Professional Documents
Culture Documents
Composite Structures
journal homepage: www.elsevier.com/locate/compstruct
a r t i c l e i n f o a b s t r a c t
Article history: The objectives of this work are to quantify the inuence of material and operational uncertainties on the
Available online 28 July 2009 performance of self-adaptive marine rotors, and to develop a reliability-based design and optimization
methodology for adaptive marine structures. Using a previously validated 3D uidstructure interaction
Keywords: model, performance functions are obtained and used to generate characteristic response surfaces. A rst-
Reliability-based design and optimization order reliability method is used to evaluate the inuence of uncertainties in material and load parameters
Adaptive marine structure and thus optimize the design parameters. The results demonstrate the viability of the proposed reliabil-
Fluidstructure interaction
ity-based design and optimization methodology, and demonstrate that a probabilistic approach is more
Composite blade
Propeller
appropriate than a deterministic approach for the design and optimization of adaptive marine structures
Turbine that rely on uidstructure interaction for performance improvement.
2009 Elsevier Ltd. All rights reserved.
0263-8223/$ - see front matter 2009 Elsevier Ltd. All rights reserved.
doi:10.1016/j.compstruct.2009.07.024
Author's personal copy
Although much progress in this eld has been made for rigid (CFRP). In addition to the well-known higher specic stiffness
and/or non-adaptive structures (see [20] for a recent review in this and higher specic strength of CFRP, the intrinsic deformation cou-
area), relatively little work focuses on exible structures that inter- pling behavior of anisotropic composites can be utilized to improve
act with the environment. Reviews of the state-of-the art methods the propeller performance by passive tailoring of the load-induced
in reliability-based design and optimization of aeroelastic struc- deformations according to the changing inow, as demonstrated in
tures can be found in [21]. As noted in [21], only limited work recent numerical (see Lee and Lin [12], Lin and Lee [13], Young
has been done on reliability analysis of structures undergoing et al. [14], Young and Liu [15], Young [16], Motley et al. [17]) and
uidstructure interactions, and most existing methods employed experimental (see Chen et al. [18]) studies. Nevertheless, all of
simplistic linear uid and uidstructure interaction models to the work thus far on self-adaptive composite marine propellers
determine the mechanical response, which introduce epistemic has been limited to deterministic analysis. Since the performance
modeling uncertainty. Hence, [21] introduced a reliability analysis of these structures is more sensitive to material or load uncertain-
method that integrates a coupled Euler ow solver with a struc- ties due to their dependence on uidstructure interaction, a reli-
tural nite element model (FEM) for the deterministic aeroelastic ability-based design and optimization method that can consider
analysis of a 3D wing structure, and employed a rst-order reliabil- natural or man-made variations is needed.
ity method (FORM) to evaluate the performance sensitivities to de-
sign parameters, operating conditions, and modeling uncertainties. 3. Problem denition
A probabilistic design assessment of smart composite structures is
presented in [22] in which sensitivity factors were developed for a To perform a reliability-based evaluation of the structure, we
series of design parameters for a composite wing based on their ef- will need to evaluate two performance measures. First, we must
fects on the angle of attack and impact response of the structure. nd the probability of unsatisfactory performance. This is done by
By improving the reliability of design parameters with the highest dening a limit state function, gX, where X consists of a vector
sensitivity factors, the failure probability was reduced for the of design variables, XD , either deterministic or random, and a vector
structure. The stochastic nature of composite properties has also of random variables, XR , representing uncertain structural proper-
been shown to lead to overestimation of structural reliability. By ties and loading conditions, and g is a function that relates the de-
using a probabilistic design methodology, improvements can be sign variables, random variables and the performance of the
made over traditional deterministic design methods [23]. In [24], structure. The function gX can either be implicit (e.g., the outcome
the reliability of a thin-walled circular composite cylinder was of a numerical BEMFEM code), or explicit (e.g., an approximate
shown to have a strong sensitivity to the applied load and to the equation obtained using the response surface method). The func-
amount of parametric scatter via multiple response surface tion gX is chosen such that gX 0 denes a boundary between
techniques. satisfactory and unsatisfactory performance (with gX < 0 indicat-
It should be noted that all of the above mentioned reliability- ing that the structure has unacceptable performance, and gX > 0
based design and optimization methods focus on adaptive/smart indicating acceptable performance). The performance state associ-
aerospace structures. Similar work is also needed for adaptive mar- ated with the boundary gX 0 is denoted as a limit state. Given
ine structures, where the uid loading tends to be much higher this formulation, the optimization problem herein can be written as
(due to the higher uid density and viscosity), the ow may be
highly unsteady (due to transient structural motion, as well as spa- maxpg obj X > 0 1
XD
tial and temporal variations in the ow eld), and may be suscep-
or minpg obj X 6 0
tible to cavitation damage. XD
16 0.8
12 0.6 Efficiency,
KT : 10 KQ :
Self-Twisting
tip (o)
Rigid
8 0.4
Torque, 10 KQ
4 Rigid 0.2
Self-Twisting - Undeformed
Self-Twisting - Deformed Thrust, KT
Theoretical Optimal
0 0
0.5 0.6 0.7 0.8 0.9 0.5 0.6 0.7 0.8 0.9
J J
Fig. 2. Comparison of the tip pitch angle (left) and performance curves (right) for the rigid and self-twisting propellers.
