Professional Documents
Culture Documents
Energy Reports
journal homepage: www.elsevier.com/locate/egyr
Research paper
article info a b s t r a c t
Article history: In order to overcome the problems of low efficiency and poor results of classical optimization methods,
Received 14 July 2022 an efficient, intelligent, and reliable method for optimizing the shape of the blades of the Savonius wind
Received in revised form 2 September 2022 turbine is proposed in this paper. Firstly, the shape of the blade is parameterized by the third-order
Accepted 14 September 2022
Bezier curve, and then a certain number of design schemes in the design space are obtained by the Latin
Available online xxxx
hypercube sampling method and the torque coefficient of each design scheme is evaluated based on
Keywords: CFD simulation, and next a radial basis function surrogate model is constructed based on these design
Blade optimization schemes and their torque coefficients, and finally the optimal design scheme is obtained by solving the
Savonius turbine surrogate model using the marine predator algorithm, and is verified using the CFD simulation. The
CFD results show that when the tip speed ratio is 1, the average torque coefficient of the optimized blade
Marine predator algorithm
obtained by the above method is about 7% higher than that of the classical semicircular blade, and
Radial basis function model
under other tip speed ratios, the average torque coefficients of the optimized blade is also significantly
Latin hypercube sampling
Bezier curve higher than that of the classical blade. Furthermore, compared with the classical semicircular blade,
about 11% materials can be saved in the manufacturing of the optimized blade.
© 2022 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND
license (http://creativecommons.org/licenses/by-nc-nd/4.0/).
https://doi.org/10.1016/j.egyr.2022.09.062
2352-4847/© 2022 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-
nc-nd/4.0/).
H. Xia, S. Zhang, R. Jia et al. Energy Reports 8 (2022) 12366–12378
12368
H. Xia, S. Zhang, R. Jia et al. Energy Reports 8 (2022) 12366–12378
Fig. 4. 2-D wind turbine simulation area and boundary condition settings.
efficiency. In general, the more grids and the shorter the time
Table 1 steps, the more accurate the simulation results will be, but at
Geometric dimensions. the same time the calculation time will be longer. In order to
Structure size parameters Values determine the appropriate number of grids and time steps, about
Diameter of the rotor of the wind turbine (D) 0.909 (m) 65,000, about 85,000, about 105,000 grids were compared in this
Diameter of the wind turbine blades (d) 0.5 (m) work, and the results are shown in Fig. 6. It can be seen that
Diameter of the rotational domain (DR) 1 (m) the differences of the Cm values corresponding to the number
Overlap ratio (s/d) 0.18 of 85,000 grids, the number of 105,000 grids, and the 65,000
Length of the static domain 21.816 (m)
Width of the static domain 18.18 (m)
are very small, so the 85,000 grid is selected as the number of
Distance from the center of the rotor to the upper wall 9.09 (m) grids in this work considering the computational efficiency. Then,
Distance from the center of the wind turbine rotor to 7.272 (m) different time step sizes corresponding to ∆θ = 0.5◦ , ∆θ = 1◦ ,
the entrance ∆θ = 2◦ , and ∆θ = 4◦ are compared, and the results are shown
Center coordinates of the rotor of the wind turbine (0, 0)
in Fig. 7. It can be seen that the results are very close when
Number of blades 2
∆θ = 0.5◦ or ∆θ = 1◦ , so the time step is finally selected as
∆θ = 1◦ .
