Spectral maps and order tracks are tools which are susceptible to improper sensor location on rotating machinery and to measurement noise. On a complex/large rotating system, the major behavior in a particular direction cannot be observed by using standard digital signal processing averaging techniques on different sensor outputs. Also, measurement noise cannot be reduced by applying averaging - due to the slew rate of the system.
A newly developed technique tested on experimental data, is presented which uses singular value decomposition (SVD) as its basis to improve the observability of rotating systems. By using data acquired from multiple accelerometers on a machine, singular values - obtained from a SVD of the cross-power matrix at each 2-D point in the frequency-RPM domain - can be plotted in a color-map format similar to a RPM spectral map. These “Singular value maps” can be used to find the best possible sensor combination on a complex rotating system like a powertrain and to observe the system as one unit in a particular direction retaining major and inherent dynamic activity and minimizing the measurement noise.
