cavity ignition in supersonic flow by spark discharge and ...cavity ignition in supersonic ﬂow by...

Available online at www.sciencedirect.comProceedings

ScienceDirect

Proceedings of the Combustion Institute 35 (2015) 2101–2108

www.elsevier.com/locate/proci

of the

CombustionInstitute

Cavity ignition in supersonic flow by sparkdischarge and pulse detonation

Timothy M. Ombrello a,⇑, Campbell D. Carter a, Chung-Jen Tam b,Kuang-Yu Hsu c

a U.S. Air Force Research Laboratory, Wright-Patterson Air Force Base, OH 45433, USAb Taitech, Beavercreek, OH 45430, USA

c Innovative Scientific Solutions, Inc., Dayton, OH 45459, USA

Available online 18 August 2014

Abstract

Ignition of an ethylene fueled cavity in a supersonic flow was achieved through the application of twoenergy deposition techniques: a spark discharge and pulse detonator (PD). High-frequency shadowgraphand chemiluminescence imaging showed that the spark discharge ignition was passive with the ignition ker-nel and ensuing flame propagation following the cavity flowfield. The PD produced a high-pressure andtemperature exhaust that allowed for ignition at lower tunnel temperatures and pressures than the sparkdischarge, but also caused significant disruption to the cavity flowfield dynamics. Under certain cavity fuel-ing conditions a multiple regime ignition process occurred with the PD that led to decreased cavity burningand at times cavity extinction. Simulations were performed of the PD ignition process, capturing thedecreased cavity burning observed in the experiments. The PD exhaust initially ignited and burned the fuelwithin the cavity rapidly. Simultaneously, the momentary elevated pressure from the detonation caused ablockage of the cavity fuel, starving the cavity until the PD completely exhausted and the flowfield couldrecover. With sufficiently high cavity fueling, the decrease in burning during the PD ignition process couldbe mitigated. Cavity fuel injection and entrainment of fuel through the shear layer from upstream injectionallowed for the spark discharge ignition process to exhibit similar behavior with peaks and valleys of heatrelease (but to a lesser extent). The results of using the two energy deposition techniques emphasized theimportance of cavity fueling and flowfield dynamics for successful ignition.Published by Elsevier Inc. on behalf of The Combustion Institute.

Keywords: Scramjet; Cavity ignition; Supersonic combustion; Pulse detonation

http://dx.doi.org/10.1016/j.proci.2014.07.0681540-7489/Published by Elsevier Inc. on behalf of The Combu

⇑ Corresponding author. Address: 1950 Fifth Street,Building 18A, Wright-Patterson AFB, OH, USA. Fax:+1 937 656 4659.

E-mail address: [email protected] (T.M.Ombrello).

1. Introduction

Successfully igniting high-speed air-breathingcombustors, such as supersonic combustionramjets, is a significant challenge because of therestrictive reactive environment. While the limitedresidence times in these combustors can be miti-gated through the use of cavities and struts [1,2],

stion Institute.

http://dx.doi.org/10.1016/j.proci.2014.07.068mailto:[email protected]://dx.doi.org/10.1016/j.proci.2014.07.068http://crossmark.crossref.org/dialog/?doi=10.1016/j.proci.2014.07.068&domain=pdf

Fig. 1. Schematic of cavity based flow path.

2102 T.M. Ombrello et al. / Proceedings of the Combustion Institute 35 (2015) 2101–2108

the low pressures and temperatures in the super-sonic flow at takeover flight speeds (Mach num-bers

T.M. Ombrello et al. / Proceedings of the Combustion Institute 35 (2015) 2101–2108 2103

flow), a step that dramatically improves imagequality. Chemiluminescence (primarily from CH*

and C2*, with soot emission inhibited by highfuel–air mixing rates) was collected from the samefield of view as the shadowgraph.

A simple automotive style spark ignition sys-tem and plug was used to provide a single10 mJ/pulse of energy for the spark discharge.The PD was comprised of a 61-cm-long and1.03-cm-ID stainless steel tube with compressionfittings. The fuel (C3H8) and oxidizer (N2O) werechosen to produce detonations in the small diam-eter tube and at the static pressure in the tunnel.The PD tube was overfilled with the reactive mix-ture to ensure that a detonation reached the cav-ity. The excess reactants that were injected intothe cavity from overfilling prior to the detonationemerging were found to not change the cavityignition process. The reactive mixture in the PDwas ignited with the same automotive spark dis-charge and each firing produced a detonationwave with near Chapman–Jouguet conditions. Incontrast to the spark discharge, the energyreleased from each detonation was �100 J. Whilethe PD could be operated at greater than 10 Hzrepetition rate, single firings were used for the cav-ity ignition study. More details are provided inRef. [10].