Based on the above settings, a numerical simulation for 10 ro-
model is shown in Eq. (9) (Menter, 1994). tation periods of a classical Savonius wind turbines is performed,
∂ ∂ ∂ ∂k as showed in Fig. 8. It can be seen that with the increase of the
[ ]
(ρ k) + (ρ kui ) = Γk + Gk − Yk − SK + Gb rotation cycle, the simulation results gradually tend to be stable,
∂t ∂ xi ∂ xj [ ∂ xj ]
∂ ∂ ∂ ∂ω and the difference between the average torque coefficient of the
(ρω) + (ρωui ) = Γω + Gω − Yω + Dω + SW + GWb tenth cycle and that of the previous cycle is very small, so in
∂t ∂ xi ∂ xj ∂ xj
the future simulation, the average torque coefficient of the 10th
(9) rotation cycle is selected as the evaluation criterion for the wind
where Gk represents the kinetic energy of the turbulent flow, Gω capture ability of the wind turbine.
is the equation of ω, Γk and Γω represent the effective diffusion Based on the above CFD process, the obtained average torque
terms of k and ω, respectively, Yk and Yω represent the divergence coefficients of the classical Savonius wind turbine under different
terms of k and ω, respectively, and Dω represents the orthogonal tip speed ratios are compared with the wind tunnel test data
divergence term. SK and SW are user-defined source terms. Gb given in the literature (R et al., 1978), as shown in Fig. 9. It
and GWb account for buoyancy terms as described in effects of can be found that, when the tip speed ratio is less than 0.9,
buoyancy on turbulence in the k-ω Models. the CFD simulation results are very close to the experimental
In this paper, ANSYS2020R1 Meshing module is used to divide results. When the tip speed ratio is equal to 1, the simulation
the grids. To ensure that the value of y+ is close to 1 and that results are slightly higher than the experimental value, but the
the turbulence characteristics of the blade near the wall surface error is not more than 5%. When the tip speed ratio is greater
are accurately captured, a total of 20 boundary layers are added than 1, as the tip speed ratio increases, the deviation of the CFD
around the blades. The height of the first layer is set to 0.1 mm simulation results and the test results increases. The reasons are:
and the inflation growth factor is set to 1.2. Meanwhile, the grids (1) With the increase of the angular velocity, the loss caused by
at the interface are also refined. The final grid division is shown the vibration of the shaft, the friction loss between the central
in Fig. 5. shaft and the bearing, and the additional loss of the bracket will
Since the topic studied is a non-steady-state problem, the time increase. But the CFD simulation does not take these losses into
step size should be considered, and the formula of the time step account, so the CFD simulation results are slightly higher than
is as Eq. (10) shows. the experimental results. (2) There may be some errors in the
experimental data.
∆θ Overall, the CFD simulation process used in this work is highly
∆t = ◦ , (10)
ω 180
π
accurate. For the aforementioned 80 set of sample schemes, the
above CFD simulation process is used to obtain their average
where ∆θ is angle relative to one time step. torque coefficients as the basis for constructing the surrogate
In CFD simulation, the number of grids and the selected time model in the following section.
steps have significant impacts on the accuracy and computational
12370
H. Xia, S. Zhang, R. Jia et al. Energy Reports 8 (2022) 12366–12378
Z · β=F (12)
ϕ (∥Z 1 − Z 1 ∥) . . . ϕ (∥Z 1 − Z N ∥)
⎡ ⎤ ⎡ ⎤
F1
⎢.. . .
. . .. . ⎥
⎥, F = ⎢.. ⎥.
⎥ ⎢
where Z = ⎢⎣. ⎦ ⎣ ⎦
ϕ (∥Z N − Z 1 ∥) . . . ϕ (∥Z N − Z N ∥) FN
Based on Eq. (12), β can be determined as β=Z −1 · F . Once
β is determined, the built RBF model can be used to predict
Fig. 9. The comparison between the CFD simulation results and the the average torque coefficient value of any scheme in the design
experimental data (R et al., 1978) under different TSR. space.
