High-Frequency Pressure Measurements for Unstart Detection in Scramjet Isolators

Transcription

1 46th AIAA/ASME/SAE/ASEE Joint Propulsion Conference & Exhibit July 21, Nashville, TN AIAA High-Frequency Pressure Measurements for Unstart Detection in Scramjet Isolators Jeffrey M. Donbar 1 and Graham J. Linn 2 Air Force Research Laboratory, Wright-Patterson AFB, Ohio, Sukumar Srikant 3 and Maruthi R. Akella 4 University of Texas at Austin, Austin, Texas, This paper describes an experimental investigation using an array of high-frequency pressure transducers located in the isolator/combustor region of a direct-connect hydrocarbon-fueled, scramjet combustor. Isolator/combustor entrance conditions were fixed and representative of a Mach-5.5 flight condition. Mean, standard deviation, and spectral content of measured pressures were similar to those measured in previous studies. A simple model relating the sum of measured pressures and the fuel flow rate delivered to the primary injectors was developed. Pressure measurements were post-processed and evaluated for use in a shock-position-control sensor to detect engine unstart. Four methods of post-processing the pressure data and corresponding detection criteria were evaluated: a) 1% pressure rise, b) 1% increase in standard deviation, c) 1% increase in power spectral density (PSD), and d) summation of pressures. Pressure summation typically provided 1 2 s more lead time in detecting unstart then any other method. d F H P P_P T T_VH W x x s Nomenclature = injector diameter = load-cell force = engine throat height = pressure = combustion-heater total pressure = temperature = combustion-heater total temperature = mass flow rate = streamwise coordinate (x = at engine throat) = axial position of shock-train leading edge = injection angle relative to combustor wall = fuel-air equivalence ratio Subscripts A = air stream (includes air, make-up oxygen, and combustion heater fuel) B2 = B2 injection site B6 = B6 injection site comb = combustion pressure distribution C3 = C3 injection site F = fuel stream tare = tare pressure distribution 1 Aerospace Engineer, AFRL/RZAS, Senior Member. 2 Electrical Engineer, AFRL/RZAS, Member. 3 Graduate Student, Department of Aerospace Engineering and Engineering Mechanics, Student Member. 4 Associate Professor, Department of Aerospace Engineering and Engineering Mechanics, Associate Fellow. 1 This material is declared a work of the U.S. Government and is not subject to copyright protection in the United States.

2 I. Introduction For many years, scramjet propulsion systems have been envisioned as the engine concept for hypersonic atmospheric-flight vehicles. Recent short duration scramjet flights of NASA s X-43 1 and the continuing multiminute scramjet flights of the USAF s X-51 2 are proof that scramjet-engine design and development are maturing. However, there are still significant areas of fundamental research to be completed in order to optimize scramjetengine operation and performance. One such area of research involves understanding engine unstart. In a typical dual-mode scramjet combustor operating in the ramjet mode, the isolator section contains a pre-combustion shock train (PCST). The shock train provides the mechanism for a supersonic flow to adjust to a combustor back pressure that is higher than the isolator inlet static pressure. 3 However, if the combustor heat release is sufficient and the subsequent pressure rise is higher than can be accommodated with the given isolator length, the PCST can be forced out of the isolator and induce an engine unstart. 3 Recovering from an unstart event could be difficult as airflow through the engine will be disrupted and the inlet airflow will need to be restarted and the combustor fuel flow re-ignited in a timely fashion to ensure adequate thrust generation to maintain vehicle altitude and flight trajectory. Several experimental and numerical studies have shown that the interaction of the PCST with the isolator boundary layers and subsequent boundarylayer separation plays an important role in unstart While significant research has been performed on shock-wave boundary-layer interactions (SWBLIs), there is still very little understanding of the dynamics of such an event. 13,14 As such, there has been little attempt to achieve some sort of flow control for a SWBLI interaction in a scramjet isolator. This limited understanding on how unstart occurs and the inability to control it has led to the current philosophy of prescribing extra isolator length to avoid an unstart event. In turn, the extra isolator length is effectively a weight and drag penalty that reduces overall vehicle performance, especially at higher Mach numbers where unstart is not a concern. An effective isolator shock-position-control sensor would prove useful for maximizing vehicle acceleration while minimizing required isolator length. A significant research effort aimed at developing a shock-position sensor for supersonic inlets was completed at NASA Lewis Research Center in the early 197s This effort used an array of pressure transducers near the terminal shock location to determine when the inlet was nearing unstart. The inlet Mach numbers for these tests were approximately 1.4 and several variations of the sensor were tested successfully on both axisymmetric and 2-D inlets. Reference 18 demonstrated a definitive correlation between the sum of the pressure transducers and the shock location and was able to create a real-time shock-location sensor. Further work using a similar array of transducers was conducted by Sajben et al. 19 in the early 199s. This work extended the NASA concept to a Mach-3 mixedcompression inlet environment. In addition to measuring pressure directly, Ref. 19 used RMS fluctuations of pressure and introduced a specific tone in the flowfield via a Hartmann generator as alternative shock-location detections schemes. The authors concluded, however, that direct measurement of pressure was the simplest, fastest, and most reliable. In scramjet environments, an array of high-speed pressure transducers have been used to study the pressure field within inlet/isolators during unstart. 2,21 In neither investigation, however, was there an attempt to utilize the measurements as part of an isolator shock-position-control sensor. Recent studies in a small-scale scramjet,.9 in. 2 (58 mm 2 ) cross-sectional area, at the University of Virginia 22,23 and a cold flow inlet/isolator experiment of comparable size, 2. in. 2 (129 mm 2 ), at the University of Texas have investigated methods for unstart detection and mitigation. In Refs. 22 and 23, an array of high-frequency pressure transducers was used to detect gross shock-train movements within the isolator under varying combustor equivalence ratios. Using techniques similar to those outlined in Ref. 19, the authors defined metrics that could be used by a scramjet engine controller to regulate shock-train position, but indicate that methods requiring spectral analysis are computationally more intense and therefore less suitable for real-time control. The University of Texas effort was in cold flow and has coupled simplified detection algorithms with complementary actuators (vortex generating jets) to arrest shock-train movement during unstart approximately % of the time. 25 Also, a more detailed spectral analysis has shown a significant increase in power spectral density for frequencies between 3 and 4 Hz after the shock has progressed upstream of a particular pressure transducer. 26 In the current work, a gaseous-hydrocarbon-fueled scramjet isolator/combustor is tested in a direct-connect facility. Low frequency (1 Hz) pressure and temperature measurements indicate the general operating characteristics of the isolator/combustor. Using an axial array of high-frequency (1 khz) pressure transducers, measurements were made in the isolator/combustor region during both started and unstarted runs of the combustor. Data acquired from the tests was subsequently used to assess potential shock-train control strategies. This study represents an initial attempt at developing a shock-position-control sensor. If sufficient instrumentation were placed in the isolator and coupled with an actuator mechanism (e.g., bleed, injection, or plasma discharge) and a suitable control algorithm were utilized, it might be possible to develop a closed-loop control system to prevent engine unstart. 2

