Multichannel spectral analysis for tube leak detection in a fossil-fueled power plant

This research describes the development of multichannel signal spectral estimation methods, which include the Classical method and the Parametric method. These spectral analysis techniques are used to detect tube leaks in a fossil-fueled power plant. The research presents the general analysis of two classical spectral estimation methods, which are the Correlogram method and the Periodogram method, and three parametric spectral estimation methods, which are the Autoregressive (AR), Moving Average (MA) and Autoregressive-Moving Average (ARMA) process models, for multichannel signals. The signals are obtained from sensors which are installed in the power plant boiler. The autospectrum for each data channel and the cross spectrum for all channel pairs are estimated. Based upon these estimates, the coherence spectrum is computed to show how much each pair of channel signals is correlated. If two channel signals have a signature that indicates a leak might have occurred, and these two channel signals are somewhat correlated around a peak frequency that is above 6 kHz, it can be assumed that both of these leaks are in fact the same leak detected on two different channels.