Basic Signal Processing For Vibration


Often when setting up measurement points for vibration data collection many choices selected such as the maximum frequency range, the lines of resolution, the window type and the integration mode are made based on rules of thumb instead of an understanding of how each of these are related to the frequency spectrum and the time waveform. A digital signal analyzer is a powerful tool that can present some problems for the uneducated user. With an understanding of signal processing basics these problems can be addressed, understood and avoided.

Fast Fourier Transform

The method used to convert time domain information to frequency domain information is the Fast Fourier Transform (FFT).

Often a frequency spectrum is referred to as an FFT. However, the FFT is the mathematical conversion from the time domain to the frequency domain. Since the signal that comes into the analyzer is an analog signal as discussed in the previous section, it must be digitally sampled by the analyzer. Therefore, the process used by digital analyzers is actually a variation of the FFT, called the Discrete Fourier Transform (DFT).

The DFT is similar as the analog time waveform is recreated in the analyzer by digitally sampling, and then the transformation into the frequency domain is done. Part of the reason the FFT process works is the assumption that the signal measured and digitally sampled is one period of a periodic signal that extends to minus infinity and to plus infinity. Normally, this is true for most vibrating pieces of equipment.

It is the digital sampling process that makes the signal processing more complicated. The information here unlocks the mysteries of digital signal processing without getting bogged down in too much theory.

In order to understand the FFT digital sampling process, you must understand the relationship between lines of resolution (LOR), Maximum frequency (Fmax), length of time waveform (Tmax), the digital sample size, aliasing, windowing, filters, and unit conversion.


Once data has been converted to the frequency domain from the time domain, view all of the frequencies of interest in as much detail as possible. Resolution is the number of parts that the spectrum is broken into, usually called lines of resolution (LOR). The number of lines of resolution selected are divided into the maximum analysis frequency (Fmax) to arrive at the bandwidth (BW).
BW = Fmax /LOR

The lines are actually the center frequencies of what are often called “bins of energy”. Each bin actually contains an infinite number of frequencies, and all of the energy in the bin is summed and represented by a single amplitude at the center frequency identified at each line of resolution.

First, identify your frequencies of interest so that enough resolution is chosen to separate closely spaced frequencies. Also, be aware that more lines of resolution affects the length of the time waveform. Increased resolution can also decrease the actual amplitude of the vibration amplitudes due to the separation of energy into more energy bins.

Time Record Length

Calculate the time record length of the time waveform, Tmax, from the following basic relationship.
Tmax = 1/BW

At face value this is a simple and often used equation. However, to understand the limitations of some analyzers, it is important to more fully investigate the relationship between the Fmax the LOR, and the Tmax.
Tmax = sample size/sample rate
Sample size = 2.56 * Lines of Resolution
Sample rate = 2.56 * Fmax

These terms have already been defined, but be aware that some analyzers have an upper limit on the sample size. Usually this number will be 1024 or 2048. Therefore, a 400 line spectrum would require 2.56*400 = 1024 samples and an 800 line spectrum would require 2.56* 800 = 2048 samples. However, if the analyzer is limited to 1024 samples, then the 800 line spectrum will be created from 1024 samples since it is the upper limit of the analyzer. This is important when discussing the Tmax in the time waveform, because, in general, raising the Fmax decreases Tmax and raising LOR increases Tmax until the point that the multiple of 2.56 * LOR reaches the sample limit in the analyzer. In this case, the sample size for anything greater than 400 lines is forced to be 1024. The table below demonstrates how this limitation affects the Tmax for a maximum 1024 sample size analyzer.
Tmax = sample size/sample rate

Fmax *2.56 =sample rate LOR*2.56 = sample size Tmax
400 1024 100 256 0.25
400 1024 200 512 0.50
400 1024 400 1024 1.00
400 1024 800 2048 1.00

The last entry in the table may seem incorrect, but remember 1024 is the maximum sample size, and anything that would be greater than 1024 is forced to be 1024 for the calculation of Tmax. This is the reason why your time waveform is not affected when following the “raise-the-LOR-to-lengthen-the-time-waveform” rule. You must be aware of the upper limit of the sample size and the number of lines of resolution to which this number corresponds.