3 II. Experimental Details The research combustor flowpath studied in this investigation is integrated into a continuous-flow supersonic combustion research facility capable of simulating flight conditions from Mach 3.5 to Mach 7. This facility is supplied with air at up to 3 lbm/s (13.6 kg/s), 7 (5.2 MPa), and 166 R (922 K), as well as 3. (2.7 kpa) continuous exhaust. Compressed natural gas is used to fuel the in-stream combustion heater and a liquid oxygen system provides oxygen replenishment to the combustion-heated air stream. Liquid- and gaseoushydrocarbon fuel systems deliver fuel to the research combustor. An electric fuel heater provides combustor fuel at the required temperatures for various simulation conditions. A recirculating cooling water system provides 2 gpm (946 lpm) at 7 (483 kpa); raw dump water at 3 (2.4 MPa) is also available. The entire flowpath is secured to a thrust stand for direct measurements of the thrust generated by the combustor. This measurement is combined with wall static pressure measurements and a performance analysis routine to deduce combustion efficiency and other performance parameters. Additional details about the facility are presented elsewhere. 28 A. Combustor Flowpath and Fuel Delivery A research combustor with a flush-wall fuel-injection system and wall-based flameholding was used in this investigation and is shown in Figure 1. The flowpath width was constant at 9 in. (229 mm) and the reference height (H) at the combustor entrance was in. (42.3 mm). Removable panels in the sidewalls enabled either optical access or conventional instrumentation. The combustor had several fuel injection sites on both body and cowl sidewalls (details of the sites used in this investigation are shown in Table 1). Each injection site had a dedicated fuel-supply manifold that was equipped with pressure and temperature instrumentation. Flameholding was provided by Table 1. Fuel injector details. B2 B6 C3 Number d (mm) (deg.) Spacing (mm) a wall cavity and a rearward-facing step on the body side in addition to a rearward-facing step on the cowl side. The cavity, upstream edge located at x/h = 12., had a depth of.85 in. (21.6 mm), a length (measured from the Body Side Cowl Figure 1. Flowpath schematic with fuel injection sites (flow direction is left to right). 3

4 Facility Nozzle Distortion Generator Isolator / Combustor Figure 1. Hardware installation (flow direction is left to right). Figure 2. Hardware installation (flow direction is left to right). separation corner to the mid-ramp location) of 3.79 in. (96.3 mm), and a 22 deg. ramp angle. It spanned the central 7 in. (178 mm) of the combustor. The body-side step flameholder was.48 in. (12.2 mm) deep and was positioned at x/h = The cowl-side step flameholder was. in. (12.7 mm) deep and was positioned at x/h = The water-cooled flowpath had several access ports in the walls for optical diagnostics. 29 Interchangeable facility nozzles were used to generate appropriate supersonic flow conditions upstream of the combustor flowpath. In this investigation, a Mach 2.84 facility nozzle/distortion generator assembly was used to simulate Mach flight conditions. Prior to combustion experiments, a series of calibrations was performed using a water-cooled traversing Pitot-pressure/stagnation-temperature probe and wall-based measurements. 3 The research combustor was installed as shown in Figure 2. The fuel control system was comprised of several devices designed to control both the total fuel flow rate and the distribution of fuel among the desired fuel injection sites. One of two Coriolis mass flow meters (Rheonik model RHM2 or RHM8, depending on the total fuel flow rate) was used to meter the supply of room-temperature fuel delivered to the combustor and to provide feedback to the primary control valve that maintained the desired overall equivalence ratio. The fuel then flowed into two fuel manifolds, one supplying the body-side injectors and the other supplying the cowl-side injectors. A second control valve was used to regulate the fuel flow rate to the primary injection site (B2 for all runs in this study). The remaining fuel was then routed to the secondary injection sites. Sonic nozzles were used to set the fuel distribution delivered to these injectors. All of the fuel injection sites were calibrated prior to combustion testing to determine discharge coefficients. This information was used in conjunction with the measured injector areas and measured fuel properties in the injector manifolds to determine the actual fuel distribution delivered to the combustor. B. Instrumentation A CAMAC-based data acquisition and control system had 48 channels of analog input, 64 channels of digital I/O, and 4 channels of analog output. The CAMAC crates were connected to a Linux workstation via a fiber optic Grand-Interconnect interface for both control and data acquisition. A Pressure Systems Incorporated (PSI) 84 pressure scanning system consisting of 4 channels with real-time display and data reduction was also used. The facility nozzle had 3 static pressure ports on its cowl wall. The distortion generator had 25 pressure taps on its cowl wall. The combustor had over 2 pressure taps instrumented on all four walls, including measurements at the combustor exit and in the base areas of each wall. A large array of thermocouples was used to monitor air, fuel, oxygen, cooling water, and hardware wall temperatures. The water temperatures, along with measured water flow rates, were used to compute the heat losses from the various components. In addition, all flows (air, fuels, water, and oxygen) were measured using orifice plates, turbine flow meters, Coriolis mass flow meters, or Venturi flow meters. All of the instrumentation described above was recorded at 1 Hz for the duration of the test. In addition to the instrumentation described above, eight 1-kHz-response-rate pressure transducers, Omega model number PX-32-1AV with -1 range, were Table 2. installed on the cowl-side centerline of the combustor flowpath ranging in location from the distortion duct to slightly upstream of Transducer the flameholding cavity. Table 2 provides the axial location of the Name high-speed transducers. For reference, the engine throat is located at x/h =. and transducers located upstream of the throat are labeled as negative. Bench tests were conducted on the Omega transducers prior to installation on the combustor and the transducers response appeared immune to the effects of spark noise, fuel-heater electric field, standoff line length, and PSI-line tee connection. Raw voltage signals from the transducers were recorded using a National Instruments PCI-671E board housed in a stand-alone 4 Measurement locations. Component Location x/h PDDB_C Dist. Duct PDDB_C21 Dist. Duct PSXB_C1 Isolator 1.24 PSXB_C3 Isolator 2.44 PCBB131 Combustor 4.73 PCBBFC1 Combustor 6.53 PCBBFC3 Combustor 7.73 PCBB133 Combustor 8.93

