REG NASA-LLIS-0830-2000 Lessons Learned Rocket Engine Failure Prediction Using an Average Signal Power Technique.pdf

1、Best Practices Entry: Best Practice Info:a71 Committee Approval Date: 2000-04-19a71 Center Point of Contact: GRCa71 Submitted by: Wil HarkinsSubject: Rocket Engine Failure Prediction Using an Average Signal Power Technique Practice: Apply a univariate failure prediction algorithm using a signal proc

2、essing technique to rocket engine test firing data to provide an early failure indication. The predictive maintenance technique involves tracking the variations in the average signal power over time.Programs that Certify Usage: This practice has been used on the Space Transportation System (STS).Cen

3、ter to Contact for Information: GRCImplementation Method: This Lesson Learned is based on Maintainability Technique number AT-5 from NASA Technical Memorandum 4628, Recommended Techniques for Effective Maintainability.Benefit:This technique will reduce unnecessary failures attributed to the traditio

4、nally used redline-based system. The average signal power algorithm can be used with engine test firing data to provide significantly earlier failure indication times than the present method of using redline limits. Limit monitoring techniques are not capable of detecting certain modes of failures w

5、ith sufficient warning to avoid major hardware and facility damage.Provided by IHSNot for ResaleNo reproduction or networking permitted without license from IHS-,-,-Implementation Method:For discrete random processes, probabilistic functions are used to describe the behavior of the rocket engine sys

6、tem. The Power Spectral Density (PSD) is computed to describe how the variation of the random process is distributed with frequency. For stationary signals, the PSD is bandlimited to 1/(2T), where T is the sampling interval in seconds.Average Signal Power CalculationsThe PSD is defined as the discre

7、te-time Fourier transform of an autocorrelation function. (The derivation of the autocorrelation function is shown in Reference 1.) When the autocorrelation function is evaluated at zero lag, then an expression for the average signal power (ASP) of a random stationary process results:refer to D desc

8、riptionD where:Pxxf ) = discrete-time Fourier Transform rxx0 = inverse discrete-time Fourier transform The average signal power for several SSME parameters is determined by calculating the autocorrelation at zero lag for the parameters provided in Table 1. The assumption is made that the signal is s

9、tationary over the computation interval. The average signal power calculations are performed over 2-second, 50-percent overlapping window for nominal test firings at both 104- and a 109-percent-rated power levels. A smaller time increment must be used to improve the failure the algorithm.Provided by

10、 IHSNot for ResaleNo reproduction or networking permitted without license from IHS-,-,-refer to D descriptionD Provided by IHSNot for ResaleNo reproduction or networking permitted without license from IHS-,-,-Table 1: Signal Threshold and Safety Factor for SSMEs The average plus three standard devia

11、tions of the average signal power are computed for all the nominal firings at both engine power levels. These values are combined to calculate the thresholds (see Reference 1).A safety factor ranging from 1.5 to 3.5 is neededto ensure no false failure indications are computed for the nominal firings

12、. The range of safety factors reflected signal behavior variations that occurred over seven nominal A2 firings. When used in the failure detection mode, failure of the average signal power of a parameter to fall outside its threshold results in a failure indication. Also shown in Table 1 are the thr

13、esholds calculated from the SSME nominal test firings based on the average signal power algorithm along with the associated safety factors.Algorithm ImplementationA system identification and signal processing software package on a RISC workstation provides the average signal power algorithm. Command

14、 and Data Simulator (CADS) data from a predetermined number of SSME test firings are used to establish the failure indication thresholds.Several system conditions must be considered to ensure that the algorithm does not erroneously indicate an engine fault. These conditions include sensor failure, p

15、ropellant tank venting and pressurization, and propellant transfer. Sensor failure detection techniques must be exercised before, or concurrently, with safety monitoring algorithms in order to eliminate the possibility of a sensor failure being interpreted as an engine problem. Typically, all parame

16、ters exhibiting sensor problems are removed prior to the application of the algorithm.Failure indication thresholds are established by applying the average signal power algorithm to a set number of nominal tests. For the SSME four anomalous firings and one nominal firing were tested using the thresh

17、olds shown in Table 1. An example of the application of the average signal power algorithm to a SSME anomalous test firing is shown in Figures 1 and 2.Figure 1 illustrates the interval over which the average signal power was computed for a single parameter, HPFP discharge pressure and one test firin

18、g.Provided by IHSNot for ResaleNo reproduction or networking permitted without license from IHS-,-,-refer to D descriptionD Figure 1: Application of the Average Signal Power Algorithm to the HPFP Discharge Pressure Figure 2 displays the resulting average signal power, as a function of time. As shown

19、, the threshold for the average signal power algorithm has been exceeded.Provided by IHSNot for ResaleNo reproduction or networking permitted without license from IHS-,-,-refer to D descriptionD Figure 2: Average Signal Power for that Interval with the Failure Indication Threshold Nomenclature:HPFP

20、high pressure fuel pump HPFT high pressure fuel turbine HPFTP high pressure fuel turbopump HPOP high pressure oxidizer pump HPOT high pressure oxidizer turbine LPFP low pressure fuel pump MCC main combustion chamber PID parameter identification SSME space shuttle main engine References:1. Meyer, C.M

21、., Zakrajsek, J.F., Rocket Engine Failure Detection Using System Identification Techniques, AIAA Paper 90-1993, July 1990.Impact of Non-Practice: Provided by IHSNot for ResaleNo reproduction or networking permitted without license from IHS-,-,-Detection of anomalous behavior is critical during the o

22、peration of the Space Shuttle Main Engine (SSME). Increasing the detectability of failures during the steady-state operation of the SSME will minimize the likelihood of costly engine damage and maintenance. The average power signal algorithm is superior to the time series algorithm because more para

23、meters contribute to the first simultaneous failure indication times. This increases the agreement between several parameters, thus increasing the likelihood that an engine anomaly has occurred. This method also reduces the number of false failure indications that can prematurely shut down the engin

24、e during testing or operation.Related Practices: N/AAdditional Info: Approval Info: a71 Approval Date: 2000-04-19a71 Approval Name: Eric Raynora71 Approval Organization: QSa71 Approval Phone Number: 202-358-4738Provided by IHSNot for ResaleNo reproduction or networking permitted without license from IHS-,-,-