Lean Blowout Precursor Detection for Gas Turbines

Abstract

A method for detecting a blowout precursor in at least one gas turbine combustor includes receiving combustion dynamic acoustic data in real time measured by an acoustic measurement device associated with the combustor, performing a wavelet analysis on the acoustic data using a simplified Mexican Hat wavelet transform analysis, and determining the presence of a blowout precursor based at least in part on the wavelet analysis. Also provided are a system and a non-transitory computer-readable medium configured to perform the method.

Description

[Background technology]

[0001] Gas turbines have become a leading technology for new power generation potential in the United States and around the world because they operate more efficiently and produce fewer pollutants than traditional power generation systems. The need for tighter emissions targets or improved fuel economy and reliability is driving the state of the art in today's gas turbine engines.

[0002] In a typical gas turbine engine, air is compressed and forced through a combustion zone where one or more fuel injectors provide a stream of fuel that is combusted by the high-pressure air in a "can" annular flame holder. Such gas turbines can have combustion temperatures in excess of 2000°C. The exhaust gases then drive a turbine to ultimately generate electricity.

[0003] Gas turbine engine pollutant emissions such as NOx can be reduced by utilizing very lean fuel-air mixtures, but this causes increased combustion instability within the gas turbine. If the gas turbine is operated under conditions where the fuel-to-air ratio is insufficient to sustain a flame within the can, the flame will go out (called a lean blowout) and the engine may need to be restarted. Summary of the Invention [Problem to be solved by the invention]

[0004] Due to today's NOx regulations, many gas turbines operate with a very thin margin (fuel-air ratio) for lean blowout (LBO), and the period from LBO onset to flameout can occur within milliseconds. When certain operating profiles are implemented with certain gas turbine models, blowout occurs without warning and without mitigation by the original equipment manufacturer's software. [Means for solving the problem]

[0005] Provided is a method for analyzing combustion dynamics and operational data leading up to or associated with a lean blowout (LBO) event. The combustion dynamics may be pressure waves of defined amplitude and frequency associated with natural acoustic modes of the combustion system. In a typical can-annular combustor in a large gas turbine, the combustion dynamics may range in frequency from less than 50 Hz to several thousand Hz. The analysis of the present invention identifies specific signals in the gas turbine operational data that indicate the onset of an LBO.

[0006] Use of the present invention may enable improved gas turbine operating flexibility, reduced power generation operating costs, and / or reduced risk of outages due to emissions compliance.

[0007] The method of the present invention utilizes a computer algorithm that analyzes combustion dynamics data (time series data) in real time. The algorithm convolves the time series data with Mexican Hat wavelet bases. The results of the convolution are called time-dependent wavelet coefficients. The wavelet time scale is set to capture high-frequency (kilohertz rate) temporal oscillations in the combustion dynamics data, which differs from known techniques that capture low-frequency dynamics.

[0008] This method also differs from other known techniques because it simplifies the wavelet convolution. These simplifications reduce the dimensionality of the convolution, which saves computational cost so that the algorithm can be implemented quickly for fast detection, as described later in this specification.

[0009] Implementation of this algorithm requires a data acquisition system capable of sampling rates of tens of kilohertz. Wavelet coefficients above a threshold are interpreted as blowout precursors, indicating insufficient blowout margin within the combustor that generated the precursor.

[0010] Gas turbines may be operated at very low power levels where blowout presents a substantial risk. The method of the present invention reduces the risk of blowout in such conditions by a) providing active monitoring for impending blowout and b) enabling counter-blowout adjustments by providing feedback regarding combustors presenting the greatest blowout risk.

[0011] Other variations of the algorithm have been explored, for example using fast Fourier transforms, but their implementations have not been fast enough for real-time monitoring.