5 computer that was connected to an SCB-1 interface box. This computer was not synchronized with the 1-Hz facility data system, but pressures were recorded over the full duration of a combustor test, approximately 6 seconds. Without synchronization to the facility data, the sharp pressure rise of the transducer (PCBB133) nearest the fueling site provides an indication of when fuel was delivered to the combustor. The transducers were sampled sequentially at an acquisition rate of 25 khz; therefore, each of the eight transducers had an effective sampling rate of 3125 Hz and a time separation between samples of successive transducers of 4 s. An in-house Visual Basic program managed the data collection, storage, and conversion to engineering units. As mentioned above, the Omega transducers were teed into a standoff used for the 1 Hz PSI pressure data system. These 1/16-in. (1.58-mm) diameter standoffs were between 9-12 in. (23-3 mm) in length for each of the 8 measurement locations. Because of the standoff mounting, an analysis of resonant frequencies was conducted. In particular, an organ pipe resonance was evaluated for the given configuration. The wavelength of the fundamental wave is equal to 4 times the length of a closed end pipe. The fundamental frequency of the pipe is obtained by dividing the local speed of sound by this fundamental wavelength. In this case, the local speed of sound is unknown as the temperature within the standoff was not measured. A reasonable estimate of the speed could be determined using the isentropic freestream static temperature of 958 R (532 K). The resulting resonant frequencies are 379 Hz and 5 Hz for the 12 in. (3 mm) and 9 in. (23 mm) standoff, respectively. Applying the NyQuist criterion, the 1 khz transducers are capable of resolving frequencies up to Hz. Given the uncertainty in the speed of sound within the standoffs, it is quite possible that a resonance frequency lies in the - Hz band. Only one of the unstart detection analyses considered herein utilizes spectral content as a parameter. Within that particular spectral analysis, the frequency information is divided by a baseline frequency spectrum which will also contain the standoff resonance, and the effect of the pipe resonance on the analysis should be minimal. The Omega transducers were calibrated prior to installation, but were also compared to their more accurate 1-Hz counterparts for a given test run. Comparisons were conducted during steady-state portions of the tests both at low pressures (tare condition prior to ignition) and high pressures (unstarted condition). For each of the transducers, a single 1-Hz sample was selected and the high-speed data were averaged for a period of 5 seconds before that time. All of the high-speed transducers were within.5 (.4%) of their 1-Hz counterpart at tare condition and within.5 (1.1%) after unstart. All tests analyzed herein were conducted during a single test night, shortly after calibration, to minimize the effects of transducer calibration drift. C. Operating Conditions Combustor inlet stagnation pressure and temperature were held fixed at 2 ± 1 (1.72 ±.7 MPa) and 2 ± 1 R (139 ± 5 K), respectively. The ethylene fuel distribution and fuel-air equivalence ratio were systematically varied as shown in Table 3. The cases listed in Table 3 are a representative subset of a more comprehensive set of tests conducted using the B2, B6, and C3 fuel-injection sites. For all cases shown in Table 3, autoignition of the fuel was observed; no external aids (spark igniters, air throttles, or other means) were required to produce ignition and flame propagation in the combustor. Pressure distributions from tare conditions (i.e., combustor at given test condition with no fuel flowing) and combustion conditions were used to determine the leading edge location of the PCST (x s /H). To minimize subjectivity in determining this location, each combusting pressure distribution was normalized by the corresponding tarepressure distribution. The start of the pre-combustion pressure rise is then identified as the location where the normalized distribution begins to depart from unity. For this work, a value of P comb /P tare = 1.1 was used to identify this axial location. In cases where the leading edge of the PCST was upstream of the engine throat, the run was considered unstarted (UNS). Typically, combustion tests were run for more than 3 seconds to achieve steady-state heat-transfer information. Occasionally, overheating of uncooled portions of the combustor flowpath made it necessary to terminate the test before steady-state conditions were established. Figure 3 shows time-based data collected from Case 774AL. Pressure and temperature data are shown from both air and fuel streams. Steady operating conditions are established prior to the introduction of fuel to the combustor. Fuel injection begins approximately 15 s into the experiment, as indicated by the changes in injection 5 Table 3. Combustor operating conditions. Case B2 B6 C3 x s /H 774AD AF AG UNS 774AH AJ UNS 774AL AN UNS 774AO AP AQ UNS

6 Pressure, Temperature, R Mass Flow Rate, lbm/s Force, lbf P_P P B2 P B6 P C3 T_VH T B2 W A W F x 1 F Figure 3. Time-based data from Case 774AL (left: pressure and temperature, right: flow rate and force). manifold conditions. Fuel injection pressures reach steady levels approximately 3 s into the test. Mass flow rate and load-cell-force data are also shown from the same experiment. The sudden increase in load-cell force near 16 s corresponds to ignition. Steady force levels are achieved approximately 3 s into the test. Fuel is turned off at approximately 56 s followed by combustor flame out. For some of the fueling conditions studied, a short time (3-5 s) was required for all of the fuel on both the body and cowl sides to ignite and burn. In Fig. 3, the initial rise in force, 16-2 s, corresponds to un-coupled combustion within the combustor (i.e., the cavity flameholder is lit, and perhaps some fuel from B6, but the cowl side fuel is not engaged). After 2 s, the flame has propagated to the cowl side and all of the fuel is participating in combustion and the force level rises to its final steady-state level. Standard 3 Hz video recording of run 774AL clearly shows initial burning along the body side; subsequently, the flame fills the entire combustor height. Note that as fuel is removed from the combustor, the initial drop in force during s is back to same level as during 16-2 s, which represents the time prior to full fuel engagement. III. Results and Discussion A collection of high-speed pressure measurements in the isolator/combustor was used to examine characteristics of unstart and identify the potential use of such an array as a shock-position-control sensor. Actual pressure, standard deviation of pressure, and spectral content are compared with ER_B2 data from previous studies that have cross sectional areas 1 times smaller than the current study. An unstart detection method using only the sum of the individual transducers is evaluated. Additionally, the.4 various detection methods are compared against one another to determine which provides the most lead time for detecting unstart..35 A. B2 Fueling and Pressure-Time History Tests were conducted on this engine configuration over several months to evaluate overall engine performance and operability with different main fuel injection sites, fuel distributions, equivalence ratios, and alternative fuels. The test cases listed in Table 3 represent a subset of the nearly test runs over that time period that used the B2/B6/C3 fueling locations and ethylene as fuel. Figure 4 shows the entire collection of B2/B6/C3 data describing shock-train position as a function of the equivalence ratio delivered to the primary, B2, fuel injection site. For this particular xs/h B x / H s Figure 4. Shock-train leading-edge position as a function of B2.

7 Pressure, 774AD Presure, plot, cases where the leading edge of the shock train begins upstream of x/h =. (engine throat) are not included. This figure clearly suggests that B2 determines shock train position within the isolator for this engine. Although some scatter exists in the data, Fig. 4 shows a fairly linear relationship between B2 and x s /H, where x s is the axial location of the shock train leading edge. The relationship identified in Fig. 4 will be used to develop a simple shock-position-control sensor based on pressure measurements within the isolator later in the paper. To confirm that B2 was indeed controlling shock train location, and therefore pressure rise within the isolator, test cases 774AD, AH, and AO and were compared. In each case, B2 =.2, but overall equivalence ratios are.9,.8, and.6 for AD, AO, and AH, respectively (see Table 3.). Figure 5 shows the 1-Hz pressure measurements from all four combustor walls during steady-state operation for each of the three x/h runs. The pressure rise for all runs begins at x/h = 4 and the shape of the pressure rise is identical. The peak pressure occurs over the cavity flameholder (x/h = 14) and is followed by a sudden decrease in pressure as the flowfield adjusts to the rapid area change created by the two step flameholders at x/h ~ A second pressure peak occurs downstream of the step flameholders; the secondary-peak magnitude increases and the location shifts slightly upstream as the overall equivalence ratio increases. This second peak is followed by a more gradual pressure reduction as the flow accelerates in the diverging portion of the combustor. While only the B2 =.2 case is shown here for brevity, the shape of the isolator pressure was found to be dependent only on B2, and not on overall equivalence ratio, for the range of B2 from.2 to.35. Figure 6 and Figure 7 show the time history of the 8 eight high-speed pressure transducers installed on the cowl side of the engine for runs 774AF (started) and 774AG (unstarted), respectively. Both of these runs have an overall equivalence ratio of.9, but B2 =.2 for 774AD and.4 for 774AG. In Figure 6, the downstream transducers (PCBBXXXX) exhibit pressure rise at ~ 1 s and begin to level off around 25 s. This portion of the run is the time where body- and cowl-side fuel manifolds reach steady-state operating condition (see Fig. 3). Note that measurements from the transducers located upstream of the engine throat (PDDB_XXX) are essentially unchanged from their pre-ignition values. Similarly, the upstream isolator pressure transducer (PSXB_C1) remains at its tare value, while the downstream isolator pressure transducer (PSXB_C3) shows a moderate rise from 2 32 s. Thus, the leading edge of the shock train for this case lies between the two isolator transducers. In contrast, Fig. 7 shows a relatively sharp rise in pressure at the PCCB transducer locations around 7 s. For the first 7 s after ignition, the AD 774AH 774AO TARE Increasing Figure 5. Combustor operation; fixed B2 variable overall fueling PCBB133, x/h= 8.93 PCBBFC3, x/h= 7.73 PCBBFC1, x/h= 6.53 PCBB131, x/h= 4.73 PSXB_C3, x/h= 2.44 PSXB_C1, x/h= 1.24 PDDB_C21, x/h= PDDB_C19, x/h= Figure 6. High-frequency pressure measurements for run 774AF. 7