12372
H. Xia, S. Zhang, R. Jia et al. Energy Reports 8 (2022) 12366–12378
where Xi represents the i-th candidate solution, NP is the popula- where CF = (1 − iter /max_iter)2·iter /max_iter , where iter represents
tion size, K is the dimension of the problem, xmin
j and xmax
j are the the current number of iterations, and max_iter represents the
upper and lower bounds of the j-th design variable, respectively, maximum number of iterations of the algorithm set.
and rand is a random number between [0, 1]. In the third stage, the algorithm employs operators as shown
After population initialization, MPA begins to iteratively try in Eq. (17).
to update the solution population. The entire iterative process is stepsizei = RL i ⊗ (RL i ⊗ Elite − X i )
evenly divided into three stages, at which the algorithm employs
w
, (17)
different operators. The operator used in the first stage is shown X ne
i = Elite + 0.5 · CF ⊗ stepsizei i = 1, 2, . . . , NP
in Eq. (14), After generating a new solution for each solution in the popula-
stepsizei = RBi ⊗ (Elite − RBi ⊗ X i ) tion, MPA employs a greedy strategy to determine the trade-offs
w
, (14) for the new solution. Specifically, if X ne
i is superior to Xi , replace
w
X ne
i = X i + 0.5 · rand(1, K ) ⊗ stepsizei i = 1, 2, . . . , NP new
Xi with X i , otherwise X inew
is discarded and Xi is retained, thus
completing the update of the solution population.
where RBi is a vector of 1×K, each dimension of it is a random
In addition, the algorithm also simulates the effects of eddies
number that conforms to the standard normal distribution; Elite
w and fish aggregation in the marine environment. In each iteration,
is the optimal solution individual in the current population; X ne
i after the population is updated in the manner described above,
is a new solution generated based on Xi ; rand(1, K ) is a vector of
the algorithm attempts to update the population again using
1×K, and its dimensions are random numbers within [0, 1] that
Eq. (18).
conform to the uniform distribution; ⊗ represents point-to-point
multiplication. X i + CF [X min + R ⊗ (X max − X min )] ⊗ U ifr ≤ 0.2
{
w
In the second stage, MPA uses two different operators. Half of X ne
i = ,
the solution population is updated using Eq. (15), X i + [FADs(1 − r) + r] (X r1 − X r2 ) else
Fig. 13. Comparison of the optimized blade and the classical blade.
Fig. 16. Comparison of average torque and power coefficients of classical Savonius turbine and optimized Savonius turbine under different tip speed ratios: (a) torque
coefficient, (b) power coefficient.
Fig. 17. Comparison of the pressure on the concave and convex surfaces of the optimized blade and the classical blade: (a) at 90◦ , (b) at 270◦ .
higher than that of the classical semicircular blade. Meanwhile, curation, Validation. Rongyuan Jia: Visualization, Writing – re-
When TSR = 1.09, the average torque coefficient of the optimized view & editing. Huadong Qiu: Investigation, Funding acquisition.
blade is about 8.2% higher than that of the classical semicircular Shuhui Xu: Methodology, Supervision, Project administration,
blade. Funding acquisition.
(2). The optimized blade has greatly improved the wind cap-
ture performance of the turbine in the range of rotation angle of Declaration of competing interest
0∼90◦ and 180∼270◦ .
(3). Similar to the classical semicircular blade, the optimized The authors declare that they have no known competing finan-
blade obtained in this paper also has the advantages of simple cial interests or personal relationships that could have appeared
structure, low cost, and easy manufacturing. Furthermore, com- to influence the work reported in this paper.
pared with the classical semicircular blade, about 11% materials
can be saved in the manufacturing of the optimized blade.
Data availability
In the research process, it is found that there is still a lot of
room for improvement. For example, the basis function of the
Data will be made available on request.
RBF surrogate model has an impact on the accuracy of the model.
In future work, better blade shapes can be found by changing
the smooth parameters of the basis functions or constructing a Acknowledgments
dynamic RBF surrogate model.
We sincerely thank the supports provided by the National
Natural Science Foundation of China (Grant No. 51905284) and
CRediT authorship contribution statement the Innovation capability improvement project of technology-
based small and medium-sized enterprises of Shandong Province,
Huaijie Xia: Software, Writing – original draft, Writing – re- China (Grant No. 2021TSGC1060), and we are very grateful to the
view & editing, Visualization. Song Zhang: Investigation, Data anonymous reviewers for their valuable comments.