max/D
0.04
structure interaction analysis using the method outlined in Section
Standard Material
4. As shown in the left plot, the adaptable propeller geometry (rep- 0.02 E 1 + 3E1
resented by the tip pitch angle, /tip ) approaches the theoretical E 1 - 3E1
0
optimal propeller geometry, which changes with the advance coef- 0 15 30 45 60 75 90
( )
o
cient (J is inversely proportional to the angle of attack). For the
possible range of J values for forward operations, the self-twisting 0.06
propeller is designed to be overpitched in its unloaded (unde-
max/D
0.04
achieve equivalent performance between the two propellers.
Standard Material
2. At all J J design , the self-twisting propeller should yield higher 0.02 G 12 + 3G12
energy efciency than its rigid counterpart. 0
G 12 - 3G12
0 15 30 45 60 75 90
( )
o
The efciency is dened as g TV=2pnQ JK T =2pK Q which
corresponds to the ratio of the thrust power to the available shaft
Fig. 3. Effect of variations in material properties on the blade tip deection for the
power, with thrust coefcient K T T=qn2 D4 and torque coefcient self-twisting propeller.
K Q Q =qn2 D5 . T and Q are the dimensional thrust and torque,
respectively. As shown in Fig. 2, the rigid and the self-twisting pro-
pellers exhibit similar performance at the design condition 45
Standard Material
J 0:66. The efciency of the self-twisting propeller is higher than 40
P (kW)
E 1 + 3E1
its rigid counterpart for all J 0:66. The efciency improvement 35
E 1 - 3E1
45
To further simplify the model, parametric sensitivity analyses of Standard Material
the random variables XR were performed (see Figs. 3 and 4). By 40 G 12 + 3G12
P (kW)
G 12 - 3G12
taking each of the material parameters and providing them with 35
a normal distribution (i.e. with mean lXR XR;design 30
J; E1 ; E2 ; G12 ; m12 ; m21 0:66; 171:42 GPa; 9:08 GPa; 5:29 GPa; 25
0 15 30 45 60 75 90
0:32; 0:32 and standard deviation rXR 0:02l XR based on
( )
o
realizations of a normal distribution are within three standard FEM model, and gX is the data generated from the response sur-
deviations of the distribution mean), and compared with the de- face method; R2 values closer to 1.0 represent higher accuracy.
sign condition (i.e. XR;design ). The contour maps of the tted response surfaces and data com-
As shown in Figs. 3 and 4, the effects of variations in the primary puted using the BEMFEM numerical solver are shown in Figs. 57.
bending modulus E1 and shear modulus G12 are negligible. The shaded contour values represent the tted equations, while
Bounds for extreme values of the secondary bending modulus the dashed contour lines represent the BEMFEM simulation data.
E2 , however, noticeably deviate from the design behavior, partic- The power requirement is more sensitive to J than to h. Lower
ularly for large h. This is because at low values of h, the primary values of J correspond to higher angles of attack and higher loads
stiffness is governed by E1 171:42 GPa, which is a much larger and thereby higher power demands. At higher loads, the change
value compared to E2 9:08 GPa. Hence, even at three standard in pitch caused by the uidstructure interaction is also greater,
deviations from the mean 161:13 GPa 6 E1 6 181:71 GPa, the ef- and hence the power demand is more sensitive to h at lower J val-
fects on the normalized bending deection and power are small ues. At high J values, the power demand is lower and is less sensi-
due to the high stiffness. On the other hand, as h approaches 90 , tive to h due to small changes in pitch caused by the hydrodynamic
the primary bending stiffness of the blade is governed by E2 . Be- load induced bendingtwisting deformation.