[0012] Embodiments of the present subject matter are disclosed with reference to the accompanying drawings, and are for illustrative purposes only. The present subject matter is not limited in its application to the details of construction or arrangement of components shown in the drawings. As used herein, "at least one" means one or more than one, and "and / or" means that the listed items may be included singly or in combination. [Brief explanation of the drawings]

[0013] [Figure 1] FIG. 10 is a graphical representation (spectrogram) of the superposition of transform coefficient spectra from the last 10 seconds before LBO, with baselines bounding two high frequency regions: between 2,000 and 3,000 Hz and between 4,000 and 6,000 Hz. [Figure 2] This is a diagram of a graphical representation of the Mexican hat basis function w(t) using σ=1. [Figure 3] FIG. 3 is a graphical representation of the spectrum of the wavelet basis shown in FIG. 2. [Figure 4a] A baseline is given at the same amplitude. a) Graphical representation of a comparison of time-dependent wavelet coefficients for an acoustic "event" but without LBO. [Figure 4b] b) Graphical representation of a comparison of time-dependent wavelet coefficients with and without LBO, where the baseline is given at the same amplitude. [Figure 5a] 4A and 4B are graphical representations of average exhaust gas path temperatures for cases 4a and 4b with a hiccup (no LBO), where dashed lines indicate acoustic events associated with the hiccup or LBO. [Figure 5b] The dashed lines indicate the acoustic events associated with the LBO or temporary interruption. b) Graphical representation of the average exhaust gas path temperature for cases 4a and 4b with LBO. [Figure 6a] The LBO point appears in the upper left part of the graph, and the temporary break point appears in the lower right part of the graph. a) Graphical representation of the RMS of wavelet coefficients for low frequency wavelets. [Figure 6b] The LBO point appears in the upper left part of the graph, and the temporary break point appears in the lower right part of the graph. b) Graphical representation of the RMS of wavelet coefficients for high frequency wavelets. [Figure 7] 1 is a block diagram illustrating certain embodiments of the method of the present invention. [Figure 8] 1 is a block diagram illustrating certain embodiments of the system of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0014] The following embodiments of the present subject matter are contemplated.

[0015] 1. A method for detecting blowout precursors in at least one gas turbine combustor, comprising: receiving combustion dynamics acoustic data in real time measured by an acoustic measurement device associated with the combustor; performing wavelet analysis on the acoustic data using simplified Mexican Hat wavelet transform analysis; and determining the presence of blowout precursors based at least in part on the wavelet analysis.

[0016] 2. The method of embodiment 1, wherein said determining the presence of a blowout precursor comprises determining an increase in amplitude of time-dependent spectral content within at least one predetermined band of high frequency dynamics.

[0017] 3. The method of embodiment 2, wherein said band of high frequency dynamics is predetermined based on identifying a band of high frequency dynamics that appeared for the same combustor about 1 second prior to a previous blowout event.

[0018] 4. The method of any one of embodiments 1 to 3, wherein performing the wavelet analysis comprises determining dominant frequencies of oscillation of the acoustic data signal as a function of time to calculate wavelet coefficients.

[0019] 5. The method of embodiment 4, further comprising using a windowed root-mean-square calculation to process the wavelet coefficients to determine the amplitude of the wavelet coefficients, and determining the presence of a blowout precursor based on increased amplitude of the wavelet coefficient oscillations.

[0020] 6. A system for acoustic detection of blowout precursors in at least one gas turbine combustor, comprising: an acoustic measurement device in communication with the combustor and generating signals indicative of acoustic combustion dynamics within the combustor in real time; and a blowout precursor monitor unit receiving the acoustic signals and performing a simplified Mexican Hat wavelet transform analysis to detect the presence of blowout precursors.

[0021] 7. The system of embodiment 6, further comprising a combustion controller configured to control at least one parameter of operation of the combustor based at least in part on the detection of a blowout precursor by the blowout precursor monitor unit.

[0022] 8. The system of embodiment 7, wherein the combustion controller is configured to generate at least one control signal upon detection of a blowout precursor to adjust the fuel-air ratio of fuel and air supplied to the combustor associated with the blowout precursor.

[0023] 9. The system of any one of embodiments 6 to 8, wherein the blowout precursor monitor unit detects the presence of a blowout precursor by determining an increase in amplitude of time-dependent spectral content within at least one predetermined band of high-frequency dynamics.

[0024] 10. The system of embodiment 9, wherein the at least one band of high frequency dynamics is predetermined based on identifying a band of high frequency dynamics that appeared for the same combustor about 1 second before a previous blowout event.