12376
H. Xia, S. Zhang, R. Jia et al. Energy Reports 8 (2022) 12366–12378
Fig. 18. Velocity and vorticity distribution of optimized turbine and classical turbine: (a) velocity distribution of classical blade, (b) velocity distribution of optimized
blade, (c) vorticity distribution of classical blade, (d) vorticity distribution of optimized blade.
Fig. 19. Pressure distribution of optimized turbine and classical turbine (a) classical blade, (b) optimized blade.
References Alom, N., Saha, U.K., 2019. Influence of blade profiles on Savonius rotor perfor-
mance: Numerical simulation and experimental validation. Energy Convers.
Abd Elaziz, M., Mohammadi, D., Oliva, D., Salimifard, K., 2021. Quantum marine Manage. 186, 267–277. http://dx.doi.org/10.1016/j.enconman.2019.02.058.
predators algorithm for addressing multilevel image segmentation. Appl. Soft Amiri, M., Anbarsooz, M., 2019. Improving the energy conversion efficiency
Comput. 110, 107598. http://dx.doi.org/10.1016/j.asoc.2021.107598. of a Savonius rotor using automatic valves. J. Solar Energy Eng. 141 (3),
Abdelaziz, K.R., Nawar, M.A.A., Ramadan, A., Attai, Y.A., Mohamed, M.H., 2022. http://dx.doi.org/10.1115/1.4042828.
Performance improvement of a Savonius turbine by using auxiliary blades. Asadi, M., Hassanzadeh, R., 2021. Effects of internal rotor parameters on the
Energy 244, http://dx.doi.org/10.1016/j.energy.2021.122575. performance of a two bladed darrieus-two bladed Savonius hybrid wind
12377
H. Xia, S. Zhang, R. Jia et al. Energy Reports 8 (2022) 12366–12378
turbine. Energy Convers. Manage. 238, 114109. http://dx.doi.org/10.1016/j. Mercado-Colmenero, J.M., Rubio-Paramio, M.A., Guerrero-Villar, F., Martin-
enconman.2021.114109. Doñate, C., 2018. A numerical and experimental study of a new Savonius
Belmili, H., Cheikh, R., Smail, T., Seddaoui, N., Biara, R.W., 2017. Study, design and wind rotor adaptation based on product design requirements. Energy Con-
manufacturing of hybrid vertical axis Savonius wind turbine for urban ar- vers. Manage. 158, 210–234. http://dx.doi.org/10.1016/j.enconman.2017.12.
chitecture. Energy Procedia 136, 330–335. http://dx.doi.org/10.1016/j.egypro. 058.
2017.10.389. Meri Al Absi, S., Hasan Jabbar, A., Oudah Mezan, S., Ahmed Al-Rawi, B., Thajeel
Chan, C.M., Bai, H.L., He, D.Q., 2018. Blade shape optimization of the Savonius Al Attabi, S, 2021. An experimental test of the performance enhancement of
wind turbine using a genetic algorithm. Appl. Energy 213, 148–157. http: a Savonius turbine by modifying the inner surface of a blade. Mater. Today:
//dx.doi.org/10.1016/j.apenergy.2018.01.029. Proc. 42, 2233–2240. http://dx.doi.org/10.1016/j.matpr.2020.12.309.
Duraklı, Z., Nabiyev, V., 2022. A new approach based on bezier curves to Mohamed, M.H., Janiga, G., Pap, E., Thévenin, D., 2011. Optimal blade shape
solve path planning problems for mobile robots. J. Comput. Sci. 58, http: of a modified Savonius turbine using an obstacle shielding the returning
//dx.doi.org/10.1016/j.jocs.2021.101540. blade. Energy Convers. Manage. 52 (1), 236–242. http://dx.doi.org/10.1016/j.
