a slight drop in the maximum value of E is seen when considering the NREL S831.
D.Lessons Learned
The analysis of the previous sections highlights the role of the blade geometry with respect to erosion.It can be con-cluded that the chordwise location of maximum E is translated into appreciable effects on the aerodynamic behavior of the blade section once erosion is initiated.On the other hand,the maximum value of E is also driven by the blade section shape.In other words,the shape of the clean blade can dictate when and where the earliest damage would ap-pear on the surface;the blade shape also dictates the performance in rough conditions due to the chordwise transition location on the upper and lower surfaces.Such an observation is an important feedback on blade geometry subject to sand erosion with respect to minimizing the loss of aerodynamic performance once the airfoil is damaged.In general, the blade designer would favor geometries that promote the lowest erosion peaks located as downstream as possible from the LE thereby allowing for better aerodynamic performance in rough conditions.From the comparison of the airfoils it can be concluded that:
?An important geometric feature that pushes the maximum-E location downstream is the LE slope.In partic-ular,it appears that a more bulbous upper side allows for the maximum E to appear more downstream,when compared with thinner LE geometries.
?A similar observation could be made for the blade lower side.However,it was observed that the curvature of the blade may play an important role on both the shape of the erosion rate curve and the maximum location of E(see Fig.6).
?The sensitivity of maximum E to variations in geometry is smaller when compared with the sensitivity of the chordwise location of maximum E.However,even small changes in maximum E translate into longer blade life,extending the clean,smooth conditions in which the turbine operate,yielding relatively higher values of GAEP over time.
E.Figure of Merit
The considerations of Section III.D pose a new problem that is quantifying the?tness of an airfoil subject to sand erosion.Generally,it would be signi?cant to combine both the maximum value of E and the corresponding chordwise location to obtain a numerical insight on the airfoil erosion performance.In particular,a designer would want to penalize high values of erosion rate,while obtaining transition locations to be as downstream as possible on the blade surface.For these reasons,if such a designer would want to create the best erosion-resilient airfoil of those previously considered,he or she would need to combine the upper airfoil geometry of the NREL S831[Fig.7]with the lower geometry of the DU96-W-212[Fig.6].
The proposed mathematical expression to obtain the?tness of an airfoil subject to sand erosion is the following ?gure of merit(FOM),that is
FOM=
1
E u max
(1.1+x u tr)ωu+
1
E l max
1+x l tr
ωl
(17)
whereωu andωl are the weights assigned to the upper and lower surface,respectively,in order to penalize large values of the chordwise location of transition x tr.Such aspect is tied with the analysis of Section III.A and represents a powerful way to favor one side of the airfoil geometry with respect to the other.Because the upper side has shown a more important effect on the airfoil performance in damaged conditions compared with the lower side,the two chosen weights areωu=5andωl=1.5.
To show the value of using FOM,a comparison of airfoils is proposed,with?xed C l=1.0.The airfoils discussed in Section III.C,along with NREL S804,S832,S813,and S810,are ranked based on the erosion performance
Table1.Figure of Merit of Eight Airfoils Subject to Sand Erosion
Airfoil t/c(%)FOM Rank
DU96-W-18018.0146084
96-W-21221.2152371
NREL S81716.0134928
S83118.0145585
S80417.9149533
S83215.0139556
S81316.1139467
S81018.0150452
as summarized in Table1.The best performing airfoil is the DU96-W-212,characterized by t/c=21.2%and representing the largest thickness amongst the tested airfoils.In light of the conclusions of Section III.D,the fact that the thickest airfoil would outperform all the competitors is not surprising.However,when selecting the blade section geometries to be used for the design of a wind turbine blade,the designer?xes the value of t/c for structural purposes, hence reducing the pool of airfoils to choose from.
A?nal consideration should be done on the aerodynamic performance of the airfoils.A proper selection of the most suitable geometries needs to be coupled with an analysis of C l and aerodynamic ef?ciency C l/C d for a range of angles of attack.This procedure would avoid selecting damage-resilient geometries that however may penalize the ultimate goal of blade design being operated at high values of aerodynamic ef?ciency.
