Nothing runs as smooth as a new machine. After time, as components wear, the hum of productivity becomes the whine of a tired collection of loose fitting parts beating out of unison. This condition is often monitored using vibration analysis of one form or other. Unfortunately, conventional techniques may not spot machine problems until it’s too late.
There is a way, however, to use filtered high-frequency analysis for catching problems that previously went unnoticed. Since its introduction, the method (referred to here as spike energy) has been used in a variety of rotating equipment to flag machine faults before they actually occur. The method relies on a signal filtering and detection process that captures the most minute influences of a defect, while greatly amplifying and exposing its fundamental frequency and multiples.
As bearings, gear teeth, and other machine components wear, they develop microscopic cracks and spalls, which in turn cause bumpier operation. The mechanical knocking produces short pulses, or spikes, of vibratory energy that excite component natural frequencies. (Side note: pump cavitation, turbulence in liquids, and control-valve noise have a similar effect.) The impacts of microscopic cracks and spalls also excite the natural frequencies of spike energy accelerometers gathering vibration signatures from around the system; acting as carrier frequencies, they lead the machine defect frequencies that flutter with them. Impact energies (labeled in acceleration units gSE of spike energy) are registered by the accelerometers as functions of spike amplitude and repetition rate, and are sent on for further analysis.
It starts with the setup
Conventional vibration parameters (displacement, velocity, acceleration) typically fall within the linear frequency response range of most transducers, and are therefore fairly easy to measure. But spike energy detects frequencies beyond the linear range of most industrial transducers. Because mounting methods affect higher frequencies, spike energy results vary with different setups.
Impact-induced resonant frequencies of industrial accelerometers typically range from 10 to 50 kHz, varying greatly with construction and mounting. If two accelerometers had the same frequency response characteristics, it would be a coincidence; thus, spike energy readings made with different accelerometers shouldn’t be compared. Because of spike energy’s great sensitivity to setup, the most meaningful way to use spike energy for machinery condition monitoring applications is to observe trends in the returned signal. For consistency, the same accelerometer, mounting method, and measurement location should be used throughout any data collection.
Mounting methods change high frequency results, and some cut signals out entirely. If an accelerometer is mounted insecurely, a mounted resonant frequency is introduced. It is always lower than the accelerometer’s inherent resonance, editing frequencies above it. When it is much lower, the usable frequency range becomes much smaller.
The best method for collecting spike energy data is with stud mounting because there is only one interface: accelerometer-tomachine. This allows greater transmission of high-frequency signals, and returns the most consistent results. Some tips:
• Any threaded holes should be perpendicular to mounting surfaces to prevent “working out.”
• Stud length should be shorter than hole length to allow direct contact between accelerometers and mounting surfaces.
• Cable connectors should be sufficiently tightened to the ac-dicelerometer to prevent rattling and erroneous readings.
• If the stud is mounted to a moving component, the extension cable should be as well; this minimizes cable wiggling during measurement.
Continue on page 2
Sometimes hand-held probes are used to measure spike energy; however, sometimes they lose high-frequency signals because of their low mounted resonances. Another alternative is magnet mounting. Used for quick periodic checks, it has two interfaces — accelerometer- to-magnet and magnet-to-machine. Flat, clean, rustfree, unpainted contact surfaces minimize loss of high-frequency vibration signals during transmission; magnet pole pieces free of dents and broken edges are best. A light coating of silicon grease or lube oil at the interface improves the transmissibility of high-frequency vibration signals, which is essential to obtaining accurate and consistent spike energy data.
After vibration signals are picked up by accelerometers, frequency band pass filters clean them up. Results above the upper limit of the spike energy detection range (65 kHz) are snipped off by a low-pass corner frequency filter. At the same time, low-band noise caused by imbalance, misalignment, and looseness is chopped off at one of six high-pass corner frequency levels — 0.1, 0.2, 0.5, 1, 2, or 5 kHz; frequencies above that value are allowed to pass through. This makes the amplitudes of bearing and gear defect frequencies, which are usually much smaller than those of low-frequency components, more prominent.
Decay time constant
The filtered signal passes through a peak-to-peak detector that applies a carefully selected decay time constant, which is rectly related to spectrum maximum frequency Fmax. It is automatically selected by either the instrument or host software, and determines the shape of the peakto- peak sawtooth signal by affecting both the overall spike energy magnitude, and harmonic terms of the spectrum. To obtain consistent overall energy readings, only one fixed decay time constant is used for the measurement in both instrument and host software.
