Measuring frequency response functions on operating systems
Anton C. Keller · The Journal of the Acoustical Society of America · 1979
In studying dynamic behavior of mechanical systems very often a three-step approach is used. In this classic approach, Step 1 involves analyzing actual running characteristics of a machine, structure, or process. Step 2 involves the determination of frequency response functions between various locations on the test system. Step 3 may then involve extraction of specific characteristics of the system such as structural mode shapes, damping parameters, stiffness values, or system resonant frequencies. Often it is quite possible to extract a good understanding of possible system design problems strictly from an analysis of operating data. However in most cases, the second step will usually involve a measure of selected frequency response functions through the use of external stimulus. In the past, this has almost certainly meant that the operating system would have to be shut down and basically be in a “static” state before artificial excitation could be furnished and the frequency response function measured. This has sometimes been impractical for several reasons. First, there are cases when a system must actually be operating before its real dynamic characteristics can be extracted. This would be the case, for example, if fluid film bearings represent a part of a total system dynamic. On the other hand. system operating characteristics may interfere with the very frequency response function which is to be determined. This could be the case in a rotating turbine blade where forced vibration responses would always be present in the output or in the case of a biological system measurement which might continually include the effects of heartbeat. In each of these cases, the system must continue to operate during a frequency response measurement but it is somehow necessary to eliminate the operating characteristics from measured frequency response functions. This paper examines the classical techniques which have been used to extract frequency response information and some of the reasons why it has not been possible or conveniento eliminate running data from the desired measurement. An alternative approach is then given which has now been implemented in hardware and which offers a significant signal-to-noise enhancement over traditional techniques making it entirely feasible to extract complete frequency response information from operating dynamic systems.