Improved performance of gas turbine diagnostics using new noise‐removal techniques

Abstract

Abstract Fault detection and identification (FDI) systems are responsible for detecting and identifying errors as fast as possible with high reliability. These systems should be robust against noise and avoid false warnings. Herein, the perspective of using wavelet filters for noise reduction in FDI systems has been investigated. To achieve that, a wavelet filter and a wavelet‐hybrid filter are presented and compared in noise reduction with conventional filters, such as linear filters (finite impulse response (FIR) and infinite impulse response), median filter, and FIR‐median hybrid filter (SWFMH). The comparison is conducted in two steps: (a) noise reduction of a noisy sample signal from a gas turbine and (b) increasing the fault detection accuracy of a gas turbine FDI system in the presence of noisy data. In step one, a conventional noisy sample signal of a gas turbine is presented, and the performances of different filters in noise reduction of the signal have been studied. In step two, considering that excessive filtering can result in the loss of useful information for an FDI system's diagnostics, the performances of an FDI system coupled with different filters have been evaluated. For this purpose, an FDI system utilising an adaptive neuro‐fuzzy inference system and gas path analysis has been designed. It is demonstrated that, in some cases, the wavelet filters have a lower denoising capability for a noisy sample signal, but when used together with the FDI system, they have better performance. Therefore, wavelet filters are better suited for use in FDI systems.

Keywords

• aerospace engines
• fault diagnosis
• filtering theory
• FIR filters
• fuzzy neural nets
• fuzzy reasoning