3. Results

3.1. Shadowgraph and chemiluminescence

The spark discharge and PD produced verydifferent cavity ignition processes because of howthey deposit energy and interact with the flow.The ignition process with a spark discharge reliesheavily upon the cavity flowfield to spread the ini-tial flame kernel and therefore is classified as apassive device. On the other hand, a detonationis disruptive to the cavity flow but still relies uponthe cavity dynamics to sustain the flame. Becauseof the interaction of the ignition devices with theflow, the cavity burning depended strongly onthe fueling condition. To best capture the ignitionprocess with these two devices, shadowgraphyallowed for interrogation of the density fluctua-tions and chemiluminescence imaging provided apseudo marker of the heat release. While the mea-surements were not taken simultaneously, goodrepeatability between each ignition sequenceallowed for meaningful comparison between theshadowgraph and chemiluminescence imaging.

A representative set of images from the high-frame-rate videos for both ignition devices areshown in Fig. 2 with a fueling rate from the slotupstream of the cavity (Qslot) of 144 slpm (stan-dard liters/min) and 104 slpm from the holes inthe cavity closeout ramp (Qcavity). Chemilumines-cence images are displayed in false-color to

provide good contrast. For the spark discharge,the ignition kernel is present in the shadowgraphimage by 0.01 ms based on the observed densitygradients but is difficult to see with chemilumines-cence because, presumably, the heat release is low.By 0.09 ms the ignition kernel had moved towardsthe front of the cavity, following the recirculationpattern, but had grown little. In stark contrast,when the PD was fired into the cavity under thesame conditions, the cavity disruption was dra-matic. Within 0.01 ms, the detonation plume hadreached the shear layer because of the high veloc-ity (�2000 m/s) exiting the tube. While the deto-nation had significant momentum flux that filledthe cavity rapidly, it was isolated to the cavityand downstream (i.e., little upstream propaga-tion). Furthermore, the initial detonation plumereached halfway across the core flow above thecavity by 0.09 ms.

Later in time, the distinct difference in the cav-ity ignition process becomes clearer (Fig. 3).Within 1 ms, the ignition kernel from the sparkdischarge had grown and moved to the front ofthe cavity and was being entrained into the shearlayer. This allowed for rapid circulation of thecombustion products towards the end of the cav-ity and then down the ramp to ignite the fuel by 2–2.5 ms. At that point, the cavity gases had com-pleted one cycle, and there was significant heatrelease. In shadowgraph images, this was shownby the thickening and movement of the shear layerthat created a shock at the front edge of the cavity(starting at �1–1.5 ms). By 3 ms, the cavityappeared to be in a stable/steady state burningprocess with a shear layer flame, as well as burn-ing within the cavity. On the other hand for thePD, the strong detonation plume had significantlydecayed by 1 ms, and there was the appearance ofcavity burning and a strong shock at the frontedge of the cavity. By 3 ms, the cavity burninghad decreased and the PD was nearing the endof its exhaust process. This resulted in less heatrelease in the cavity, as shown by the weakershock at the front edge of the cavity. At this point,the large disruption from the detonation had sig-nificantly affected the cavity cycling and thereforechanged the burning process.

The cavity burning for the spark dischargechanged little between 3 and 4 ms (Figs. 3 and 4)and also did not change significantly thereafter.Therefore, 3 ms after the spark discharge, the cav-ity had settled into a quasi-stable burning processthat appeared to be independent of the ignitionevent. This makes sense since the cavity had cycleda couple of times (� 1–1.5 ms/cycle). For the PD,on the other hand, the cavity burning steadilydecreased between 3 and 5 ms and indeed appearednearly extinguished at 5 ms before achieving quasi-stable burning at approximately 8 ms. While thecavity was successfully ignited using both devices,the processes were vastly different.

Fig. 2. Shadowgraph and chemiluminescence images of spark discharge and PD cavity ignition for Qslot/Qcavity = 144/104 slpm from 0.01 to 0.09 ms.

