Application of Silicon Carbide Photodiode Flame Temperature Sensors in an Active Combustion Pattern Factor Control System

Carl A. Palmer
e-mail: [email protected]

Royce L. Abel
Impact Technologies, LLC, 200 Canal View Boulevard, Rochester, NY 14623

Peter Sandvik
GE Global Research, Niskayuna, NY 12309

This paper describes the development and initial application studies for an active combustion pattern factor controller (APFC) for gas turbines. The system is based around the use of a novel silicon carbide optical ultraviolet dual diode ame temperature sensor (FTS) developed by General Electric Co. The APFC system determines combustion ame temperatures, validates the values, and integrates an assessment of signal and combustion hardware health to determine how to trim the fuel ow to individual fuel nozzles. Key aspects of the system include the following: determination of each ames bulk temperature using the FTS, assessment of the reliability of the ame temperature data and physical combustion hardware health through analysis of the high-frequency output of the sensor, validation of the ame temperature signal using a data-driven approach, fusion of sensor health indices into the APFC to alter the trim control signal based on the health (or believability) of each sensor and fuel nozzle/combustor, fault-tolerant peak/valley detection and control module that selects individual fuel valves to target for reducing pattern factor while simultaneously balancing the overall fuel ow. The authors demonstrated feasibility of the approach by performing simulations using a quasi-2D T700 turbine engine model. Tests were run on the simulated platform with no faults, simulated sensor faults, and on a system with underlying combustion hardware issues. The nal APFC system would be applicable for aviation, naval, and land-based commercial gas turbines, and can be used in closed-loop control or adapted as an open-loop advisory/ diagnostic system. DOI: 10.1115/1.4001942 Keywords: gas turbine, combustion pattern factor, control, ame temperature, sensor, silicon carbide

Introduction

The combustion pattern factor in a gas turbine is a measurement of the difference between the peak and average temperature at the turbine inlet. The denition of combustor pattern factor is given as follows: PF = Tpeak 1 Tavg

where Tpeak is the measured peak temperature at the combustor exit and Tavg is the average of the measured temperatures. The objective of active pattern factor control APFC is to minimize the pattern factor by modulating the fuel ow to each fuel nozzle while maintaining a constant engine ring temperature. Reducing the burner pattern factor through active control has many potential advantages, including elimination of hot streaks at the turbine inlet, more efcient fuel burning, decreased emissions, and increased life of expensive hot-zone parts such as turbine nozzles and blades. As a secondary effect, APFC also can potentially enable an overall increase in average ring temperature and, thus, engine efciency by reducing required ring temperature operating margins that have been necessary because of large combustion pattern factors. A cost effective, practical, and capable APFC has not been created to dated because of the difculty of sensing or accurately
Contributed by the International Gas Turbine Institute IGTI of ASME for publication in the JOURNAL OF ENGINEERING FOR GAS TURBINES AND POWER. Manuscript received June 19, 2009; nal manuscript received February 9, 2010; published online September 13, 2010. Editor: Dilip R. Ballal.

estimating the actual ame temperature from each fuel nozzle. Typical temperature sensors cannot function at the extreme temperatures that exist in the combustion zone. Therefore, temperature sensors i.e., those used for control are placed in cooler downstream locations-usually after several rows of turbine blades. The limitation of using these downstream measurements for closed-loop pattern factor is that the correlation between which nozzle to control and the location of the hot spot is not exact. This approach also has fundamental accuracy limitations. Therefore, a means to measure each ames actual temperature directly would signicantly enhance the potential for APFC because each ame could be controlled based on a direct measurement. Between 2001 and 2003, General Electric Co. GE modied the technology used in their silicon carbide SiC ame detector for gas turbines to perform as a ame temperature sensor FTS . The FTS proved that it was capable of tracking the ame temperature from a premixed natural gas ame LM6000 premixed fuel nozzle in a laboratory setting. The basic SiC ame temperature sensor also has a fast dynamic response that can be utilized for diagnostics of the sensor and system health.

