Professional Documents
Culture Documents
Sdarticle
Sdarticle
Sdarticle
www.elsevier.com/locate/actaastro
Received 17 June 2008; received in revised form 21 December 2008; accepted 11 February 2009
Available online 2 April 2009
Abstract
Operation pattern recognition/classification is one of the most important issues of the hypersonic inlet and is also the foundation
of protection control of a scramjet. To solve this problem, the 2-D inner steady flow of the hypersonic inlet was numerically
simulated at different freestream conditions and backpressures, and two different inlet unstart phenomena were analyzed. The
pattern classifications of hypersonic inlet start/unstart, especial the inlet backpressure unstart/low Mach number unstart were
performed based on “numerical experimental” data by the support vector machine algorithms. The physical significance of the
classification criterions was explained, and the discussions of the classification criterion used in NASA/CIAM flight test were
presented. In conclusion, the operation pattern classification of hypersonic inlet can be solved well by using the support vector
machine algorithms.
Crown Copyright © 2009 Published by Elsevier Ltd. All rights reserved.
0094-5765/$ - see front matter Crown Copyright © 2009 Published by Elsevier Ltd. All rights reserved.
doi:10.1016/j.actaastro.2009.02.011
458 J. Chang et al. / Acta Astronautica 65 (2009) 457 – 466
Nomenclature
k turbulent kinetic energy p2 static pressure at x2 , pressure under the lip
turbulent dissipation rate p3 static pressure at x3 , pressure at the exit of
M∞ Mach number of the freestream the isolator
T∞ static temperature of the freestream p4 static pressure, shown in Table 3
p∞ static pressure of the freestream p5 static pressure, shown in Table 3
angle of attack of the freestream p6 static pressure, shown in Table 3
ps surface static pressure of the inlet pmin mass-weighted averaged static pressure at
x axis location of the inlet the entrance of the isolator
y vertical location of the inlet pmax static pressure larger than the maximal sus-
x1 shown in Fig. 1 tainable backpressure of the isolator
x2 shown in Fig. 1 pb static pressure at the exit of the isolator
x3 shown in Fig. 1 mass-captured coefficient
p1 static pressure at x1 , pressure behind total-pressure recovery coefficient
of the first oblique shock
backpressure of the isolator, etc.), and the impact of [13] analyzed the major factors that influence the inlet
these factors individually or jointly make the hypersonic operation modes and studied the start/unstart character-
inlet be different operation modes (inlet start, low Mach istic of 2-D hypersonic inlet. Reinartz [14] and Saied
unstart, backpressure unstart, etc.). The state detection Emami [15] studied the variation of isolator geometry
is the base of the unstart control of the hypersonic in- and its influence on the overall inlet compression effi-
let, where the state detection is referred to the pattern ciency. Tan [16] investigated the flow pattern of a hy-
recognition/classification of different hypersonic inlets personic inlet at an unstable–unstarted condition exper-
operation modes. The different operation modes of hy- imentally, and analyzed the oscillations of the shock
personic inlets must be distinguished by unstart control system and the unsteady process of the duct flow. Wag-
system, otherwise the control system can not work ef- ner [17] studied the unstart dynamics of an inlet/isolator
fectively, such as the inlet in low Mach number unstart model at a Mach 5 flow experimentally. However, few
cannot be restarted by only regulating the backpressure. numerical and experimental investigations have been
So the classification of different hypersonic inlet opera- reported on the detection of hypersonic inlet opera-
tion modes, especial the classification of the inlet back- tion modes. The detection [18] has been proposed and
pressure unstart/low Mach number unstart, is necessary several measurement techniques provided useful infor-
to the unstart control of the hypersonic inlet. mation by which to evaluate the inlet operating mode,
Unstart phenomena of the hypersonic inlet have been that is the flow visualization technique using the shad-
very active fields of research in last decades, there have owgraph/schlieren windows, and the visualization tech-
been many investigations devoted to this subject [5–19]. nique is complex and not practical to the flight test. Yu
Mayer and Paynter [5,6] used an Euler solver and sim- [19] discussed the classification of the inlet start/unstart
ulated an axisymmetric inlet unstart due to the varia- based on “numerical experimental” data, while the clas-
tion of freestream variables such as temperature, veloc- sification of the inlet unstart (backpressure unstart and
ity and pressure. Neaves and McRae [7] developed dy- low Mach number unstart) has not been studied.