8 standard deviation, standard deviation, Mean Pressure, Mean Pressure, Pressure, PCBB133, x/h= 8.93 PCBBFC3, x/h= 7.73 PCBBFC1, x/h= 6.53 PCBB131, x/h= 4.73 PSXB_C3, x/h= 2.44 PSXB_C1, x/h= 1.24 PDDB_C21, x/h= PDDB_C19, x/h= Figure 7. High-frequency pressure measurements for run 774AG. PSXB and PDDB transducers remain at their pre-ignition values. As the fuel manifolds fill and are pressurized, more of the fuel is reacting in the combustor and the PCST moves ahead of the PSXB transducers at ~ 14 s and passes the PDDB transducers (which are upstream of the engine throat) around 18 s. The steady-state position is achieved between 18 and 34 s for this run. The subsequent pressure drop off after 34 s occurs as the main fuel supply valve is closed and the fuel within the lines exhausts into the combustor and the cavity fuel remains lit. Note in Figure 7 that transducer PSXB_C3 is responding slowly to large pressure rises. This trend was true for all runs analyzed herein; therefore, that particular transducer was excluded from any further analysis. B. Mean, Standard Deviation, and Frequency Analysis of Measured Pressures Substantial changes in mean pressure, standard deviation of pressure, and frequency content of measured pressure have all been proposed and utilized as potential inputs for a shock-train-control sensor 19,21 in both smaller reacting facilities and in cold-flow studies. Each of these methods will be evaluated using the current data set in a larger, combusting environment. Figure 8 shows the pre-combustion mean and standard deviation for each of the 7 high-speed transducers in test 774AG. The pre-combustion region for this run was defined to be from -6 s (see Figure 7). Figure 8a shows the influence of the distortion-duct-generated shock pattern as the mean pressure varies from 3 to 13 over the range of installed transducers. In a typical direct-connect test without distortion, the wall pressures in the isolator would be relatively equal. As would be expected in the pre-combustion phase, Figure 8b indicates that each of the standard deviations is extremely low, on the order of.1. This type of information can be used to determine thresholds (% of increase) that could then be used as a control sensing input. Section D will detail specific values of pressure and standard deviation that will be used for a potential control algorithm. Also, these pre-combustion mean and standard deviation values compare relatively well with those measured in the small-scale scramjet of Ref. 22 and the cold x/h Figure 8. Pre-combustion flow: a) mean pressure, and b) standard deviation x/h Figure 9. Post combustion flow: a) mean pressure, and b) standard deviation. 8

9 db db flow inlet/isolator configuration of Ref. 24. Figure 9 shows the post-combustion mean and standard deviation for each of the 7 transducers in test 774AG. For this run, the post-combustion region was defined as s (see Fig. 7). In this unstarted condition, Figure 9a shows that the mean value is nearly identical for all of the transducers as the leading edge of the shock train is upstream of all transducers. Figure 9b shows a significant increase (factors of 5-9) in standard deviation when compared with the pre-combustion values of Fig. 8b. These trends are also consistent with measurements in Ref. 24. Frequency analysis of the high-speed pressure measurements shown previously was conducted. A Fast Fourier Transform (FFT) was computed from the collected signal and then analyzed using power spectral density (PSD). The FFT and PSD computations were completed using a MATLAB script developed at the University of Texas. 26 This particular script contained the capability to compute the PSD and other relevant parameters over a running window. Of particular interest are the frequency characteristics both upstream and downstream of the shock. Figure 1 shows sensor PDDB_C19 (far upstream) from run 774AG prior to shock passage. For this particular figure, the PSD is computed over the time to 6 s, which is prior to ignition. There is a significant amount of DC signal, but no dominant frequency out to Hz. Conversely, Figure 11 shows the same transducer after the shock has moved upstream. In general, the post-shock PSDs indicate a general increase of energy in the - Hz range than the preshock PSD. The energy is distributed equally over the band, but some significant peaks are noted between and 125 Hz. Whether or not those peaks are general in nature or specific to this configuration is difficult to determine. Qualitatively, the PSD results look similar to those in Refs. 22 and 24. A spectral content analysis similar to that in Ref. 26 was also completed using the PSD information. Specifically, Frequency, Hz Figure 1. PSD of transducer PDDB_C19 for run 774AG prior to ignition Frequency, Hz Figure 11. PSD of transducer PDDB_C19 for run 774AG after shock has passed. the authors of Ref. 26 looked at energy distribution across frequency range from 1 Hz for two regions of the flow: pre-combustion and around the unstart event. Two regions are used here as well; however, one of the regions chosen is different from Ref. 26. The PSD values used were calculated by selecting a single steady-state portion of the run prior to ignition and a similar steady-state region after the shock has passed, as opposed to around the unstart event as in Ref. 26. The frequency range was divided into 1-Hz bins and the ratio of post- to pre-shock energy within each band was determined. The results for transducer PDDB_C21 in runs 774 AD, AF, AG, and AJ are shown in Fig. 12. The plot indicates that there is significant increase in energy content for frequencies less than 4 Hz for unstarted runs AG and AJ. The range of frequencies showing energy rise is similar to that of Ref. 26; however, the increase in magnitude by a factor of 3- is substantially lower than that found for transducers in a similar location in Ref. 26. Differences that may contribute to such a dramatic difference are hot flow vs. cold flow and that boundary layer in Ref. 26 was 65% of the duct height. While the ratio is lower in the present study, the increase of 2- times the pre-combustion value is still sufficient for use in a shock position-control sensor. For started runs AD and AF, the ratio remains near unity indicating that the effects of the combustion pressure rise are not communicated upstream of the shock train leading edge. C. Summation of Pressures As mentioned previously, the idea of using the sum of individual pressure measurements as a shock-location sensor has been used effectively in supersonic inlets. 18 The concept would seem to be applicable for estimating the shock-train leading-edge location within scramjet combustors as well. The axial array of transducers effectively 9