Fig. 3. Shadowgraph and chemiluminescence images of spark discharge and PD cavity ignition for Qslot/Qcavity = 144/104 slpm from 1 to 3 ms.


In an attempt to go beyond simple qualitativeimages, the total chemiluminescence from the cav-ity versus time was plotted. A control volume ofthe cavity only was used for the measurement,and Fig. 5 shows the results for the spark dis-charge and the PD with Qslot/Qcavity = 0/66 slpm.The images were normalized to the peak chemilu-minescence coming from the PD. The total emis-sion, and hence heat release in the cavity,followed different paths for the two ignitiondevices, but it converged to the same level by

approximately 8 ms. The heat release from sparkdischarge ignition was gradual with a local peakat 4 ms, while the PD had five distinct regimes(shown by the numbering in Fig. 5). Initially therewas significant emission (regime I), from PDchemiluminescence and soot incandescence, fol-lowed by a rapid decay and then a diminishedand relatively constant level of emission (regimeII). The duration of regime II was approximatelyone cavity cycle time that led to a “cavity ignitiondelay time.” In regime III, there was a rapid

Fig. 4. Chemiluminescence images of spark discharge and PD cavity ignition for Qslot/Qcavity = 144/104 slpm from 4 to12 ms.

Fig. 5. Average integrated chemiluminescence in cavityduring ignition versus time with Qslot/Qcavity = 0/66 slpmfor spark discharge and PD.

Fig. 6. Average integrated chemiluminescence in cavityduring ignition versus time for PD at different Qslot andQcavity.


increase in heat release from burning the fuel inthe cavity. Once the fuel was consumed, therewas a long lull in integrated emission of 3–4 ms(regime IV) that was needed to replenish a com-bustible mixture. At that point (regime V), theheat release in the cavity increased to a quasi-steady value that was independent of the ignitionevent.

When different fuel flow rates in the cavity wereused, the distinct burning regimes changed dramat-ically. In Fig. 6, the result with Qslot/Qcavity =0/66 slpm was compared to that with Qslot/Qcavity= 0/59 slpm, 0/136 slpm, and 126/38 slpm. WithQslot/Qcavity = 0/59 slpm, regime I and II werepresent, but regime III was not, and the cavityextinguished. When Qcavity was increased to136 slpm or when Qcavity decreased to 38 slpmand Qslot was 126 slpm, there was not as muchdecay in regime II, regime III was much longer intime, and regime IV was suppressed. The resultsindicated that sufficiently high cavity fueling(either directly into the cavity or through entrain-ment via the shear layer) could mitigate thedecrease in heat release caused by the disruptionfrom the detonation. Furthermore, it was alsofound that while the PD was disruptive to the cav-ity flow, the high level of energy deposition allowed

for ignition across a wider range of T0 and P0 in theflowpath when compared to the spark discharge.To wit: the PD was capable of igniting the cavityat a T0 of 500 K, while the spark discharge couldnot achieve ignition until 570 K for the same pres-sure and fueling conditions.

The spark discharge also demonstrated similardecreased heat release behavior when using bothlower and higher fueling rates. In Fig. 7 with Qslot/Qcavity = 0/52 slpm, the heat release process wasmore gradual, and there were two peaks beforethe cavity achieved steady burning. The first localpeak in heat release came from the first cycling ofthe cavity after ignition. This preconditioned thecavity for the second heat release peak which wassignificantly larger. The time between these peaksaligned with the 1–1.5 ms cavity cycling time.Interestingly, when the Qcavity was decreased to45 slpm and upstream fuel injection (Qslot = 115 -slpm) was used, the same multi-peak behaviorwas present, but there was a lull in emission afterthe second larger heat release peak. There was thena gradual rise in heat release until steady cavityburning was achieved. This cavity ignition behav-ior was similar to that with the PD where therewas a peak in heat release from burning a signifi-cant amount of fuel in the cavity followed by a lull

Fig. 7. Average integrated chemiluminescence in cavityduring ignition versus time for spark discharge atdifferent Qslot and Qcavity.

Fig. 8. Initial conditions for PD exhaust model.


in heat release. While the exact nature of the sparkdischarge ignition process was dependent on Qcav-ity, cavity flame extinction was not observed for theentire range of Qcavity (42–110 slpm) and Qslot (0–126 slpm) employed here.

