The development of random vibration tests from non-stationary data

The theory of random vibrations is based on the assumption that the data being used for test development is stationary. Laboratory tests can be developed readily for components in a stationary environment by taking time averages of many samples of data. However, when this method is applied to non-stationary data, the results are very inaccurate. Because of this limitation, random vibrations - although common in the aerospace industry - have found only limited acceptance in the automotive industry. There are two methods which have been used for non-stationary data. The time average method gives results which are too low since the peaks are averaged away. The peak hold method gives the worst case conditions, but loses all reference to duration. A procedure has been developed which eliminates both of these problems. The time history data is reviewed for patterns of vibration. Power Spectral Density (PSD) lots re made of each pattern and the total time duration of each pattern is determined. Then the PSDs are grouped, with the time durations of the lower level ones being decreased by increasing the input levels to that of the maximum level of the group - based on the material properties. The result is a vibration specification which is based on real-world data, consists of several PSDs per axis, can be run in a matter of hours, and does not exceed the real-world levels. This procedure has been successfully used to test an automotive bracket. Two configurations of bracket were tested. The first configuration had a failure mode in the laboratory identical to that of a similar bracket in a vehicle undergoing road testing. The second configuration successfully passed both the laboratory and road tests.