10 Sum of Pressures, Energy (Post Shock / Pre Shock) 6 4 AD (no unstart) AF (no unstart) AG (unstart) AJ (unstart) Hz 1-2 Hz 2-3 Hz 3-4 Hz 4- Hz Frequency Bands Figure 12. Power sensitivity for PDDB_C21 during runs 774 AD, AF, AG, and AJ. samples the shape of the shock train at any instant in time and provides an indication of where the leading edge is located. As the pressure behind the shock train is significantly larger than upstream (~ vs. ~ 1 ), the method also provides a wide dynamic range. As opposed to any shock-position-control sensor using some type of spectral analysis, this method involves only addition and would be computationally efficient. For the evaluation conducted herein, only the 7 functioning transducers were used. To determine the effectiveness of the method, data from only the cases listed in Table 3 were used. For a single steady-state time with all fuel flow stabilized, the sum of the high-frequency pressure transducers was plotted against B2 ; this result is shown in Fig. 13. Both started and unstarted runs are shown in the plot, with the unstarted runs having sums greater than 275 and B2 greater than.35. Conversely, those runs where the shock train is located far downstream have sums around 8 psi and B2 near.2. Those points with the shock approximately 2 duct heights from the engine throat have sums near 1 and B2 of roughly.3. To create a simple model for control evaluation, a linear curve fit was created for the data, as shown in Fig. 13. The curve fit represents a reasonable fit to the data, but could be improved if all of the runs plotted in Fig. 4 were used. However, using only the data from a single test night also eliminates transducer drift and other complexities that may complicate the analysis. The effectiveness of the linear fit was evaluated for several runs by comparing a time history of the summed pressures to predict B2 using the curve fit of Fig. 13 with the experimentally measured B2. Figure 14 shows predicted and actual B2, as well as error, for a started run (774AF) with the shock train far downstream of the throat. Figure 15 and Figure 16 show similar trends for two unstarted cases. It should be noted that for the line 3 Curve Fit: y = Data Points B2 Figure 13. Data points and curve fit of summation of pressures as a function of B2. 1

11 labeled predicted B2, the values near the beginning and end of the runs are not realistic since the measured manifold pressures and temperatures are still being used to compute B2 even though there is no fuel flowing at these times. This, in turn, leads to significant error in those areas. However, in the steady-state portion of the run, the error between predicted and actual B2 is 1% or less. Again, if all of the data points shown in Figure 4 were used, it is likely the fit could be improved. Regardless, this relatively simple model appears to be adequate as a proof-ofconcept for a shock-position-control sensor. D. Comparison of Unstart Detection Methods An attempt was made to compare the various unstart detection methods using some of the acquired data. In particular, four unstart detection methods were chosen for comparison: 1. 1% of the steady pressure level prior to ignition (Ref. 22) 2. 1% of the steady level of pressure standard deviation prior to ignition (Ref. 22) 3. 1% of the steady PSD value in the frequency range from 1-4 Hz (similar to Ref. 26) 4. Summation of 7 high-speed pressure transducers exceeds 1 The criteria used for the first three methods represent the criteria selected by the respective authors and only method 3 was altered (1% vs. 2%) for this comparison. The rolling window for method 2 was set to points, which corresponds to.16 s, while the rolling window for method 3 was set to 2 points, which corresponds to.64 s. These window sizes do not indicate a delay in unstart detection as the window moves forward by only one sample in each iteration and represent window sizes similar to those used previously. 26 The selection of 1 for the summation of pressures was based on the results shown in Fig. 13. All test runs with B2 of nearly.3 (and sum of pressures of ~ 1) have shock-train leading edges located approximately 2 duct heights from the engine throat; current design practice is to include that margin to account for uncertainties in air flow, fuel flow, inlet geometry, and differences between ground and flight test operation. A persistence criterion was also used for each of the 4 methods. Each method was considered to have detected an unstart if 1 consecutive samples met the required threshold for that methodology. Requiring 1 consecutive positive triggers helps eliminate false detections of unstart when spurious data is received by the control algorithm. In this set of tests, each of the 4 methods is applied to a single transducer, PSXB_C1, located 1.24 duct heights downstream of the engine throat. Note that if transducer PSXB_C3 had been functioning reliably (x s /H = 2.44), it would have been an acceptable choice as well, as would a combination of the two. Figure 17 shows a comparison of the four methods outlined above for detecting unstart in run 774AJ. The baseline for methods 1, 2, and 3 was determined from the data in the first 5 s of the test and the threshold is shown above each data trace. For this run, the sum of pressure detection method signals an unstart at 8.9 s into the run, while the standard deviation and PSD methods both signal an unstart at 1.5 s. The actual pressure rise threshold is not met until ~ 11.5 s. For this run, the sum of pressures method provides 1.5 seconds more notice of unstart than any of the other methods chosen for analysis. Figure 18 shows the same comparison of unstart methods as Fig. 17, but for run 774AM, which was not a run listed in Table 3 and therefore not used to generate the curve fit in Fig. 13. Consequently, this run serves as a validation case for the sum of pressures method. The fueling conditions for this run were B2 =.4, B6 = C3 =.15. As was the case in Fig. 17, the first 5 s of data are used to compute the baseline values and determine unstart thresholds for methods 1, 2, and 3. Similar to Fig. 17, the sum of pressures method detects unstart at 8.7 s, while the standard deviation and PSD methods do not detect unstart until 1.7s, followed by the actual measured pressure at 11.7s. Again, the sum of pressures provides nearly 2 seconds more lead time to actuate an appropriate flow control to mitigate unstart. Figure 19 shows the unstart detection method comparison for run 774AG. All baseline values were determined in the manner described above. For this run, however, the standard deviation method signals unstart at 8.9 s, which is a full second before the sum of pressures at 9.9 s. The PSD detects unstart at 11 s and finally the measured pressure at 13.7 s. Close inspection of the PSXB_C1 pressure trace (top) indicates significant noise on the trace between 5 and 1 s. Whether or not this is due to electrical interference or a result of combustor ignition is difficult to determine. The early detection by the standard deviation in this run does point out that the 1% thresholds used may need to be refined or at least a range of thresholds tried to determine the best value for a particular isolator/combustor combination. Also, the sum of multiple pressure transducers should distribute the unstart detection over several transducers and therefore minimize the influence of one particularly noisy transducer that may falsely detect an unstart. 11