Elsakka, M.M., Ingham, D.B., Ma, L., Pourkashanian, M., 2019. CFD analysis of
enconman.2010.06.070.
the angle of attack for a vertical axis wind turbine blade. Energy Convers.
Mohammadi, M., Lakestani, M., Mohamed, M.H., 2018. Intelligent parameter
Manage. 182, 154–165. http://dx.doi.org/10.1016/j.enconman.2018.12.054.
optimization of Savonius rotor using artificial neural network and ge-
Eshagh Nimvari, M., Fatahian, H., Fatahian, E., 2020. Performance improvement
netic algorithm. Energy 143, 56–68. http://dx.doi.org/10.1016/j.energy.2017.
of a Savonius vertical axis wind turbine using a porous deflector. Energy
10.121.
Convers. Manage. 220, http://dx.doi.org/10.1016/j.enconman.2020.113062.
Nedelkova, Z., Lindroth, P., Stromberg, A.B., Patriksson, M., 2016. Integration of
Faramarzi, A., Heidarinejad, M., Mirjalili, S., Gandomi, A.H., 2020. Marine preda-
tors algorithm: A nature-inspired metaheuristic. Expert Syst. Appl. 152, expert knowledge into radial basis function surrogate models. Opt. Eng. 17
http://dx.doi.org/10.1016/j.eswa.2020.113377. (3), 577–603. http://dx.doi.org/10.1007/s11081-015-9297-7.
Fatahian, E., Ismail, F., Hafifi Hafiz Ishak, M., Shyang Chang, W, 2022. An R, E, Sheldahl, L., V, Feltz, B., F, Blackwell, 1978. Wind tunnel performance
innovative deflector system for drag-type Savonius turbine using a rotating data for two- and three-bucket Savonius rotors. J. Energy 2:3 (3), 160–164.
cylinder for performance improvement. Energy Convers. Manage. 257, http: http://dx.doi.org/10.2514/3.47966.
//dx.doi.org/10.1016/j.enconman.2022.115453. Roy, S., Saha, U.K., 2013. Computational study to assess the influence of overlap
Ferdoues, M.S., Ebrahimi, S., Vijayaraghavan, K., 2017. Multi-objective optimiza- ratio on static torque characteristics of a vertical axis wind turbine. Procedia
tion of the design and operating point of a new external axis wind turbine. Eng. 51, 694–702. http://dx.doi.org/10.1016/j.proeng.2013.01.099.
Energy 125, 643–653. http://dx.doi.org/10.1016/j.energy.2017.01.070. Roy, S., Saha, U.K., 2015. Wind tunnel experiments of a newly developed two-
Ferrari, G., Federici, D., Schito, P., Inzoli, F., Mereu, R., 2017. CFD study of Savonius bladed Savonius-style wind turbine. Appl. Energy 137, 117–125. http://dx.
wind turbine: 3D model validation and parametric analysis. Renew. Energy doi.org/10.1016/j.apenergy.2014.10.022.
105, 722–734. http://dx.doi.org/10.1016/j.renene.2016.12.077. Sagharichi, A., Zamani, M., Ghasemi, A., 2018. Effect of solidity on the perfor-
González-Gorbeña, E., Qassim, R.Y., Rosman, P.C.C., 2018. Multi-dimensional mance of variable-pitch vertical axis wind turbine. Energy 161, 753–775.
optimisation of tidal energy converters array layouts considering geometric http://dx.doi.org/10.1016/j.energy.2018.07.160.
economic and environmental constraints. Renew. Energy 116, 647–658. http: Tahani, M., Rabbani, A., Kasaeian, A., Mehrpooya, M., Mirhosseini, M., 2017.
//dx.doi.org/10.1016/j.renene.2017.10.009. Design and numerical investigation of Savonius wind turbine with dis-
Kaldellis, J.K., 2022. 2.01 - Introduction to wind energy. In: Letcher, T.M. (Ed.), charge flow directing capability. Energy 130, http://dx.doi.org/10.1016/j.