[0025] 11. The system of any one of embodiments 6 to 10, wherein the blowout precursor monitor unit performs wavelet analysis by determining dominant frequencies of vibration of the acoustic signal as a function of time to calculate wavelet coefficients.

[0026] 12. The system of embodiment 11, wherein the blowout precursor monitor unit uses a windowed root-mean-square calculation to process the wavelet coefficients to determine the amplitude of the wavelet coefficients, and detects the presence of a blowout precursor based on increased amplitude of the wavelet coefficient oscillations.

[0027] 13. The system of any one of embodiments 6 to 12, wherein the blowout precursor monitor unit, upon detection of a blowout precursor, sends an alarm signal to the electronic device and / or sends a signal to the combustion controller indicating the detection of a blowout precursor.

[0028] 14. A non-transitory computer-readable storage medium having executable program code encoded thereon for performing a method for detecting blowout precursors in at least one gas turbine combustor, the method comprising: receiving, in real time, combustion dynamics acoustic data measured by an acoustic measurement device associated with the combustor; performing a wavelet analysis on the acoustic data using a simplified Mexican Hat wavelet transform analysis; and determining the presence of a blowout precursor based at least in part on the wavelet analysis.

[0029] 15. The non-transitory computer-readable medium of embodiment 14, wherein the step of determining the presence of a blowout precursor comprises determining an increase in amplitude of time-dependent spectral content within at least one predetermined band of high-frequency dynamics.

[0030] 16. The non-transitory computer-readable medium of embodiment 15, wherein the band of high frequency dynamics is predetermined based on identifying a band of high frequency dynamics that appeared for the same combustor about 1 second before a previous blowout event.

[0031] 17. The non-transitory computer-readable medium of any one of embodiments 14 to 16, wherein performing the wavelet analysis includes determining dominant frequencies of oscillation of the acoustic data signal as a function of time to calculate wavelet coefficients.

[0032] 18. The non-transitory computer-readable medium of embodiment 17, wherein the method further comprises using a windowed root-mean-square calculation to process the wavelet coefficients to determine the amplitude of the wavelet coefficients, and determining the presence of a blowout precursor based on the increased amplitude of the wavelet coefficient oscillations.

[0033] The method starts with combustor acoustic data and utilizes wavelet-based analysis configured to provide computational efficiency beyond traditional Fourier transform and other wavelet-based approaches to detect high-frequency blowout precursors in combustion dynamics data, identify combustors experiencing an impending blowout, and perform the detection in sufficient time to detect and respond to the precursors. Additionally, the method identifies which combustors experienced LBO issues during operation so that corrective adjustments to those combustors can be made after the fact.

[0034] The method may be implemented through the use of a non-transitory computer-readable medium having encoded thereon executable program code for performing the method or including instructions configured to be executed by a processor of a system for acoustic detection of blowout precursors in at least one gas turbine combustor, the instructions including instructions configured to cause the processor to perform the steps of the method.

[0035] The LBO detection algorithm consists of a wavelet-based analysis. Wavelet analysis is applied in real time to the acoustic data from each combustor. Wavelet analysis allows for simultaneous data analysis in the time and frequency domains. Additionally, wavelet analysis can be applied to select frequencies of interest without wasting computational effort on other frequencies. The output of the wavelet analysis is a vector of time-dependent coefficients for each frequency of interest. The coefficients in each vector represent the acoustic amplitude at that frequency and time. The values ​​of these coefficients oscillate around zero and have peak-to-peak amplitudes related to the acoustic amplitude. These coefficients are post-processed using a windowed root-mean-square (RMS) calculation to "flatten" their oscillatory nature. This RMS value is monitored as an indicator of LBO precursors, with elevated values ​​indicating an impending LBO event.

[0036] An initial analysis may be performed to characterize at least one band of high-frequency dynamics that typically appears during the first second or so before an LBO and that typically appears less frequently before a "hiatus" or "near-LBO" event. Fourier analysis is one possible tool for characterization. Wavelet analysis would target these same frequency ranges, but requires better time resolution and overall more efficient and faster algorithms.

[0037] The purpose of the wavelet transform is to determine the frequency or spectral content of a signal as a function of time. The spectral content of a signal can never be perfectly separated in time. However, the wavelet transform is well suited to approximating the time-dependent spectral content. Wavelets can do this better than traditional Fourier analysis.