In spike energy spectrum measurement, smaller decay time constants are selected for higher frequency measurements, since defect impulses occur more rapidly. Plus, the period of impact is more evident by using a shorter constant.
Monitoring with only spike energy
Depending on machine dynamic characteristics, certain machines can be sufficiently monitored by only observing overall spike energy magnitude trends. Monitoring sealless pumps is one example. There are two kinds of sealless pump problems: process-related problems (from dry running, cavitation, flow change, and internal recirculation) and mechanical problems (from rotor rub, or excessive wear of thrust and journal bearings). Conventional vibration measurements have never been very successful in detecting these problems because the internal rotor mass of a sealless pump is relatively small compared to the rest of the pump. Also, internal fluids often create confusing vibration signals.
Spike energy can detect both mechanical and process problems. Spike energy magnitude trends and sealless pump problems have been linked through experiments, making for great reductions in sealless pump damage and downtime.
Other vibration parameters
In most applications, spike energy alone doesn’t sufficiently monitor machine conditions. Concurrently observing it with other vibration parameters (such as acceleration, velocity, or temperature) is helpful to establish useful correlations.
When spike energy increases, it usually means that bearing, gear, or other component problems are developing. It also means that acceleration and velocity trends should be more closely observed for changes; if acceleration readings exceed their allowable vibration limits but velocity readings are still acceptable, vibration spectrum analysis should be performed to confirm the problem. Repairs should be scheduled for a convenient future time.
When velocity, acceleration, and spike energy readings all exceed allowable levels, the observed machine is approaching the end of its useful life. Sometimes, spike energy readings may decrease and, just prior to failure, increase again to excessive values; if this happens and is seen in time, the machine should be shut down to prevent more avoidable damage.
Continue on page 3
Harmonics are integer multiples of rotation-related frequencies like shaft running frequency, vane pass frequency (number of vanes times shaft speed), and gear mesh frequency (number of gear teeth times shaft speed). Harmonics are produced by events that repeat during one revolution, or by distortions of sinusoidal signals. In the spike energy spectrum all harmonics caused by low-frequency excitations are filtered out by highpass filters, so if harmonics of some signal do appear, they are high-frequency. One example: if the spectrum of a spinning shaft showed harmonics, it would indicate a problem with a high-frequency interaction, such as gear mesh. One possible diagnosis: a gear might be riding on a bent shaft.
Sidebands are (theoretically symmetrical) alterations to carrier frequencies. There are two kinds of sidebands — amplitude modulations are associated with loading changes, and frequency modulations are associated with changes in speed.
In many cases, amplitude and frequency sidebands coexist. For example, frequency modulation may occur in a gear riding on a bent shaft, because the tooth space measured on the pitch circle will vary where the shaft bends. Since modulating frequencies are caused by certain bearing, gear or other machine component problems, spike energy spectrum is great for diagnosing these faults.
In rolling-element bearing applications, sidebands are usually multiples of one bearing defect frequency; in other words, amplitude modulation signals. Bearing defect frequencies include ball pass (inner and outer race), ball spin, and fundamental train frequencies. Vibration amplitudes vary when the defects on inner race or rolling elements enter and exit the bearing load zone.
In gear applications, sidebands represent either the shaft rotational speed or one of its multiples (n x rpm). Amplitude modulations are present when gear meshes have eccentric gears, or when gears ride on bent or misaligned shafts. In this case, a cyclic loading pattern occurs because of the periodic forcing of teeth into mesh. A minimum and maximum meshing force occur once per shaft revolution. As the eccentricity increases, the sideband amplitudes increase. If there are faults in individual gear teeth or small groups of teeth, the gear vibrates when the defective teeth are in mesh. Local gear faults include tooth space error, cracked or broken teeth, tooth surface damage, and hunting tooth problems. With local faults, changes in a gear’s angular velocity as a function of rotation are possible. From speed variation, frequency modulations occur and generate many sideband pairs.
Thanks to Joseph M. Shea of Vibtec, Grant D. Mayers of BASF, Julien Le Bleu, Jr. of Lyondell Chemical, James Lobach of Crane Pumps, Donn Stoutenburg and Aaron Hipwell of Rockwell Automation, Integrated Condition Monitoring for their support and assistance during this study.