IV.Blade Section Shape Optimization
The objective of the blade section optimization is to maximize FOM.This is translated into a compromise between maximum erosion rate E max and chordwise location of?xed transition x tr.Modern optimization algorithms include gradient-based and genetic schemes.Gradient-based optimization schemes may be often challenging to implement as they require the exact mathematical model of the physics behind the problem.Because of the interplay between aerodynamics and particle dynamics,such approach is disregarded in favor of a genetic algorithm(GA)scheme.50 Based on the code developed for estimating wind turbine damage due to airborne particles(BugFoil),22,25a MAT-LAB wrapper has been implemented to perform GA optimization.Because the authors required complete freedom in the selection of the GA parameters,the wrapper has been written completely in-house,thus not using the available toolboxes of MATLAB.
Two main steps were involved in the optimization method implemented in this paper,with incremental degrees of complexity.The early phase is based on perturbations of existing airfoil geometries since it was expected to provide insight into the optimum airfoil geometries while being simple to implement.Such an approach can be regarded as a direct airfoil design GA optimization,aimed at maximizing FOM.In this initial approach the aerodynamic perfor-mance of the airfoils are not considered,as the scope is to obtain a?rst indication of the optimal geometries.The second part of the study deals with the more general and?exible theory of inverse airfoil design51,52coupled with genetic optimization.53In this phase,the aerodynamic performance of the airfoils are considered,hence resulting in a multi objective optimization.In particular,the objectives of the optimization are(C l/C d)max and FOM.
A.Genetic Optimization Through Geometry Perturbations
The?rst approach implemented for optimization makes use of random tournament selections to march through gener-ations.The basic scheme of the genetic optimization through geometry perturbations follows these steps:
1.A pool of suitable airfoils is selected from the existing literature,
2.The geometric properties of two airfoils are combined in a random fashion to obtain a new generation(crossover),
3.Random perturbations on the airfoil properties may be inserted to increase species diversity(mutation),
4.The constraints are imposed:typically C l orα,and(t/c)max,
5.The particle code BugFoil is executed(evaluation),
6.FOM is computed and data is written into a log?le,
7.The best n specimens are selected for crossover for the next step(selection),
8.The algorithm proceeds back to step2)until the number of maximum generations is reached or convergence is
obtained.
Several questions arise on how to optimize a blade section shape.In particular,a designer may be interested in using existing geometries and modifying only a few aspects of those,for example the leading edge curves,the airfoil camber,or the airfoil thickness.Since the present study represents a novelty in wind turbine airfoil optimization, multiple approaches to the optimization process were investigated.
1.Optimization of the Leading Edge
The particle analysis carried out in previous studies highlighted the LE as the primary location of particle impact.22,25 For such reasons,an initial investigation of solely the LE geometry for a given airfoil is performed.The idea here is to perform subsequent crossovers of an airfoil with a mutated version of itself,for a given number of generations.
However,a practical issue involving the LE optimization is?nding suitable families of geometric curves to generate smooth,naturally-looking airfoils.In fact,even if the choice of algebraic n-degree polynomials may seem natural,it has several issues as far as continuity in the?rst and second derivative(C1and C2),once the modi?ed branch has to merge with the pre-existing part of the geometry.For these reasons,the Bezier polynomials were chosen to optimize the LE geometry.Such polynomials allow to?x the start and end points of a curve,while varying the curve path through the position of intermediate control points.53–56The number of Bezier control points dictates the degree of the Bezier polynomial.It should be noted that the higher the degree of the Bezier polynomial,the higher the number of convexities and concavities that a curve may have.Such considerations reduce the practical number of control points considerably.From a practical standpoint,once the LE curves of a given airfoil are obtained,a complete replacement of such a geometry may be performed while still maintaining continuity and shape regularity with the existing curves of the airfoil.
To obtain shape continuity and smoothness,two control points are?xed at the start and end points of the upper and lower LE geometries(P1and P5in Fig.8),while two control points are placed strategically next to them,and named P2and P4.P2and P4control the second derivative of the curve in the region where the modi?ed LE has to merge with the existing geometry.To investigate the optimal LE shape with the GA scheme,the coordinates of the last control point P3are randomly placed by the algorithm on a mesh of possible positions.The mesh is de?ned as a grid of coordinates within the most forward LE point,and the location of maximum thickness of the airfoil.By operating in such a way,the regularity of the LE curve is mantained to a reasonable extent even for extreme positions of the random control point P3.Moreover,it avoids the possibility of concave curves or very?at leading edges.Figure9 shows the results in applying such an approach to the upper and lower curves of the DU96-W-180LE.As it can be
Figure8.Modi?cation of the LE geometry.Black solid line:original geometry;red line:modi?ed geometry;red circles:Bezier control points;black crosses:possible positions of the random control point.