[0038] Wavelet analysis generally consists of a discrete inner convolution of a wavelet basis function with a signal. The two waveforms are convolved in time over a short time window (typically a few periods of oscillation of the basis functions), and this convolution is repeated at each time step where the signal is "updated." In other words, the two waveforms are cross-correlated in time at each time step. This identifies how "similar" a short portion of the signal is to the wavelet.

[0039] This analysis uses a "Mexican hat" basis function, which provides a good approximation of the dominant frequency of a signal's oscillations as a function of time with good time resolution. The Mexican hat basis function is named for the "sombrero" shape of the function. This basis function is plotted in FIG. 2, and its functional form in the variant used herein is expressed in Equation 1. This wavelet offers a good tradeoff between time and frequency separation. TIFF0007776447000001.tif2484 formula 1

[0040] The wavelet analysis used in this algorithm is performed on the dominant frequency of each band of high-frequency dynamics selected during characterization. Targeted analysis at only a few frequencies, sometimes referred to herein as "simplified wavelet analysis" or "simplified Mexican hat wavelet analysis," is more efficient than analysis at many frequencies. In certain embodiments, the wavelet analysis used in this algorithm may be performed on two frequencies (i.e., two values ​​of σ).

[0041] In addition, the algorithm performs a dot product between the basis functions and the signal samples only once per time step; there is no convolution of the basis functions with the samples (i.e., no "shifting"). This adds significant computational efficiency. However, this approach produces oscillatory behavior in the wavelet coefficients, which can be handled in post-processing. As the frequency of the signal approaches the frequency of the wavelet basis, the peak-to-peak amplitude of these oscillating wavelet coefficients increases. Because the basis functions are held stationary and the signal is shifted relative to the basis at each time step, the period of oscillation of the wavelet coefficients is equal to the period of oscillation of the signal samples. In other words, the wavelet coefficient values ​​are repeated each time the signal "looks the same."

[0042] Windowed root mean square (RMS) is used for post-processing of the wavelet coefficients. RMS is the sum of the wavelet coefficient amplitude (A) and the peak-to-peak amplitude (A PTP ) and is related by the following equation for sinusoidal vibration: amplitude; TIFF0007776447000002.tif1142 Peak-to-peak amplitude: TIFF0007776447000003.tif1249

[0043] This RMS calculation requires at least one period of the frequency of interest to capture the amplitude of the wavelet coefficients. For example, if a signal is being examined for its 1,000 Hz frequency content with a wavelet centered at 1,000 Hz, the RMS calculation would require post-processing the wavelet coefficients with a window for the RMS calculation of at least 1 millisecond (calculated as the reciprocal of the frequency).

[0044] That is, the present method differs from other known techniques because it involves simplified wavelet analysis and simplified wavelet convolution. These simplifications reduce the dimensionality of the convolution, saving computational overhead so that the algorithm can be implemented quickly for high-speed detection. Typical wavelet convolutions include shifting the basis frequency, shifting the basis phase, and time-shifting the signal. These simplifications eliminate the basis frequency and phase shift as follows: The present method formulates wavelet bases at only one or two predetermined frequencies instead of sweeping the base through a wide range of frequencies. The present method establishes a base with constant phase. The final result of the convolution is an oscillating wavelet coefficient. The present method calculates the root mean square (RMS) of the oscillating wavelet coefficient to indicate its amplitude.

[0045] Example Acoustic "events" without LBO will be referred to herein as "hiatuses." The cases used in the experimental method showed a reproducible 4-fold difference in wavelet coefficient RMS for the LBO case versus the hiccups.

[0046] The method has been historically validated against archived combustion dynamic acoustic data from a can-annular gas turbine with multiple combustors. In the present test example, the can-annular gas turbine had 10 combustors, but the applicability of the method is not limited to that number of combustors, but rather is applicable for use with gas turbines with more or fewer than 10 combustors.

