Non-par[a]metric analysis for tube leak detection

Because forced outages of utilities in the United States are primarily the result of boiler tube failures in fossil-fueled power plant--failures that have placed tremendous financial burdens on the industry--it is no surprise that the interest in spectral analysis methods for tube leak detection has reached an epic high.

Therefore, in this thesis, non-parametric spectral estimation methods were developed to investigate their effectiveness for tube leak detection. Such methods included the following: (1) the minimum variance method and (2) eigenanalysis methods, utilizing both the EV and MUSIC techniques. Findings suggest that although all of the PSD estimation methods performed well enough to detect leaks, eigenanalysis methods proved to be far superior over both classical and parametric methods, and the EV method showed the most promise regarding early leak detection.