Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
Energy
journal homepage: www.elsevier.com/locate/energy
a r t i c l e i n f o a b s t r a c t
Article history: In our previous study, it has been shown that the use of protuberances covering the whole leading edges
Received 10 May 2016 of B-blades will enhance the performance of horizontal axis wind turbine (HAWT) under low speed
Received in revised form rotations. However, a great loss of performance shall occur after a critical tip speed ratio is reached. The
30 August 2016
research aims to optimize the protuberance design of HAWT blades for the system to gain optimal
Accepted 16 September 2016
performance under low speed rotations, whereas no significant drop of performance happens at high tip
speed ratios. The optimization study was performed not only for baseline blades but also for those
designed by the blade element momentum theory. For verifications, wind tunnel experiments were
Keywords:
Small variable-speed HAWT
carried out to provide comparative data; additionally, numerical simulations were also conducted to
Optimization of leading-edge protuberances further visualize the flow fields. The results show that the optimal leading-edge protuberances would
significantly increase the HAWT performance under low speed rotations, yet no serious performance
payback occurs at high tip speed ratios.
© 2016 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.energy.2016.09.100
0360-5442/© 2016 Elsevier Ltd. All rights reserved.
C.-C. Huang et al. / Energy 115 (2016) 1156e1167 1157
c=6cm
La
R=0.35m
b
Lw
Lp
r
8cm
Rated power Watt 200 Experiments were performed in an open-type wind tunnel with
Rated wind speed m/s 12 a test section dimensioned by 914 mm (height) 4440 mm
Design tip speed ratio 4.5 (length) 1220 mm (width). For the present case, the area ratio,
Number of blades 3 defined by the ratio of the swept area of turbine blade to the cross
Design angle of attack degree 6
section area of wind tunnel, is about 34%. According to [13], no
C.-C. Huang et al. / Energy 115 (2016) 1156e1167 1159
Fig. 3. Experimental setups for testing the HAWT system in wind tunnel.
a ¼ 6
1 14500 0.721 0.0305 0.830 0.0192 3.2. Numerical simulations
2 28500 0.783 0.0266
3 40600 0.802 0.0238 The 2D/3D flow fields were simulated by solving the steady-
4 58300 0.806 0.0238 state incompressible Navier-Stokes equations coupled with the
5 72000 0.803 0.0236
Shear-Stress Transport (SST) k-u turbulence model for all cases.
Case CFD simulations Experiment [14] As reported by Krogstad and Lund [14], the best yþ value for the
Grid size Cl Cd Cl Cd turbulence model can be found by 2D simulations in order to
a ¼ 10 reduce the number of grid points, yet the yþ value is still applicable
1 14500 0.947 0.0503 1.100 0.0460 to 3D cases. This approach was also applied to our simulations. At
2 28500 0.988 0.0451 beginning, the yþ value around the 2D airfoil and 3D turbine blade
3 40600 1.069 0.0423 was targeted on unity for generating the boundary layer. However,
4 58300 1.072 0.0423
the yþ value did not maintain at this value in the subsequent
5 72000 1.070 0.0422
analysis, where a distribution of 0.8e3 was found due to the highly
curved surface. At any rate for yþ smaller than 3, the simulation is
valid for using the SST model. Fig. 4 shows the 2D grid system of the
1160 C.-C. Huang et al. / Energy 115 (2016) 1156e1167
Table 4
Control factors and the level values.
Fig. 10. Power coefficient of the full protuberance model with la ¼ 0.04, lw ¼ 0.145. In order to verify our CFD simulations, analysis of the airfoil for
its lift and drag coefficients were conducted first. Next, the power
coefficients of B-blades were also computed using the CFD simu-
lations from the boundary layer construction of airfoil. All results
were compared with experimental data.
Shown in Fig. 6 are the computed lift and drag coefficients,
plotted as functions of a for Re ¼ 1 105.
