Downloaded from https://www.everyspec.com ADS-79D-HDBK FIGURE D-6. An example of method of physical and functional modeling D.5.2.4 Identify candidate feature extraction/CI algorithms. With the large number of vibration sources on an aircraft, the data collected by any one sensor has a tremendous amount of noise. The first set of algorithms to be developed are those that can enhance the signal to noise ratio, giving the algorithms the best chance of extracting the characteristics, or features, which describe the fault through sensor readings. There are a number of possible techniques for de-noising. Three popular methods are listed below (not inclusive or exclusive): a. Soft Thresholding b.
Wavelet shrinking c. Adaptive Thresholding The methods should be tested with the sample data to determine which technique works best. The signal conditioning for feature extraction continues with some technique for signal compression that can save as much of the true ―information‖ in the signal as possible. For vibration analysis, the most common compression technique is Synchronous Time Averaging (STA). Figure D-7 identifies an example of typical signal processing steps from data collection to CI comparison. STA is possible whenever there is a means to indicate the start of an individual revolution, by means of a pulse signal or other means.
The STA takes the readings for a number of individual revolutions and averages them, resulting in an averaged data segment with a length corresponding to a single rotation. STA results enhance the vibration frequencies that are multiples of the shaft frequency. The feature or CI to be extracted from the signal is the basis for accurate diagnosis. The CI should be capable of detecting the fault prior to its causing significant damage or injury and it should be reliable and consistent enough to merit the trust of maintenance personnel. Appendix E of ADS-79D lists a 87