[0047] During historical data validation testing of the example, gas turbine characterization Fourier analysis was performed on combustion dynamics data consisting of 60-second recordings sampled at 12,500 Hz for each of 10 combustors. The Fourier analysis was performed by scanning the Fourier transform in a 1-second window through each of the 10 signals. The resulting time-dependent Fourier coefficients were plotted as a function of time and frequency for each combustor. Discrete spectra were plotted and overlaid for each second during the 10 seconds leading up to LBO. An example of the spectral overlay is shown in Figure 1, which overlays 10 spectra from each of the 10 seconds leading up to LBO. Near LBO, high-frequency dynamics are evident in two bands, near 2,500 Hz and 5,000 Hz. These bands are enclosed by vertical reference lines in Figure 1. Specific target frequencies may vary depending on the gas turbine equipment used and other factors, as discussed below.

[0048] In the example of Figure 1, two wavelet analyses would be performed, one targeting a frequency of 2500 Hz and the other targeting a frequency of 5000 Hz. The frequency is targeted by adjusting the value of σ in Equation 1 so that the peak of the basis function is centered at the desired frequency. This center frequency is approximately related to the scaling parameter by Equation 2: TIFF0007776447000004.tif2238 formula 2

[0049] Figure 2 shows an example of a basis function with σ = 1. Figure 3 shows the Fourier analysis spectrum of this basis function.

[0050] Figures 4a and 4b compare wavelet coefficients from two cases: one where there was an acoustic event before RMS post-processing (the system recovered and there was no LBO) and one where there was an LBO. For comparison, a dashed set of reference lines of the same amplitude is shown in both cases. This figure reveals two things. First, it reveals the oscillatory nature of the wavelet coefficients. Second, it reveals the increased amplitude of the wavelet coefficient oscillations preceding the LBO.

[0051] For acoustic "events" (hiccups) without LBO, an exhaust gas path analysis was performed to confirm the absence of LBO. The exhaust gas path analysis for the two cases in Figures 4a and 4b is shown in Figures 5a and 5b, which are graphical representations of the average exhaust gas path temperature for a) the hiccup and b) the LBO case. The dashed lines indicate the acoustic events associated with the hiccup or LBO event. The figure shows that there is no temperature disturbance in the hiccup case, but that there is a temperature dip associated with the LBO and re-ignition in the LBO case.

[0052] The entire method was tested against several sets of historical data, all containing "acoustic events," some of which followed LBOs and some of which only involved hiccups. The results of the historical testing are shown in Figures 6a (low-frequency wavelet) and 6b (high-frequency wavelet), which plot the wavelet coefficient amplitude RMS for each case, with the LBO point appearing in the upper left portion of the graph and the hiccup point appearing in the lower right portion. Figures 6a and 6b show a reproducible four-fold difference in the wavelet coefficient RMS for the LBO case versus the hiccup. Note that this factor may vary depending on the specific gas turbine device in use.

[0053] High-frequency dynamics offer better forecasting opportunities because many cycles of high-frequency oscillations can be observed in a short period of time compared to low-frequency oscillations. Also, high-frequency dynamics can better distinguish "hits" from true LBO events.

[0054] In certain embodiments, the combustion dynamics monitoring system collects dynamic data in the time domain. The time domain data may include pressure oscillations, acoustic data, electromagnetic emissions from the flame (e.g., chemiluminescence or thermal emissions), velocity oscillations, or any other observable related to combustion dynamics. Operational data, such as engine power, inlet guide vane angle, ambient temperature, and other operational data, may also be collected. Relevant data may be obtained, without limitation, from the combustion dynamics monitoring system, other device data collection systems, or directly from sensors associated with the device.

[0055] According to certain embodiments, the method may include receiving real-time combustor fuel split data and fuel gas temperature data. In some embodiments, the fuel gas temperature may be up to approximately 150° C. In some embodiments, the fuel split may range from 0% to 100% fuel and 0% to 100% air, with the combination of fuel and air being 100%. In some embodiments, the method may include comparing the real-time combustor fuel split data and fuel gas temperature data to data in a reference database, the reference database including at least one data set selected from the group consisting of normalized load data, wheelspace temperature data, compressor discharge temperature data, dynamic amplitude data, and dynamic frequency data.

[0056] Wheelspace temperatures and compressor discharge temperatures can be collected using thermocouples. Dynamic amplitude and frequency data can be collected using acoustic or acoustic pressure sensors and transformed from the time domain to the frequency domain as described above.