The term aliasing brings up the idea of someone hiding as someone else. Criminals often do this to elude law enforcement agencies. In the world of vibration analysis, an aliased frequency is one that appears as something it is not. Fortunately, it is not trying to hide from us but was forced into existence by an incorrect digital sampling process.

Now, in a digital analyzer, the time domain data is digitally sampled at the sampling rate. We already know that the sampling rate is controlled by the maximum analysis frequency, Fmax.

In order to get good data, the sampling rate must be set higher than two times the maximum frequency of interest, Fmax. Two times Fmax is known as the Nyquist frequency. A very common sampling rate is set at 2.56 times the Fmax.

Now, an aliased frequency occurs when frequencies higher than the Fmax are present in the signal. The sampling rate under samples this high frequency and creates a low (aliased) frequency equal to the high frequency minus sampling frequency.

For example, if the Fmax is 1000 Hz, then the sampling frequency will be 1000 * 2.56 = 2560. If a frequency is in the data at 2768 Hz, then an aliased frequency can be expected in the data at 2768-2560 = 208 Hz.

To avoid aliasing, digital analyzers use anti-aliasing filters which remove all frequencies above 40% of the sampling rate before the time data is converted to frequency data.

Therefore, with most digital analyzers, aliasing is no longer a concern.


Leakage is a problem that can occur when complete cycles of the vibration signal in the time waveform are not captured in the time record. Instead of a discrete frequency in the spectrum, the result is energy spilling into adjacent bands or lines of resolution. A rectangular window, which is really not a window at all, can sometimes allow time data to be captured that can not be transformed well into the frequency domain. If a non-integer multiple of the periods of vibration appear in the time waveform for all the frequencies present, then the frequencies will smear over many lines of resolution.

Digital analyzers use windowing functions to avoid this problem. Typical window choices are Uniform, which is really no window at all, Hanning (often called a cosine taper), Flat top, Kaiser Bessel, Force, and Exponential as well as many others not listed here. The purpose of windowing data is to eliminate leakage problems. For most vibration analysis applications, the window of choice is the Hanning window. The Hanning window does a great job of forcing the beginning and the end of the time record to a zero amplitude. This allows for the reconstructed time waveform to be continuous with no amplitude variations.

A sinewave acquired with a uniform window that has an integer number of periods in the waveform will transform into the frequency spectrum with all of the energy contained in one spectral line of resolution. However, this condition is difficult to achieve without a lot of effort.

If the time waveform contains a non-integer multiple of periods, then the frequency data will smear over many lines of resolution.

The most typical case involves using a Hanning window on a non-integer multiple of periods in the time waveform. In this case, the data transformed to the frequency domain is contained primarily in one line of resolution with some frequency data smeared into the adjacent bands on either side of the primary frequency. This, however, tends to be the most acceptable method of data collection.

The Hanning window will still spread a pure discrete tone over three lines of resolution. To separate very closely spaced frequencies, increase the lines of resolution a little more. However, this can lead to another problem called the Picket Fence Effect.

Picket Fence Effect

The Picket Fence Effect occurs when the frequencies transformed to the frequency domain fall between the lines of resolution so that the amplitudes observed smear between adjacent lines of resolution. It’s like looking at the data though a picket fence with all of the peak amplitudes not necessarily showing up.

Some analyzers and some analysis software has the ability to look between the lines of resolution and pick the peak value. This is often called a peak detection routine or a locate feature.

Overlap Averaging

This is the process of re-using data stored in the input time buffer of the analyzer.

When using a Hanning window on our vibration data, most of the data has been multiplied by a factor of one at the center of the window to zero at the beginning and the end of the time window. In order to fully use all of the data being acquired in the analyzer, use overlap averaging. Fifty percent overlap averaging reuses 50% of the time data saving acquisition time and generally improving amplitude accuracy.

Integration and Differentiation

The vibration data that is the input signal into the analyzer is a time varying voltage proportional to the vibration measured by the transducer. In other words, an accelerometer produces a voltage that varies over time relative to the acceleration measured by the transducer. The voltage amplitude in the time waveform is converted to the desired amplitude units based on the sensitivity and conversion factor of the transducer.

Most analyzers have the ability to convert from the measurement units of the transducer to either of the other two units in the time domain or the frequency domain. At CSI integration of the time signal is called analog integration and integration of the frequency domain is called digital integration.

The process of integration is that of converting from acceleration to velocity or displacement, or converting from velocity to displacement.

The process of differentiation is that of converting from displacement to velocity or acceleration, or converting from velocity to acceleration.

Mathematically, how are these unit types related?

D (Displacement) = distance traveled by vibrating object
V (Velocity) = change in Displacement/change in Time
A (Acceleration) = change in Velocity/change in Time

Mathematically these terms are often represented with the following equations:

A = X/T/T = V/T

Therefore, if any one of these terms has been measured, integration and differentiation allows any of the other terms to be calculated provided the analyzer or software used is capable of this conversion process.

One drawback to integration is a flare-up of the lower frequency data due to the integration process. This effect is often called “integration noise” or a “ski-slope effect.” This is typically only noticeable when integrating from acceleration to displacement and tends to affect only the lower 1% of the frequencies. This may cause the overall vibration level to be higher than usual, and it may be excluded from the calculation of the overall vibration level.

Normally, this is not of any real concern unless the analyst is concerned with frequency analysis below 2-4 Hz. If this is the case, an investment in a low frequency accelerometer may be beneficial.


The vibration industry generally uses three types of filters:
Low Pass
Band Pass
High Pass

Each one filters data out of a signal, which you may find useful when analyzing signals with large dynamic ranges. For example, some spectra have both large and small amplitudes relative to each other. Because of the dynamic range of the analyzer, however, you cannot analyze the low amplitude vibration in the same plot as the high amplitude vibration. You would then use a filter to resolve the problem.

As shown above, the low pass filter removes data above the selected frequency. Use this filter during low speed (<600 RPM) frequency analysis. Data collection times can exceed 2-10 minutes per average. Therefore, you can increase data collection times with a low pass filter data by retaining only those events that take place in the selected frequency range.

You have two types of bandwidth filters from which to choose: Constant Percentage Bandwidth and Constant Bandwidth. These filters primarily serve the same function. The Constant Percentage Bandwidth filter changes width depending on the selected frequency.

Note the difference between the two types of filters in the example above. As shown, the filter to choose is the Constant Bandwidth Filter, because it provides the best resolution between both high and low frequency components.

The high pass filter displayed above gives you the ability to filter out low frequency components for detailed analysis. This proves useful when low frequency, high amplitude data swamps the high frequency, low amplitude data you want to see. This situation often occurs when high frequency events appear in the same plot as Run Speed and its relative harmonics.

Frequency Demodulation

Several publications have described the application of demodulation to vibration measurements for machinery defect analysis. As an application tool, demodulation proves helpful in a wide variety of applications.

The demodulation primarily increase the effective dynamic range of the analyzer for certain types of low level measurements. This increased range enhances defect indicators for fault analysis.

There are three primary functions for a demodulator.
A demodulator lets you use a low noise pre-amplifier to improve performance with very small input signals.
It provides the ability to filter input signals for specific analysis requirements.
It gives you envelope demodulation as an input signal processor for amplitude modulation analysis.


The key to collecting worthwhile data is understanding how the analyzer collecting the data performs its job. When setting up for data collection, remember that the spectral Fmax, the lines of resolution, the waveform Tmax, and the waveform size are all related. Don’t forget to choose the correct analysis window for each measurement application.

The conversion from analog data to digital data can alter the way that the data appears. Remember to recognize the limitations of the measurement process. Using the information in this section should help unlock the mysteries of digital signal processing.


Machinery Vibration Diagnostics 2


Normal Gear Spectrum

Typical Spectrum

Normal Spectrum shows 1x and 2x RPM, along with Gear Mesh Frequency (GMF). GMF commonly will have running speed sidebands around it relative to the shaft speed which the gear is attached to. All peaks are of low amplitude and no natural gear frequencies are excited.

Gear Tooth Wear
Typical Spectrum

A key indicator of gear tooth wear is excitation of the Gear Natural Frequency, along with sidebands around it spaced at the running speed of the bad gear. Gear Mesh Frequency (GMF) may or may not change in amplitude, although high amplitude sidebands surrounding GMF usually occur when wear is noticeable. Sidebands may be a better wear indicator than Gear Mesh Frequencies themselves.

Tooth Load
Typical Spectrum

Gear Mesh frequencies are often very sensitive to load. High GMF amplitudes do not necessarily indicate a problem, particularly if sideband frequencies remain low and no gear natural frequencies are excited. Each analysis should be performed with the system at maximum operating load.

Gear Eccentricity and Backlash
Typical Spectrum

Fairly high amplitude sidebands around GMF often suggest gear eccentricity, backlash or non-parallel shafts which allow the rotation of one gear to “modulate” the running speed of the other. The gear with the problem is indicated by the spacing of the sideband frequencies. Improper backlash normally excites GMF and Gear Natural Frequencies, both of which will be sidebanded at 1x RPM. GMF amplitudes will often decrease with increasing load if backlash is the problem.

Gear Misalignment
Typical Spectrum

Gear Misalignment almost always excites second order or higher GMF harmonics which are sidebanded at running speed. Often will show only small amplitude 1x GMF, but much higher levels at 2x or 3x GMF. Important to set the Fmax high enough to capture at least 2 GMF harmonics if the transducer has the capability.

Cracked or Broken Gear Tooth
Typical Spectrum

A Cracked or Broken Tooth will generate a high amplitude 1x RPM of this gear, plus it will excite the gear natural frequency (fn) sidebanded at its running speed. It is best detected in Time Waveform which will show a pronounced spike every time the problem tooth tries to mesh with teeth on the mating gear. Time between impacts (delta.gif (67 bytes)) will correspond to 1/speed of gear with the problem. Amplitudes of impact spike in Time Waveform will often be much higher than that of 1x Gear RPM in FFT.

Hunting Tooth Problems
Typical Spectrum

Hunting Tooth Frequency (fHT) is particularly effective for detecting faults on both the gear and the pinion that might have occurred during the manufacturing process or due to mishandling. It can cause quite a high vibration, but since it occurs at low frequencies, predominantly less than 600 CPM, it is often missed. A gear set with this tooth repeat problem normally emits a “growling” sound from the drive. The maximum effect occurs when the faulty pinion and gear teeth both enter mesh at the same time (on some drives, this may occur once every 10 or 20 revolutions, depending on the fHT formula). Note the TGear and TPinion refer to the number of teeth on the gear and pinion respectively. Na = number of unique assembly phases for a given tooth combination which equals the product of prime factors common to the number of teeth on each gear.


Blade Pass & Vane Pass

Typical Spectrum Machine Diagram 

bpf.gif pump.gif

Blade Pass Frequency (BPF) = number of blades (or vanes) x RPM. This frequency is inherent in pumps, fans and compressors and normally does not present a problem. However, large amplitude BPF (and harmonics) can be generated in the pump if the gap between the rotating vanes and the stationary diffusers is not kept equal all the way round. Also, BPF (or harmonics) sometimes coincide with with a system natural frequency causing high vibration. High BPF can be generated if the wear ring seizes on the shaft or if welds fastening diffuesers fail. Also, high BPF can be caused by abrupt bends in linework (or duct), obstructions which disturb the flow path, or if the pump or fan rotor is positioned eccentrically within the housing.

Flow Turbulence
Typical Spectrum

Flow turbulence often occurs in blowers due to variations in pressure or velocity of the air passing through the fan or connected linework. This flow disruption causes turbulence which will generate random, low frequency vibration, typically in the range of 20 to 2000 CPM.

Typical Spectrum

Cavitation normally generates random, higher frequency broadband energy which is sometimes superimposed with blade pass frequency harmonics. Normally indicates insufficient suction pressure (starvation). Cavitation can be quite destructive to pump internals if left uncorrected. It can particularly erode impeller vanes. When present, it often sounds as if “gravel” is passing through the pump.


Stator Eccentricity, Shorted Laminations and Loose Iron

Typical Spectrum

Stator problems generate high vibration at 2x line frequency (2FL). Stator eccentricity produces uneven stationary air gap between the rotor and the stator which produces very directional vibration. Differential air gap should not exceed 5% for induction motors and 10% for synchronous motors. Soft foot and warped bases can produce an eccentric stator. Loose iron is due to stator support weakness or looseness. Shorted stator laminations cause uneven, localised heating which can significantly grow with operating time.

Eccentric Air Gap (Variable air gap)
Typical Spectrum

Eccentric Rotors produce a rotating variable air gap between rotor and stator which induces pulsating vibration (normally between (2FL) and closest running speed harmonic). Often requires “zoom” spectrum to separate the (2FL) and the running speed harmonic. Eccentric rotors generate (2FL) surrounded by Pole Pass frequency sidebands (FP) as well as FP sidebands around running speed   FP appears itself at low frequency (Pole Pass Frequency = Slip Frequency x # Poles). Common values of FP range from approximately 20 to 120 CPM (.30 – 2.0 Hz)

Rotor Problems
Typical Spectrum

Broken or Cracked rotor bars or shorting rings, bad joints between rotor bars and shorting rings, or shorted rotor laminations will produce high 1x running speed vibration with pole pass frequency sidebands (FP). In addition, cracked rotor bars will often generate  FP sidebands around the 3rd, 4th and 5th running speed harmonics. Loose rotor bars are indicated by 2x line frequency (2FL) sidebands surrounding the rotor bar pass frequency (RBPF) and/or its harmonics (RBPF = Number of rotor bars x RPM). Often will cause high levels at 2x RBPF with only small amplitude at 1x RBPF.

Phasing Problems
Typical Spectrum

Phasing problems due to loose or broken connectors can cause excessive vibration at 2x Line frequency (2FL) which will have sidebands around it at 1/3rd Line Frequency (1/3 FL). Levels at (2FL) can exceed 25 mm/s (1.0 in/s) if left uncorrected. This is particularly a problem if the defective connector is only sporadically making contact and periodically not.

Synchronous Motors
Typical Spectrum

Loose stator coils in synchronous motors will generate fairly high vibration at Coil Pass Frequency (CPF) which equals the number of stator coils x RPM (# Stator Coils = Poles x # Coils/Pole). The coil pass frequency will be surrounded by 1x RPM sidebands.

DC Motor Problems
Typical Spectrum

DC motor problems can be detected by higher than normal amplitudes as SCR firing Frequency (6FL) and harmonics. These problems include broken field windings, bad SCR’s and loose connections. Other problems including loose or blown fuses and shorted control cards can cause high amplitude peaks at 1x through to 5x line frequency (3,600 – 18,000 CPM).


Rotor Rub

Typical Spectrum Phase Relationship
Type ‘A’

Rotor Rub produces similar spectra to Mechanical Looseness when rotating parts contact stationary components. Rub me be either partial or throughout the whole revolution. Usually generates a series of frequencies, often exciting one or more resonance’s. Often excites integer fraction sub harmonics of running speed (1/2, 1/3, 1/4, 1/5, ….1/n), depending on location of rotor natural frequencies. Rotor rub can excite many higher frequencies (similar to wide-band noise when chalk is drug along a blackboard). It can be very serious and of short duration if caused by shaft contacting bearing Babbitt; but less serious when the shaft is rubbing a seal, an agitator blade rubbing the wall of a vessel, or a coupling guard pressing against a shaft.