Technical Approach

Figure 1 provides an overview of the technical approach to control the turbines pattern factor using the FTS. The APFC is made up of several interacting data analysis modules that can be separated into three major components. the ame temperature sensors the combustor and sensor health models that interpret the signals from the FTS JANUARY 2011, Vol. 133 / 011601-1

Journal of Engineering for Gas Turbines and Power Copyright 2011 by ASME

Downloaded 12 Jan 2011 to 210.212.195.230. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm

Fig. 1 Overview of APFC system layout

the peak/valley detector and proportional-integral controller that use the temperature signals and results from the combustor and sensor health models to control individual fuel valves

Other turbine engine operating parameters are used as APFC system inputs. In this simulation study, a T700 engine model is used to produce values for these measured parameters. 2.1 Flame Temperature Sensor. The basic temperature sensing technology utilizes the phenomena that the relative amount of ultraviolet UV light emitted mostly from OH molecules at different wavelengths during a combustion process changes as a function of temperature Fig. 2 . GE exploited this physics by creating a silicon carbide UV photodiode photograph in Ref. 1

in which one half of the area is optically ltered to block certain wavelengths of UV light primarily region 2 of Fig. 2 and the other half is unltered to detect all UV wavelengths to which SiC responds. The ratio of outputs from the two diodes is a function of ame temperature 2 . Entitlement testing using a commercial UV spectrometer determined that the potential performance of the sensor could be 20 F / 11 C. The initial prototype SiC-based device provided a UV light ratio measurement that yielded a performance of 50 F / 28 C in a low-NOx natural gas laboratory combustor. 2.1.1 Flame Temperature Sensor Test Data and Simulation of the FTS Raw Output for Use in Study. The sensor data used in this project is comprised of raw data from the above sensor testing in the low-NOx burner. The high-speed data was comprised of 2000 data points per run at data acquisition frequencies up to 2000 Hz. In the current project, this data was used to simulate the raw signal from the sensorboth the raw high bandwidth output and calculated low bandwidth sensor outputs that are used within the pattern factor controller. To test the overall APFC concept, the authors created a ame temperature simulator FTSim module to simulate the high bandwidth, noisy raw signal that would come directly from the ame temperature sensor over a full range of engine operating conditions. The simulator also models the initial interpretation of the raw signal and calculation of physical values from the sensor i.e., ame temperature . The approach used was to recreate high-speed sensor output at any given temperature based on the frequency spectrum from the raw test data. To simulate the signal at a temperature where no measured data exists, a fast Fourier transform FFT is computed for both the ltered and unltered diodes at that temperature by interpolating between the two spectrums of the available measured temperatures that bound the temperature of interest. The inverse of the spectrum is then calculated and used. The measured temperature is based on the ratio of the means of the ltered diode and unltered diode outputs of the FTS, as preTransactions of the ASME

Fig. 2 UV spectrometer outputUV light spectrum changes with temperature

011601-2 / Vol. 133, JANUARY 2011

Downloaded 12 Jan 2011 to 210.212.195.230. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm

Fig. 3 FTSim processed output

viously discussed. The simulated instantaneous FTS outputs are passed through a double exponential smoothing algorithm as a low-pass lter to determine the signal means. This procedure produces an extremely smooth calculation but introduces a response time of 5 s which can be tuned in the long run . The sensor simulator is demonstrated in Fig. 3 step function is the input; smooth curve is the response . The slow response time is not a signicant negative to the overall project goal, as the APFC is meant to be a trim type control at steady state conditions. In fact, it would be desirable to have a relatively slow controller to ensure that the stability or dynamics of the ame is not affected by the introduction of a new fuel nozzle valve position. 2.2 Modeling the T700 Engine and Planar Temperature Distributions. A simplied linear T700 turbine engine as developed by Duyar et al. 3 was created to test the APFC within an engine system. The inputs to this model are fuel ow rate and power turbine speed. The outputs are generator speed, interturbine temperature T4.5 , shaft torque, and compressor discharge temperature P3 . The T700 engine model outputs a partial 1D prole of the temperature through the engine ow path. To be able to use the model for testing the active pattern factor control, a 2D temperature prole i.e., a planar/annular temperature cross-section must also be simulated at all locations in the engine where multiple temperature sensors exist so that individual temperature sensor response can be derived from turbine nozzle inlet temperature. Two temperature cross-sections are needed by the techniques used in the APFC. These two cross-sections are T4.5 and T4 i.e., at the exit of each of the 12 combustors locations. The T4.5 average plane temperature comes directly from the T700 model. The T700 model used does not calculate an average combustor exit temperature. In lieu of a more exact thermodynamic model, a simple offset of 900 F is added to the T4.5 average plane temperature to obtain the average T4 plane temperature. A more exact model may use pressure ratios, turbine efciency, turbine maps, etc. Within the goals of this effort, this assumption does not have a signicant effect on the resulting conclusions. To simulate an initial pattern factor or misbalance of the system, disturbances are added to the T4 average at the exit of each of the 12 combustors. These disturbances are chosen such that the overall average T4 value would not be altered i.e., overall sum of the disturbances= 0 . Given this T4 prole, the expected temperature distribution prole at the interturbine T4.5 location can be calculated. After the hot combustion gas leaves, the annular combustor the gas diffuses spatially and swirls through two stators and the two rotating turbine stages before the gas reaches the seven T4.5 thermocouples. The swirl and diffusion effects are simulated with an inuence Journal of Engineering for Gas Turbines and Power

Fig. 4 T700 fuel nozzle and T4.5 thermocouple circumferential locations

matrix equation shown below derived from Ref. 4 . This has a swirl of 45 and diffusion standard deviation of 30, where is the angular difference degrees between the ith thermocouple T4.5 location and the jth nozzle location shown in Fig. 4.
i + 2 2 j 45 2

mi,j = 0.3989e

The gain of the inuence matrix equation was calculated to eliminate the error between the expected T4.5 mean from the T700 engine model and the mean from the T4.5 distribution calculated via this manner. 2.3 Combustion and Sensor Health Model. The combustion and sensor health model interprets the raw signals from the FTS and calculates parameters used by the controller. This is made of ve different subsystem modules. 1. 2. 3. 4. 5. FTS high-frequency health analysis general low frequency sensor health analysis combustion/nozzle diagnostic health analysis using FTS data direct FTS temperature computation fusion of the FTS computation and the health indicators

The rst three subsystems are used to nd any measurable faults in the engine or sensors that should change the way the controller should respond. The fourth subsystem is the computation of ame temperatures with a linear transform using the smoothed ratio of the ltered and unltered outputs of the FTS. The smoothing algorithm ensures that high-frequency dynamics inherent in combustion processes observed by the FTS is not transferred to the controller. The fth subsystem fuses all the data that is calculated within the combustion and sensor health model. The output of this health model is sent to the controller to trim the fuel ow to each nozzle. 2.3.1 FTS High Frequency Health Analysis. The highfrequency signal analysis module acquires a stream of data directly from the FTS i.e., the FTSim in this analysis and processes that information to calculate certain numerical features that are related to the health of the signal and/or health of the combustion occurring from the burner monitored by that sensor. The module calculates the mean, range, random anomaly, signal clipping, and instantaneous temperature of the ltered and unltered FTS output within a time window. See Palmer et al. 5 for details on these algorithms . These values are then used in the model-based validation MBV routine to determine if the FTS is JANUARY 2011, Vol. 133 / 011601-3

Downloaded 12 Jan 2011 to 210.212.195.230. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm

Fig. 5 Numerical features processed from high-frequency data from ltered and unltered diodes

functioning according to how a healthy system would. As a demonstration of the high-frequency signal processing module, raw FTS sensor data associated with a constant temperature of 2700 F / 1482 C was the input to the FTSim and the outputs of the ltered and unltered diodes were calculated. The results of this simulation for healthy data are shown in Fig. 5. The outputs shown in this module are periodic over 1 s only because this is the length of the snapshots of raw high-frequency data taken during the combustor tests used as inputs to the FTS simulator for this study. This 1 s is repeated continuously for longer time simulations. The outputs of the various numerical features calculated from the high-frequency data are inputs used in the overall model-based validation routine. 2.3.2 MBV Routines. Signal validation using data-driven models is used in two different modules in the APFC. The rst MBV model is based around the FTS itself. The model estimates the steady state FTS output based on all other values from that single FTS and only that FTS . For this, the inputs are the features calculated from the high-frequency health analysis. This works because these numerical features i.e., high-frequency behavior are typically consistent with the current temperature of the ame being monitored. The results of this module on data from a healthy sensor are shown in Fig. 6. The MBV estimate of temperature is bounded by 12 F. As an example, the high-frequency component of the range max minus min of the FTS output can be used as a good diagnostic feature. The range of the sensor represents the limits of the oscillating intensity of the ame. If some sort of physical change in the actual ame were to occur e.g., from coking or local lean ow conditions , this range will most likely change. If the range changes in a manner that is inconsistent with the actual tempera-

Fig. 7 Neural network performance for predicting individual fuel nozzle combustion temperature from downstream thermocouple readings

ture changing, then the MBV algorithm will output an estimated temperature reading that is different than the temperature being sensed by the FTS. That is, the error between the sensed and model-based estimates of the temperature increases the uncertainty in the health of the system. The controller should adjust its outputs to compensate for this uncertainty. The second instance where a model-based validation technique is used is to verify individual ame temperature sensor readings based on other sensor readings across the overall T700 engine. Sensor readings include each separate downstream thermocouple as well as changes in nozzle fuel valve position. Basically, this is data-driven modeling of the quasi-2D ow path from the combustors to the seven T4.5 thermocouples. This relationship was modeled using an articial neural network ANN . The performance of the ANN that was developed to estimate the burner temperatures from the downstream interturbine thermocouple temperatures is shown in Fig. 7. The linear trend of the overall data is very good R value of 0.97 . The variance around each output value is occurring because the target values were developed from the instantaneous FTS outputs, which are very dynamic because of the uctuation of the combustion ame. The points that drift away from the line occur during operating transients of the engine. The fact that these points are not predicted well is not a concern because the APFC will not and should not be active during transient operations. 2.3.3 Low Frequency Sensor Health Analysis. The low frequency signal processing module is used to calculate indices for the health of every sensor not just the FTS being used by the APFC system. The module receives a data window of current and recent historical data. For example, a sensor measured at 10 Hz with a desired health evaluation every second would have a 20 point data window representing the past 2 s delivered to the module. This module is used to assess the health of the signals for the T4.5 temperatures, P3 pressure, generator speed, power turbine shaft torque, and calculated steady state ame temperature from each FTS. The low frequency signal processing module calculates ve outputs that assess different aspects regarding the health of each sensor. These include the following. Noise estimation: Is the signal moving around randomly without physical signicance? Transience: Is the signal moving too quickly? Faster than reasonably expected for this sensor? Saturation: Is the signal outside its normal range of operation? Transactions of the ASME

Fig. 6 Model based validation of FTS at steady state conditions based on correlations of high-frequency sensor response characteristics to ame temperature

011601-4 / Vol. 133, JANUARY 2011

Downloaded 12 Jan 2011 to 210.212.195.230. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm

Fig. 8 Flowchart for estimating fuel nozzle health

Self-correlation: Is the signal close to where it has been in the recent past? Filter health: This gives a broad view of high-speed transients in the signal.

