0
Research Papers: Gas Turbines: Controls, Diagnostics, and Instrumentation

# A Comparison of Two Trending Strategies for Gas Turbine Performance Prediction

[+] Author and Article Information
Rajat Sekhon, Hany Bassily

Energy Systems Laboratory, Department of Mechanical Engineering, Clemson University, Clemson, SC 29634

John Wagner

Energy Systems Laboratory, Department of Mechanical Engineering, Clemson University, Clemson, SC 29634jwagner@clemson.edu

For the GE 7EA gas turbine, over 180 signals were collected at $Δt=60s$ during system operation. Faster data logging would exceed electronic storage capacity.

J. Eng. Gas Turbines Power 130(4), 041601 (Apr 29, 2008) (10 pages) doi:10.1115/1.2898838 History: Received March 08, 2007; Revised September 24, 2007; Published April 29, 2008

## Abstract

Complex multidomain dynamic systems demand reliable health monitoring to minimize breakdowns and downtime, thereby enabling cost savings and increased operator safety. Diagnostic and prognostic strategies monitor a system’s transient and steady-state operations, detecting deviations from normal operating scenarios and warning operators of potential system anomalies. System diagnostics detect, identify, and isolate a system fault while prognostics offer strategies to predict system behavior at a future operating time to define the useful period before failure criterion is reached. This paper presents the development and the experimental application of two methods to predict the system behavior based on trends in performance. Statistical regression concepts have been used to analyze dynamic plant signals, and based on these results, future plant operation was estimated. Wavelet transforms were used to condition the signal, and the denoised signals were subsequently forecast. The case study presented here applies the two methodologies to the operational data from a simple cycle $85MW$ General Electric gas turbine. Those operating data were used to train and validate the algorithms. A comparison of the two methodologies reveals that the wavelet forecast is better than the statistical strategy with lower forecasting error. The developed approaches may be used in parallel with a diagnostic algorithm to monitor gas turbine system behavior.

<>

## Figures

Figure 5

Wavelet prediction strategy: (a) approximation coefficients for turbine inlet temperature for learning window (0<t<22,200min), (b) wavelet signal forecast (22,200<t<32,200min) with reduced display density, and (c) signal and wavelet forecast mapped back into the operating domain for 0<t<32,200min

Figure 6

Comparison between the calculation complexity of the statistical and wavelet methods

Figure 7

Application of the wavelet prediction methodology

Figure 1

Statistical and wavelet prediction strategies with forecasting

Figure 2

Application of the statistical trending strategy to estimate a useful remaining time based on a selected threshold of two standard deviations satisfying a 95% confidence limit

Figure 3

GE 7EA stationary gas turbine inlet temperature profile: (a) raw data signal for 0<t<59,000, (b) restricted operating range of 600–800°C for 0<t<32,200min and (c) filtered data showing the learning 0<t<22,200min and validation windows 22,200<t<32,200min

Figure 4

Statistical prediction strategy: (a) regression curve for turbine inlet temperature during learning window (0<t<22,200min) minutes with reduced display intensity, (b) statistical signal forecast (22,200<t<32,200min) minutes with reduced display intensity, and (c) signal and statistical forecast mapped back into the operating domain for 0<t<32,200min

## Errata

Some tools below are only available to our subscribers or users with an online account.

### Related Content

Customize your page view by dragging and repositioning the boxes below.

Related Journal Articles
Related Proceedings Articles
Related eBook Content
Topic Collections