Anda di halaman 1dari 19

Theoret. Comput.

Fluid Dynamics (1998) 11: 4967

Theoretical and Computational Fluid Dynamics


Springer-Verlag 1998

Direct Numerical Simulation of Hypersonic Boundary-Layer Flow on a Flared Cone1


C. David Pruett
Department of Mathematics, James Madison University, Harrisonburg, VA 22807, U.S.A.

Chau-Lyan Chang
High Technology Corporation, Hampton, VA 23666, U.S.A. Communicated by M. Y. Hussaini Received 20 March 1997 and accepted 21 May 1997

Abstract. The forced transition of the boundary layer on an axisymmetric ared cone in Mach 6 ow is simulated by the method of spatial direct numerical simulation (DNS). The full effects of the ared afterbody are incorporated into the governing equations and boundary conditions; these effects include nonzero streamwise surface curvature, adverse streamwise pressure gradient, and decreasing boundary-layer edge Mach number. Transition is precipitated by periodic forcing at the computational inow boundary with perturbations derived from parabolized stability equation (PSE) methodology and based, in part, on frequency spectra available from physical experiments. Signicant qualitative differences are shown to exist between the present results and those obtained previously for a cone without afterbody are. In both cases, the primary instability is of second-mode type; however, frequencies are much higher for the ared cone because of the decrease in boundary-layer thickness in the ared region. Moreover, Goertler modes, which are linearly stable for the straight cone, are unstable in regions of concave body are. Reynolds stresses, which peak near the critical layer for the straight cone, exhibit peaks close to the wall for the ared cone. The cumulative effect appears to be that transition onset is shifted upstream for the ared cone. However, the length of the transition zone may possibly be greater because of the seemingly more gradual nature of the transition process on the ared cone.

1. Introduction
The computation documented in the recent papers of Pruett et al. (1995), Pruett and Chang (1995), and Pruett and Zang (1995), represents a rst attempt to simulate numerically laminarturbulent transition in a highspeed boundary-layer ow for Reynolds numbers and congurations of engineering interest. Specically, their work examined the laminar breakdown of a Mach 8 boundary-layer ow on an axisymmetric sharp cone in the absence of a streamwise pressure gradient. The computation, an approximate analog of the sharp-cone experiment of Stetson et al. (1983), was motivated by the hope that, when experiment and computation are both brought to bear on the same physical problem, the results will be more enlightening than when either approach is used alone. In this spirit, in this paper we take a further step in the direction of conguration
1 The rst author gratefully acknowledges the nancial support of the Air Force Ofce of Scientic Research through Grant F49620-95-1-0146 (monitored by Dr. Leoridas Sakell).

49

50

C.D. Pruett and C.-L. Chang

DNS by simulating transitional hypersonic ow on a ared cone; again the choice of test case is motivated by previous experimental investigations, which we discuss briey below. It is now well established both experimentally and computationally that the high-frequency secondmode disturbances predicted by Mack (1984) of the basis of linear stability theory (LST) provide the primary instability mechanism for transition to turbulence in high-speed boundary-layer ows. Specically, second-mode instability waves have been observed experimentally in the hypersonic boundary layers of right circular cones by several experimentalists, among them Stetson and Kimmel (1992), who summarize a body of work in their 1992 paper. That second-mode disturbances can lead to transition on a cone has been veried computationally by the present authors (Pruett and Chang, 1995). Because of a generally stabilizing trend as Mach number is increased (Mack, 1984), in high-speed boundary layers, the transition zone (i.e., the region between transition onset and fully turbulent ow) can be quite long. It is natural to wonder how the results for straight cones translate to generalized axisymmetric forebodies. In particular, how are the primary instability, the transition mechanism, the location of transition onset, and the length of the transition zone altered by variations in parameters such as surface curvature? Recently, several experiments have been conducted on cones in hypersonic wind tunnels to assess the effects on boundary-layer stability and transition of streamwise surface curvature (Kimmel, 1993; Lachowicz et al., 1996a,b), of wall cooling (Blanchard and Selby, 1996), and of angle of attack (Doggett et al., 1997). Of these inuences, surface curvature is of primary interest here. In general, the effects of streamwise surface curvature are both overt and subtle. The dominant effects of body curvature on transition are the induced changes in the mean ow. Concave curvature, for example, produces a positive (adverse) streamwise pressure gradient, a decrease in mean edge Mach number, an increase in mean wall temperature, and a decrease in boundary-layer thickness. More subtle effects include Goertler instabilities, possible competition between second-mode and Goertler instabilities, and the presence of additional curvature terms in the equations of motion that must be considered in any theoretical or computational context. Of the experiments mentioned above, Kimmels (1993) was strictly a transition experiment, in which the primary goal was determination of the effects of parameter variation on the location of transition onset and the length of the transition region. Specically, Kimmel examined the boundary-layer ow on ve different conical congurations at several values of freestream unit Reynolds number. For each conguration, an ogive or ared afterbody was appended to a (right circular) conical forebody, which remained the same for all congurations. By varying the geometry of the afterbody, a range of nearly linear favorable (for ogival afterbodies) and adverse (for ared afterbodies) pressure gradients was obtained. Transition locations were determined solely on the basis of surface heat transfer. In general, Kimmel found that although favorable pressure gradients delay transition, and adverse pressure gradients promote transition, favorable pressure gradients caused a shorter transition zone. For high-speed boundary-layer ows, the transition zone can be quite long, and the effect on the length of the transition zone of an adverse pressure gradient was inconclusive. Kimmels transition experiments were conducted in the Mach 8 wind tunnel at the Arnold Engineering Development Center, a conventional facility with relatively high levels of freestream turbulence. In contrast, the stability and (natural) transition experiments of Lachowicz et al. (1996a,b), Doggett et al. (1997), and Blanchard and Selby (1996) are unique in that they were conducted in a quiet tunnel, the Mach 6 pilot facility at the NASA Langley Research Center, which was designed specically to reduce freestream turbulenc signicantly. The characteristics of the facility itself are documented in Blanchard et al. (1997). The experiment of Lachowicz et al. (1996a,b), the rst conducted after the facility study, examined boundary-layer stability on an axisymmetric ared cone with a nominally adiabatic wall. In this experiment the goal was to identify the primary mechanism of transition as well as the transition location and the extent of the transition region. Transition onset was estimated from surface temperature measurements, for which the model was instrumented with thermocouples. Run times were sufciently long so that the model surface was considered to be in thermal equilibrium. Remarkably different surface temperature distributions were obtained dependent upon whether or not the tunnel was operating in its quiet mode (with tunnel-wall boundary-layer bleed valves open). (See Figure 2 of Lachowicz et al. (1996b).) Boundary-layer totaltemperature and mass-ux proles and disturbance spectra were also obtained by hot-wire anemometry. In addition, a few images were obtained by schlieren interferometry. The experiment involved a single aredcone conguration, for which the ared afterbody produced a nearly linear adverse pressure gradient, the effects of which were of primary concern. The body are was also essential to the experiment in another aspectwithout it transition could not have been achieved within the quiet core of the tunnel. The later

