Much of the structural vibration and aeroacoustic noise of aerospace vehicles is nonstationary random. Methodology for nonstationary data analysis exists, but the procedure requires measurements from repeated flights, which is often impractical. An alternative parametric procedure can be applied to individual records under the assumption that the data have a specific type of nonstationarity. A type of parametric model, called the product model, and a special case, called the locally stationary model, are described and applied to space vehicle launch data. A nonstationary procedure is presented and evaluated using launch vibration data measured at a Shuttle Orbiter/payload interface.
The procedure permits a longer sampling duration to be used for spectral analysis as compared with the traditional short-duration method, significantly reducing random errors. In this evaluation, only modest differences were observed between the two types of "raw" spectra. However, when random errors were taken into account (large for the traditional method and small for the nonstationary procedure), the short-duration spectra exceeded those for the nonstationary procedure by a significant amount. Thus unnecessarily high vibration spectra can be avoided using the nonstationary procedure.