MULTIPLE FEATURE ANALYSIS OF SODIUM BOILING NOISE DATA

In the context of acoustic detection of sodium boiling in LMFBRs the results of using more than one signal feature are evaluated. In particular, two signal features derived from an acoustic signal are considered, namely the mean square of the 6th difference and the mean square of the prediction error based on estimates of the conditional expectation. Both these features are shown to exhibit superior sensitivity to that obtained with simple variance. Further, it is shown that when using more than one feature improvements in diagnostic performance will depend on the nature of a chosen decision boundary in the feature space. Experimental results using data from the KNS test rig are presented.