ABSTRACT
Diagnostic applications typically involve the detection and isolation of a system fault based on comparison between a good baseline system and a damaged system. Many diagnostic systems are designed based on mathematical models for the good and bad systems using methods that fall under the broad class of model-based diagnostics. A health signal can be interpreted as a measurement delta between the damaged measurement z(d) and undamaged measurement z(u) and written as Δ = z(d) − z(u). Under ideal conditions, the system has no fault, i.e., Δ = 0. When a fault occurs, Δ assumes a nonzero value whose magnitude depends on the size and location of the fault. In this idealized system, the nonzero value of the measurement deviation, along with other measurement deviations, can be used to detect and isolate the fault. For commercial aircraft engines, only few data points are received for each flight. Therefore, it is important to keep the forward data point requirement to minimum. In Chapter 2, we looked at the “gentle” center weighted idempotent median (CWIM) filter for gas turbine diagnostics. In this chapter, we explore other filters with a low time delay for gas turbine applications.