12 B2 and (% error)/1 B2 and (% error)/1 B2 and (% error)/ Predicted B2 Actual B2 Error Figure 14. Comparison of predicted and actual B2 for run 774AF..6 Predicted B2.5.4 Actual B2 Error Figure 15. Comparison of predicted and actual B2 for run 774AG Predicted B2 Actual B2 Error Figure 16. Comparison of predicted and actual B2 for run 774AJ. 12

13 db Sum of Pressures standard deviation Pressure Pressure, Run J, Threshold = 1.5*baseline = 18.7 Pressure UNS detection: (11.5 s) Standard Deviation, Threshold = 1.5*baseline =.1 STD UNS detection: (1.5) Sum of Pressure, Threshold = 1. Sum P UNS detection: (8.9 s) Power Spectral Density, (1-4 Hz band), Threshold = 1.5*baseline = 27.9 db PSD UNS detection: (1.5 s) Figure 17. Comparison of unstart detection methods for run 774AJ. 13

14 db Sum of Pressures standard deviation Pressure 5 2 Pressure, Run M, Threshold = 1.5*baseline = 18.7 Pressure UNS detection: (11.7 s) Standard Deviation, Threshold = 1.5*baseline =.1 STD UNS detection: (1.7) Sum of Pressure, Threshold = 1. Sum P UNS detection: (8.7 s) Power Spectral Density, (1-4 Hz band), Threshold = 1.5*baseline = 27.9 db PSD UNS detection: (1.7 s) Figure 18. Comparison of unstart detection methods for run 774AM. 14

15 db Sum of Pressures standard deviation Pressure Pressure, Run G, Threshold = 1.5*baseline = 18.6 Pressure UNS detection: (13.7 s) Standard Deviation, Threshold = 1.5*baseline =.1 STD UNS detection: (8.9) Sum of Pressure, Threshold = 1. Sum P UNS detection: (9.9 s) Power Spectral Density, (1-4 Hz band), Threshold = 1.5*baseline = 27.6 db PSD UNS detection: (11. s) Figure 19. Comparison of unstart detection methods for run 774AG. 15

16 IV. Summary and Conclusions An array of high-speed pressure transducers was successfully utilized in the isolator/combustor region of a direct-connect gaseous-hydrocarbon-fueled scramjet. The transducers were used to measure pressure rise for a series of tests that included both started and unstarted runs. Fueling sites were fixed for this investigation, but overall equivalence ratio and fuel distributions between injector sites were varied. Isolator entrance conditions were fixed and represented a Mach 5.5 flight condition. Additionally, a distortion duct was placed upstream of the isolator entrance to simulate the oblique shock that would be generated by the inlet in flight. For this isolator/combustor configuration, the shock-train leading edge location was only a function of the fuel flow rate at the B2 ( B2 ) injection site. Over the range of B2 from , the shock-train leading edge is downstream of the engine throat. For values above.35, the engine was unstarted. The relationship between B2 and shock position was linear and this allowed a control strategy based on an array of pressure transducers to be developed. Mean and standard deviations of pressure for the pre- and post-shock high frequency measurements correlated well with previous measurements in small scramjet combustors and cold-flow back-pressured facilities. Power-spectral-density analysis of the pressure measurements showed relatively low energy content in the pre-shock region. In the post shock region, the measurements show an increase in energy content and a broad plateau from -4 Hz. Comparison of pre- and post-shock PSDs showed an increase in energy content for the post-shock PSD of approximately 2 - over the range of -4 Hz. This value was significantly lower than that reported in a cold flow facility for transducers in a similar location. A summation of individual-pressure measurements was evaluated as a potential shock-position-control indicator. A simple linear model between B2 and the sum of pressures was used to determine when the engine was approaching unstart. In all cases, the predicted value of B2 was within 1% of the actual B2 using the simple model. In the particular isolator/combustor configuration studied, the summation-of-pressure technique was compared against previously established methods for detecting unstart. Detection criteria were established for each method and the time-series data were then post-processed to determine which method provided the most lead time prior to unstart. In cases with minimal transducer noise, the summation of pressure method provided 1-2 s more lead time prior to unstart than the other methods. Acknowledgments The authors acknowledge the AFRL/RZA management (Dr. T. Jackson, Mr. R. Mercier, and Dr. A. Auslender) for their financial and technical support of this effort. Technical discussions with Dr. M. Gruber, Dr. T. Mathur, Mr. D. Barone, and Mr. C. Smith regarding experimental details are also greatly appreciated. References 1 McClinton, C. R., X-43 Scramjet Power Breaks the Hypersonic Barrier Dryden Lectureship in Research for 26, AIAA Paper 26-1, Jan Hank, J. M, Murphy, J. S., and Mutzman, R. C., The X-51A Scramjet Engine Flight Demonstration Program, AIAA Paper , Apr Heiser, W.H. and Pratt, D. T., Hypersonic Air Breathing Propulsion, AIAA Education Series, AIAA, Washington, D.C., Waltrup, P. J. and Billig, F. S., Structure of Shock Waves in Cylindrical Ducts, AIAA Journal, Vol. 11, No. 1, 1973, pp Waltrup, P. J. and Billig, F. S., Prediction of Precombustion Wall Pressure Distributions in Scramjet Engines, Journal of Spacecraft and Rockets, Vol. 1, No. 9, 1973, pp Sullins, G. and McLafferty, G., Experimental Results of Shock Trains in Rectangular Ducts, AIAA Paper , Dec