Typical Spectrum Phase Relationship
resonancespec.gif critical.gif

Resonance occurs when a Forcing Frequency coincides with a System Natural Frequency, and can cause dramatic amplitude amplification which can result in premature or even catastrophic failure. This may be a natural frequency of the rotor but can often originate from a support frame, foundation, gearbox or even drive belts. If a rotor is at or near resonance, it will be almost impossible to balance due to the great phase shift it experiences (90° at resonance; nearly 180° when it passes through). Often requires changing natural frequency location. Natural Frequencies do not change with a change in speed, this helps facilitate their identification.



Typical Spectrum

A Beat Frequency is the result of two closely spaced frequencies going into and out of synchronisation with one another. The wideband spectrum normally will show one peak pulsating up and down. When you zoom into this peak (lower spectrum), it actually shows two closely spaced peaks. The difference in these two peaks (F2 – F1) is the beat frequency which itself appears in the wideband spectrum. The beat frequency is not commonly seen in normal frequency range measurements since it is inherently low frequency. Usually ranging from only approximately 5 to 100 CPM.

Maximum vibration will result when the time waveform of one frequency (F1) comes into phase with other frequency (F2). Minimum vibration occurs when waveforms of these two frequencies line up 180° out of phase.


Worn, Loose or Mismatched Belts

Typical Spectrum

Belt frequencies are below the RPM of either the motor or the driven machine. When they are worn, loose or mismatched, they normally cause 3 to 4 multiples of belt frequency. Often 2x belt frequency is the dominant peak. Amplitudes are normally unsteady, sometimes pulsing with either driver or driven RPM. On timing belt drives, wear or pulley misalignment is indicated by high amplitudes at the timing belt frequency.

Belt / Sheave Misalignment
Typical Spectrum

Misalignment of sheaves produces high vibration at 1x RPM predominantly in the Axial direction. The ratio of amplitudes of driver to driven RPM depends on where the data is taken as well as on relative mass and frame stiffness. Often with sheave misalignment, the highest axial vibration will be at the fan RPM.

Eccentric Sheaves
Typical Spectrum

Eccentric and/or unbalanced sheaves cause high vibration at 1x RPM of this sheave. The amplitude is normally highest in line with the belts, and should show up on both driver and driven bearings. It is sometimes possible to balance eccentric sheaves by attaching washers to taperlock bolts. However, even if balanced, the eccentricity will still induce vibration and reversible fatigue stresses in the belt.

Belt Resonance
Typical Spectrum

Belt Resonance can cause high amplitudes if the belt natural frequency should happen to approach or coincide with either the motor or the driven machine RPM. Belt natural frequency can be altered by either changing the belt tension or the belt length. Can be detected by tensioning and the releasing belt while measuring response on sheaves or bearings.

Machinery Vibration Diagnostics 1


Force Unbalance

Typical Spectrum Phase Relationship

Force Unbalance will be in-phase and steady. Amplitude due to unbalance will increase by the square of speed (3x speed increase = 9x higher vibration. 1x RPM always present and normally dominates the spectrum. Can be corrected by placement of only one balance weight in one plane at rotor centre of gravity (CG).

Couple Unbalance
Typical Spectrum Phase Relationship
forcespec.gif couple.gif

Couple Unbalance tends toward 180° Out-of-phase on same shaft. 1x always present and normally dominates the spectrum. Amplitude varies with square of increasing speed. May cause high axial vibrations as well as radial. Correction requires placement of balance weights in at least 2 planes. Note that approx. 180° phase difference should exist between Outboard and Inboard horizontals as well as Outboard and Inboard verticals.

Overhung Rotor Unbalance
Typical Spectrum Phase Relationship
overhungspec.gif overhungdraw.gif

Overhung Rotor Unbalance causes high 1x vibration in both Axial and Radial directions. Axial readings might be unsteady. Overhung rotors often have both force and couple unbalance, each of which will likely require correction.

Eccentric Rotor
Typical Spectrum Phase Relationship
eccentric.gif eccendraw.gif

Eccentricity occurs when the centre of rotation is offset from the geometric centreline of a sheave, gear, bearing, motor armature, etc. The largest vibration occurs at 1x RPM of eccentric component in a direction through the centres of the two rotors. Comparative horizontal and vertical phase readings usually differ either by  0° or by 180° (each of which indicate straight line motion). Attempts to balance an eccentric rotor often results in reducing the vibration in one direction, but increasing it in the other radial direction (depending on the amount of eccentricity).