[0057] The reference database can be configured to be updated with real-time combustion dynamics and fuel split data. According to certain embodiments, data can be streamed to the reference database or manually uploaded through batch upload. In some embodiments, a user-defined variable time sliding window determines when relevant values ​​in the reference database are updated.

[0058] In some embodiments, the non-transitory computer-readable medium can include instructions for updating the reference database with acoustic data as a function of gas turbine combustion dynamics. According to certain embodiments, the reference database can be updated in real time. The reference database may be included on the same non-transitory computer-readable medium or on a separate non-transitory computer-readable medium. According to certain embodiments, the reference database is configured to be updated with new data when directed by a user.

[0059] According to certain embodiments, the non-transitory computer-readable medium may be included in a computer system comprising at least one processor coupled to a memory. In some embodiments, a gas turbine system comprising at least one gas turbine may provide data to a computer system comprising the non-transitory computer-readable medium. The computer system may be configured to receive data input from at least one acoustic sensor in the gas turbine, wherein the data input may be performed in real time. The computer system may be configured to receive data input from at least one pressure sensor, wherein the data input may be performed in real time. The computer system may be configured to receive data input from at least one thermocouple, wherein the data input may be performed in real time.

[0060] As used in this application, the terms "module" and "system" can refer to a computer-related entity, which can be either hardware, a combination of hardware and software, software, or software in execution. A module may, in certain embodiments, include manually executed steps or processes. For example, a module can be, but is not limited to, a process running on a processor, a processor, an object, an executable file, a thread of execution, a program, or a computer. Illustratively, both an application running on a server and the server can be a module. One or more modules can reside within a process and / or thread of execution, and modules may be localized on one computer or processor and / or distributed among two or more computers or processors. The systems and methods can be implemented for real-time control, for example, using a digital signal processor (DSP) or similar embedded device.

[0061] A method 100 for detecting blowout precursors is depicted in the block diagram flowchart of FIG. 7. The method 100 receives as a data input 108 sensor data and device-related data 104 associated with a plurality of gas turbine engine devices and sensors 102. In an embodiment, the sensor data 104 may include combustion dynamics data and operational data, such as the amplitude and frequency of combustion pressure oscillations within the gas turbine engine. It should be noted that combustion dynamics depend on a significant number of factors. For example, engine manufacturer or model, engine power level, ambient temperature, fuel composition, inlet guide vane angle, machine operating hours, and numerous other parameters may affect the combustion dynamics data.

[0062] Sensors useful in the systems and methods of the present invention are known in the art as being capable of providing signals representative of combustion dynamics data associated with the operation of a gas turbine device. Conventional sensors that report signals to a Combustion Dynamics Monitoring System (CDMS) program may be used. The method 100 may receive the sensor data 104 from, for example, a CDMS 106. The method may receive the sensor data 104 directly from a sensor associated with the device 102.

[0063] Sensor data 104, such as combustion dynamics data in the time domain, is subjected to simplified Mexican hat wavelet analysis 110 to arrive at a series of wavelet coefficients for lower and higher frequencies, typically in the kilohertz range. Both the lower and higher frequency wavelet coefficients are compared 113 against previous wavelet coefficient thresholds. If both exceed the wavelet coefficient thresholds simultaneously for the same combustor, a blowout precursor is identified and an LBO signal is output for that combustor 114. The LBO signal can be output to initiate a warning signal or alarm 116 or can send an actuation signal to a controller to address operating parameters to avoid or respond to an LBO event.

[0064] 8 illustrates a system for detecting a blowout precursor in a gas turbine engine combustor. The system includes a sensor 120 associated with and / or in communication with at least one turbine engine combustor to provide sensor data 122, an input data module 124, a processing module 126, a database 128, an output data module 130, and an interface module 132.

[0065] In an embodiment, the sensors 120 are configured to generate sensor data 122 representative of operating conditions of the gas turbine engine 120. The sensor data may include, for example, combustion dynamics data and operational data associated with the gas turbine engine 120. The input data module 124 may receive the sensor data 122 directly from the sensors. In other embodiments, the sensor data may be received by the input data module 124 from another program or system, such as a combustion dynamics monitoring system or other data acquisition system, or may itself include a combustion dynamics monitoring system.