Figure9.Bezier curves applied on the LE geometry of a DU96-W-180airfoil.Red circles:upper Bezier control points;cyan circles:lower Bezier control points.
seen in the lower part of the?gure,a natural-looking airfoil is obtained for random positions of the upper and lower P3points.
2.Results
Three airfoils are considered:the DU96-W-180,the DU96-W-212,and the DU96-W-250by moving from the blade outboard region toward the hub.The three airfoils differ in thickness(t/c).In fact,the outboard airfoil has a18% thickness,followed by21.2%,and?nally a25%thick airfoil is used.Each optimization is run by iteratively perform-ing a crossover of the airfoil with a mutated version of itself followed by the computation of FOM.The optimization is stopped once the maximum number of generations is reached,which in this case is100.
The selected airfoils are optimized by mantaining the same in?ow velocity and chord length,so that the results between airfoils are comparable,and the effects of the in?ow velocity on the erosion rate E are the same.A37-m wind turbine blade,with a tip-speed ratio of8.7,and operating at23rpm is considered.The diameter of the sand grain used for the optimization is d=200μm,the in?ow velocity is V∞=69.4m/s,the airfoil chord length is c=1.7m,and
Figure10.LE optimization:history of FOM for the three airfoils.
(a)
(b)
(c)
Figure11.Optimized LE of(a)DU96-W-180,(b)DU96-W-212,and(c)DU96-W-250;sand grain diameter d=200μm,C l=1.
the airfoils are tested at C l=1.By doing so,the airfoils operate with the same amount of circulation per unit span, hence the aerodynamic effects on the particle trajectories are comparable away from the surface.All simulations were performed with BugFoil22,25nested in a newly-written MATLAB GA routine.The evaluation of a single individual requires≈6sec computational time on a8Gb RAM Core i7Linux machine running LinuxMint OS.On average,the runtime for each optimization was3hr.
Figure10shows the history of FOM with respect to the number of generations for the three airfoils.It can be observed that convergence is reached with ample margin on the maximum number of generations.The optimized leading edges are presented in Fig.11for the three airfoils.Despite the different thickness,the?rst common charac-teristic to all three airfoils is related to the curvature of the upper side of the LE.The optimized upper side appears more bulbous,especially for the DU96-W-180[Fig.11(a)]and for the DU96-W-180[Fig.11(c)]as compared with the original geometries.The second common characteristic is related to the?at,slanted lower side of the LE,visible in all three airfoils.
A comparison of FOM is reported in Table2.The airfoil with the highest FOM is the DU96-W-250,which is also the thickest airfoil of the three.Again,a thick airfoil allows the sand particles to slow down more,and offering near-normal particle impacts on the lower surface,thus mitigating the erosion effects when compared with thinner airfoils,as mentioned in Section III.E.Also,it can be seen that a consistent increase in FOM exists for the three airfoils when compared to their baseline.
The results for the LE optimization show that the observations of Section III.C are repeated.In particular,a high FOM can be obtained by allowing the sand particles to hit at a steep angle on the lower side,while offering moderate angles of impact on the upper side.Moreover,the lower side of the optimized airfoil,characterized by a?at,slanted surface may allow for a stronger decrease in particle velocity compared with more curved surfaces.
Figure12.An example of the crossover operator:weighted average of a NREL S804and DU96-W-212airfoils.
3.Global Airfoil Optimization
The optimization of the LE geometry only is restricting and prevents the analysis of larger design spaces.In this sec-tion,the optimization is performed by a crossover operator within a selected pool of airfoils with comparable thickness. The idea is to blend the airfoils by computing the random weighted average of the upper and lower geometries of the parent airfoils.Figure12shows the result of operating crossover on two fairly different geometries,the NREL S804 and the DU96-W-212geometry.The weighted average crossover allows for the child geometry to lie in between the geometry of the parents,while considerably expanding the design space of the geometries with respect to the LE optimization.Such approach makes it possible to only consider feasible geometries,enhancing the effectiveness of the optimization process by reducing the possibility of faulty airfoils.
