Three- and Seven-Point Optimally Weighted Recursive Median Filters for Gas Turbine Diagnostics

Abstract

engine components. Zedda and Singh [3] estimated the performance of the gas turbine through the optimization of an objective function by means of genetic algorithm. From the literature, it can be seen that algorithms based on least-squares, Kaiman filters, and soft computing have been widely used for gas turbine diagnostics. A comparative study between Kaiman filter and neural network method was done in Ref. [4]. Recently, the Bayesian approach to gas turbine diagnostics has been used [5].In this note, a study of the effect of integer weight on threepoint weighted recursive median (WRM) filter and seven-point WRM filter is done. Optimal weights for different health signals are obtained. The three-point filter is useful for engines where data are obtained slowly and the seven-point filter where there are more data available. This note expands the earlier work of Uday and Ganguli [6], which was limited to a WRM filter of length tive. The effect of window length on the optimal weights of WRM filters developed for gas turbine diagnostics is brought out.Measured health signais incorporate significant details about any malfunction in a gas turbine. The attenuation of noise and removal of outliers from these health signals while preserving important features is an important problem in gas turbine diagnostics. The measured health signals are a time series of sensor measurements such as the low rotor speed, high rotor speed, fuel flow, and exhaust gas temperature in a gas turbine. In this article, a comparative study is done by varying the window length of acausal and unsymmetrical weighted recursive median fllters and numerical results for error minimization are obtained. It is found that optimal filters exist, which can be used for engines where data are available slowly (three-point filter) and rapidly (sevenpoint filter). These smoothing filters are proposed as preprocessors of measurement delta signals before subjecting them to fault detection and isolation algorithms.

Symbolic Dynamic Analysis of Transient Time Series for Fault Detection in Gas Turbine Engines

“…For example, Lipowsky et al [1] made use of Bayesian forecasting for change detection and performance analysis of gas turbine engines. Guruprakash and Ganguli [2] reported optimally weighted recursive median filtering for gas turbine diagnostics from noisy signals. However, transient operational data (e.g., from takeoff, climb, or landing) can be gainfully utilized for early detection of incipient faults.…”