On the use of hypothesis tests as statistical indexes for frequency response analysis of electric machinery

- Statistical indexes add more objectivity to FRA results.
- A new index based on hypothesis tests is proposed.
- This index is capable to detect small variations between spectra, even in cases with high variability in measurements.

This work proposes a new statistical index, based on hypothesis tests, to be used on frequency response analysis (FRA) of electric machines. This index is, particularly, interesting on the detection of early failures (which appear as small deviations from the baseline spectrum), even when the measurements have high variability (such as the case of online measurements in rotating machines). The majority of the indexes proposed so far in the FRA literature only compares two spectra at a time and mix data from the measurements at different frequencies. The index proposed in this paper calculates one hypothesis test for each frequency (in order to assure randomness, independence and normality) in the spectra of several measurements (in order to take into account the variability on the measurements) and performs the normalized sum of the individual tests. Experimental results are presented on a synchronous generator (both offline and online) to validate the proposed index.