namic solution-adaptive gird algorithm introduced by In this paper, the study is focused on the operation
Benson and McRae [8] and simulated the 3-D inlet un- pattern classification of hypersonic inlets based on the
start caused by a combustor perturbation. Zha [9,10] in- simulation results, especially the classification of hyper-
vestigated unstart transient mechanism of a typical ax- sonic inlet unstart (backpressure unstart and low Mach
isymmetric HSCT (high speed civil transport) inlet at number unstart). To establish the classification crite-
angle of attack using CFD. Cox [11] presented several rions, the inlet unstart phenomena (backpressure un-
mechanisms of the hypersonic inlet unstart, including start and low Mach number unstart) were studied firstly.
backpressure unstart, overcontraction unstart and angle Based on the “numerical experiment” data, the SVM
of attack unstart. The second approach is based on the (support vector machine) algorithms are introduced to
experimental investigation. Schmitz [12] and Van Wie solve the operation pattern classification of hypersonic
J. Chang et al. / Acta Astronautica 65 (2009) 457 – 466 459
discrepancy between the three mesh levels is less than Comparison of a schlieren picture [14,21] without throt-
5 percent, where the performance parameter is referred tling and corresponding Mach number contour lines of
to the one (mass-weighted averaged) at the exit of the the computation is shown in Fig. 3, and it reveals an
isolator. Out of this analysis, the medium grid was overall good agreement. The shock wave pattern, the
selected, and all results shown are computed applying separation, and the approximate boundary-layer thick-
this resolution. The use of a medium grid resolution ness of the schlieren picture are also present in the simu-
greatly saves the CPU time. lation results. The surface pressure distributions shown
The accuracy of the current numerical investigation is in Fig. 4, allow for a more quantitative comparison be-
evaluated by comparison with the experimental results. tween numerical and experimental results. Here, a dis-
crepancy in the ramp pressure distribution can be seen
in the expansion region with subsequent separation. The
computed separation appears smaller than experimen-
tally observed, and thus, the separation shock is weaker
and impinges downstream of the measured location on
the cowl surface. The reason for the discrepancy is prob-
ably the deficiency of the turbulence model, the differ-
ences between experiment and computation conditions,
or the measurements error of the sensor. In a word, the
computation results of the hypersonic inlet accord with
Table 2
Performance parameter of hypersonic inlets at different grid-
refinement levels.
Grid level p3 /p0
Fig. 3. Comparison of a schlieren picture (bottom) without throttling and corresponding Mach number contour lines (top) of the computation.
J. Chang et al. / Acta Astronautica 65 (2009) 457 – 466 461
3.1. Backpressure unstart pb /p ∞ is achieved. After even a slight shift of the oper-
ating point (pb /p ∞ = 37.5), the pressure rise is pushed
Firstly, we discuss the backpressure unstart of hyper- forward into the contracting part of the inlet. A fur-
sonic inlets. In scramjet, a precombustion shock system ther increase of the backpressure causes a severe flow
is developed inside of the isolator because of the sub- blockage and results in a strong decrease of the cap-
sequent high-pressure combustion zone. To produce a tured mass flow. In this condition, the inlet flowfield is
similar shock wave system in the test, the effect of the unstable and the inlet is no longer started. This class of
operating engine is simulated by a specified backpres- unstart phenomenon is the backpressure unstart.
sure. Fig. 5 shows surface pressure distributions at dif-
ferent backpressure ratios. The high backpressure leads
3.2. Low Mach number unstart
to the separation of the boundary layer. The pressure
buildup proceeds continuously due to the rapidly grow-
Then we analyze the low Mach number unstart of
ing boundary layer. As the backpressure increases, the
hypersonic inlets. The internal contraction ration of
onset of pressure buildup moves upstream into the isola-
the hypersonic inlet is about 1.9, which is above the
tor. At pb /p ∞ = 37.5, the complete isolator contributes
Kantrowitz limit for self-starting. Fig. 6 shows the
to the pressure buildup and the maximum pressure ratio
variations of mass-captured coefficient, total-pressure
recovery coefficient and kinetic energy efficiency with
Mach number of freestream, where the performance
index is referred to the one (mass-weighted averaged)
at the exit of the isolator. The Mach number contours
of hypersonic inlets at different M ∞ are shown in
Fig. 7. As M ∞ decreases, the mass-captured coeffi-
cient and kinetic energy efficiency gradually decreases
and the total-pressure recovery coefficient increases.
When M ∞ decreases to Mach 3.2, the performance
parameter varies abruptly because the separated flow
appears and the backward shock is formed, thus the
inlet is no longer started. This class of unstart phe-
nomenon is the low Mach number unstart, and the
starting Mach number of the hypersonic inlet is about
3.2 at the attack of angle zero.
Fig. 5. Surface pressure distributions with different backpressures, There probably exist two stable solutions at the same
M ∞ = 4 and = 0. boundary conditions, at this condition the solutions
462 J. Chang et al. / Acta Astronautica 65 (2009) 457 – 466
Table 3
Axis location of feature attributions.
