To identify unwanted vibrations in a machine, I am looking for sharp peaks in spectra calculated from a measured acceleration time series. The acceleration data were recorded during a 10-second test consisting of multiple 2-second periods in which conditions were constant, interrupted by 0.2-second periods of transition. The spectra should only be estimated during the constant-condition periods, but I am wondering how to combine the data from subsequent constant-condition periods.
- I could extract the constant-condition periods, concatenate them, then estimate the spectra of the concatenated time series, but each 2-second period would start at a different point in any cycles that are there. If a particular 0.1-second cycle occurs at 0.1, 0.2, 0.3,... seconds in the first 2-second period, but at 0.175, 0.275, 0.375,... seconds into the next 2-second period, would the spectra still contain a sharp peak at 0.1 seconds? Would the concatenation introduce an artificial peak at 2 seconds due to the transition between 1 segment and the next?
- Or, I could estimate the spectra from each 2-second period, then average the resulting spectra together.
Which approach is best for identifying sharp resonant peaks in a spectrum?