A starting pool of six airfoils is selected with t/c≈18%.The pool is composed by the following airfoils:DU96-W-180,NREL S804,S810,S813,S817,and S831(see Table1).Two different optimization approaches are pursued. The simplest involves the crossover of unmutated parents only.Unmutated refers to geometries that are a linear combi-nation of previous geometries(pure crossover).By doing so,an insight is expected regarding the?ttest shapes for the upper and lower airfoil surfaces.The second approach involves also LE edge perturbations(crossover and mutation) in the form of Bezier curves,as explained in Section IV.A.1.The second approach is expected to achieve better FOM values and give a clearer view of the optimal airfoil shape.
4.Results
Figure13shows the optimization history of FOM and geometry without LE perturbations[Figs.13(a)and13(b)], and with LE perturbations[Figs.13(c)and13(d)].The convergence for the optimization with no LE perturbation [Fig.13(a)]occurs at a faster rate than the convergence for the LE perturbations[Figs.13(c)].The presence of LE perturbations also allows for prominent jumps in the FOM curve.
Because the starting pool of airfoils is unchanged,a striking similarity between the two histories of geometry can be seen[Figs.13(b)and13(d)].Due to its simplicity,the approach with no LE perturbations[Fig.13(b)]shows fewer optimal shapes through the evolution when compared with the LE perturbation approach[Fig.13(d)].In fact,by focusing on the LE region,a more crowded scenario can be observed when the LE perturbations are used[Fig.13(d)].
A comparison of the optimal geometries can be seen in Fig.14.The LE appears more skewed on the upper side when the LE perturbations are used,as opposed to the fuller and more rounded LE of the other case.Such a shape, which can be thought of as a knee shape,allows for the peak of erosion rate E to move considerably downstream on the upper surface.Due to this fact,FOM can achieve considerably larger values when compared with the no LE perturbations approach as shown in Table3.However,from an aerodynamic point of view,a knee shape is responsible Table2.Figure of Merit of the DU Family Airfoils Before and After Leading Edge Optimization
Airfoil FOM original FOM optimized%?FOM
DU96-W-18014,60819,79935.5
DU96-W-21215,23720,38033.7
DU96-W-25015,95221,02431.8
(a)(b)
(c)(d)
Figure13.Global airfoil optimization:(a)and(c)best FOM,(b)and(d)best airfoils at each selection,using no LE perturbations,and using LE perturbations respectively;red airfoils:best individuals;t/c=18%,sand grain diameter d=200μm,C l=1.
for strong and localized LE suction peaks.An unfavorable boundary layer growth is expected in that region with possible early separation and a signi?cant drop in aerodynamic performance(increase in C d and decrease in C l).
An important piece of information derived from Figs.13and14relates to the airfoil camber.Since the current approach is based on airfoil blending,the optimal shapes are free to vary in camber through the GA optimization method.Both optimal airfoils have a pronounced camber with respect to the optimal airfoils of Section IV.A.2.As the GA optimization progresses,the increase in camber is due to an upward shift of both the lower and upper surfaces of the airfoils[Figs.13(b)and13(d)].It can be concluded that cambered airfoils allow for weaker suction peaks on the upper side for a given lift coef?cient C l when compared with non-cambered airfoils operating at the same C l.Such feature allows for the particles to impact the upper surface at a reduced velocity,hence promoting smaller values of E. On the lower side,a?at surface allows for the particle to slow down more,hence reducing the erosive effect.
The optimization of wind turbine airfoils through geometry perturbations shows its?aws due to the limited design space and to the aerodynamic performance tied to that.In other words,since the design space is small,also is the capability of investigating good airfoil geometries.Finally,it has to be noted that by selecting different pools of airfoils,the algorithm would have different outcomes of the optimal airfoil.
https://www.wendangku.net/doc/415462556.html,parison of global airfoil optimization.