2.3.4 Combustion/Nozzle Diagnostic Health Analysis Using FTS Data. The concept behind this module is that if there are any anomalies within the dynamic signature of a fuel nozzle that does not correspond with all the other nozzles, the nozzle is behaving differently, and therefore, the health of the specic nozzle is most likely in a degraded state. The procedure for estimating fuel nozzle hardware health is outlined below and shown graphically in Fig. 8. Acquire raw high bandwidth data from each FTS. Compute the FFT from the dynamic signal to get a frequency domain ngerprint of each fuel nozzles behavior. Normalize each frequency spectrum to have equal power. This reduces the potential problem of one optical sensor being dirtier than the rest. Break each FFT into 4 separate frequency bins: 10 100 100400 400600 and 600900 Hz. Compute the power of each bin. Compute the average power of each frequency bin for each sensor. Compute a percent difference of each frequency bin for a single nozzle compared with the average value for that bin from all the nozzles and normalize to a value between 0 and 1. Take the mean of all health gauges calculated within a nozzle to get a single value for the health of that nozzle.

system FTS and T700 operating conditions and outputs a single sensor health indicator for each FTS. Each FTS health indicator is averaged with the health estimate of each T700 parameter generator speed, shaft torque, P3 pressure, and T4.5 thermocouples to achieve a nal summary health indicator for each FTS Fig. 9 . 2.3.6 Fusion of the Computed FTS Temperature and the Health Indicators. This process gives the controller the actual combustion temperature imbalance on which to act. This involves subtracting the average nozzle temperature from each individual nozzle temperature. If the FTS worked perfectly and was perfectly accurate and the nozzles did not degrade over time there would not be any other processing required. The measured imbalance could then be sent to the controller to actuate the valves. In application, the health of the FTS and nozzle are changing and the health estimates should be used to augment the estimate of the imbalance and affect how to use the sensor information in the controller. The approach of using the health estimates to augment the calculated error signal sent to the controller is to scale each ame temperature estimate by the FTS health estimate 01 and the health of the nozzle holding the ame 01 . Figure 10 gives an example. If the FTS and nozzle health estimate are both 1, the

The drawback to this approach is that if all the fuel nozzles were to degrade at the same time, there would be no indicators. This failure, although possible, is unlikely and would most likely be a greater problem for the engine than the APFC. This procedure is a good starting point, and as feature or failure signatures are identied in the long term, each can be separated and processed to improve the analysis of hardware health in the faulttolerant controller. 2.3.5 Sensor Health Estimate Fusion. The sensor health estimate fusion module receives all the health indicators from the
Fig. 9 Sensor health estimate fusion owchart

Fig. 10 Flowchart for combustion temperature imbalance estimate compensation for sensor and hardware health

Journal of Engineering for Gas Turbines and Power

JANUARY 2011, Vol. 133 / 011601-5

Downloaded 12 Jan 2011 to 210.212.195.230. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm

temperature imbalance will not be altered. If one or both of the values are equal to 0 then the controller will not target that valve for control. If the health values are between 0 and 1, there will be different amounts of control effort. When a nozzle/FTS pair is believed to be unhealthy for example, ame 4 in Fig. 10 , the corresponding valve is not targeted for actuation in the overall pattern factor. As more diagnostics are implemented into the combustion and sensor health model and greater knowledge of the FTS in real use on a platform is gained, different adjustments and weighting factors can be made on the imbalance. For example, when the estimated health of a ame temperature sensor is lowered, the uncertainty in its value is increased. The controller may be set to intentionally slightly limit the fuel to that fuel nozzle maintaining a lowered measured temperature while taking a penalty of higher fuel ows to the other nozzles . The concept behind this would be to equalize the probability of every nozzle being in an unacceptably high temperature region. Additionally, as a protection mechanism, if a ame starts to become unstable on the lean side, then that valve could be adjusted to increase the fuel ow to stabilize the ame and a diagnostic reading can be given to the operator/maintainer about the situation. This is not currently the base APFC functionality but should be included if ammability and stability limits are an issue. 2.4 Active Pattern Factor Controller. A number of algorithms for controlling the individual fuel valves were evaluated. These included a decentralized Proportional plus Integral controller and a neural-network based adaptive controller. In the end, the team developed a controller, which consists of a fault-tolerant peak/valley detection/switching module integral controller. The fault-tolerant peak/valley detection/switching module rst uses the condence of nozzles and sensors to determine whether the nozzle and/or sensor is healthy. If the condence of the nozzle/sensor is less than a threshold e.g., 0.85 , this nozzle/sensor is declared to be faulty, and is disconnected from the closed-loop control system. Only the healthy sensors/nozzles are then used for the closed-loop pattern factor control. The system then nds a healthy temperature sensor with the maximum temperature reading. The corresponding fuel modulator is selected for the fuel ow modulation. The peak temperature deviation is used as a feedback through a constant gain to the integrator associated with that particular fuel modulator. In the next processing cycle the minimum ame temperature reading is found and the corresponding fuel modulator is selected.