Also, all experimental data recorded from the wind tunnel ex-
periments are shown in Fig. 6 for comparisons. Experimental data
of the Cp distributions as a function of various tip speed ratios at
V ¼ 10 m/s were determined by t-distribution, while the experi-
mental errors were calculated within 5%. Clearly, it can be seen that
the simulations agree pretty well with the experimental data for
a ¼ 6 ~10 . Another interesting phenomenon to note is the
nonlinearity of Cl for a ¼ 1 e2 ; in addition, the simulations have
successfully captured the stall angle. In the stall region a ¼ 11 e13 ,
the simulations tend to yield smaller values as compared with the
experimental data; nevertheless, the trend is still consistent. Also,
the slight off-accuracy of Cd for a ¼ 3 e9 is observed, where its
maximum error is about 12%. As a whole in general, all simulated
Fig. 11. Flowchart of the analysis by Taguchi Method. data are in satisfactory agreements with the experimental data.
1162 C.-C. Huang et al. / Energy 115 (2016) 1156e1167
Table 5
Analysis for power coefficient by Taguchi Method.
A B C 1.833 2.199 2.566 2.932 3.299 3.665 4.032 4.398 4.765 5.131
1 1 1 1 0.010 0.019 0.042 0.066 0.100 0.150 0.208 0.259 0.289 0.304 0.145 1444.634 31.598
2 1 2 2 0.011 0.020 0.040 0.070 0.113 0.166 0.228 0.278 0.302 0.299 0.153 1197.348 30.782
3 1 3 3 0.017 0.034 0.064 0.108 0.150 0.200 0.252 0.285 0.294 0.274 0.168 462.091 26.647
4 1 4 4 0.027 0.048 0.079 0.119 0.160 0.207 0.228 0.263 0.269 0.265 0.167 217.048 23.366
5 1 5 5 0.028 0.045 0.073 0.102 0.133 0.159 0.189 0.197 0.192 0.204 0.132 225.111 23.524
6 2 1 2 0.011 0.022 0.034 0.059 0.104 0.161 0.214 0.264 0.337 0.297 0.150 1157.390 30.635
7 2 2 3 0.017 0.033 0.069 0.096 0.145 0.192 0.249 0.273 0.271 0.260 0.160 482.800 26.838
8 2 3 4 0.030 0.053 0.083 0.123 0.163 0.199 0.223 0.241 0.247 0.238 0.160 181.959 22.600
9 2 4 5 0.031 0.045 0.077 0.106 0.139 0.166 0.187 0.191 0.194 0.192 0.133 197.988 22.966
10 2 5 1 0.010 0.020 0.042 0.074 0.112 0.157 0.214 0.268 0.294 0.297 0.149 1289.019 31.103
11 3 1 3 0.016 0.028 0.048 0.094 0.143 0.156 0.236 0.271 0.272 0.272 0.154 581.900 27.648
12 3 2 4 0.031 0.055 0.086 0.127 0.163 0.204 0.236 0.269 0.273 0.239 0.168 171.043 ¡22.331
13 3 3 5 0.031 0.058 0.084 0.112 0.144 0.182 0.199 0.203 0.216 0.189 0.142 172.767 22.375
14 3 4 1 0.013 0.019 0.041 0.072 0.109 0.158 0.212 0.265 0.294 0.295 0.148 938.873 29.726
15 3 5 2 0.013 0.025 0.050 0.086 0.127 0.179 0.226 0.275 0.284 0.277 0.154 794.154 28.999
16 4 1 4 0.022 0.043 0.075 0.113 0.159 0.188 0.196 0.198 0.209 0.222 0.143 293.235 24.672
17 4 2 5 0.027 0.048 0.077 0.111 0.136 0.178 0.175 0.181 0.189 0.178 0.130 225.914 23.539
18 4 3 1 0.010 0.023 0.042 0.071 0.107 0.147 0.214 0.261 0.296 0.297 0.147 1341.430 31.276
19 4 4 2 0.016 0.030 0.053 0.081 0.118 0.165 0.216 0.267 0.285 0.274 0.150 563.014 27.505
20 4 5 3 0.020 0.037 0.063 0.091 0.127 0.171 0.227 0.251 0.254 0.237 0.148 370.012 25.682
21 5 1 5 0.021 0.034 0.063 0.099 0.124 0.156 0.179 0.192 0.186 0.189 0.124 376.555 25.758
22 5 2 1 0.014 0.025 0.047 0.071 0.112 0.167 0.192 0.226 0.241 0.237 0.133 750.995 28.756
23 5 3 2 0.018 0.030 0.054 0.082 0.119 0.165 0.188 0.227 0.236 0.226 0.134 502.473 27.011
24 5 4 3 0.024 0.044 0.072 0.105 0.120 0.174 0.193 0.220 0.220 0.233 0.140 279.739 24.468
25 5 5 4 0.023 0.038 0.057 0.085 0.116 0.148 0.166 0.181 0.181 0.192 0.119 324.763 25.116
Ave. 0.020 0.035 0.061 0.093 0.130 0.172 0.210 0.240 0.253 0.247
Note: Values in boldface correspond to the optimal case with the best power coefficient.