Direct Numerical Simulation of Hypersonic Boundary-Layer Flow on a Flared Cone

51

(and complementary) experiments of Doggett et al. (1997) and Blanchard and Selby (1996) expanded on the initial study of Lachowicz et al. (1996a,b) by examining the effects on stability of angle-of-attack and wall cooling, respectively. All three experiments involved ared-cone models whose boundary layers were characterized by adverse pressure gradients. To complement this experimental work, theoretical studies were conducted by Balakumar and Malik (1994), who used LST to predict transition locations on the basis of the integrated growth rates (N-factors) of selected disturbances. For the ared-cone conguration of Lachowicz et al. (1996a,b), Balakumar and Malik (1994) predicted that second-mode disturbances with frequencies in the neighborhood of 230 kHz were most likely to lead to transition. The interested reader is referred to the recent paper of Wilkinson (1997) that reviews the NASA Langley quiet-tunnel experiments. For our present work, a computational test case has been designed based on the congurationa ared cone in nominal thermal equilibriumand ow conditions of the experimental investigation of Lachowicz et al. (1996a,b). This choice of test case affords a rare opportunity to examine a single high-speed transitional boundary-layer ow by theoretical (Balakumar and Malik, 1994), experimental (Lachowicz et al., 1996a,b), and computational means. We confess at the outset that such comparisons are never exact, and the reader should be cautioned as to the limitations of each viewpoint. The N-factor analysis attempts to predict transition, which is a fundamentally nonlinear phenomenon, on the basis of linear theory. The limitations of the experiment include a focus only on the primary instability mechanism (i.e., no information is available regarding three dimensionality of the ow) and a lack of specicity in regard to the disturbance environment. Moreover, as the rst of the three quiet-tunnel experiments, the experiment of Lachowicz et al. (1996a,b) experienced some problems with the signal-to-noise ratio of the hot-wire anemometer, and there was difculty in calibrating data collected very near the model surface. Whereas the physical experiment examined natural transition, for reasons of computational efciency, the numerical experiment examines forced transition, for which the boundary layer is seeded with disturbances of specic rather than random frequencies. Unfortunately, the resolution requirements for the simulation of natural transition rendered such a computation impractical at the time due to the computational resources available. Moreover, also because of practical limitations in computer resources, we simulate only the early stages of laminarturbulent transition on the ared cone; the simulation of end-stage transition is not attempted. Consequently, the reader should consider these theoretical, experimental, and computational results as different perspectives on similar, but not identical, objects. Despite the limitations and inconsistencies, our hope is that the composite picture of transition on a ared cone is clearer than that of each of the individual views. To summarize, our motivations are to advance the state-of-the-art in conguration DNS and to initiate a computational study of the effects of surface curvature on transitional high-speed ows. To these ends, we adapt the approach and algorithm of Pruett et al. (1995). The effects of streamwise surface curvature, mean streamwise pressure gradient, and varying boundary-layer edge conditions are incorporated fully into the governing equations and boundary conditions. The particular test case affords an opportunity to compare the present results not only with theoretical and experimental results but also with computational results of transitional ow on a straight cone (Pruett and Chang, 1995; Pruett and Zang, 1995). In the next section we briey discuss the governing equations for ow along axisymmetric bodies with streamwise curvature. The computational test case is summarized in Section 3. Section 4 addresses the numerical methodology, in general, and the changes to the approach of Pruett et al. (1995) necessitated by the ared-cone geometry, in particular. In Section 5 we exploit nonlinear parabolized stability equation (PSE) methodology to develop a viable transition mechanism based on second-mode primary instability. Results are presented in Section 6, and, where appropriate, comparisons are made with previous theoretical, experimental, or computational results. Conclusions are offered in Section 7.

2. Governing Equations
We consider the compressible ow of air, for which p, , T , , and denote its pressure, density, temperature, viscosity, and thermal conductivity, respectively. We assume that the ow is exactly governed by the compressible NavierStokes equations (CNSE), which can be found in many references on uid mechanics. If we restrict attention to axisymmetric bodies and dene x as the surface arc length, as the azimuthal angle, and z as the coordinate normal to the wall, then in body-tted coordinates, the CNSE assume the

52

C.D. Pruett and C.-L. Chang

form given in Pruett et al. (1995), where u, v , and w denote the velocity components in the streamwise, azimuthal, and wall-normal directions, respectively. We note that the equations given in Pruett et al. include both transverse and streamwise curvature terms and incorporate the notation R(x) and (x) to denote the body radius and the angle of the surface tangent relative to the axis, respectively, as functions of arc length. (See Figure 2 of Pruett et al. (1995).) Four dimensionless parameters arise from nondimensionalization of the governing equations: Mach number M , Reynolds number Re, Prandtl number P r, and the ratio of specic heats, . Throughout this work, P r and are assumed to be constants with values of 0.72 and 1.4, respectively. Finally, the dependency of viscosity on temperature is modeled by Sutherlands law, and = /P r.

3. Test Case
The experiment of Lachowicz et al. (1996a,b), on which the numerical computation is based, examines Mach 6 ow on a 20-in-long ared-cone model at a (nominal) zero angle of attack. The model consists of a 5 half-angle axisymmetric sharp cone to which a ared afterbody is appended. The afterbody, which begins 10 axial inches from the apex of the cone, is dened by a circular-arc are with a constant radius of curvature of 93.07 in; a curvature discontinuity exists at the are interface. The nominal preshock freestream conditions, which we use for the numerical experiment, are M T (Tt p (pt Re 1 = = = = = = 6.0, 98.78R 810.0R), 11.86 psf 18,720 psf), 2,728,087 ft1 ,

(1)

where Re 1 is the freestream unit Reynolds number. Throughout this work, asterisks denote dimensional quantities, and the subscripts , t, and e denote freestream, total, and boundary-layer edge conditions, respectively. The actual freestream conditions as reported by Lachowicz et al. (1996a,b), which differ 6 slightly from the nominal values above, are M = 5.93 and Re 1 = 2.82 10 per foot. To facilitate later comparisons with the experimental results of Lachowicz et al. (1996a,b) and the theoretical results of Balakumar and Malik (1994), we dene as follows the root Reynolds number based on freestream values of density, velocity, and viscosity: ReL =
Re 1x = u x .

(2)

A precise thermal boundary condition at the model surface cannot be ascertained on the basis of information reported about the experiment. Run times were long, and it is believed that the wall behaved approximately adiabatically under equilibrium conditions. However, the model was thin-skinned and uninsulated, and mounting hardware prevented the attainment of thermal equilibrium near the models base. Given the uncertainty as to the experimental thermal condition and the fact that we consider only the early stages of transition for the DNS, for which the mean ow has laminar characteristics, we have adopted the boundary condition of Pruett and Chang (1995). The use of this thermal condition is advantageous in that it allows comparison of simulated ared-cone and straight-cone results for the same boundary conditions. The boundary conditions are specied precisely later.