Table3.Figures of Merit of the Global Optimization and Comparison with State-of-the-Art
Airfoil FOM%?FOM
DU96-W-18014608-
no LE perturbations1984035.8
LE perturbations2096543.5
B.Genetic Optimization Through Airfoil Inverse Design
However complex,an optimization that makes use of pre-existing geometries is intrinsically limited.In the current section,the optimization is performed using airfoil inverse design so that this issue is overcome.The algorithm with such a capability is PROFOIL,51,52which is included in the GA wrapper and called at each airfoil generation in or-der to obtain the geometry of the new individual.PROFOIL makes use of a conformal mapping technique to design airfoils.In particular,it allows the user to prescribe the velocity distribution over an airfoil by means of theα?-φdistribution.A detailed explanation of the inverse design method is beyond the scope of this paper,and the authors suggest referring to the published literature for further details.51,52
In order to assess the aerodynamic properties of the airfoil,the aerodynamic polars are included in the current ap-proach.By doing so,the optimization allows to maximize FOM without penalizing(C l/C d)max,effectively becoming a two-objective optimization.The classic approach for these type of problems is by using the Pareto front criteria.50 The aerodynamic polars are evaluated by also including XFOIL in the GA wrapper.34
Since a few steps in the GA scheme are now different from the direct airfoil design,the new steps of the code can be summarized as follows:
1.A population of randomly generated airfoils is initalized.In particular,PROFOIL is initialized with random
values ofα?,whileφare kept constant,
2.α?of two airfoils(parents)are combined in a random fashion to obtain a new generation(crossover),
3.Random perturbations on the airfoil properties may be included randomly to increase specie diversity,in the
form of?α?(mutation),
4.The constraints are imposed:typically C l orα,and(t/c)max,
5.The aerodynamic polar of the airfoil is computed through XFOIL.The value of(C l/C d)max is computed(evalu-
ation),
6.The particle code BugFoil is executed and FOM is also computed(evaluation),
7.FOM and(C l/C d)max are written into a log?le,
8.The Pareto front is computed,and the parents for the following generation are selected amongst the individuals
lying on the Pareto front,
9.The algorithm proceeds back to step2)until the number of maximum generations is reached or convergence is
obtained.
Table4.Values of theα?-φDistribution Used for the Airfoil Inverse Design
Pairα?(deg)φ(deg)
110.015.5
2unconstrained19.5
3unconstrained25.5
4unconstrained32.2
5unconstrained45.5
6 4.060.0
It can be immediately seen that the computational time of this approach is increased due to the aerodynamic compu-tations of XFOIL.In fact,the evaluation of C l and C d with reasonable increments in angle of attack(?α=0.3deg) increases considerably the runtime of the code.Also,it should be noted that on modern machines PROFOIL can be executed in less than1/10of a second,thus not hurting the overall computational time.In general,an average of10 seconds were required to evaluate each airfoil.
1.Results
An extensive test of PROFOIL was performed to decide the parameters to be used for the GA scheme.By doing so, the goal was to reduce the computational time of the optimization and avoid expanding the design space to unfeasible airfoil geometries or to airfoils with limited aerodynamic performance.Also,because the most prominent effects of particle erosion are seen toward the blade tip,the airfoil t/c and moment coef?cient C m were set equal to those of an outboard-region airfoil–the DU96-W-180(t/c=18%,C m=-0.063).
Particular care was taken to select the number of pairs in theα?-φdistribution.It was observed that increasing the number of such pairs increased the computational time noticeably.Moreover,a large number ofα?-φwould only allow for modest improvements in(C l/C d)max.For these reasons,the number of pairs was set to six,andα?were set unconstrained on the interval(-1–13)deg.Table4shows the parameters used for the inverse design GA.Since the approach used in this phase encompasses the widest of the design spaces investigated thus far,the number of individ-uals for each generation was raised signi?cantly,and the maximum number of iterations was raised to120.At the end of each optimization an average of5,000individuals were sampled,yielding a total computational time that varied between12and20hr.