Fig. 12. The S/N ratio for each set of simulated torque.
Since the major contribution to Cp is made by Cl, the slight off- that all simulated values are in excellent agreements with the
accuracy of Cd will not significantly affect our later simulations experimental data for u ¼ 500 rpme1400 rpm (l ¼ 1.8e5). Despite
for 3D turbine blades. the agreements of results, one might still hold a conservative
Fig. 7 plots the resulting power coefficients of manner about the modeling of B-blades with LE-prots.
Model_1 at V ¼ 10 m/s. Once again, to our satisfactions, it turns out To be more confirmative, CFD simulations were also conducted
for the other two B-blades (Fig. 8) studied by Huang et al. [10].
Likewise, both simulations and experiments were carried out.
Table 6
Shown in Fig. 9 are plots of Cp for the protuberant model with
Results of the variance analysis of Cp. la ¼ 0.04, lw ¼ 0.404.
Corresponding Cp data of the other model with la ¼ 0.04,
Factor SS DOF Var F Confidence Significant
lw ¼ 0.145 are plotted in Fig. 10. As can be observed from these plots,
A 3.44 4 0.86 No our CFD simulations are shown to yield very reliable results as
B 18.06 4 4.51 5.26 93.15% Yes
compared with experiments. With confirmation of our CFD simu-
C 193.32 4 48.33 56.27 99.91% Yes
error 3.44 4 0.86 lations, the study of optimization can be continued, and this pro-
Total 214.81 12 53.70 cess will be elaborated next. Before details of the optimization are
C.-C. Huang et al. / Energy 115 (2016) 1156e1167 1163
Table 7
Comparison of Cp for Model_1~Model_3.
1.833 2.199 2.566 2.932 3.299 3.665 4.032 4.398 4.765 5.131
Model_1 0.010 0.019 0.036 0.067 0.106 0.159 0.221 0.268 0.292 0.314 1395.21 31.446
Model_2 0.031 0.055 0.086 0.127 0.163 0.204 0.236 0.269 0.273 0.239 171.043 22.331
Model_3 0.033 0.056 0.088 0.131 0.172 0.204 0.239 0.265 0.267 0.268 155.346 21.913
Table 8
Comparison of Cp for BEMT-blades.
1.833 2.199 2.566 2.932 3.299 3.665 4.032 4.398 4.765 5.131
Model_4 0.125 0.169 0.220 0.294 0.388 0.422 0.439 0.443 0.455 0.455
Model_5 0.134 0.185 0.244 0.303 0.362 0.418 0.415 0.413 0.400 0.388
4.3. Analysis of variance and confirmation run analysis has applied F-test with F defined by
In Eq. (5), Varp is the factor variance and Vare the error variance.
From its definition, the factor variance is given by DOFP ¼ LP 1; (7b)
parameters of the LE-prots for an optimum design of protuberant B- improve the power coefficient for l 4.032 (rotation speeds under
blades. The twenty-five sets of power coefficients obtained for 1100 rpm); however, the optimal model does not gain performance
Model_1 ~ Model_3 are tabulated in Table 5, where the tip speed at higher rotational speeds.
ratios are for u ¼ 500 rpme1400 rpm with an increment of
100 rpm. Also displayed in the table are the corresponding signal
noises of all cases. From the outputs, it is clear that Case 12 is the 6. Comparisons of Cp with experimental data
best among all, having the maximum Cp ¼ 0.273 at l ¼ 4.765. Using
the characteristic of “the-bigger-the-better”, one may sort out the 6.1. Baseline models
optimal parameters.
As a result of substituting all data in Table 5 into Eq. (3), one may For verifications of our CFD simulations, Model_1~Model_3
obtain the S/N ratios of power coefficients, plotted in Fig. 12. From were also experimented in the wind tunnel to provide comparison
this S/N ratio graph, one may clearly see the dominant order of the data. Fig. 13 plots the experimental data of power coefficients along
three, given by lp, lw, and la. Hence, as a result of picking the with the simulation results. Indeed, the simulations turn out to be
maximum S/N ratio of levels for each control factor, the optimal in excellent agreements with all measured data. Interestingly, one
parameters are determined to be lp ¼ 62.5%, lw ¼ 1/16, and may spot the critical tip speed ratio (l ¼ 4.398), after which the
la ¼ 0.166, which defines the B-blades with optimized pro- power coefficient starts to drop for Model_2 and Model_3.
tuberances (Model_3 in Fig. 1) at different wind speeds. To ensure To have an insight into this phenomenon, the Cp calculated for
the accuracy of our Taguchi analysis, its variance analysis and cases with the parameters (la ¼ 0.166c, lw ¼ R/16, and lp ¼ 25%, 50%,
confirmation run were also performed. The results are demon- 75%, and 100%) are plotted as a function of l in Fig. 14. As can be
strated next. observed from the comparison shown in this figure, although LE-
As a post-process, variance analysis of the data obtained pre- prots may greatly enhance Cp at low tip speed ratios, the perfor-
viously was carried out to calculate values of SS and Var using Eqs. mance will drop fast at high speed ratios when full protuberance
(6) and 7(a)e7(b). First, the analysis was performed to get the (lp ¼ 100%) is applied. This is mainly due to the disturbance of flow
field on most blade surfaces, where no stall occurs at high rotation
C.-C. Huang et al. / Energy 115 (2016) 1156e1167 1167
speeds. However, under low speed rotations, stall shall take place protuberances with optimized parameters applied to only a part of
near the root and thus, LE-prots may help increase the perfor- the leading edges. Before the optimization analysis, CFD simula-
mance. At this point, it is learned that more favorable use of LE- tions were first performed for different combinations of parameters
prots is the case when proper lp is employed. to obtain a database of Cp values calculated for different l. Using
For visualizing the flow difference between Model_1 and these data, Taguchi Method was applied to sort out the optimal
Model_3, their surface streamlines are plotted in Fig. 15 for the parameters of LE-prots with its variance analysis performed. As a
sampling tip speed ratios 1.833, 3.55, and 5.236. It can be seen that result, it turns out that span-percentage is the dominant factor for
the most obvious difference of flows between Model_1 and improving the performance and the wave-amplitude has the least
Model_3 is the presence of vortices accumulated at the protuber- effect. Sequentially, the optimal parameters were determined to be
ance regions. For this, additional pressure drop on the upper blade lp ¼ 62.5%, lw ¼ 1/16, and la ¼ 0.166. Both simulations and experi-
surface occurs, resulting in extra torque. When the speed ratio is ments were conducted. To our satisfactions, the simulations were
increased, stall area will be reduced. At this moment, accompanied in good agreements with all measured data. Indeed, the present
with the vortices is the presence of unsteady flows leading to work has provided an optimal design of sinusoidal LE-prots for
reduction of Cp values. small variable-speed HAWT turbine blades. By this optimal design,
the stall phenomenon at low tip speed ratios can be efficiently
6.2. BEMT models improved, whereas no serious performance loss is present at high
speed ratios.
From the previous study for B-blades, it has been learned that
the factor la has the least effect on the overall performance. Thus, Acknowledgements
one may make a plausible assumption that similar performance
gain can also be obtained for the BEMT-blades when the optimal The authors gratefully acknowledge the financial support from
parameters of LE-prots are applied. For verifying this, both BEMT the Ministry of Science and Technology, Taiwan (No. 105-2221-E-
models, i.e. Model_4 and Model_5, were investigated for their Cp 006-113).