3.2. Numerical simulations

With the limited set of measurements that werepossible to examine the cavity ignition process,numerical simulations were utilized. Here, thefocus was on the PD. Specifically, an explanationof the decay in heat release after the initial cavityignition process was sought. Fortunately, a previ-ous investigation of the PD exhausting into a M-2cross-flow provided validation through high-frame-rate shadowgraph and planar laser-inducedfluorescence (of NO) [10]. For the simulations, 3-D unsteady calculations using the CFD++ codewere used [11]. The turbulence was modeled usingthe two-equation cubic j-e model, which has non-linear terms that account for normal-stress anisot-ropy, swirl, and streamline-curvature effects.Three turbulent Schmidt numbers were employed,Sct = 0.5, 0.7, and 1.0, to bound the problem andobserve any ignition/mixing differences. The deto-nation and reaction inside the PD were not com-puted [10], but rather the exhaust process wassimulated with the boundary conditions shownin Fig. 8. This was reasonable since the majorityof the chemical reactions from the PD were occur-ring inside the tube. In order to replicate the prod-ucts of the detonation from the experiments, thegas composition was derived from the Chap-man–Jouguet calculations with the species N2,H2O, CO, CO2, O2, H2, O, H, and OH. For thechemical reactions inside the cavity with C2H4,the TP2 reduced kinetic model was used and noturbulence-chemistry interaction was taken intoaccount [12]. The computational domain con-sisted of the full width and height of the test

section with a total of 12.4 million cells. Thenumerical approach was divided into two steps.First a steady-state solution was performed forthe supersonic flow path without the interactionof the PD tube. Second, the PD flow field was ini-tiated at the cavity floor, similar to a shock tubeproblem. At this time, the simulation was per-formed in a time-accurate manner.

Initially the simulations were validated in thecavity geometry in terms of capturing the timescales and mixing processes. A comparison ofshadowgraph images (integrated) and density gra-dients (synthetic shadowgraph in the plane of thePD) showed some variation across the range ofSct (0.5–1.0). Specifically, the computation withSct = 0.5 showed too much mixing, while that atSct = 1.0 showed too little mixing (relative toexperimental observation), therefore boundingthe problem. Nevertheless, all three Sct were stillused for the ignition simulations to see the effecton reactivity. A Sct of 0.7 appeared to producethe best agreement with experiments (and alsomatched the previous validation studies [10]). Fig-ure 9 shows the comparison between the experi-ments and numerical simulations with no fuelinjection. While the early time scales (first0.12 ms) agreed well, there appeared to be dis-agreement at later times. Specifically, by 2.8 msthe simulations had relaxed to steady state whilethe experiments showed the PD still exhausting.In fact, the simulations at 1.4 ms agreed with theexperiments at 2.8 ms. This result emphasized adeficiency in the PD model developed in the previ-ous study [10]: agreement between the simulationsand experiment were sought only for the first0.4 ms where it was most difficult to match the den-sity gradients and plume structure. Consequently,the simulations predicted PD blow-down timesapproximately half that of the experiments. Whilethe initial conditions of pressure, temperature, andvelocity of the PD model could be adjusted to cap-ture the overall timescale of the exhaust processmore accurately, the initial exhaust time scalesand structure were in good agreement, and there-fore the PD evolution was deemed reasonable forthe current ignition simulations.

Fig. 9. Comparison of shadowgraph and density gradients from simulations of cavity flow without fueling.

Fig. 10. Psuedo-integrated T, P, and YC2H4 in cavityversus time from simulations with inset simulated imagesof YC2H4.


