ABSTRACT
In this chapter, we look at the myriad filter as a substitute for the moving average filter, which is widely used in the gas turbine industry. In contrast to previous chapters, which used steady-state signals, the current chapter considers transient signals. Typically, most gas turbine diagnostics are conducted with steady-state measurement data. Some gas turbine problems such as misscheduled nozzle and compressor blade movement due to control system faults appear only during transient processes [45]. The three ideal test signals used in this study include the step signal, which simulates a single fault in the gas turbine, while ramp and quadratic signals simulate long-term deterioration. Further, an adaptive weighted myriad filter algorithm that adapts to the quality of incoming data is studied. The filters are demonstrated on clean and simulated deteriorated engine data obtained from an acceleration process from idle to maximum thrust condition. These data were simulated using a transient performance prediction code. The deteriorated engine had single-component faults in the low-pressure turbine (LPT) and intermediate-pressure compressor (IPC). The signals are obtained from T2 (IPC total outlet temperature) and T6 (LPT total outlet temperature) engine sensors with their nonrepeatability values, which were used as noise levels. The weighted myriad filter shows greater noise reduction and outlier removal when compared to the sample myriad and FIR filter in the gas turbine diagnosis. Adaptive filters such as those considered in this chapter are also useful for online health monitoring, as they can adapt to changes in quality of incoming data.