values under different tip speed ratios at V ¼ 10 m/s by simulations,
yielding results listed in Table 8. References
The simulations also show that the optimized LE-prots may
enhance the power coefficient at low tip speed ratios (l 2.932), [1] Miklosovic DS, Murray MM, Howle LE, Fish FE. Leading-edge tubercles delay
stall on humpback whale (Megaptera novaeangliae) flippers. Phys Fluids
whereas performance drops shall occur for high l. Fig. 16 shows the 2004;16(5):L39e42.
surface streamlines on the both models. [2] Fish FE, Weber PW, Murray MM, Howlw LE. The tubercles on humpback
For verifications of our simulations, wind tunnel tests were also whales' flippers: application of bio-inspired technology. Integr Comp Biol
2011;51(1):203e13.
carried out for the both models (Fig. 17) to collect their experi- [3] Fish FE, Battle JM. Hydrodynamic design of the humpback whale flipper.
mental data of Cp. Comparisons between the simulations and all J Morphol 1995;225:51e60.
measured data for the Cp values of Model_4 and Model_5 are shown [4] Johari H, Henoch C, Custodio D, Levshin A. Effects of leading-edge pro-
tuberances on airfoil performance. AIAA J 2007;45(11):2634e42.
in Fig. 18.
[5] Miklosovic DS, Murray MM. Experimental evaluation of sinusoidal leading
From the comparisons, one may clearly see the agreements edges. J Aircr 2007;44(4):1404e7.
between our simulations and the measured data for both of the [6] Guerreiro JLE, Sousa JMM. Low-reynolds-number effects in passive stall con-
BEMT-models. trol Using sinusoidal leading edges. AIAA J 2012;50(2):461e9.
[7] Zhang RK, Wu JZ. Aerodynamic characteristics of wind turbine blades with a
sinusoidal leading edge. Wind Energy 2012;15:407e24.
7. Conclusive remarks [8] Bai CJ, Lin YY, Lin SY, Wang WC. Computational fluid dynamics analysis of the
vertical axis wind turbine blade with tubercle leading edge. J Renew Sustain
Energy 2015;7. 033124e1-24-14.
This article presents a study of the optimal design of sinusoidal [9] Lee MH, Shiah YC, Bai CJ. Experiments and numerical simulations of the rotor-
LE-prots on B-blades and BEMT-blades used for small variable- blade performance for a small-scale horizontal axis wind turbine. J Wind Eng
speed HAWT. Learned from the previous study by Huang et al. Ind Aerodyn 2016;149:17e29.
[10] Huang GY, Shiah YC,Bai CJ, Chong WT. Experimental study of the protuber-
[10], the use of LE-prots covering full blade-spans would improve ance effect on the blade performance of a small horizontal axis wind turbine.
the stall phenomenon for B-blades, yet the performance would J Wind Eng Ind Aerodyn 2015;147:202e11.
drop significantly once the rotation reached a critical speed. Un- [11] Bai CJ, Wang WC, Chen PW, Chong WT. System integration of the horizontal-
axis wind turbine: the Design of turbine blades with an axial-flux permanent
doubtedly, this performance drop at high l will certainly reduce the
magnet generator. Energies 2014;7(11):7773e93.
average Cp value of a variable-speed HAWT system. This has [12] Bai CJ, Hsiao FB. Code development for predicting the aerodynamic perfor-
motivated the present research to seek an optimal design of sinu- mance of a HAWT blade with variable-operation and verification by numerical
simulation. In: 17th national computational fluid dynamics conference; 2010.
soidal LE-prots for BEMT-blades under low rotation speeds, yet no
p. 29e31 [July; Taoyuan, Taiwan].
significant drop of Cp is present under high speed rotations. The [13] Ryi J, Rhee W, Hwang UC, Choi JS. Blockage effect correction for a scaled wind
previous study [10] was only performed for (un-tapered/un- turbine rotor by using wind tunnel test data. Renew Energy 2015;79:227e35.
twisted) baseline blades with sinusoidal protuberances applied full [14] Krogstad PA, Lund JA. An experimental and numerical study of the perfor-
mance of a model turbine. Wind Energy 2012;15:443e57.
leading edges. To improve the design by Huang et al. [10], the [15] Selig MS, McGranahan BD, Broughton BA. Summary of low-speed airfoil data.
BEMT-blades of small-scale HAWT have employed the sinusoidal Virginia, USA: SoarTech Publications; 1995.