Since there was no direct methodology to com-pare the chemiluminescence images to the simula-tions, the time dependence of the T, P (staticvalues) and C2H4 mass fraction (YC2H4) were usedto monitor the ignition process. These values wereextracted from the numerical simulations inplanes at every 6.4 mm spanning the cavity, pro-ducing 23 planes. At each time step, the T, P,and YC2H4 were averaged to produce a pseudo-integrated view of the cavity, to compare to thechemiluminescence images of PD ignition inFigs. 5 and 6. For the case with Qslot/Qcavity = 0/66 slpm, there was an initial rapid decay ofYC2H4 in the cavity. This aligned with regime Iin Figs. 5 and 6. Here, the detonation forced a sig-nificant amount of C2H4 out of the cavity (asshown by inset image (b) above the plot inFig. 10), and T and P increased because of the ele-vated values within the detonation plume. Inregime II, there were relatively constant valuesof T, P, and YC2H4 in the cavity. This could beassociated with a “cavity ignition delay time” tocycle hot detonation products and ignite the fuelthat remained in the cavity. The fuel then startedto burn rapidly in regime III with a decrease inYC2H4 and increase in T while maintaining a con-stant P. The rapid burning in regime III is shownby the deficiency of C2H4 in image (c) in Fig. 10and also by the spike in heat release shown bythe increased chemiluminescence in Fig. 5. It isimportant to note that up to this point the cavityflow had been significantly disrupted by the deto-nation, specifically by elevated P from the detona-tion. While this was beneficial to reaction rates toenhance ignition, it was also detrimental becauseit decreased and, at times, stopped the ramp fuel-ing, since the fuel injection from the closeout rampwas not choked. The simulations were, of course,designed to replicate the experimental fuel injec-tion set-up by defining the fuel mass flow rateand temperature approximately 5 cm upstreamof the exit of the fuel injection into the cavity. Thislocation was approximately where the fuel plenumwas located, and therefore allowed for pressurevariations in the cavity to change Qcavity at the

ramp face. Indeed, the approximate 15% increasein the average cavity P from the detonation signif-icantly decreased Qcavity. In regimes I, II, and III,Qcavity was reduced, the cavity cycling and flowwas significantly disrupted, and the detonationwas able to ignite and burn most of the fuel withinthe cavity. This led to decreased heat release inregime IV because most of the fuel and air wereconsumed and there was not sufficient time toreplenish a reactive mixture. Simultaneously, thePD was at the end of its exhaust cycle, and thusP began to decrease in the cavity. This in turn(i) allowed for more fuel to flow into the cavityas shown in regime IV of Fig. 10 and after a per-iod of time (ii) increased heat release and chemilu-minescence as shown in Fig. 5. After a fewmilliseconds, the cavity flow was able to cycle afew times and regain its normal steady state burn-ing. Image (d) in Fig. 10 shows this behavior witha distribution of fuel throughout the cavity front,along the floor, and near the leading edge of theshear layer. As noted above, the time scales dif-fered by a factor of �2, due to the PD behaviors(simulated vs. observed). Nevertheless, the overall


trends were correctly captured and gave insightinto mechanisms for the observed ignitionregimes. The high-pressure of the detonation withseveral milliseconds of duration disrupted the cav-ity flow and the non-choked fuel injection led toless fuel.

4. Summary and conclusions

Cavity ignition in a supersonic flow wasachieved using two vastly different forms of energydeposition devices that caused different levels offlow disruption. Spark discharge ignition was amore passive process, following the flow to achievesteady cavity burning. Conversely, PD ignitionwas extremely disruptive to the cavity flow dynam-ics with its high-pressure and -temperature plume.The disruption was found to be advantageousunder certain conditions (such as at lower T0 andP0), but was also seen to be somewhat detrimentalbecause of decreased cavity burning and nearextinction from the flow field interaction. High-frame-rate shadowgraph and chemiluminescenceimaging allowed for an unprecedented look atthe transient ignition process and provided ameans for comparison to simulations. While thesimple model developed of the PD exhaust processpredicted shorter time scales, it correctly capturedall burning regimes of the ignition processobserved in the experiments. Most importantly,it provided insight to the mechanism of decreasedheat release because of the high-pressure PDexhaust momentarily diminishing fuel flow to thecavity. Therefore, a smaller PD with shorterblow-down time that produces less pressure risein the cavity, as well choked fuel injection to thecavity has the potential to mitigate any decrease

in burning and therefore a greater chance for suc-cessful ignition over a wide range of conditions.

Acknowledgements

This work was partially supported by the AirForce Office of Scientific Research under Dr. Chi-ping Li.

References

[1] M.R. Gruber, R.A. Baurle, T. Mathur, K.-Y. Hsu,J. Propul. Power 17 (1) (2001) 146–153.

[2] M.R. Gruber, J.M. Donbar, C.D. Carter, K.-Y.Hsu, J. Propul. Power 20 (5) (2004) 769–778.