[0066] In certain embodiments, the combustion dynamics monitoring system collects dynamic data in the time domain. The time domain data may include pressure oscillations, electromagnetic emissions from the flame (e.g., chemiluminescence or thermal emissions), velocity oscillations, or any other observable related to combustion dynamics. As noted above, operational data may also be collected.

[0067] In an embodiment, the input data module 124 makes combustion dynamics data and operational data 122 associated with the gas turbine engine 120 available to the processing module 126. The processing module 126 may analyze and process the data, performing a simplified Mexican Hat wavelet analysis to detect the presence of a blowout precursor. In an exemplary but non-limiting embodiment, the processing module may perform the following operations in accordance with the systems and methods of this invention: Parameters such as sampling frequency, number of burners, wavelet parameters, wavelet coefficient threshold, and wavelet duration are set. Typically, two wavelets will be pre-computed, called low frequency and high frequency, but the raw data for both will be in the kilohertz range, so two sets of wavelet parameters will be defined. The number of samples that will be within the wavelet duration is calculated. The wavelet functions are pre-generated the first time the code is run. It runs an explicit analytical equation with the wavelet parameters as input and the vector (wavelet as output). This is run twice: once with the "low" frequency wavelet and once with the "high" frequency wavelet. The following sequence of steps is repeated each time a new data sample is acquired: Create a data buffer consisting of high-speed CDMS data from each combustor as new data samples are available. Data enters the buffer on a first-in, last-out basis. The buffer vector must be the same length as the wavelet vector. For each combustor, multiply the corresponding data entry by each wavelet vector entry and add them all together (wavelet entry 1 x data entry 1, + wavelet entry 2 x data entry 2, etc.). Do this twice for each combustor, once with the low frequency wavelet and once with the high frequency wavelet. These are the ''wavelet coefficients''. If both the low frequency and high frequency wavelet coefficients are simultaneously higher than the wavelet coefficient threshold for the same combustor, an LBO signal (warning signal or actuation signal) is generated for that combustor.

[0068] In an embodiment, database 128 is configured to store and make available to processing module 126 data related to operating conditions of the turbine engine, including signals generated by the sensors. In yet another embodiment, database 128 is configured to store and make available to processing module 126 historical sensor data associated with the turbine engine and sensors 120, including, but not limited to, frequency oscillations and / or wavelet coefficients associated with hiccups and lean blowout events.

[0069] Output module 130 may report results identified by processing module 126 to interface module 132 for presentation or notification to a user. Output module 130 may report the results to interface module 132 in raw form or may be configured to perform additional processing of the results identified by processing module 126.

[0070] The interface module 132 can communicate the results reported by the output data module 130. The results can be communicated to a user through an electronic device, display, or printout, or can be used for control purposes. In embodiments, the results may be made available in real-time for real-time monitoring or control, or may be stored and made available for later use. In aspects of the present disclosure, the output data module 130 can communicate the results in the form of an alert, an audible indicator, an email, a text message, an instant message, a social media message, a pager notification, or may utilize other communication methods.

[0071] In other embodiments, the output data module 130 may forward the results to another program or system, such as a combustion dynamics monitoring system, for further processing or control purposes. For example, the interface module may generate at least one control signal upon detection of a blowout precursor to adjust the fuel-air ratio of fuel and air supplied to the combustor associated with the blowout precursor. In certain embodiments, the output data module 130 and the interface module 132 may be integrated.

[0072] Previous practice required tuning all combustors to avoid LBO, unlike the present system and method which can tune only the combustors that have an LBO problem.

[0073] It will be understood that the embodiments described herein are merely illustrative and that variations and modifications may be made by those skilled in the art without departing from the spirit and scope of the invention. All such variations and modifications are intended to be included within the scope of the invention as described and claimed herein. Furthermore, all embodiments disclosed are not necessarily in the alternative, as various embodiments of the invention may be combined to provide desirable results.

