Vibration Analysis of Rotating Machinery by Joe Gerhardstein

Jim Fortuna, a research fellow from MeadWestvaco gave a presentation at NI Week 2005 on vibration analysis of rotating machinery. Jim's talk centered around measurements made on one of MeadWestvacos paper machines. For those of you who have not had the opportunity to visit a paper mill, imagine a machine 30 feet across and several hundred feet long that takes dilute paper fibers suspended in water (about 2% fiber and 98% water) at one end and spits out 6-foot diameter (by 30-foot across) rolls of paper at the other in a continuous process running at 60 miles-per-hour. These machines are run 24-hours a day producing several thousand tons of paper per day for months on end.

Jim had been called in to diagnose a vibration problem at a mill. The operator wanted to know whether to shut down the machine soon, or wait until the next maintenance cycle in three months. Jim set up a monitoring system on several of the rotating components to record shaft torque vs. time, rotation speed and rotation angle. His analysis looked at both periodic/harmonic and discrete/stochastic components.

Harmonic Analysis
Harmonic events are those that appear every rotation of a device. These typically build over time due to wear of machine components, and may appear as a modulation of the fundamental or as a periodic impulse in the time measurement. Because events are harmonic, Fourier analysis is a key tool here. Modulated harmonic events (e.g. a rotating object being out of balance) and periodic impulse events (e.g. a gear with a sticky tooth) have very different power spectrums, as shown in the four figures below.



Figure 1. 10-Hz sine wave modulated on a 2-Hz sine wave (time data).



Figure 2. Power spectrum of 10-Hz sine wave modulated on a 2-Hz sine wave. Note the peaks in the FFT at 2 and 10 Hz corresponding to the fundamental and modulation.



Figure 3. 10-Hz sine wave with periodic impulse (time data).



Figure 4. Power spectrum of 10-Hz sine wave with periodic impulse. Note the fundamental peak at 10 Hz, and the many harmonic peaks caused by the periodic impulse.

Discrete/Stochastic Analysis
Discrete/stochastic events are characterized as inconsistent, often changing frequency and power unexpectedly with changing system inputs. These events typically involve mechanical systems that are about to fail catastrophically. Because they are not synchronous, the events can be hard to capture and because the frequency content changes, they can be hard to analyze. Statistical parameters of the time data are often useful to characterize these types of events. These include:

- skewness
- kurtosis
- variance
- total energy in the event
- crest factor

If Fourier analysis is also performed, the amount of power in the sidebands is often more meaningful than signal amplitude or vibration signature analysis.