[3] S.B. Leonov, D.A. Yarantsev, A.P. Napartovich,I.V. Kochetov, 44th AIAA Aerospace SciencesMeeting, Reno, Nevada, AIAA-2006-563, 2006.

[4] H. Do, M.A. Cappelli, M.G. Mungal, Combust.Flame 157 (9) (2010) 1783–1794.

[5] T.C. Wagner, W.F. O’Brien, G.B. Northam, J.M.Eggers, J. Propul. Power 5 (5) (1989) 548–554.

[6] Y. Sato, M. Sayama, K. Ohwaki, et al., J. Propul.Power 8 (4) (1992) 883–889.

[7] L.S. Jacobsen, C.D. Carter, R.A. Baurle, et al., J.Propul. Power 24 (4) (2008) 641–654.

[8] S. Tabejamaat, Y. Ju, T. Niioka, AIAA J. 35 (9)(1997) 1441–1447.

[9] M.R. Gruber, A.S. Nejad, 32nd AIAA AerospaceSciences Meeting, AIAA-94-0544.

[10] T. Ombrello, C.D. Carter, J. McCall, F. Schauer,C.-J. Tam, A. Naples, J. Hoke, K.-Y. Hsu, 50thAIAA Aerospace Sciences Meeting, AIAA-2012-123.

[11] Metacomp Technologies, http://www.metacomp-tech.com/index.html, 2009.

[12] J. Liu, C.-J. Tam, T. Lu, C.K. Law, 42nd JointPropulsion Conference, AIAA 2006-4862.

http://refhub.elsevier.com/S1540-7489(14)00378-2/h0005http://refhub.elsevier.com/S1540-7489(14)00378-2/h0005http://refhub.elsevier.com/S1540-7489(14)00378-2/h0005http://refhub.elsevier.com/S1540-7489(14)00378-2/h0005http://refhub.elsevier.com/S1540-7489(14)00378-2/h0005http://refhub.elsevier.com/S1540-7489(14)00378-2/h0010http://refhub.elsevier.com/S1540-7489(14)00378-2/h0010http://refhub.elsevier.com/S1540-7489(14)00378-2/h0010http://refhub.elsevier.com/S1540-7489(14)00378-2/h0010http://refhub.elsevier.com/S1540-7489(14)00378-2/h0010http://refhub.elsevier.com/S1540-7489(14)00378-2/h0020http://refhub.elsevier.com/S1540-7489(14)00378-2/h0020http://refhub.elsevier.com/S1540-7489(14)00378-2/h0020http://refhub.elsevier.com/S1540-7489(14)00378-2/h0020http://refhub.elsevier.com/S1540-7489(14)00378-2/h0025http://refhub.elsevier.com/S1540-7489(14)00378-2/h0025http://refhub.elsevier.com/S1540-7489(14)00378-2/h0025http://refhub.elsevier.com/S1540-7489(14)00378-2/h0025http://refhub.elsevier.com/S1540-7489(14)00378-2/h0025http://refhub.elsevier.com/S1540-7489(14)00378-2/h0030http://refhub.elsevier.com/S1540-7489(14)00378-2/h0030http://refhub.elsevier.com/S1540-7489(14)00378-2/h0030http://refhub.elsevier.com/S1540-7489(14)00378-2/h0030http://refhub.elsevier.com/S1540-7489(14)00378-2/h0035http://refhub.elsevier.com/S1540-7489(14)00378-2/h0035http://refhub.elsevier.com/S1540-7489(14)00378-2/h0035http://refhub.elsevier.com/S1540-7489(14)00378-2/h0035http://refhub.elsevier.com/S1540-7489(14)00378-2/h0040http://refhub.elsevier.com/S1540-7489(14)00378-2/h0040http://refhub.elsevier.com/S1540-7489(14)00378-2/h0040http://refhub.elsevier.com/S1540-7489(14)00378-2/h0040http://www.metacomptech.com/index.htmlhttp://www.metacomptech.com/index.html

Cavity ignition in supersonic flow by spark discharge and pulse detonation1 Introduction2 Experimental setup3 Results3.1 Shadowgraph and chemiluminescence3.2 Numerical simulations

4 Summary and conclusionsAcknowledgementsReferences

cavity ignition in supersonic flow by spark discharge and ...cavity ignition in supersonic ﬂow by...

Documents