cause E2 is comparatively small with respect to E1 , the system The maximum deection is a strong function of both J and h.
behavioral effects for variations in E2 are magnied, though only This is because, as the ber orientation angle becomes larger, the
marginally. For the purposes of this paper, however, it is assumed blades are less stiff along their primary (longitudinal) axis (which,
that the material parameters, except for h, have negligible effect on at h 45 becomes oriented more as the secondary axis). As a re-
the efciency, power requirement, and tip displacement of the pro- sult, the blade tip deections have nonlinear growth with ber ori-
peller blades. As such, the random variable vector XR can be simpli- entation angle. The increasing of the tip deection with decreasing
ed to only contain the advance coefcient, J. J is also expected due to increasing longitudinal load.
The efciency is highest near the design values J J design
0:66; h hdesign 32 , which means that the original design
7. Response surface methodology
90
The fully-coupled boundary element method-nite element
28
method (BEMFEM) model [25,16] summarized in Section 4 is 36
24
used for the design and analysis of adaptive composite marine ro-
tors. Although the coupled BEMFEM analysis method is relatively
fast, it can still be computationally expensive to use, with wait 60
32
28
large enough to successfully achieve a reliable optimization, this
20
24
becomes impractical. Since the behavior of the performance 30
(power, deection, and efciency) is expected to be smooth func-
40 36 32
tions of J and h, the response surface methodology is a reasonable
analysis alternative. Data points obtained from the BEMFEM 44
DJ; h
0 .0 4
06
60 0.
0:000005Jh2 0:00031J2 h 0:000001J 2 h2 9
gJ; h 0:2358 2:2626J 0:0015h 0:0058Jh
(o)
watts for the power surface. The goodness-of-t of the surfaces can 0 .0 2
be represented by the coefcients of determination for the power
0 .0 1
demand, blade tip deection, and efciency, which are 0.997,
0.997, and 0.988, respectively, where 0 0.0 1
0.4 0.5 0.6 0.7 0.8 0.9
Rg BEMFEM gBEMFEM 2
J
2
R 1 11
max / D
Rg BEMFEM gX2 Shaded Contours - Fitted
Dashed Lines - BEM-FEM 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09
where g BEMFEM is the data obtained from the numerical BEMFEM Fig. 6. Normalized blade tip deection, Dmax =D, contour map for the self-twisting
model, gBEMFEM is the mean of all data obtained from the BEM propeller.
Author's personal copy
90 50000
0.66
0.7 0
40000
60
0.70
(o)
P rigid (W)
30000
0 .5 4
30
0 .5 8
0 .6 2
0.
70
0 .6
20000
0 .5
6
0
0
46
2
Eq. (1) denes the probability of unacceptable performance, and
computing this probability requires integration of the probability
density function of X over the domain of x values that would result
in unacceptable performance
Z
0 pf pgX < 0 fX x dx 13
0 15 30 45 60 75
gX60
( )
o
20 3
g(X) < 0 h(U) < 0
FAILURE FAILURE h(U)=0
g(X) = 0
15 1.5
x* u*
U1
X2
10 0
5 -1.5
g(X) > 0 h(U) > 0
SAFE SAFE
0 -3
0 5 10 15 20 -3 -1.5 0 1.5 3
X1 U2
Fig. 10. Generalization of the transformation from the original X space (Left) into the U space (Right).
U1
trated in Fig. 10. The latter integral can still not be solved
analytically, unless the limit state function hU is linear. We thus
linearize hU at the so-called design point u , dened as
follows:
h(U) > 0
u arg minkUk jhU 0 16 SAFE
0.06 Normal
stresses. Further, the natural frequency limit is approximately
Lognormal 60 Hz, or 3600 rpm, which is about 4.6 times higher than the de-
0.04
PDF
0.75
to also be an order of magnitude smaller than the primary axis
0.5
0.25
0
600 180 0 30 60 90
design
160 1
500
0.75
p(gf1 <0)
max
v
140 0.5
400
0.25
(MPa)
120 Serviceability
n (Hz)
0
300 0 30 60 90
design
max
100
v
200 1
Safety
80 0.75
p(gf2 <0)
100 n 0.5
60
FORM
0.25
MonteCarlo
0 0
0.02 0.04 0.06 0.08 0 30 60 90
max / D design
0.75
0.5 normal, and Gumbel distributions. The Gumbel distribution was
considered to be more appropriate physically, and thus was used
0.25
as the advance coefcient distribution.
0
0 30 60 90 First-order reliability methods (FORM) were shown to be an
design adequate design tool instead of the more time consuming Monte
1 Carlo simulations for probabilistic propeller optimization. Through
0.75 Serviceable Operating Range FORM analysis, it was shown that the optimal ber orientation an-
p(gf1 < 0)
0.5
gle for the adaptive propeller is 29 6 h 6 31 , which will yield a
93% probability of acceptable performance based on three criteria:
0.25
a serviceability limit state based on propeller power requirement, a
0 safety limit state based on blade tip deection, and an objective
0 30 60 90
design function that ensures the maximum overall energy efciency.
1
The serviceability limit state and constraint functions are designed
such that, on average, the power requirement of the adaptive pro-
0.75 Safe Operating Range
p(gf2 < 0)
peller is less than its rigid counterpart. The safety limit state and
0.5
constraint functions are designed to limit the blade tip deection
0.25 to a specied value to prevent excessive deections, stresses, and
0 to reduce the susceptibility to hydroelastic instability failures. Fi-
0 30 60 90
design nally, the objective function was used to determine the optimal -
ber orientation angle that will maximize the average energy
Fig. 15. Results of the FORM methodology for the probability of failure of the efciency of the self-twisting propeller for all forward operating
objective and limit state functions. conditions. With the knowledge of the optimized equivalent single
layer ber orientation, a series of possible layup sequences can be
developed that will provide an equivalent optimal loaddeforma-
perfect agreement between the two methods, suggesting that tion behavior of the blade.
FORM is valid for this example, where h is assumed to follow a nor- The results show that a probabilistic approach is more appropri-
mal distribution and J is assumed to follow a Gumbel distribution. ate than a deterministic approach for the design and optimization
of adaptive composite structures that rely on uidstructure inter-
9.3. Results action. This is because such structures are inherently more sensi-
tive to random variations in material properties, geometric
The results of the objective function and limit states are shown congurations, and loading conditions. It is important to note that,
in Fig. 15. Again, h is assumed to follow a normal distribution while while the self-twisting propeller can be optimized for efciency, it
J is assumed to follow a Gumbel distribution. According to the top is possible that the limit state functions would provide no viable
gure, the optimal ber orientation angle in terms of the objective design option that would be both serviceable and safe. Should a de-
function is about 59 ; however, the objective function showed lit- sign singularity occur, changes in the blade geometry, material
tle variation in the failure probability across the entire range of h. properties, and material congurations may be necessary. Addi-
The limit states play a very important role beyond the objective tional work is needed to assess the effect of material, geometry,
function. The constraint functions each have denitive boundaries and load uncertainties on the initiation and evolution of failure
for acceptable performance. The serviceability constraint is that modes. This is more complex due to the need to consider the many
the power requirement of the self-twisting propeller is, on average, layers of laminates and the many possible modes of failure, as well
lower than that of the rigid propeller, which can only be satised if as uncertainties in the failure modeling of CFRP. It should be
29 6 h 6 83 . Second, the safety constraint limits the ber orien- emphasized here that although the methodologies presented here-
tation angle to h < 31 . Hence, what seemed initially to be a wide in focus on adaptive composite marine propellers, the framework
range of viable options for the design variable based on the objec- is also generally applicable to other exible structures that under-
tive function is limited to a small range of 29 6 h 6 31 . This go uidstructure interactions, including wind or tidal turbines.
places the optimal design based on FORM methodology to be very
close to the optimal deterministic design. In this case, the probabil-
ity of failure of the objective function ranges between 6.6% and Acknowledgements
6.9%, which represents approximately 93% condence that the
self-twisting propeller will exhibit safe and improved performance The authors are grateful to the Ofce of Naval Research (O N R)
over the rigid propeller for a realistic range of operating conditions. and Dr. Ki-Han Kim (program manager) for their nancial support
through grant numbers N00014-07-1-0491 and N00014-08-1-
10. Conclusions 0475.
[4] Ganguli R, Chopra I. Aeroelastic tailoring of composite couplings and blade [18] Chen B, Neely S, Michael T, Gowing S, Szwerc R, Buchler D, et al. Design,
geometry of a helicopter rotor using optimization methods. J Am Helicopter fabrication and testing of pitch-adapting (exible) composite propellers. In:
Soc 1997;42(3):21828. The SNAME propeller/shafting symposium, Williamsburg, VA; 2006.
[5] Soykasap O, Hodges DH. Performance enhancement of a composite tilt-rotor [19] Nicholls-Lee RF, Turnock SR. Enhancing performance of a horizontal axis tidal
using aeroelastic tailoring. J Aircraft 2000;37:8508. turbine using adaptive blades. In: OCEANS 2007 Europe, Aberdeen, Scotland;
[6] Glaz B, Friedmann P, Lu L. Helicopter vibration reduction throughout the entire 2007.
ight envelope using surrogate-based optimization. J Am Helicopter Soc [20] Frangopol DM, Maute K. Reliability-based optimization of civil and aerospace
2009;54(1). systems. Engineering design reliability handbook. CRC Press; 2005. 24-124-
[7] Yang Y. Structural analysis and multidisciplinary design of exible uid loaded 32 [chapter 24].
composite canard. Ph.D. thesis, Iowa State University; 2005. [21] Allen M, Maute K. Reliability-based shape optimization of structures
[8] Lee AT, Flay RGJ. Compliant blades for passive power control of wind turbines. undergoing uidstructure interaction phenomena. Comput Method Appl M
Wind Eng 2000;24:311. 2005;194:347295.
[9] Lobitz DW, Veers PS, Eisler GR, Laino DJ, Migliore PG, Bir G. The use of twist- [22] Chamis CC. Design of smart composite structures in the presence of
coupled blades to enhance the performance of horizontal axis wind turbines. uncertainties. J Chin Inst Eng 2004;27:77181.
Tech rep, Sandia National Laboratories; 2000. [23] Lekou DJ, Philippidis TP. Mechanical property variability in FRP laminates and
[10] Lobitz DW, Veers PS. Load mitigation with bending/twist-coupled blades on its effect on failure prediction. Compos Part B Eng 2008;39:125674.
rotors using modern control strategies. Wind Energy 2003;6:10517. [24] Rais-Rohani M, Singh MN. Comparison of global and local response surface
[11] Gowing S, Cofn P, Dai C. Hydrofoil cavitation improvements with elastically techniques in reliability-based optimization of composite structures. Struct
coupled composite materials. In: Proceedings of 25th American towing tank Multidiscip O 2004;26:33345.
conference, Iowa City, IA; 1998. [25] Young YL. Time-dependent hydroelastic analysis of cavitating propulsors. J
[12] Lee Y, Lin C. Optimized design of composite propeller. Mech Adv Mater Struct Fluid Struct 2007;23:26995.
2004;11:1730. [26] Huang TT, Groves NC. Effective wake: theory and experiment. Tech rep, David
[13] Lin C, Lee Y. Stacking sequence optimization of laminated composite Taylor Naval Ship Research and Development Center, Bethesda, MD;
structures using genetic algorithm with local improvement. Compos Struct 1981.
2004;63:33945. [27] Choi JK, Kinnas SA. Prediction of non-axisymmetric effective wake by a three-
[14] Young YL, Michael TJ, Seaver M, Trickey ST. Numerical and experimental dimensional Euler solver. J Ship Res 2001;45:1333.
investigations of composite marine propellers. In: 26th symposium on naval [28] ABAQUS. ABAQUS version 6.5 documentation. ABAQUS, Inc.; 2005.
hydrodynamics, Rome, Italy; 2006. [29] Young YL, Liu Z, Motley MR. Inuence of material anisotropy on the
[15] Young YL, Liu Z. Hydroelastic tailoring of composite naval propulsors. In: 26th hydroelastic behaviors of composite marine propellers. In: Proceedings of
International conference on offshore mechanics and arctic engineering, San the 27th symposium on naval hydrodynamics, Seoul, Korea; 2008.
Diego, CA; 2007. [30] Plucinski M, Young YL, Liu Z. Optimization of a self-twisting composite marine
[16] Young YL. Fluidstructure interaction analysis of exible composite marine propeller using a genetic algorithm. In: Proceedings of 16th international
propellers. J Fluid Struct 2008;24:799818. conference on composite materials, Kyoto, Japan; 2007.
[17] Motley MR, Liu Z, Young YL. Utilizing uidstructure interactions to improve [31] Melchers RE. Structural reliability analysis and prediction. 2nd ed. Chichester
energy efciency of composite marine propellers in spatially varying wake. (NY): John Wiley; 1999.
Compos Struct 2009;90(3):30413.