4. Methodology
Roughly speaking, spatial DNS is a four-step process. In the rst step a laminar and steady base state is dened, whose stability is to be investigated. The base state is then subject to time-dependent disturbances, which are imposed at or near the computational inow boundary. The uctuations are usually obtained from linear or nonlinear stability theory; specication of the disturbances comprises the second step. The third

Direct Numerical Simulation of Hypersonic Boundary-Layer Flow on a Flared Cone

53

step involves computing the evolution of the disturbances by numerically solving the three-dimensional and unsteady CNSE. The temporally and spatially evolving numerical solution is sampled and stored periodically in time for the subsequent postprocessing and analysis that constitute the nal step. In this section we address steps one and three; steps two and four are addressed in the next two sections, respectively. To obtain the laminar base state for their stability calculations, Balakumar and Malik (1994) computed the ow over the ared-cone conguration by both Euler and NavierStokes techniques. The NavierStokes solution was sensitive to the choice of limiter (dissipation) and required very high resolution within the wall layer to provide accurate rst and second wall-normal derivatives, which were needed for accurate linear-stability analysis. They found that the pressure distributions computed by the two methods agreed very well except near the tip of the cone. Moreover, N-factors computed by the two methods agreed to within approximately 10%. Because the difference in N-factors was small, for computational efciency, they recommended an Euler/boundary-layer approach over the solution of the full CNSE. For the present work, we also favor a boundary-layer equation approach. Specically, we exploit the spectrally accurate boundary-layer code of Pruett and Streett (1991) with the modication for varying edge conditions proposed by Pruett (1993). The postshock edge pressure distribution p e (x ) needed by the boundary-layer code is obtained from the NavierStokes solution of Balakumar and Malik (1994). In the vicinity of the tip (a region excluded in the DNS calculation), the pressure distribution is modied by linearly extrapolating upstream from downstream values. For self-consistency, all other boundary-layer edge conditions are inferred from the pressure, based on the assumption of isentropic postshock ow. Although this assumption is not strictly valid in the ared region because of the gradually increasing shock angle, the shock is highly oblique and weak, and errors committed by assuming isentropic postshock ow are small, as will be conrmed shortly. Figure 1 presents the radius R of the cone as a function of surface arc length x . The edge Mach number, edge temperature, and edge pressure distributions are also provided as functions of x . Note that the edge Mach number decreases almost linearly from nearly 5.6 at the beginning of the are to 4.91 at the end of the model. The close agreement between our present end-of-model Mach number and that reported by Balakumar and Malik (1994) (who observe a Mach number of 4.93 at the end of the are) serves to validate the approximation of isentropic postshock ow. In contrast to the edge Mach number, edge pressure and temperature both increase signicantly and nearly linearly in the are region. The streamwise evolution of the laminar boundary-layer thickness is shown in Figure 2. Although for experimentalists the boundary-layer thickness is the conventional measure of boundary-layer growth, for computational

Figure 1. Radius and boundary-layer edge conditions versus surface arc length x for axisymmetric ared-cone model in Mach 6 ow. Units for temperature and pressure are degrees Rankine and psf, respectively.

54

C.D. Pruett and C.-L. Chang

Figure 2. Computationally (dashed line) and experimentally (symbols) derived values of boundary-layer thickness versus surface arc length for ared-cone model. For reference, computa (solid line) is also protionally derived displacement thickness 1 vided. Boundary-layer growth reverses just downstream of body are, which begins at the location indicated by vertical dashed line.

purposes the displacement thickness 1 , an integral quantity given for axisymmetric bodies in White (1974) (or in Equation (13) of Pruett et al. (1995)), is preferred. Both quantities are presented in Figure 2. For the Mach-number range of the are region, is approximately 1.2 1 . For comparison, measured values of taken from the essentially laminar-ow region of the experiment of Lachowicz et al. (1996a) are also presented. The experimental and computational values agree remarkably well; both manifest thinning of the boundary layer in the are region. Dimensional temperature and streamwise velocity proles for an adiabatic wall are shown in Figure 3. The proles, which correspond to axial station x = 1.083 ft (13.0 in) in the are region, are virtually indistinguishable from the adiabatic-wall results shown in Figure 13b of Balakumar and Malik (1994), obtained at the same station. It is also instructive to view the same proles in normalized units, as shown in Figure 4. For this gure, lengths have been normalized by the local boundary layer displacement thickness 1 and temperature and velocity have been normalized by their respective edge values. Note that the velocity distribution is nearly linear across the boundary layer, a characteristic of high-speed boundary layers. The shoulder of the boundary layer lies approximately 1.25 1 from the wall. If we bear in mind that second-mode disturbances typically travel at 9095% of the boundary-layer edge velocity, it can be inferred from Figure 4 that the critical layer (where u and disturbance phase velocity are equal) lies approximately one boundary-layer thickness from the wall. Also note that, because of frictional heating, the wall temperature for this high-speed ow is nearly six times that at the boundary layer edge. The boundary-layer solution is interpolated onto a DNS grid using a spectrally accurate interpolation code, which extrapolates outside the boundary-layer region by solving the continuity equation exactly in the asymptotic limit as z . The asymptotic outer solution is matched to the inner solution at the boundarylayer edge. Special attention is paid to computing the wall-normal velocity accurately, for which we use the method of Pruett (1993). A typical wall-normal velocity distribution versus z is shown in Figure 5. For axisymmetric cones, the wall-normal velocity is negative, which implies that the far-eld computational boundary should be treated as an inow boundary.

Figure 3. Dimensional temperature and streamwise velocity distributions versus wall-normal coordinate z at axial station x = 13 in. Proles are virtually indistinguishable from those shown in Figure 13 of Balakumar and Malik (1994).

Direct Numerical Simulation of Hypersonic Boundary-Layer Flow on a Flared Cone

55

Figure 4. Normalized temperature and streamwise velocity distributions versus locally normalized wall-normal coordinate z at axial station x = 13 in.

