Many studies have attempted to detect structural damage by examining differences in the frequency response functions of a structure before and after damage. In an experimental setting, this variation can not be attributed solely to the addition of damage. Other sources of variation include testing and structure variation. Examples of testing variation include the error introduced by modal parameter extraction, measurement noise, and the mass loading of the accelerometer. Structure variability is due to slight differences in the supposedly identical structures. Dimensional tolerancing is one example.

This study began with six "identical" undamaged aluminum cylinders, of which three were later damaged to varying extents. The frequency response functions of the undamaged and damaged cylinders were measured. Also, the frequency response function of the same undamaged cylinder was measured multiple times to investigate testing variation. The contributions of testing, cylinder, and damage variation to the differences between cylinder responses was elucidated by specifically examining their frequency response functions in two ways: comparing the natural frequencies and directly investigating the entire frequency response function. The curvature of the frequency response functions was then used to determined the presence, location, and severity of the imparted damage.