Comprehensive Renewable Energy, second ed. Elsevier, Oxford, pp. 1–12. energy.2017.04.125.
Kamoji, M.A., Kedare, S.B., Prabhu, S.V., 2009. Experimental investigations on Thiyagaraj, J., Rahamathullah, I., Anbuchezhiyan, G., Barathiraja, R., Ponshanmu-
single stage modified Savonius rotor. Appl. Energy 86 (7), 1064–1073. http: gakumar, A., 2021. Influence of blade numbers, overlap ratio and modified
//dx.doi.org/10.1016/j.apenergy.2008.09.019. blades on performance characteristics of the Savonius hydro-kinetic turbine.
Kim, S., Cheong, C., 2015. Development of low-noise drag-type vertical wind tur- Mater. Today: Proc. 46, 4047–4053. http://dx.doi.org/10.1016/j.matpr.2021.
bines. Renew. Energy 79, 199–208. http://dx.doi.org/10.1016/j.renene.2014.
02.568.
09.047.
Tian, W., Mao, Z., Zhang, B., Li, Y., 2018. Shape optimization of a Savonius wind
Li, S., Li, Y., Yang, C., Wang, Q., Zhao, B., Li, D., Zhao, R., Ren, T., Zheng, X., Gao, Z.,
rotor with different convex and concave sides. Renew. Energy 117, 287–299.
Xu, W., 2021. Experimental investigation of solidity and other characteristics
http://dx.doi.org/10.1016/j.renene.2017.10.067.
on dual vertical axis wind turbines in an urban environment. Energy Convers.
Tian, X., Pang, W., Wang, Y., Guo, K., Zhou, Y., 2019. LatinPSO: An algorithm
Manage. 229, 113689. http://dx.doi.org/10.1016/j.enconman.2020.113689.
for simultaneously inferring structure and parameters of ordinary differ-
Marinić-Kragić, I., Vučina, D., Milas, Z., 2020. Computational analysis of Savonius
ential equations models. Biosystems 182, 8–16. http://dx.doi.org/10.1016/j.
wind turbine modifications including novel scooplet-based design attained
via smart numerical optimization. J. Cleaner Prod. 262, http://dx.doi.org/10. biosystems.2019.05.006.
1016/j.jclepro.2020.121310. Wenehenubun, F., Saputra, A., Sutanto, H., 2015. An experimental study on the
Masdari, M., Tahani, M., Naderi, M.H., Babayan, N., 2019. Optimization of airfoil performance of Savonius wind turbines related with the number of blades.
based Savonius wind turbine using coupled discrete vortex method and salp Energy Procedia 68, 297–304. http://dx.doi.org/10.1016/j.egypro.2015.03.259.
swarm algorithm. J. Cleaner Prod. 222, 47–56. http://dx.doi.org/10.1016/j. Yadav, S., Saha, S.K., Kar, R., Mandal, D., 2022. EEG/ERP signal enhancement
jclepro.2019.02.237. through an optimally tuned adaptive filter based on marine predators
Mckay, M., Beckman, R.J., 1979. Comparison of three methods for selecting algorithm. Biomed. Signal Process. Control 73, 103427. http://dx.doi.org/10.
values of input variables in the analysis of output from a computer code: 1016/j.bspc.2021.103427.
Technometrics. Technometrics 21 (2), http://dx.doi.org/10.1080/00401706. Yahya, W., Ziming, K., Juan, W., Qurashi, M.S., Al-Nehari, M., Salim, E., 2021.
1979.10489755. Influence of tilt angle and the number of guide vane blades towards the
Menter, F.R., 1994. Two-equation eddy-viscosity turbulence models for Savonius rotor performance. Energy Rep. 7, 3317–3327. http://dx.doi.org/10.
engineering applications. Aiaa J. 32, http://dx.doi.org/10.2514/3.12149. 1016/j.egyr.2021.05.053.
12378