The special care used to match inner and outer solutions consistently is necessary to prevent discontinuities in higher derivatives of the basic state, which would contaminate the subsequent stability analysis. As a consistency check, we typically compute the residuals of the time-independent and dimensionless CNSE by using the interpolated boundary-layer solution as an initial state. Residuals that are quite small, say of order 104 , as shown in Figure 6, indicate that the boundary-layer solution closely approximates the exact solution. Because the residuals reect the sum of approximation errors, interpolation errors, numerical errors, and blunders, small residual errors provide a stringent test of self-consistency among the various numerical tools exploited by DNS. Indeed, on several occasions we have uncovered and corrected algorithmic bugs simply by examining the steady-state residuals. For the DNS calculation, we adapt the techniques of Pruett et al. (1995). Their fully explicit collocation scheme exploits the third-order low-storage RungeKutta scheme of Williamson (1980) for time advancement and both spectral and central compact-difference methods for spatial discretization. Specically, spectral-collocation methods (Canuto et al., 1988) are used in the azimuthal direction, in which the ow is periodic, whereas the sixth-order compact-difference methods of Lele (1992) are exploited in the aperiodic streamwise and wall-normal directions. Modest ltering is used to correct the innate tendency of central-difference approximations toward oddeven decoupling of the solution in the far eld whenever resolution is marginal. Filtering also signicantly reduces numerical reections at the outow boundary. A discussion of the necessity for and implementation of ltering can be found in Pruett et al. (1995). With re-

Figure 5. Normalized wall-normal velocity versus locally normalized wall-normal coordinate z at axial station x = 13 in.

Figure 6. Residuals of governing equations along inow boundary of computational domain of DNS.

56

C.D. Pruett and C.-L. Chang

gard to the spectral-collocation scheme used in the azimuthal direction, we implement dealiasing according to the 1 2 -rule to suppress high-frequency oscillations that could lead to numerical instability. As in Pruett et al. (1995), we formulate the energy equation in terms of the pressure, and, for computational efciency, we enforce symmetry with respect to the plane = 0. Minor modications to the scheme of Pruett et al. are necessitated by particulars of the current aredcone problem. Because boundary-layer edge conditions vary for the current problem, ow quantities are nondimensionalized by the preshock freestream temperature, velocity, and density, etc., given or implied in (1). Previously, ow quantities were scaled by appropriate postshock values at the boundary-layer edge, which were presumed constant along the axisymmetric cone without are. For both the work of Pruett et al. (1995) and the current work, lengths are scaled by the boundary-layer displacement thickness 1 at the computational inow boundary xin . Previously, R(x) was computed internally and (x) was constant; for the present problem, both quantities are computed externally by the boundary-layer code and read as input to the DNS algorithm. As was done in Pruett et al., grid spacing is uniform in the x direction, and, in the wall-normal direction, grid points are tightly clustered near the wall and near the critical layer. For the present problem, the DNS computation includes only a region of ow aft of the are, where the boundary-layer thickness diminishes. (Refer to Figure 2.) This is in contrast to the growing boundary layer along the cone without are. In either case, to account for the variation in boundary-layer thickness, we exploit a mapping of the form z = f (x). (3) However, for the present problem, the mapping function is linear and is given by f (x) = 1 0.05 x xout , xin xout (4)

where xout is the coordinate of the computational outow boundary. For all practical purposes, = 1 corresponds to a distance one (local) boundary-layer displacement thickness from the surface. The boundary conditions for the present problem are taken without modication from Pruett et al. (1995). In brief, all ow variables are specied along the inow boundary, a buffer domain approach (Streett and Macaraeg, 1989/90) is exploited at the outow boundary, and Thompson (1987) conditions are adapted along the far-eld boundary = max . At the wall, all velocity components vanish; the thermal boundary condition may be either adiabatic or isothermal. As mentioned previously, to facilitate comparison with the straight-cone results of Pruett and Chang (1995), we adopt their hybrid thermal boundary condition whereby wall temperature is held xed at its laminar adiabatic-wall value. Density at the wall is computed directly from the governing equations.

5. Primary Instability and Transition Mechanism


As mentioned previously, an advantage in considering the present ared-cone conguration is that both theoretical and experimental data regarding the primary instability are available. As observed by Balakumar and Malik (1994), for the ared cone, the gradual reduction in boundary-layer thickness in the are region allows unstable disturbances of xed frequency to experience amplication over longer regions than for the straight cone, whose boundary layer continuously thickens. Their linear stability analyses predicted that transition was likely to result from a primary second-mode disturbance of frequency 230 kHz, which attained the largest integrated growth rate (N-factor). Remarkably, the frequency data obtained by Lachowicz et al. (1996a,b) using hot-wire anemometry are in very close agreement with the predictions of Balakumar and Malik (1994). Specically, the experimental results show that the most unstable frequency F in the are region lies in the range of 218228 kHz. Unfortunately, the signal-to-noise levels of the data of Lachowicz et al. (1996a,b) are too low in the region ahead of the are to provide useful information there. As mentioned in the Introduction, we believe that well-resolved DNS of natural transition is currently impractical for existing supercomputers; consequently, we limit our consideration to forced transition with monochromatic forcing. Ultimately, transition is a three-dimensional and nonlinear process, and additional information beyond the frequency content of the primary disturbance is needed to identify a viable transition mechanism. Specically, information regarding disturbance three dimensionality is essential. To date, we

Direct Numerical Simulation of Hypersonic Boundary-Layer Flow on a Flared Cone

57

know of no hypersonic stability or transition experiment in which the disturbance environment is well dened, with the sole exception of the stability experiment of Cavalieri (1995) in which a glow-discharge device was used to seed the boundary layer of a cone with linear two- and three-dimensional disturbances of specied frequencies. Unfortunately, the results of this pioneering experiment were not available at the time of our computation. Neither is such information provided by the experiment of Lachowicz et al. (1996a,b) or the theoretical study of Balakumar and Malik (1994), on which the present numerical experiment is based. Therefore, to construct a viable transition scenario, we must turn to numerical experimentation, for which we exploit parabolized stability equation (PSE) methodology as adapted to compressible ows by Chang et al. (1991). Unlike LST, nonlinear PSE methodology treats both meanow nonparallelism and moderately nonlinear wave interactions. In the PSE approach, disturbances are Fourier-transformed in the periodic temporal and azimuthal dimensions. Each harmonic is associated with an integer pair (l, m), where the integers l and m identify the harmonic with respect to the fundamental values of temporal frequency and azimuthal wave number , respectively. Harmonics (l, m) are axisymmetric if m = 0, oblique (helical) if m = 0, and stationary if l = 0. The Fourier component (0,0) represents the nonlinear distortion of the base state. Because of the azimuthal periodicity of the ow on a cone, the wavenumber of helical modes must satisfy R = n, where n is an integer. For conical bodies, this implies that necessarily varies with x. In the ared region, modes of zero temporal frequency are referred to as Goertler modes and are associated with concave streamwise surface curvature. We mention in passing that, because computations are performed in Fourier space in PSE methodology, subharmonics of fundamental disturbances arise only if explicitly introduced. In contrast, in physical experiments or DNS, subharmonic frequencies may emerge spontaneously from low-level physical or numerical noise. Parameter studies based on nonlinear PSE methodology conrmed, as expected, that a fundamental (1, 0) second-mode disturbance of circular frequency = 2F was highly unstable for frequency F = 230 kHz. Moreover, helical modes of the same frequency were also unstable for a wide range of values of n. In particular, n = R = 15 was shown to lead to highly unstable disturbances. On the basis of this study, a transition scenario was constructed based on the nonlinear interactions of a triad of nite-amplitude Fourier harmonics (1, 0), (1, 1), and (1, -1) (and their complex conjugates), where all three components have the same rms amplitude of 1% in their maximum temperature uctuation (when averaged in time and in the azimuthal dimension). The effect of summing the equal and opposite nite-amplitude helical components (1, 1) and (1, -1) is to produce a spectrum of three-dimensional disturbances, including the (0, 2) harmonic, which is known to promote laminar breakdown. It should be noted that the axisymmetric second-mode (1, 0) component alone is insufcient to trigger transition, an inherently three-dimensional phenomenon. The Fourier amplitudes (not rms) of selected harmonics from the PSE calculation are shown in Figure 7. The PSE computation spanned the region x = 0.58 ft (ReL = 1258), ahead of the are, to x = 1.4 ft (ReL = 1954) in the are zone. At the inow boundary, the calculation was initialized with the disturbance triad dened above. The initial amplitudes of the axisymmetric and helical components were chosen arbitrarily to establish transition on the ared afterbody. All other harmonics shown in the gure emerged spontaneously from the nonlinear interactions among the fundamental triad. Figure 7 shows saturation of the fundamental components of the disturbance near x = 1.1 ft occurring simultaneously with