Fig. 11 How health gauges compensate t imbalancehealthy system with initial imbalance. Each trace represents a separate fuel nozzle

3.1 Healthy System. In the simulation of the healthy system presented in Figs. 11 and 12, the low frequency health gauge is the primary health gauge in effect middle left of Fig. 12 . This is to be expected, because a portion of this health gauge is monitoring the change in the downstream thermocouples. During an engine transient, or when the system is controlling the individual fuel nozzle valves, there will be changes in the thermocouple readings. Any change in the thermocouple readings away from steady state causes the system to downgrade the estimated health of the thermocouples. At rst, this may not make sense because the system is performing as expected but if the thermocouples are changing rapidly this could indicate a problem in the pattern factor control valves. Therefore, by reducing the estimated health of the thermocouples a portion of the general low frequency sensor health model the amount of control sent to the individual fuel valves will effectively be reduced as well. This has the effect that the controller does not work excessively fast. Once the controller is almost done acting on the individual fuel nozzle trim valves, the condence in the thermocouple readings approach 1.0. Other than the low frequency health gauge, the rest of the

The process brings the highest temperature nozzle to the mean temperature reading before altering the subsequent nozzles. Note that as the peak nozzle temperature is reduced by T, all the other nozzle temperatures should increase by roughly T / #nozzles 1 , as this APFC trim control is outside of the overall global fuel control scheme.

Simulation Results

To test the APFC, the system was run through a number of simulated scenarios. Two are presented here. In these two scenarios, the T700 engine repeated the same engine operating cycle. The system is given an initial unbalanced pattern factor. The engine cycle starts with a throttle step input at t = 0. When the engine reaches a steady state at around the 9 s point , a mode detector turns on the pattern factor controller. The rst simulation is a of a health system. The second simulation introduces a nozzle malfunction. Other simulations run but not included in this paper included dirty optics e.g., viewport , sensor failures, and sensor biases. 011601-6 / Vol. 133, JANUARY 2011

Fig. 12 Calculated health gauges in healthy system

Transactions of the ASME

Downloaded 12 Jan 2011 to 210.212.195.230. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm

Fig. 13 How health gauges compensate t imbalance irregular high-frequency combustion nozzle malfunction

Fig. 15 Effect of sensor accuracy x axis on eventual achievable pattern factor

health indicators in Fig. 12 are above 0.9 in this simulation. The closer the gauges are to 1.0, the more likely it is that the parameters the gauges monitor are healthy. 3.2 Nozzle Malfunction: Irregular High Frequency Dynamics. In this simulation, high-frequency noise is introduced to the FTS ltered and unltered outputs for one sensor. This simulates a change in the ame dynamics as seen by UV light coming from the rst nozzles sensor that may come from degradation in the combustion, such as possible coking on the nozzle. In this case it is uncertain how the temperature reading of the sensor would respond. It is hypothesized that the temperature reading from the FTS could go in any direction depending on the shape of the nozzle spray that develops after the nozzle problem. For this simulation, we kept the temperature reading unchanged. The nozzle health gauge did appropriately indicate a possible failure in Fig. 13 upper right and Fig. 14 middle right . 3.3 Assessment of Sensor Accuracy Requirement. A Monte Carlo study was conducted to determine what accuracy requirements need to be imposed on the FTS to make APFC viable. In this study, a hard bias is introduced to each FTS simulation module. The errors introduced to each sensor were from random values selected from a normal distribution curve. The error band for the normal distribution was dened as a 95% 2 sigma condence that the sensor values would be within. For this study, the

