ABSTRACT
The last few chapters focused on noise removal from gas turbine signals. The signal processing algorithms ranged from simple “gentle smoothers” like the CWIM filter suitable for trend plots to the sophisticated adaptive myriad filter for real-time transient data analysis. In this chapter, we address the problem of locating the discrete time or epoch when the sudden change in the gas turbine signal occurs. Trend shift detection is posed as a two-part problem: filtering of the gas turbine measurement deltas followed by the use of edge detection algorithms. Measurement deltas are deviations in engine gas path measurements from a good baseline engine and are a key health signal used for gas turbine performance diagnostics. Just like in the previous chapters, the measurements used in this chapter are exhaust gas temperature, low rotor speed, high rotor speed, and fuel flow, which are called cockpit measurements and are typically found on most commercial gas turbines. In this chapter, a cascaded recursive median (RM) filter, of increasing order, is used for the purpose of noise reduction and outlier removal, and a hybrid edge detector that uses both gradient and Laplacian of the cascaded (RM) filtered signal is used for the detection of step change in the measurements. Simulated results with test signals indicate that cascaded RM filters can give a noise reduction of more than 38% while preserving the essential features of the signal. The cascaded RM filter also shows excellent robustness in dealing with outliers, which are quite often found in gas turbine data, and can cause spurious trend detections. Suitable thresholding of the gradient edge detector coupled with the use of the Laplacian edge detector for cross-checking can reduce the system false alarms and missed detection rate. Further reduction in the trend shift detection false alarm and missed detection rate can be achieved by selecting gas path measurements with higher signal-to-noise ratios.