Figure 7. Streamwise evolution of temperature maxima of selected harmonics of primary disturbances obtained by PSE method. First and second indices identify temporal and spanwise harmonics with respect to fundamental temporal frequency and spanwise wave number , respectively, where R = 15.

58

C.D. Pruett and C.-L. Chang

Figure 8. Growth rates of Goertler modes versus fundamental azimuthal wavenumber n = R at ReL = 1630 (x = 0.974 ft) obtained by linear PSE methodology. (Growth rates are nondimensionalized according to Mack using local boundary-layer length scale.)

signicant distortion of the mean ow (0, 0) and rapid growth of higher harmonics to signicant amplitudes. These conditions indicate that the ow is transitioning from a laminar to a turbulent state. Of particular interest here are the stationary (Goertler) modes (0, m). For the ared cone with an adverse pressure gradient, the PSE method predicts linearly unstable Goertler modes for a broad range of values m. Specically, as shown in Figure 8, Goertler modes are highly unstable for fundamental azimuthal wavenumbers n = R ranging from less than 10 to more than 80. In contrast, Goertler modes are linearly stable for the cone without are. Whereas, stationary harmonics are indeed observed for both the straight cone and the ared cone in the region ahead of the are, they are not eigenfunctions of the ow and thus owe their existence to other instability mechanisms. Specically, such harmonics emerge initially due to linear spatial transient effects, but their growth is sustained only through nonlinear interactions (as observed for the straight cone by Pruett and Chang (1995)). In contrast, true Goertler modes are able to extract energy directly from the mean ow. Given (potentially) different energy transfer mechanisms, signicant qualitative differences in the respective transition processes of straight and ared cones should not be surprising.

6. Results
The computational domain of the DNS encompassed a section along the body downstream of the beginning of the are. Specically, the domain spanned the region
x in = 0.9 x 1.2 = xout ft, 2 (n = 15), 0 n 0 7.5.

(5)

In terms of ReL , the computational domain of the DNS spanned 1567 ReL 1809. To facilitate comparisons of the DNS results with those of Lachowicz et al. (1996a,b), Figure 9 shows the dependence of ReL on x . The length scale (x in ) = 0.00439 in. The time-periodic inow condition at x in = 0.9 was derived from the PSE results at the corresponding station. At this station, the fundamental components ((1, 0), (1, 1), and (1, -1)) of the disturbance had attained relatively large amplitudes, and numerous harmonics of the disturbance showed signicant amplitudes. All signicant harmonics were incorporated into the DNS inow condition. Relative to the wavelength of the fundamental components of the disturbance, the computational domain encompassed approximately 26 wavelengths in streamwise extent. The resolution of the computational grid was 2304 18 128 in the streamwise, azimuthal, and wall-normal directions, respectively. At this streamwise grid density, each fundamental wavelength was resolved by approximately 90 grid points, sufcient, according to Pruett et al. (1995), to resolve at least the rst six streamwise harmonics. The azimuthal resolution, signicantly less than that of Pruett et al. (1995), was sufcient only for the initial stages of laminar breakdown, and was dictated largely by available computational resources. Consequently, the simulation was halted prior to endstage transition.

Direct Numerical Simulation of Hypersonic Boundary-Layer Flow on a Flared Cone

59

Figure 9. Root Reynolds number versus surface arc length for ow conditions of test case. DNS computation spans 1567 ReL 1809, as indicated by vertical dashed lines.

Figure 10. Comparison of DNS and PSE results for selected harmonics.

The DNS computation was performed on the Waterways Experiment Station (WES) Cray C90 at Vicksburg, Mississippi. Approximately 250 hours of single processor time and 115 megawords of main memory were required for the computation. The code is optimized to run in parallel, and during the production runs, we typically requested three CPUs. In physical terms, the computation proceeded for 30 periods of oscillation at the fundamental frequency. The dominant instability waves were observed to propagate downstream at approximately 90% of the boundary-layer edge velocity; consequently, the physical computation time was approximately that necessary for the fundamental instability wave to complete one traverse of the computational domain. During the computation, the numerical solution was periodically written to output and archived at a rate of 64 samples per period of the fundamental frequency. These data were later postprocessed to extract quantitative and qualitative information about the evolution of the ow. Most of the results to follow were obtained by Fourier time-series analysis of the nal period (2930) of the computation. Figure 10, obtained by Fourier transforms in the periodic temporal and azimuthal dimensions, displays the streamwise evolution of the amplitudes of the temperature uctuation of selected harmonics of the DNS calculation. The corresponding PSE results are also provided for comparison. The DNS and PSE results agree well for approximately the rst third of the computational domain of the DNS. Downstream of the rst third, the DNS and PSE results gradually diverge, but the methods continue to indicate agreement in the qualitative behavior of the selected harmonics. Downstream of x = 1.15 ft, some transient effects of the leading wavefront remain; here the DNS results cannot be considered fully developed and should be disregarded. Figures 11 and 12 display the structure (amplitude envelope) of the velocity and temperature components, respectively, of selected harmonics at x = 1.05 ft, the midpoint of the computational domain of the DNS. For comparison, PSE results are also presented. In general, the largest uctuation amplitude in the temperature is experienced at the critical layer, whereas, for the dominant harmonics, velocity uctuations are strongest nearer to the wall. The agreement between DNS and PSE methods can be considered good only in a qualitative sense. The magnitude of the differences comes somewhat as a surprise, given the closer agreement between PSE and DNS obtained by Pruett and Chang (1995) for the cone without are. We have considered a number of possible explanations, none of them satisfactory. Rather than to speculate, we prefer to admit simply that the origin of the disagreement remains unknown. Whatever the explanation, it should be kept in mind that the DNS is initiated well into the transition process, in a region of moderately nonlinear interactions not far upstream of saturation of the fundamental. That PSE and DNS agree even moderately well at this relatively late stage is testament to the potential of PSE as a transition predictor. Figure 13 displays the streamwise evolution of various mean quantities computed by averaging both temporally and spatially (over the azimuthal coordinate). Here, where used, overbars denote averaged

60

C.D. Pruett and C.-L. Chang

Figure 11. Amplitudes of velocity uctuations of selected harmonics versus scaled wall-normal coordinate at x = 1.05 ft.