error bands used for the sensor were 25 F, 50 F, 100 F, 200 F, and 400 F 14 C, 28 C, 56 C, 111 C, and 222 C, respectively . The resulting range of nal pattern factor after control for ten runs using each of these error bands is shown in Fig. 15. Summarizing these results, the two sigma precision of the FTS measurement should be within 100 F / 56 C to have signicant impact in reducing the pattern factor. Sensors with larger measurement errors could still have some positive end-effect but can produce slightly worse pattern factors increased peak temperatures during the APFC transient processes not shown . Note that this 100 F / 56 C measurement requirement has already been achieved by the prototype FTS. Therefore, it is expected that a production sensor with even lower measurement error, has strong potential to perform very well for APFC.

Conclusions

The APFC system determines combustion ame temperatures, validates the values, and fuses the calculated temperature values with combustion/sensor hardware health estimates to control fuel ow to individual fuel nozzles to minimize combustion pattern factor. Simulations performed on a T700 turbine engine model have demonstrated the concept and the feasibility of the approach. Simulations indicate that ame temperature sensors with a two sigma accuracy of 100 F / 56 C coupled as part of this selfvalidating fault-tolerant APFC would enable substantial potential improvement in combustion pattern factor.

Acknowledgment
This APFC work by Impact Technologies was supported by the Air Force Research Laboratory under Small Business Innovative Research under Contract No. FA8650-06-M-2656. The Flame Temperature Sensor work by GE was supported under DOENETL Cooperative Agreement Contract No. DE-FC2601NT41021. The authors would like to thank AFRL employee Ken Semega for his input and support. The authors would also like to give special thanks to the creators of the SiC ame detector, Leo Lombardo, and Dale Brown for sharing their knowledge and passion over the years. JANUARY 2011, Vol. 133 / 011601-7

Fig. 14 Health gaugesirregular high-frequency combustion

Journal of Engineering for Gas Turbines and Power

Downloaded 12 Jan 2011 to 210.212.195.230. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm

Nomenclature
ANN APFC FFT FTS FTSim GE MBV P3 SiC T T4 T4.5 T700 TC articial neural network active pattern factor control fast Fourier transform ame temperature sensor ame temperature simulator General Electric Co. model-based validation compressor discharge pressure silicon carbide UV sensitive photodiode temperature combustion temperature in combustor interturbine temp after gas generator expander turbine type used for study thermocouple

UV

ultraviolet light

References
1 Sandvik, P., Brown, D., Fedison, J., Matocha, K., and Kretchmer, J., 2005, Dual-SiC Photodiode Devices for Simultaneous Two-Band Detection, J. Electrochem. Soc., 152 3 , pp. G199G202. 2 Brown, D. M., Fedison, J. B., Hibshman, J. R., Kretchmer, J. W., Lombardo, L., Matocha, K. S., and Sandvik, P. M., 2005, Silicon Carbide Photodiode Sensor for Combustion Control, IEEE Sens. J., 5 5 , pp. 983988. 3 Duyar, A., Gu, Z., and Litt, J., 1992, A Simplied Dynamic Model of the T700 Turboshaft Engine, NASA Glenn Technical Report No. NASA/TM105805. 4 McCarty, B., Tomondi, C., and McGinley, R., 2004, Reliable and Affordable Control Systems Active Combustor Pattern Factor Control, NASA Glenn Technical Report No. NASA/CR-2004-213097. 5 Palmer, C. A., Mackos, N., and Roemer, M. J., 2007, Approach to Monitor and Assess the Quality of Sensor Data in Support of Calibration and Health Maintenance, ASME Paper No. GT2007-28251.

011601-8 / Vol. 133, JANUARY 2011

Transactions of the ASME

Downloaded 12 Jan 2011 to 210.212.195.230. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm