You are on page 1of 11

Computational Fluid Dynamics JOURNAL

vol.12 no.1

April 2003

(pp.)

Inuence of Inow Models on Helicopter Aeroelastic Optimization


Senthil Murugan

Ranjan Ganguli

Abstract This paper presents a multi-objective optimization study for helicopter vibration reduction in forward ight using an evolutionary algorithm. The objective is to reduce the vibratory forces acting on the rotor hub which are the main sources of helicopter vibration. A nonlinear aeroelastic model of the helicopter is used and the numerical results are obtained for a 4-bladed hingeless rotor using nite elements in space and time. The optimization is performed using two dierent multiobjective formulations: 1) an Euclidean norm method and 2) a min-max method. A real-coded genetic algorithm capable of nding the global minima is used for aeroelastic optimization. The main rotor blade elastic stinesses and mass are taken as design variables. The min-max method shows better reduction in the objective functions than the Euclidean norm based multi-objective formulation. The inuence of linear inow model and free wake aerodynamic models on aeroelastic optimization is then studied. The aeroelastic optimization results with free wake model show a quite dierent optimal design when compared with the linear inow model based optimization. A vibration reduction of 8 15 percent is achieved with blade elastic stiness and mass as design variables. The real-coded genetic algorithm is computationally ecient and is recommended for aeroelastic optimization with advanced aerodynamic models. Key Words: Helicopter vibration, Free wake models, Real-coded genetic algorithm, Min-max method. hensive rotor analysis code, mathematical modeling of the rotor ow-eld which plays a key role in the calculation of vibratory air loads have not been yet matured. This is due to the fact that the rotor blade pass through various ow-elds such as transonic ow in the advancing side, stalled and reversed ow in the retreating side and swept ow in the fore and aft region of the rotor disk. The aerodynamic models developed in the last two decades ranges from the lifting line vortices with a deformed helix to the full NavierStokes solutions [3]. The computational time and storage data for an aeroelastic simulation increases enormously when the full Navier-Stokes solutions are used. For example, the Euler/Navier-Stokes simulation for aerodynamic analysis often consume weeks of simulation time to map the entire ight regime [4]. Therefore, a full CFD analysis for aeroelastic optimization in forward ight is dicult. A tightly coupled nonlinear aeroelastic simulation with the CFD for aerodynamics and CSD (computational structural dynamics) for structural dynamics is still in the research stage [5]. However, hybrid methodologies developed in the recent years show a promising feature to couple the CFD analysis with a comprehensive aeroelastic analysis code [4]. In hybrid methodologies, an enormous reduction in the computational time is achieved by

INTRODUCTION

In contrast to the xed wing aircraft, ow-eld around the helicopter main rotor is highly inuenced by the wake eects. The tip vortex shed from a helicopter rotor blade causes a rapid change in the local ow condition of the successive blades. These blade vortex interactions lead to a high level of vibration and aerodynamic noise for the vehicle. Aeroelastic optimization studies have been carried out to minimize these vibrations [1, 2]. The main elements needed for a rotorcraft aeroelastic optimization process are: a comprehensive rotor analysis code and a robust optimizer. The rotorcraft aeroelastic analysis code used in the optimization should represent the blade aeroelasticity and rotor aerodynamics accurately to predict the blade response and performance parameters. Compared to the structural modeling of the rotor blades in compreReceived on. Graduate Student, Indian Institute of Science, India Associate Professor, Indian Institute of Science, India

c Murugan & Computational Fluid Dynamics JOURNAL 2002

Ganguli

2 using Navier-Stokes analysis only in the unsteady viscous ow regions and potential ow analysis in the other regions. Therefore, with the increasing computer power, it will be quite possible to couple the hybrid CFD analysis with comprehensive aeroelastic analysis code for aeroelastic optimization in the near future. Other than the modeling issues in aeroelastic analysis, a robust optimizer is essential for a successful aeroelastic optimization. The rotorcraft optimization problems are highly nonlinear and nonconvex in nature. Most of the previous aeroelastic optimization studies have used gradient based methods for designing a rotor blade for low vibration [1]. The gradient based rotorcraft aeroelastic optimization studies encounter the problems such as selection of initial design, local minima and diculties and errors in the calculation of gradients. However, gradient based methods are preferred than the evolutionary algorithms in the previous studies due to the computationally expensive nature of the helicopter aeroelastic analysis [1]. The time requirement is due to the fact that the rotor governing equations are function of both radial and temporal coordinates and need to be solved together with the nonlinear trim equations for accurate load predictions. At least, four loops are involved in rotorcraft aeroelastic optimization with the optimization loop as the outermost [6]. For a set of design variables from the optimizer, the trim equations are solved in the inner second loop and for each set of trim values, aerodynamic load distribution over the blades for a full revolution of the rotor is calculated as the third loop. The wake analysis form the innermost fourth loop as nding the wake geometry and induced velocities for each airload distribution. Because of these multiple loops, evolutionary algorithms are generally avoided in aeroelastic optimization to reduce computation time. Ribera and Celi [6] derived sensitivity derivatives for the free wake models to reduce the computation time of an aeroelastic optimization. In recent years, the power of computers has increased dramatically. Computer codes which used to take hours to run ten years ago now run in minutes. Therefore, it appears possible to use stochastic optimization methods such as genetic algorithms for rotorcraft optimization [7, 8]. Lee and Hajela used the parallel genetic algorithm for rotor blade design [9]. The use of high delity aerodynamic models such as free wake models is restricted in the rotorcraft optimization studies because of the large number of functional evaluations. Real-coded genetic algorithms (RCGA) developed in the recent years show a reduction in the computation time when compared to the conventional binary coded genetic algorithms [10, 11].

Murugan and Ganguli One more interesting issue in the rotorcraft aeroelastic optimization is the multi-objective nature of the problems. Therefore, proper multi-objective formulation is needed rather than the simple scalar addition of objective functions. A weighted linear combination of vibratory forces or norm of the forces is used as the objective function in most of the previous vibration reduction optimization studies [1]. The main disadvantage of this type of multi-objective formulation is nding the proper weights for each sub-objective function. Therefore, alternative multi-objective optimization methods have to be investigated to combine these objectives and constraints. A survey of multiobjective optimization methods is given in Reference [12]. This paper focus on the development of a robust optimization technique and multi-objective formulations for rotorcraft optimization studies. Rotorcraft optimization is carried out for vibration reduction with two multi-objective formulations. Then, the inuence of linear and free wake inow models on aeroelastic optimization is studied. Real-coded genetic algorithm is used as the optimization tool to design a low vibration rotor blade. 2 ROTOR AEROELASTICITY In this work, a comprehensive aeroelastic analysis code, based on nite element method is used to evaluate the helicopter vibration. The rotorcraft structure is modeled as a nonlinear representation of elastic rotor blades coupled to a rigid fuselage. The blade is modeled as a slender elastic beam undergoing ap bending, lag bending, elastic twist, and axial deection. The eect of moderate deections is included by retaining second order non-linear terms. The blade is discretized into beam nite elements, each with fteen degrees of freedom. The nite element equations are reduced in size by using normal mode transformation. This results in the non-linear ordinary dierential equation with periodic coecients as given below. M + Cp + Kp = F(, p, p) p (1)

Here M, C, K and F represents the nite element mass matrix, structural stiness matrix, damping matrix and nite element force vector, respectively. Also, p() is the modal coordinate and = t is the azimuth angle. Here, is the rotor rotational speed. These equations are then solved using nite element in time in combination with the Newton-Raphson method. The solutions to the equations are then used to calculate rotor blade loads using the force summation method, where aerodynamic forces are added to the inertial forces. The blade loads are integrated over the blade length and transformed to the xed frame

Inuence of Inow models on Helicopter Aeroelastic Optimization to get hub loads. The steady hub loads are used to obtain the forces acting on the rotor and combined with fuselage and tail rotor forces to obtain the helicopter rotor trim equations. F() = 0 (2) ky = 2
C i = T /2 2 (1 + kx x cos + ky x sin) 2 +

kx =

4 3

(1 1.82 ) 1 + ( )2

The above nonlinear trim equations are also solved using the Newton-Raphson method. The helicopter blade response equations and rotor trim equations in (1) and (2) are solved simultaneously to obtain the blade steady response and hub loads. This coupled trim procedure is important for capturing the aeroelastic interaction between the aerodynamic forces and the blade deformations. Further details of the analysis are available from Ref. [13]. 3 AERODYNAMICS The aerodynamic modeling used in this study consists of two parts : a local blade element model and a global wake model. The blade element model consists of a linear attached ow model and a separated ow model. In the attached ow model, the unsteady lift, drag and pitching moment consists of components from circulatory and non circulatory loads. The eect of near shed wake is considered in the circulatory part. The trailing edge ow separation is modeled using the Kircho model which accounts for the nonlinear lift variations [14]. The wake behind the rotor disk determines the induced inow distribution over the disk and plays a key role in the calculation of blade response and loads. At low speed ight condition, the wake stays close to the rotor disk and has a dominating inuence on blade airloads. The comprehensive analysis code used in this work has three types of inow models: 1) linear inow model, 2) prescribed wake model, and 3) free wake model. These models are briey discussed below. The simplest rotor wake models assume a uniform or linear inow distribution over the rotor disk. These simple inow models may capture the global eects of the rotor wake and are usually satisfactory for high speed ight condition where large portions of the near wake are carried away by a high incoming velocity. The linear inow model used in this study is based on the Drees model in which the rotor induced inow is calculated by [14] = tan s + i

where , s and CT are the advance ratio, rotor shaft tilt and rotor thrust coecient. The above inow model is useful for performance predictions but greatly unpredict vibratory loads. The linear wake model becomes less accurate at low speed and hovering ight conditions when the inow distribution becomes highly nonuniform over the rotor disk. The prescribed wake and free wake models considers the spatial and temporal variations of circulation around the rotor blade to predict the induced inows. The wake is divided into three regions: 1) Near wake, 2) rolling up wake and 3) far wake. In the wake analysis, the induced inow depends on the blade motion, free stream velocity and vorticity strength and its geometry. The free wake model used in this study is briey given below. The main part of the wake analysis is to nd the geometry of the wake. The free wake model iteratively solves for wake geometry accounting for both free stream and self induced velocity elds. The wake analysis is said to be converged when the geometry has attained the steady state solution. The distortion of wake geometry is considered only for blade tip vortex where as the rigid or prescribed wake is considered for blade inboard vorticity. The geometry of the tip vortex behind the reference blade is given by

r(, ) =rb ( ) + + D (, )(3) where rb is the position vector of the blade tip, is the free stream velocity vector, is the wake age and D is the distortion due to wake self induced velocity. Once the geometry of the wake is solved, the inuence coecients are calculated using the Biot-Savart law. The induced velocity due to the vortex line segment is given as v = 1 4 r d r3 (4)

where r is the vector from the vortex line segment d to the point P. For the vortex sheet element, the induced velocity is expressed as v = 1 4 r dA r3 (5)