Figure 12. Amplitudes of temperature uctuations of selected harmonics versus scaled wall-normal coordinate at x = 1.05 ft.

Direct Numerical Simulation of Hypersonic Boundary-Layer Flow on a Flared Cone

61

Figure 13. Streamwise evolution of selected mean quantities.

quantities. For comparison, the corresponding quantities are shown also for the undisturbed initial laminar state. Clear indications of transitioning ow are evident: principally a dramatic drop in the shape factor H = 1 /2 below its laminar value (where 2 is the boundary-layer momentum thickness). We note that the laminar shape factor, which is nearly constant for the cone without are (Pruett and Zang, 1995), diminishes with x in the ared region for the present conguration because of the decrease in Mach number. At this stage of the transition process, the decrease in H is due almost entirely to an increase in momentum thickness 2 rather than to a decrease in displacement thickness 1 . Near the end of the computational domain, the skin-friction coefcient Cf has increased approximately 25% over its laminar value, where Cf =
wall 2wall , 2 e ue u = |z=0 . z

(6)

However, at the same station, the coefcient Ch of thermal stress at the wall has doubled. Taken together, the plots for Cf and Ch call attention to an ambiguity in dening the location x tr of transition onset for high-speed ows. On the basis of minimum wall shear, x 0 . 95 ft. On the basis of minimum heat transfer, tr transition onset apparently occurs upstream of the computational inow boundary of the DNS. Figure 14 presents the streamwise evolution of the principal components of the Reynolds stress tensor, namely u u , v v , and w w , where primes denote uctuations about the mean state. The streamwise and wall-normal stresses are observed to be of nearly equal magnitudes and peaked quite close to the wall. Similarly, turbulent Mach number Mt and (density-weighted) turbulent kineetic energy K (Figure 15) also

62

C.D. Pruett and C.-L. Chang

Figure 14. Streamwise evolution of principal components of Reynolds stress.

Figure 15. Streamwise evolution of turbulent Mach number, turbulent kinetic energy, and rms density uctuation, each averaged temporally and azimuthally.

exhibit peaks close to the wall. (Turbulent kinetic energy is dened as one-half the trace of the Reynolds stress tensor.) The same gure presents evolution of the rms uctuation in density. Large rms density uctuations, as high as 30% of the reference density, are observed at the critical layer and also at the wall. 6.1. Comparisons with the Experiment of Lachowicz et al. Because the paths to transition differ fundamentally for the computational and physical experiments (i.e., forced versus natural transition), caution must be exercised in drawing comparisons and contrasts. However, a few quantitative and qualitative comparisons can be attempted. First, as discussed previously (Figure 2), the experimentally and computationally derived valued of boundary-layer thickness compare very well in

Direct Numerical Simulation of Hypersonic Boundary-Layer Flow on a Flared Cone

63

the essentially laminar region. Second, in Figure 9 of Lachowicz et al. (1996a) rms mass-ux and totaltemperature uctuation maxima are presented for a station that lies in the transition zone. Mass-ux spectra through the boundary layer are also shown in Figure 12 of that paper. For comparison, we present Figure 16, which shows the streamwise evolution of both the raw and normalized rms mass-ux uctuations. The shapes of the computationally and experimentally derived mass-ux proles are remarkably similar, although the computational results attain higher amplitudes (perhaps because energy in the physical ow is distributed over a range of frequencies). Both show peaks at the critical layer, one boundary-layer thickness from the surface. Because of hot-wire calibration difculties near the wall, the experimental results are given only for 2 3 z . (The reader is cautioned that the variable has a different usage in this paper from its usage in Lachowicz et al. (1996a), for which 6 .) Third, in Figure 9 of Lachowicz et al. (1996b), disturbances of frequencies in the 200 kHz range are shown to grow linearly over the region 1880 ReL 2050, in close agreement with LST, and to saturate downstream of ReL = 2050. In particular, the dimensional mass-ux growth rate of the disturbance of 234 kHz is calculated to be 12.9 per foot, compared with a rate of 13.9 predicted by LST. For comparison, Figure 17 presents the streamwise evolution of the (1, 0) component of the fundamental disturbance obtained by PSE methodology. The growth is approximately linear in the region immediately downstream of the body are; however, due to a relatively large initial amplitude, saturation begins considerably further upstream than in the experiment, at approximately ReL = 1700. The average growth rate at the beginning of the are region is approximately 9.2 per foot, somewhat lower than either the experimental and theoretical values, presumably because nonlinear effects are already signicant at the higher amplitude. Fourth, Figure 24 of Lachowicz et al. (1996a) shows signicant energy in the rst harmonic (430 kHz) of the fundamental disturbance (230 kHz) at a station near the end of the model. At this location the amplitude of the rst harmonic is roughly one order of magnitude below that of the fundamental. Similarly, Figure 10 of the present paper shows considerable energy in the rst axisymmetric harmonic (2, 0).

Figure 16. Streamwise evolution of raw and normalized rms mass-ux uctuations. For legend, see Figure 14.

Figure 17. Maximum amplitude versus ReL of temperature of fundamental (1, 0) component of disturbance as obtained by PSE (dashed line) and DNS (solid line). Location corresponding to beginning of ared afterbody identied by the vertical dashed line.

64

C.D. Pruett and C.-L. Chang

Figure 18. Computational ow image of instantaneous wall-normal density gradient showing rope-like appearance of dominant instability waves.