Claims

1. 1. A method for detecting a blowout precursor in at least one gas turbine combustor, comprising: receiving combustion dynamics acoustic data in real time measured by an acoustic measurement device associated with the combustor; performing wavelet analysis on the acoustic data using simplified Mexican Hat wavelet transform analysis; determining the presence of a blowout precursor based at least in part on said wavelet analysis; and The method, wherein determining the presence of a blowout precursor comprises determining an increase in amplitude of time-dependent spectral content within at least one predetermined band of high frequency dynamics in the kilohertz range.

2. The method of claim 1 , wherein the band of high frequency dynamics is predetermined based on identifying a band of high frequency dynamics that was present for the combustor about 1 second prior to a previous blowout event.

3. performing the wavelet analysis includes determining dominant frequencies of oscillations of the acoustic data signal as a function of time to calculate wavelet coefficients; The method of claim 1.

4. The method of claim 3, further comprising the steps of using a windowed root mean square calculation to process the wavelet coefficients and determine the amplitude of the wavelet coefficients, and determining the presence of a blowout precursor based on the increased amplitude of wavelet coefficient oscillations.

5. 1. A system for acoustic detection of blowout precursors in at least one gas turbine combustor, comprising: an acoustic measurement device in communication with the combustor and generating a signal indicative of acoustic combustion dynamics within the combustor in real time; a blowout precursor monitor unit that receives signals indicative of the acoustic combustion dynamics and performs wavelet analysis on the acoustic data using simplified Mexican Hat wavelet transform analysis to detect the presence of the blowout precursor; Equipped with The system, wherein the blowout precursor monitor unit detects the presence of a blowout precursor by determining an increase in amplitude of time-dependent spectral content within at least one predetermined band of high frequency dynamics in the kilohertz range.

6. The system of claim 5 , further comprising a combustion controller configured to control at least one parameter of operation of the combustor based at least in part on detection of a blowout precursor by the blowout precursor monitor unit.

7. The system described in claim 6, wherein the combustion controller is configured to generate at least one control signal upon detection of a blowout precursor to adjust the fuel-air ratio of fuel and air supplied to the combustor associated with the blowout precursor.

8. The system of claim 7 , wherein the at least one band of high frequency dynamics is predetermined based on identifying a band of high frequency dynamics that was present for the combustor about 1 second prior to a previous blowout event.

9. 6. The system of claim 5, wherein the blowout precursor monitor unit performs the wavelet analysis by determining a dominant frequency of oscillations of the signal indicative of the acoustic combustion dynamics as a function of time to calculate wavelet coefficients.

10. 10. The system of claim 9, wherein the blowout precursor monitor unit uses a windowed root-mean-square calculation to process the wavelet coefficients to determine amplitudes of the wavelet coefficients, and detects the presence of a blowout precursor based on increased amplitude of wavelet coefficient oscillations.

11. The system of claim 5 , wherein the blowout precursor monitor unit transmits an alarm signal to an electronic device upon detection of a blowout precursor and / or transmits a signal to a combustion controller indicative of the detection of the blowout precursor.

12. 1. A non-transitory computer readable storage medium encoded with executable program code for performing a method for detecting a blowout precursor in at least one gas turbine combustor, the method comprising: The method comprises: receiving combustion dynamics acoustic data in real time measured by an acoustic measurement device associated with the combustor; performing wavelet analysis on the acoustic data using simplified Mexican Hat wavelet transform analysis; determining the presence of a blowout precursor based at least in part on said wavelet analysis; and determining the presence of a blowout precursor comprises determining an increase in amplitude of time-dependent spectral content within at least one predetermined band of high frequency dynamics in the kilohertz range; A non-transitory computer-readable storage medium.

13. The non-transitory computer-readable medium of claim 12 , wherein the band of high frequency dynamics is predetermined based on identifying a band of high frequency dynamics that was present for the combustor about 1 second before a previous blowout event.

14. 13. The non-transitory computer-readable medium of claim 12, wherein the encoded executable program code for performing the wavelet analysis includes determining dominant frequencies of vibration of an acoustic data signal as a function of time to calculate wavelet coefficients.

15. A non-transitory computer-readable medium as described in claim 14, wherein the encoded executable program code for performing the method further includes the steps of using a windowed root-mean-square calculation to process the wavelet coefficients and determine the amplitude of the wavelet coefficients, and determining the presence of a blowout precursor based on increased amplitude of wavelet coefficient oscillations.

Images