4 In the nal stage of wake analysis, the induced velocity at the discrete points on the rotor disk i (ri , j ) is calculated as the product of bound circulation and inuence coecients and summing the contributions from all the vortex laments. Further details of the rotor wake model are given in Ref [13]. Use of the free wake model in the aeroelastic analysis helps in predicting the magnitude of the helicopter vibration more accurately [15]. 4 REAL-CODED GENETIC ALGORITHM Helicopter vibration reduction problem considered in this study is optimized using a real coded genetic algorithm. GAs are stochastic optimization techniques based on the Darwins theory of survival of the ttest [8]. GA is a search algorithm based on the mechanism of natural selection that transforms a population (a set of solutions) into a new population (i.e., next generation) using genetic operators such as crossover, mutation and reproduction [7]. A survival of the ttest strategy is adopted to identify the best strings and subsequently genetic operators are used to create a new population for the next generation. The GAs dier from the conventional optimization methods and search methods in the following way [8]: 1. Genetic algorithms work with a coding of the parameter set, not the parameters themselves. 2. Genetic algorithms search from a population of solution points instead of from a single solution point. 3. Genetic algorithms use objective function information, not derivatives or other auxiliary knowledge in their search for the optimal or best solution. 4. Genetic algorithms use probabilistic transition rules i.e., randomized operators and not deterministic rules for information exchange among the strings. More details about how genetic algorithms work for a given problem can be found in the literature [7]. A good representation scheme for the solution is very important in obtaining the best solution for a given problem using GA. The most commonly used solution representation is the binary vector (0 and 1). Recently in [16], it is shown that a string can consist of binary digits, integers and oating point numbers. It is also shown in [16] that a natural representation of strings is more ecient and produces better results. The advantage of the real-coded GA are 1. The binary coding of the design variables in the algorithm discretizes the design space as a set of design points whereas most of the design variable

Murugan and Ganguli values are continuous in the design space. This preassigned constraint for the design space may result in a local optimum. 2. A major disadvantage of the binary representation is the length of the chromosome. The length of the chromosome increases exponentially with the required decimal accuracy which in turn reduces the eciency of the GA. 3. The mapping between the real space and the binary space creates a problem for the crossover operator used in GAs. In the discrete variables case, the GA operators may produce an invalid ospring and therefore repair algorithms are used to avoid the osprings falling outside the discrete regions. The quality of solution in RCGA depends on the type of genetic operators used in the problem. A hybrid genetic algorithm employs dierent crossover and mutation operators at dierent stages of the genetic process and can provide an eective solution to many practical problems. A detailed study on eects of hybrid operators for real-coded genetic algorithm is presented in [17]. The genetic operators used in real-coded genetic algorithm of the present study is discussed in the following paragraphs. Let X be the possible solution X = [x1 x2 ] where x1 and x2 are real numbers. The parents in the population are randomly initialized and also satisfy the geometric constraints. The main genetic operators for this problem are dened such that the solution generated always satises the geometric constraints. Next, we describe the genetic operators and the tness function. 4.0.1 Genetic operators: The crossover and mutation operators used in this study are described below. Let X and Y be the two parents selected for the crossover operation. X = [x1 x2 ] Y = [y1 y2 ] where xi and yi are real numbers. a) Averaging Crossover: This crossover operator selects a scalar value in the range of 0 1 and osprings are produced by averaging as follows X = X + (Y X) Y = Y + (X Y ) (7) (6)

where X is less than Y . This operator always generates the ospring that satises the geometric

Inuence of Inow models on Helicopter Aeroelastic Optimization constraints. The new ospring will have values in between their parents. b) Heuristic Crossover: This crossover operator selects a random value in the range of 1 1 and osprings produced by this operator are X = X + c Y = Y + c

the longitudinal (Fx ), lateral (Fy ) and vertical (Fz ) forces. The 4 moments are the rolling (Mx ), pitching (My ) and yawing (Mz ) moments. The 4 forces are normalized by the rotor steady thrust, and the 4 moments are normalized by the rotor steady yawing moment. Therefore, for vibration reduction problem, the six objective functions can be given by M inimize : J(x) = [Fx , Fy , Fz , Mx , My , Mz ] (11)

(8)

where is half of the dierence between the upper and lower bounds of the individual design variables and c is the crossover rate. The mutation operators are used to introduce diversity in the population.The mutation operators used in this study are explained below. Let X be the parent selected for mutation operation. X = {x1 , x2 } a) Heuristic mutation : In this paper, a single mutation point is randomly selected. The ospring produced in this operation is described below.

The rotor blade structural stinesses, i.e, ap bending (EIy ), lag bending (EIy ) and torsional stiness (GJ), and blade mass per unit length (m) are considered as design variables. Move limits are placed on these design variables to ensure that the design does not become impractical.
low high EIy EIy EIy low high EIZ EIZ EIz

GJ low GJ GJ high mlow m mhigh

(12) (13)

The above optimization problem is solved using a real-coded genetic algorithm as discussed below. 6 RESULTS A four bladed soft-inplane hingeless rotor similar to the BO-105 rotor is considered for the numerical study. The baseline blade properties considered in this study are given in Table (1). The initial values Table 1: Baseline Hingeless Blade Properties Number of blades Radius, R (m) Advance ratio, Rotational speed, rad/s Lock number mo (kg/m) solidity CT / c/R 4 4.94 0.3 40.10 5.2 6.46 0.07 0.07 0.055

X = X + m

(9)

where m is the mutation rate. In this work, hybrid operators are used in the realcoded genetic algorithm. In hybrid real-coded genetic algorithm, the same parents are used to generate more osprings using dierent crossover and mutation operators. The ospring with the better tness value will replace the parents in the next generation. 5 OPTIMIZATION The general unconstrained multi-objective optimization problem is of the form M inimize : F (x) = [F1 (x), F2 (x), ..., Fk (x)] with xL x xU (10)

The superscripts L and U refer to the lower and upper bounds on vector of design variables x, respectively. The subscript k refers to the number of objective functions. An Nb -bladed helicopter rotor transmits Nb forces and moments to the fuselage as the principal source of vibration. Here, refer to the rotational speed of the rotor. For vibration reduction of a fourbladed rotor, the objective is to reduce the 4 forces and the 4 moments transmitted to the rotor hub from the rotating rotor system. The 4 forces are

of design variables are non-dimensionalized and are given below Xbaseline = [EIy /(m0 2 R4 ), EIz /(m0 2 R4 ), (14) GJ/(m0 2 R4 ), m/m0 ] Xbaseline = [0.0108, 0.0268, 0.00615, 1.0]

6 Initially the optimization is performed with two dierent multi-objective formulations: 1) Euclidean norm method with linear inow model, case I and 2) Min-max method with linear inow model, case II. The optimization results of both the methods are compared. Then, the inuence of free wake model on the aeroelastic optimization is studied in the case III. The three cases are discussed in the following section. 6.1 Case I: Euclidean norm method with linear inow model

Murugan and Ganguli six vibratory forces or sub-objective functions given in Eqn. (11) is shown in Fig. 1. The vertical force (Fz ) acting on the rotor hub which is considered as the dominant force in causing vibration is reduced by about 20 %. The remaining loads show a vibration reduction of about 20% except for the yawing moment Mz which shows a reduction of about 60%. The optimal design values are given as Xoptimal = [0.0122, 0.0189, 0.00689, 1.09] (18) The optimal design shows an increase in the blade mass, ap and torsional stiness whereas lag bending is reduced when compared to the baseline values given in Eqn. (14). The main disadvantage of this method of scalarizing the multi-objectives is that of the loss of individuality of objective functions. The yawing moment Mz show a higher reduction when compared to the other ve vibratory forces as shown in the Fig. 1. This is because the net eect of forces is considered by the single objective function given in equation (17). The increase in one objective function can be hidden by the decrease in other objective functions. Therefore, alternative methods are needed to perform the multiobjective optimization such that individual forces or objective functions are optimized as per the designer preferences. 6.2 Case II: Min-max method with linear inow model

One way of solving the multi-objective problem is by minimizing the dierence between the potential optimal point and a utopia point (also called as an ideal point) [12]. The Euclidean distance N (X) between the current objective function values and utopia points is given by N (X) = |F (X) F o | (15) where F o is the vector of utopia points for the vector F (X). For the vibration reduction problem considered, the Euclidean distance N (X) for vibratory forces and vibratory moments can be written as N1 = N2 =
o o o (Fx Fx )2 + (Fy Fy )2 + (Fz Fz )2

o o o (Mx Mx )2 + (My My ) + (Mz Mz )2

where the design vector X is given in the Eqn. (14) The norm of the forces and moments are added with weights W1 and W2 to form a single objective function as given below J = W1 N1 + W2 N2 (16)

In this case, min-max approach is used for scalarizing the multi-objectives [12]. The basic min-max formulation can be posed as Minimize: F (x) = max [F1 (x) F1 (x) ... Fk (x)] where the maximum value of objective functions Fi (x) is taken for functional or tness evaluation in GA. The multi-objective function for vibration reduction can then be written as Minimize: F (x) My (x) J1 (x) = max [ Fx (x) y b Fz (x) Mx (x) M b Mz (x) ] Fb F Fb Mb Mb
x y z x y z

Generally, the utopia points for forces and moments are considered as zero in the vibration reduction optimization studies since zero vibration is the ideal goal. And the weights of objective functions W1 and W2 are given a value of one which is equivalent to giving equal weights for both forces and moments. Therefore, the objective function can be written as [1] J=
2 2 2 Fx + Fy + Fz + 2 2 2 Mx + My + Mz

(17)

Now, the optimization is performed with the above single objective function, J. The upper and lower limits on the blade stinesses are taken as thirty percent above and below the baseline values given in the Eqn. (14) and twenty percent above and below the baseline value for mass per unit length. The real-coded GA is run with the above objective function with limits on the design variables. A vibration reduction of 18% is achieved by the optimization for the objective function J. The percentage reduction in each of the

where the superscript b refer to the baseline values of objective functions. This objective function tries to minimize the objective function with maximum value at the current design value. By this approach, the GA tries to minimize all the objective functions simultaneously. A reduction of 15 30% is achieved for loads other than Mz and 50% reduction is achieved for Mz with this min-max method. The percentage of reduction in each of the forces are given in Fig. 2. In the min-max method, each of the objective functions show an almost equal amount of reduction. In the Euclidean norm method, the objective function Mz show

Inuence of Inow models on Helicopter Aeroelastic Optimization a reduction of 60% whereas other objective functions Fx , Fy , Fz , Mx , My show a reduction of 20%. The optimal design values of min-max method are given as Xoptimal = [0.0120, 0.0206, 0.005965, 1.15] (19)

The optimal design shows an increase in ap stiness and blade mass whereas lag bending and torsional stiness are reduced from their baseline values. However, the optimization results are more evenly distributed. 6.3 Case III: Min-max method with free wake model

In the previous optimization cases I and II, a linear inow model is used to perform airload calculations. The rotorcraft aeroelastic optimization is often performed with the linear inow aerodynamic model with an implicit assumption that the same optimal design or at least the same direction in the design space will result even with the high delity aerodynamic model. To check this assumption, the objective functions are evaluated at the optimal design values of the case II with the free wake model in the aerodynamic analysis and the results are shown in Fig. 3. In contrast to the assumption, the objective functions Fz and Mz show an increase from their baseline values when the free wake model is used whereas the same optimal design values show a reduction in all objective functions with the linear inow model as shown in Fig. 2. Therefore, it is necessary to use the proper high delity aerodynamic models in the aeroelastic optimization. In case III, a free wake model is used for predicting the vibratory loads instead of linear inow model as in the case II. The GA is run with the same parameters as in case II. The min-max method is used to formulate the multi-objective function as in case II. It is found that the vibratory forces are more sensitive to design variables while using the free wake model. The GA needed more runs to nd an optimal design when free wake models are used. The percentage of reduction in rotor hub forces and moments is given in Fig. 4. The hub shear, pitching and rolling moments show a reduction of about 8 15%. The optimal values of the design variable are given as Xoptimal = [0.0124, 0.0306, 0.00633, 1.16] The optimal blade stiness and mass per unit length of this case show an increase from their baseline design values. Now, the six vibratory forces are evaluated with the linear inow model for the current optimal design and the percentage of reduction is shown in Fig. 5. All the six vibratory forces show a higher percentage of reduction with the linear inow model

when compared to the free wake model results shown in Fig. 4. Using the free wake model therefore gives a robust design and is much more realistic. The optimal design values for cases I, II and III are shown in Fig. 6. The stiness values EIy and m show a similar trend whereas the EIz and GJ diers with the multi-objective formulation and aerodynamic models in aeroelastic optimization. In particular, using the free wake model (case III) results in a completely dierent blade design in terms of the lag stiness which is not accurately predicted by the linear inow models. The ap, lag and torsion response of the helicopter rotor blade for optimal results of the case II and case III are shown in Fig. 7 to Fig. 9. The blade response for the optimal design values with linear inow model and free wake model show a similar change from its baseline value. However, while the inow models do not cause much change in the ap response, they signicantly aect the lag and torsion response. The pitching and roll moments and hub shear are particularly important for helicopter vibration. These loads for the optimal and baseline design values of case II and III are shown in Fig. 10 to 12. It is clear from the gures that these loads are underpredicted with linear inow model. Therefore, aeroelastic optimization results overestimate the vibration reduction if the low delity aerodynamic models are used and can lead to the actual designs not being able to deliver the vibration reduction levels predicted by the analysis. 7 CONCLUSIONS The following conclusions are drawn from this optimization study: 1) The min-max method for multi-objective formulation is found to be an eective and practical way of combining the multi-objectives of the helicopter vibration reduction problem. In contrast to the Euclidean norm method, the min-max method reduces all the six hub loads in an equitable manner. 2) The aerodynamic inow models play a key role in the prediction of vibratory loads. The optimization results obtained using free wake analysis are quite different from that using linear inow model. Vibration reduction of 15 - 30 percent obtained using linear inow model and min-max optimization reduces to 8 15 percent when free wake model is used. 3) Real-coded genetic algorithm is ecient for solving aeroelastic optimization problems with free wake modeling. Therefore, real-coded genetic algorithm can be also a candidate to the hybrid CFD methods based rotorcraft design optimization.

8 REFERENCES [1] R. Ganguli, A Survey of Recent Developments in Rotorcraft Design Optimization, Journal of Aircraft, 41, 3, 493-510, 2004. [2] M. S. Murugan, R. Ganguli, Aeroelastic Stability Enhancement and Vibration Suppression in a Composite Helicopter Rotor, Journal of Aircraft, 42, 4, 1013-1024, 2005. [3] A. T. Conlisk, Modern Helicopter Rotor Aerodynamics, Progress in Aerospace Sciences, 37, 5, 419-476, 2001. [4] Z. Yang, L. N. Sankar, M. J. Smith, O. Bauchau, Recent Improvements to a Hybrid Method for Rotors in Forward Flight, Journal of Aircraft, 39, 5, 804-812, 2002. [5] M. Potsdam, H. Yeo, W. Johnson, Rotor Airloads Prediction Using Loose Aerodynamic/Structural Coupling, Proceedings of the 60th Annual Forum of the American Helicopter Society, Baltimore MD, 2004. [6] M. Ribera, R. Celi, Maneuvering Free Wake Sensitivity for Design Optimization Applications, Proceedings of the 57th Annual Forum of the American Helicopter Society, Washington, DC, May 2001. [7] H. J. Holland, Adaptation in Natural and Articial Systems, University of Michigan Press, Ann Arbor, 1975. [8] Z. Michalewicz, Genetic Algorithms + Data Structures = Evolution Programs, Series, Springer-Verlag, New York, 1994. [9] J. Lee, P. Hajela, Parallel Genetic Algorithm Implementation in Multidisciplinary Rotor Blade Design, Journal of Aircraft, 33, 5, 962-969, 1996. [10] A. H. Wright, Genetic algorithms for real parameter optimization, Foundations of Genetic Algorithms-1, San Mateo, CA: Morgan Kaufman, 205-218, 1991. [11] K. Deb, A. Anand, and D. Joshi, A computationally ecient evolutionary algorithm for real-parameter optimization. Evolutionary Computation Journal, 10, 4, 371-395, 2002. [12] R. T. Marler, J. S. Arora, Survey of multi-

Murugan and Ganguli objective optimization methods for engineering, Structural and Multidisciplinary Optimization, 26, 369-395, 2004. [13] G. Bir, et al, University of Maryland Advanced Rotorcraft Code(UMARC) Theory Manual, UMAERO Report, University of Maryland, 1992. [14] J. G. Leishman, Principles of Helicopter Aerodynamics, Cambridge University Press, New York, 2003. [15] R. Ganguli, I. Chopra, W. H. Weller, Comparison of Calculated Vibratory Rotor Hub Loads with Experimental Data, Journal of the American Helicopter Society, 43, 4, 312-318, 1998. [16] V. Bharadwaj, H. F. Li, and T. Radhakrishnan, Scheduling Divisible Loads in Bus Networks with Arbitrary Processor Release Times, Computer Mathematics Application, 32, 57-77, 1996. [17] F. Herrera, M. Lozano, A. M. Snchez, A taxonomy for the crossover operator for real-coded genetic algorithms: An experimental study, International Journal of Intelligent Systems, 18, 309-338, 2003.

60

60 50

50

40

Vibration reduction (%)

Vibration reduction (%)

30 20 10 0 10 20 30 40

40

30

20

10

50

Fx

Fy

Fz

Mx

My

Mz

Fx

Fy

Fz

Mx

My

Mz

Fig. 1: Euclidean norm method with linear inow model, case I.

Fig. 3: Case II optimal design with free wake model.

60

60

50

50

Vibration reduction (%)

Vibration reduction (%)

40

40

30

30

20

20

10

10

Fx

Fy

Fz

Mx

My

Mz

Fx

Fz

My

Mz

Fig. 2: Min-max method with linear inow model, case II.

Fig. 4: Min-max method with free wake model, case III.

60

0.01

Lag response (nondimensional)

Vibration reduction (%)

50

0.005 0 0.005 0.01 0.015 0.02 0.025 0

Baseline ( linear inflow) Optimal ( linear inflow) Baseline ( Free wake ) Optimal ( Free wake )

40

30

20

10

Fx

50

100

150

200

250

300

350

My

Azimuth angle (deg)

Fig. 5: Case III optimal design with linear inow model.


0.3

Fig. 8: Blade tip lag response.

50

% change from baseline value

Torsion reponse ( deg )

40 30 20 10 0 10 20

Case I Case II Case III

0.2 0.1 0 0.1 0.2 0.3 0.4 0.5 0.6 0 50 100 150

Baseline ( Linear inflow ) Optimal ( Linear inflow ) Baseline ( Free wake ) Optimal ( Free wake )

200

250

300

350

30

Azimuth angle (deg)

EIy

EIz

GJ

Fig. 9: Blade tip torsional response.


1 x 10
3

Fig. 6: Optimal design values for cases I, II and III.


Baseline ( Linear inflow ) Optimal ( Linear inflow ) Baseline ( Free wake ) Optimal ( Free wake )

Pitching moment ( nondimensional )

Flap response (nondimensional )

0.08 0.07 0.06 0.05 0.04 0.03 0.02 0

0.5

Baseline (Linear inflow) Optimal (Linear inflow) Baseline ( Free wake) Optimal ( Free wake)

0.5

1.5

50

100

150

200

250

300

350

2 0

50

100

150

200

250

300

350

Azimuth angle (deg)

Azimuth angle (deg)

Fig. 7: Blade tip ap response. 2

Fig. 10: Reduction in pitching moment My .

20

x 10

Rolling moment, Mx ( nondimensional)

Baseline ( Linear inflow) Optimal ( Linear inflow) Baseline (Free wake) Optimal (Free wake )

15

10

50

100

150

200

250

300

350

Azimuth angle (deg)

Fig. 11: Reduction in rolling moment Mz .

Vertical hub shear, Fz ( nondimensional )

0.105 0.1 Baseline ( Linear inflow ) Optimal ( Linear inflow ) Baseline ( Free wake ) Optimal ( Free wake )

0.095 0.09

0.085 0.08

0.075

50

100

150

200

250

300

350

Azimuth angle (deg)

Fig. 12: Reduction in vertical hub shear Fz .

You might also like