The ratio of the amplitudes of the fundamental to the rst harmonic at the location of saturation of the fundamental is about 5:1. (However, as cautioned by Wilkinson (personal communication), the frequency of the rst harmonic lies considerably outside the bandwidth (350 kHz) of the hot-wire anemometer in the experiment; thus, amplitude data at this frequency cannot be considered accurate.) Finally, we call attention to Figure 3 of Lachowicz et al. (1996a), a schlieren photograph that shows a wave-like appearance at the boundary-layer edge in the ared-cone experiment. Similar structures have been observed in hypersonic boundary-layer ows by several other experimentalists and have been referred to previously as rope-like waves. See, for example, the excellent review by Smith (1994). Computational ow imaging (CFI) of the present numerical experiment also clearly reveals instability waves of rope-like appearance (Figure 18). To approximate crudely the physics involved in schlieren imagery, the instantaneous wall-normal density gradient has been averaged over the azimuthal coordinate and projected as a twodimensional image in greyscale. The numerically derived image is qualitatively similar in appearance to the actual schlieren image obtained by Lachowicz et al. (1996a). In particular, the wavelength of the instability is very nearly twice the boundary-layer thickness, in keeping with the nature of second-mode disturbances and the observations of Lachowicz et al. (1996a). 6.2. Comparisons with the Computations of Pruett and Coworkers It is well known from experience with incompressible and low-speed boundary-layer ows that an adverse pressure gradient destabilizes Goertler modes, and Goertler instability was anticipated for the present aredcone test case. This suspicion was veried by PSE methodology; Goertler modes were indeed found to be linearly unstable for a broad range of n = R (Figure 8). In contrast, Goertler modes are linearly stable for the cone without are and grow only through nonlinear interactions. As shown in Figure 11, the amplitude of the streamwise velocity uctuation of the principal Goertler harmonic (0, 1) is strongly peaked close to the wall. Surprisingly, the fundamental (1, 0) and (1, 1) components also exhibit their strongest streamwise velocity uctuations very close to the wall. In contrast, Pruett and Chang (1995) observe the (1, 0) harmonic to have approximately equal peaks in velocity amplitude near the wall and near the critical layer. Perhaps the most profound difference between the present results and previous computational experiments on the straight cone concerns the rms density uctuation. For the ared cone, density uctuations as large as 30% of the reference density are observed at the wall (Figure 15), whereas in Figure 7 of Pruett and Zang (1995), duplicated here for completeness as Figure 19, density uctuations are nonzero but small at the wall. Both congurations show rms density uctuation peaks near the critical layer. We speculate that the large density uctuation near the wall is driven by the streamwise gradient in mean wall temperature in the are region (Figure 20). The differences in velocity and density uctuations between the straight-cone and ared-cone cases apparently combine to result in a striking contrast in the behavior of the Reynolds stresses relative to the observations of Pruett and Zang (1995). As observed in Figures 8 and 9 of Pruett and Zang, the dominant peaks in Reynolds stress occur initially about a displacement thickness from the wall and gradually migrate toward the wall as nonlinear effects become stronger. In distinct contrast, as shown in Figures 14 and 15 of the present paper, the turbulent kinetic energy develops and remains close to the wall along the ared cone.

Direct Numerical Simulation of Hypersonic Boundary-Layer Flow on a Flared Cone

65

Figure 19. Streamwise evolution of temporally and azimuthally averaged rms density uctuation for cone without are adapted from Figure 7 of Pruett and Zang (1995). Note: Reynolds numbers here are based on local (not freestream) reference values and do not correspond directly with (2).

Figure 20. Laminar adiabatic-wall temperature versus arc length.

We also observe that, for the cone without are, the Reynolds stresses are dominated by their streamwise component, whereas along the ared cone the streamwise and wall-normal stresses are of nearly equal magnitudes. It is also instructive to compare the present results with the earlier computational studies of Pruett and Zang (1995), who examined the laminar breakdown of hypersonic boundary-layer ow on axisymmetric bodies by means of temporal, rather than spatial, DNS. In particular, they considered external ows of Mach 4.5 along a cylinder and Mach 6.8 along a cone. Initial disturbances in the form of primary and subharmonic secondary instability waves were used to trigger transitional states. Despite the fact that the transition mechanism was fundamentally different from that later used by Pruett and Zang (1995) for spatial DNS, the qualitative nature of the transition processes for both the Mach 4.5 and Mach 6.8 test cases was similar to that described above for the cone without are. These results (and others) strongly suggest that the qualitative nature of a transitioning hypersonic boundary-layer ow is not terribly sensitive to changes in either Mach number or moderate transverse curvature (Pruett et al., 1991). Moreover, once triggered, the transition process seems to be relatively insensitive to the triggering mechanism. In contrast, the present results suggest that nonzero streamwise curvature has a marked effect on the nature of transition. In summary, transition is qualitatively different for the straight and ared cones. Because most of the turbulent activity is conned initially to the critical layer in the boundary layer of the straight cone, transition onset, as manifest by surface changes to the mean state, is delayed but relatively abrupt. As shown in Figures 6 and 11 of Pruett and Zang (1995), signicant changes in skin friction and shape factor are evident only considerably downstream of the station where the fundamental harmonics experience saturation. In contrast, for the ared cone, signicant deviations from laminar values are observed to occur in shape factor, skin friction, and thermal stress at stations upstream of saturation of the fundamental. Qualitatively, it appears that transition on the ared cone is, for lack of a better term, mushy. Transition onset occurs earlier, but the transition process is gradual. 6.3. Comparisons with the Experiment of Kimmel Although the results of Kimmel (1993) are admittedly inconclusive, and the present computational results are admittedly incomplete, we attempt to draw a parallel. As mentioned previously, because different researchers use different measures of transition onset, it is difcult to draw denitive comparisons between computation and experiment, and in our judgement, the transition community would benet by consensus in the denition of transition onset for high-speed boundary-layer ows. However, our overall impression that, relative to the straight cone, transition onset occurs earlier on the ared-cone, is consistent with Kimmels observation that

66

C.D. Pruett and C.-L. Chang

an adverse pressure gradient moves the onset of transition forward. Moreover, the impression that transition occurs more gradually on the ared cone is not inconsistent with his observations that favorable pressure gradients cause a shorter transition zone and that adverse pressure gradients may possibly lengthen the transition zone.

7. Conclusions
The laminarturbulent transition of the boundary layer on an axisymmetric ared cone in Mach 6 ow has been simulated by nonlinear PSE and spatial DNS techniques. The full effects of streamwise surface curvature have been taken into account in the governing equations and boundary conditions. These effects include adverse streamwise pressure gradient, decreasing edge Mach number, decreasing boundary-layer thickness, and increasing wall temperature. Wherever possible, results from the present computation have been compared with those obtained from previous physical and numerical experiments. Physical experiments (Lachowicz et al., 1996a,b), N-factor studies based on compressible LST (Balakumar and Malik, 1994), nonlinear PSE methodology, and DNS all conrm that disturbances of very high frequency (on the order of 230 kHz), provide the primary instability mechanism that precipitates transition. Disturbances in this frequency range are classied as second-mode disturbances in the nomenclature of Mack (1984). A further indication that the primary disturbances are of second-mode type is the appearance of so-called rope-like waves in schlieren images from the physical experiment or in computational ow imagery from the numerical experiment. These waves manifest a wavelength that is approximately twice the boundary-layer thickness, in keeping with the predictions of LST for second-mode disturbances. Because an adverse pressure gradient results in diminishing boundary-layer thickness in the are region, the disturbance frequencies observed on the ared cone are higher than for a cone without are. The results of the present numerical ared-cone experiment differ fundamentally from those of Pruett and Chang (1995), who simulated laminar breakdown on a right circular cone without a ared afterbody. In particular, for the present conguration, the dominant Reynolds stress components, turbulent kinetic energy, and turbulent Mach number all manifest maxima close to the wall, rather than near the critical layer as observed previously. In addition to a large uctuation amplitude in density at the critical layer, as observed also for the straight cone, large density uctuations are observed at the wall for the ared-cone conguration. Moreover, for the ared-cone, Goertler modes of instability (i.e., harmonics of zero frequency) are shown to be linearly unstable, whereas they are stable for the cone without are. In general, these differences result in qualitatively different transition processes on straight and ared cones. Relative to the straight cone, transition on the ared cone appears to begin earlier and to proceed more gradually. The present computational results would seem to be consistent with the experimental results of Kimmel (1993).
Acknowledgement. The rst author is indebted to Dr. Jason Lachowicz, for helpful discussions and data from the experiment of Lachowicz et al. (1996a,b), and to Dr. Steve Wilkinson of NASA Langley Research Center, for many clarifying comments regarding the Langley quiet-tunnel experiments.