The?rst inverse design optimization is performed at the same conditions as for the direct airfoil design(d=200μm, V∞=69.4m/s,c=1.7m,and C l=1),equivalent to Re=6.88×106for a blade section located at r/R=0.75along the bladespan.The results of the optimization are reported in Fig.15.The?rst information that appears is related to the high values of(C l/C d)max as shown in Fig.15(a).A steady(C l/C d)max=324.62is reached as early as31generations, while the FOM plateaus earlier[Fig.15(b)].However,the most relevant information comes from analyzing the history of the best geometries shown in Fig.15(c).Similar to the direct airfoil design optimization,a progressive increase in airfoil camber is also observed here.In fact,the lower surface reaches an almost?at con?guration,followed by a slightly cambered shape.The upper surface moves upward accordingly,and the upper portion of the leading edge appears bulbous.These observations are in good agreement with the earlier phases of the study.The authors also found in the recent published literature a similar optimization study by He et al.57The shape optimization of a wind turbine airfoil to maximize(C l/C d)max was investigated,and similar trends in the displacement of upper and lower surfaces were observed.The reason behind such high values of(C l/C d)max observed in the current study has to be found in the recovery region behind the location of maximum thickness on the airfoil upper surface.This portion of the airfoil acts as a Stratford recovery region which represents a technique used for high lift airfoil design.58 As most of blade erosion occurs toward the blade tip,a second optimization was performed for a blade section located in the tip region at r/R=0.95.At such location,the wind turbine blade has a small chord(c=1.0m),while the local velocity is increased to V∞=84.9m/s.The resulting Reynolds number is now Re=5.84×106,but the relevant information comes from the nondimensional mass AK of the sand particle.In fact,by keeping the diameter to d=200μm,AK is now greater than for a blade section located at r/R=0.75(see Eq.13).In other words,a short blade section has less aerodynamic in?uence on a particle than a large blade section.Therefore,the optimization is expected to show a smaller in?uence of the blade shape with respect to erosion.
The optimization of the blade section located at r/R=0.95is shown in Fig.16.It can be seen that both(C l/C d)max
(b)
(a)
(c)
Figure15.History of inverse airfoil optimization:(a)(C l/C d)max,(b)best FOM,and(c)best airfoils at each selection;r/R=0.75,t/c=18%, C m=-0.063,sand grain diameter d=200μm,C l=1.
[Fig.16(a)]and FOM[Fig.16(b)]reach a plateau later than for the r/R=0.75optimization[see Figs15(a)and15(b)]. Also,both(C l/C d)max and best FOM are smaller than for the airfoil located at r/R=0.75.Finally,the geometry his-tory shown in Fig.16(c)shows smaller variations in the best airfoils as compared with the more inboard section [Fig.15(c)].This result can be seen as a smaller in?uence of the airfoil on the particle trajectory;therefore,the best (C l/C d)max plateaus at311.03.It should be noted that due to the high particle impact velocity at r/R=0.95,FOM is also lower,since it is a function of1/E max.
A comparison of both(C l/C d)max and FOM with the DU96-W-180airfoil is presented in Table5.The two airfoils designed with inverse design process outperform the DU96-W-180airfoil both in FOM and in(C l/C d)max.However, such a margin becomes smaller as the considered blade section is moved outboard.Again,this is due to the reduced effect of the?ow?eld on the particle at such locations.
V.Conclusions
This paper presents the optimization of wind turbine airfoils subject to sand erosion.The effect of sand impingement on the blade surface is translated into two peaks of maximum erosion rate E located at a distance from the LE.Both erosion peaks are subject to shifts due to angle of attack,particle diameter,and in?ow velocity.Moreover,it was seen that the aerodynamic performance of a damaged airfoil due to sand erosion depends on the location of the damage along the chord length.
The role of different airfoil shapes with respect to erosion rate was also investigated.It was found that a bulbous upper leading edge promotes the erosion rate peak to move downstream;on the other hand,a?at slanted LE lower surface allows for the particles to slow down more and hit the surface at near-normal angles.The erosion rate of plastic
(a)
(b)
(c)
Figure16.History of inverse airfoil optimization:(a)(C l/C d)max,(b)best FOM,and(c)best airfoils at each selection;r/R=0.95,t/c=18%, C m=-0.063,sand grain diameter d=200μm,C l=1.
https://www.wendangku.net/doc/415462556.html,parison of FOM and(C l/C d)max with State-of-the-Art Airfoils.
Airfoil r/R FOM%?FOM(C l/C d)max%?(C
l/C d)max
DU96-W-1800.7514608-174.18-
DU96-W-1800.9510815-172.14-
Inverse Design0.752118044.9324.6286.4
Inverse Design0.951242214.9311.0380.7