17 7 Carroll, B. F. and Dutton, J. C., Characteristics of Multiple Shock Wave/Turbulent Boundary Layer Interactions in Rectangular Ducts, Journal of Propulsion and Power, Vol. 6, No. 2, 199, pp Carroll, B. F. and Dutton, J. C., Multiple Normal Shock Wave/Turbulent Boundary-Layer Interactions, Journal of Propulsion and Power, Vol. 8, No. 2, 1992, pp Carroll, B. F., Lopez-Fernandez, P. A., and Dutton, J. C., Computations and Experiments for a Multiple Normal Shock/Boundary-Layer Interaction, Journal of Propulsion and Power, Vol. 9, No. 3, 1993, pp Matsuo, K., Miyazato, Y., and Kim, H.-D., Shock Train and Pseudo-Shock Phenomena in Internal Gas Flows, Progress in Aerospace Sciences, Vol. 35. No. 1, 1999, pp Neaves, M. D., McRae, S., and Edwards, J. R., High-Speed Inlet Unstart Calculations Using An Implicit Solution Adaptive Mesh Algorithm, AIAA Paper , Jan McDaniel, K. S. and Edwards J. R., Three-Dimensional Simulation of Thermal Choking in a Model Scramjet Combustor, AIAA Paper , Jan Dolling, D.S., Fifty Years of Shock-Wave/Boundary-Layer Interaction Research: What Next?, AIAA Journal, Vol. 39, No. 8, 21, pp Clemens, N. T. and Narayanaswamy, V., Shock/Turbulent Boundary Layer Interactions: Review of Recent Work on Sources of Unsteadiness, AIAA Paper , June Cole, G. L., Neiner, G. H., and Crosby, M. J., Design and Performance of a Digital Electronic Normal Shock Position Sensor for Mixed-Compression Inlets, NASA TN D-566, Dec Dustin, M. O., Cole, G. L., and Wallhagen, R. E., Determination of Normal Shock Position in a Mixed-Compression Supersonic Inlet, NASA TM X-2397, Nov Dustin, M. O. and Cole, G. L., Performance Comparison of Three Normal-Shock Position Sensors for Mixed-Compression Inlets, NASA TM X-2739, March Dustin, M. O., Cole, G. L., and Neiner, G. H., Continuous-Output Terminal-Shock-Position Sensor for Mixed- Compression Inlets Evaluated in Wind-Tunnel Tests of YF-12 Aircraft Inlet, NASA TM X-3144, Dec Sajben, M., Donovan, J. F., and Morris, M. J., Experimental Investigation of Terminal Shock Sensors for Mixed- Compression Inlets, Journal of Propulsion and Power, Vol. 8, No. 1, 1992, pp Rodi, P. E., Emami, S., and Trexler, C. A., Unsteady Pressure Behavior in a Ramjet/Scramjet Inlet, Journal of Propulsion and Power, Vol. 12, No. 3, 1996, pp Parrot, T. L., Jones, M. G., and Thurlow, E. M., Unsteady Pressure Loads in a Generic High-Speed Engine Model, NASA TP 3189, Dec Le, D. B, Goyne, C. P., and Krauss, R. H., Shock Train Leading Edge Detection in a Dual-Mode Scramjet, Journal of Propulsion and Power, Vol. 24, No. 5, 28, pp Le, D. B, Goyne, C. P., Krauss, R. H., and McDaniel, J. C., Experimental Study of a Dual-Mode Scramjet Isolator, Journal of Propulsion and Power, Vol. 24, No. 5, 28, pp Wagner, J. L., Yuceil, K. B., Valdivia, A., Clemens, N. T., and Dolling, D. S., Experimental Investigation of Unstart in an Inlet/Isolator Model in Mach 5 Flow, AIAA Journal, Vol. 47, No. 6, 29, pp Valdivia, A., Yuceil, K. B., Wagner, J. L., Clemens, N. T., and Dolling, D. S., Active Control of Supersonic Inlet Unstart Using Vortex Generating Jets, AIAA Paper , June Srikant, S., Wagner, J. L., Valdivia, A., Akella, M. R., and Clemens, N., Unstart Detection in a Simplified-Geometry Hypersonic Inlet-Isolator Flow, Journal of Propulsion and Power (to be published). 27 Wagner, J. L, Yuceil, K. B., and Clemens, N. T., PIV Measurements of Unstart of an Inlet-Isolator Model in a Mach 5 Flow, AIAA Paper , June Gruber, M., Donbar, J., Jackson, K., Mathur, T., Baurle, R., Eklund, D., and Smith, C., Newly Developed Direct-Connect High-Enthalpy Supersonic Combustion Research Facility, Journal of Propulsion and Power, Vol. 17, No. 6, 21, pp Gruber, M., Carter, C., Ryan, M., Rieker, G. B., Jeffries, J. B., Hanson, R. K., Liu, J., and Mathur, T., Laser-Based Measurements of OH, Temperature, and Water Vapor Concentration in a Hydrocarbon-Fueled Scramjet, AIAA Paper 28-7, July Gruber, M. R., Hagenmaier, M. A., and Mathur, T., Simulating Inlet Distortion Effects in a Direct-Connect Scramjet Combustor, AIAA Paper , July