References
P. Balakumar and M.R. Malik, Effect of Adverse Pressure Gradient and Wall Cooling on Instability of Hypersonic Boundary Layers, High Technology Report No. HTC-9404, 1994. A.E. Blanchard and G.V. Selby, An Experimental Investigation of Wall Cooling Effects on Hypersonic Boundary-Layer Stability in a Quiet Wind Tunnel, NASA Contractor Report 198287, 1996. A.E. Blanchard, J.T. Lachowicz, and S.P. Wilkinson, NASA Langley Mach 6 Quiet Wind-Tunnel Performance, AIAA J., Vol. 35, No. 1, 1997, pp. 2328. C. Canuto, A. Quarteroni, M.Y. Hussaini, and T.A. Zang, Spectral Methods in Fluid Dynamics, Springer-Verlag, Berlin, 1988. D. Cavalieri, On the Experimental Design for Instability Analysis on a Cone at Mach 3.5 and 6 Using a Corona Discharge Perturbation Method, M.Sc. Thesis, Illinois Institute of Technology, 1995. C.-L. Chang, M.R. Malik, G. Erlebacher, and M.Y. Hussaini, Compressible Stability of Growing Boundary Layers Using Parabolized Stability Equations, AIAA Paper No. 91-1636, 1991.

Direct Numerical Simulation of Hypersonic Boundary-Layer Flow on a Flared Cone

67

G.P. Doggett, N. Chokani, and S.P. Wilkinson, Effect of Angle of Attack on Hypersonic Boundary-Layer Stability, AIAA J., Vol. 35, No. 3, 1997. R.L. Kimmel, Experimental Transition Zone Lengths in Pressure Gradient in Hypersonic Flow, in Transitional and Turbulent Compressible Flows, eds. L.D. Kral and T.A. Zang, FED, Vol. 151, ASME 1993, New York, pp. 117128. J.T. Lachowicz, N. Chokani, and S.P. Wilkinson, Hypersonic Boundary-Layer Stability Over a Flared Cone in a Quiet Tunnel, AIAA Paper No. 96-0782, 1996a. J.T. Lachowicz, N. Chokani, and S.P. Wilkinson, Boundary-Layer Stability Measurements in a Hypersonic Quiet Tunnel, AIAA J., Vol. 34, No. 12, 1996b, pp. 24962500. S.K. Lele, Compact Finite-Difference Schemes with Spectral-Like Resolution, J. Comput. Phys., Vol. 103, 1992, pp. 1642. L.M. Mack, Boundary-Layer Stability Theory, in Special Course on Stability and Transition of Laminar Flow, ed. R. Michel, AGARD Report No. 709, 1984, pp. 3.13.81. C.D. Pruett, On the Accurate Prediction of the Wall-Normal Velocity in Compressible Boundary-Layer Flow, Internat. J. Numer. Methods Fluids, Vol. 16, 1993, pp. 133152. C.D. Pruett and C.-L. Chang, Spatial Direct Numerical Simulation of High-Speed Boundary-Layer FlowsPart II: Transition on a Cone in Mach 8 Flow, Theoret. Comput. Fluid Dynamics, Vol. 7, No. 5, 1995, pp. 397423. C.D. Pruett and C.L. Streett, A Spectral Collocation Method for Compressible, Non-Similar Boundary Layers, Internat. J. Numer. Methods Fluids, Vol. 13, No. 6, 1991, pp. 713737. C.D. Pruett and T.A. Zang, Direct Numerical Simulation of Laminar Breakdown in High-Speed Axisymmetric Boundary Layers, Theoret. Comput. Fluid Dynamics, Vol. 3, No. 6, 1992, pp. 345367. C.D. Pruett and T.A. Zang, On Simulation and Analysis of Instability and Transition in High-Speed Boundary-Layer Flows, Comput. Systems Engrg., Vol. 6, No. 6, 1995, pp. 563575. C.D. Pruett, L.L. Ng, and G. Erlebacher, On the Nonlinear Stability of a High-Speed Axisymmetric Boundary Layer, Phys. Fluids A, Vol. 3, No. 12, 1991, pp. 29102926. C.D. Pruett, T.A. Zang, C.-L. Chang, and M.H. Carpenter, Spatial Direct Numerical Simulation of High-Speed Boundary-Layer Flows Part I: Algorithmic Considerations and Validation, Theoret. Comput. Fluid Dynamics, Vol. 7, No. 1, 1995, pp. 4976. L.G. Smith, Pulsed-Laser Schlieren Visualization of Hypersonic Boundary-Layer Instability Waves, AIAA Paper No. 94-2639, 1994. K.F. Stetson and R.L. Kimmel, On Hypersonic Boundary-Layer Stability, AIAA Paper No. 92-0737, 1992. K.F. Stetson and R.L. Kimmel, On the Breakdown of a Hypersonic Laminar Boundary Layer, AIAA Paper No. 93-0896, 1993. K.F. Stetson, E.R. Thompson, J.C. Donaldson, and L.G. Siler, Laminar Boundary-Layer Stability Experiments on a Cone at Mach 8. Part 1: Sharp Cone, AIAA Paper No. 83-1761, 1983. C.L. Streett and M.G. Macaraeg, Spectral Multi-Domain for Large-Scale Fluid Dynamic Simulations, Appl. Numer. Math., Vol. 6, 1989/90, pp. 123139. K.W. Thompson, Time Dependent Boundary Conditions for Hyperbolic Problems, J. Comput. Phys., Vol. 68, 1987, pp. 124. F.M. White, Viscous Fluid Flow, McGraw-Hill, New York, 1974, pp. 347350. S.P. Wilkinson, A Review of Hypersonic Boundary-Layer Stability Experiments in a Quiet Mach 6 Wind Tunnel, AIAA Paper No. 97-1819. Presented at the 28th Fluid Dynamics Conference, Snomass, CO, June 29 to July 2, 1997. J.H. Williamson, Low-Storage RungeKutta Schemes, J. Comput. Phys., Vol. 35, 1980, pp. 4856.

Anda mungkin juga menyukai