Table 4
Axis location of feature attributions.
Fig. 12. Train and test of the classification between hypersonic inlet Fig. 13. p6 and p5 at different M ∞ , p ∞ and .
start and unstart.
[5] D. Mayer, G.C. Paynter, Prediction of supersonic inlet unstart [18] W.R. Hawkins, E.J. Marquart, Two-dimensional generic inlet
caused by freestream disturbances, AIAA Journal 33 (2) (1995) unstart detection at Mach 2.5–5.0, AIAA Paper 1995-6019,
266–275. April 1995.
[6] D. Mayer, G.C. Paynter, Boundary conditions for unsteady [19] D. Yu, J. Chang, W. Bao, Z. Xie, Optimal classifications
supersonic inlet analyses, AIAA Journal 32 (6) (1994) criterions of hypersonic inlet start/unstart, Journal of Propulsion
1200–1206. and Power 23 (2) (2007) 310–316.
[7] M.D. Neaves, D.S. McRae, J.R. Edwards, High-speed inlet [20] D. Drikakis, Advances in turbulent flow computations using
unstart calculations using an implicit solution adaptive mesh high-resolution methods, Progress in Aerospace Science (39)
algorithm, AIAA Paper 2001-0825. (2003) 405–424.
[8] R.A. Benson, D.S. McRae, Numerical simulations of the unstart [21] C.D. Herrmann, W.W. Kosvhel, Experimental investigation of
phenomenon in supersonic inlet/diffuser, AIAA Paper 1993- the internal compression of a hypersonic intake, AIAA Paper
2239. 2002–4130.
[9] G.C. Zha, D. Knight, D. Smith, Numerical investigations of [22] J. Chang, D. Yu, W. Bao, L. Qu, Dimensionless analysis of
high speed civil transport inlet unstart transient at angle of the unstart boundary for 2-D mixed hypersonic inlets, The
attack, Journal of Aircraft 35 (6) (1998) 851–856. Aeronautical Journal 112 (1135) (2008) 547–555.
[10] G.C. Zha, D. Knight, D. Smith, Numerical simulation of high [23] T. Cui, D. Yu, J. Chang, W. Bao, Topological geometry
speed civil transport inlet operability with angle of attack, interpretation of hypersonic inlet start/unstart-catastrophe,
AIAA Journal 36 (7) (1998) 1223–1229. hysteresis and bifurcation, AIAA Journal of Aircraft 45 (4)
[11] C. Cox, C. Lewis, R. Pap, Prediction of unstart phenomena in (2008) 1464–1468.
hypersonic aircraft, AIAA Paper 1995-6018. [24] H. Liu, L. Yu, M. Dash, H. Motoda, Active feature
[12] D.M. Schmitz, N.C. Bissinger, Design and testing of fixed- selection using classes, Pacific–Asia Conference on Knowledge
geometry hypersonic intakes, AIAA Paper 1998-1529. Discovery and Data Mining (2003) 474–485.
[13] D.M. Van Wie, F.T. Kwok, Starting characteristics of supersonic [25] M. Dash, H. Liu, Feature selection for classification, Intelligent
inlets, AIAA Paper 1996-2914. Data Analysis 1 (3) (1997) 131–156.
[14] B.U. Reinartz, C.D. Herrmann, Aerodynamic performance [26] Q. Hu, D. Yu, Z. Xie, Information-preserving hybrid data
analysis of a hypersonic inlet isolator using computation and reduction based on fuzzy-rough techniques, Pattern Recognition
experiment, Journal of Propulsion and Power 19 (5) (2003) Letters 27 (5) (2006) 414–423.
868–875. [27] I. Guyon, N. Matic, V. Vapnik, Discovering Informative Patterns
[15] S. Emami, C.A. Trexler, Experimental investigation of inlet- and Data Cleaning, Advances in Knowledge Discovery and
combustor isolators for a dual-mode scramjet at a Mach number Data Mining, MIT Press, Cambridge, MA, 1996 pp. 181–203.
of 4, NASA Technical Paper 3502, May 1995. [28] I. Guyon, J. Weston, S. Barnhill, V. Vapnik, Gene selection for
[16] H. Tan, R. Guo, Experimental study of the unstable–unstarted cancer classification using support vector machines, Machine
condition of a hypersonic inlet at mach 6, Journal of Propulsion Learning 46 (1–3) (2002) 389–422.
and Power 23 (4) (2007) 783–788.
[17] J.L. Wagner, A. Valdivia, An experimental investigation of
supersonic inlet unstart, AIAA Paper 2007-4352, June 2007.