Bent Shaft
Typical Spectrum Phase Relationship
bentshaftspec.gif bentdraw.gif

Bent Shaft problems cause high axial vibration with axial phase differences tending toward 180° on the same machine component. The dominant vibration ins normally at 1x if bent near the shaft centre, but at 2x if bent near the coupling. (Be careful to account for the transducer orientation for each axial measurement if you reverse probe direction).


Angular Misalignment

Typical Spectrum Phase Relationship
Spectra Drawing

Angular Misalignment is characterised by high axial vibration, 180° Out-of-phase across the coupling. Typically will have high axial vibration with both 1x and 2x rpm. However, not unusual for either 1x, 2x or 3x to dominate. These symptoms may also indicate coupling problems as well.

Parallel Misalignment

Typical Spectrum Phase Relationship
Spectra Drawing

Offset Misalignment has similar vibration symptoms to Angular, but shows high radial vibration which approaches 180° Out-of-phase across the coupling. 2x often larger than 1x, but its height relative to 1x is often dictated by coupling type and construction. When either Angular or Radial Misalignment becomes sever, it can generate either high amplitude peaks at much higher harmonics (4x – 8x) or even a whole series of high frequency harmonics similar in appearance to mechanical looseness. Coupling construction will often greatly influence the shape of the spectrum when misalignment is severe.

Misaligned Bearing Cocked On Shaft

Typical Spectrum Phase Relationship
Spectra bearing

Cocked Bearing will generate considerable axial vibration. Will cause twisting motion with approximately 180° phase shift top to bottom and/or side to side as measured in the axial direction of the same bearing housing. Attempts to align the coupling or balance the rotor will not alleviate the problem. The bearing must be removed and correctly installed.


Mechanical Looseness

Typical Spectrum Phase Relationship
Type ‘A’
Type ‘B’
typeb.gif typebdraw.gif
Type ‘C’
typec.gif typecdraw.gif

Mechanical Looseness is indicated by either type A, B or C spectra. Type ‘A’ is caused by structural looseness/weakness of machine feet, baseplate or foundation, also by deteriorated grouting, loose hold-down bolts at the base and distortion of the frame or base (i.e Soft Foot). Phase analysis may reveal approx. 180° phase difference between vertical measurements on the machine foot, baseplate and base itself. Type ‘B’ is generally caused by loose pillowblock bolts, cracks in the frame structure or bearing pedestal. Type ‘C’ is normally generated by improper fit between component parts which will cause many harmonics due to non linear response of loose parts to dynamic forces from the rotor. Causes a truncation of time waveform. Type ‘C’ is often caused by a bearing liner loose in its cap, excessive clearance in either a sleeve or rolling element bearing or a loose impeller on a shaft. Type ‘C’ phase is often unstable and may vary widely from one measurement to the next, particularly if the rotor shifts position on the shaft from one start-up to the next. Mechanical looseness is often highly directional and may cause noticeably different readings if you compare levels at 30° increments in the radial direction all the way around one bearing housing. Also note that looseness will often cause subharmonic multiples at exactly 1/2 or 1/3 x rpm (.5x, 1.5x, 2.5x etc.)


Rolling Element Bearings
(4 Failure Phases)

Stages of Progressive Deterioration
spacer20v.gif (51 bytes)
Stage 1
Stage 1 : Earliest indications of bearing problems appear in ultrasonic frequencies ranging from approximately 20,000 – 60,000 Hz (1,200,000 – 3,600,000 CPM). These are frequencies evaluated by spike energy (gSE), HFD(g) and shock pulse (dB). Foe example, spike energy may first appear at about .25 gSE in stage 1 (actual value depending on measurement location and machine speed.

Stage 2stage2.gif Stage 2 : Slight bearing defects begin to “ring” bearing component natural frequencies (fn) which predominantly occur in the 30K – 120K CPM range. Sideband frequencies appear above and below natural frequency peak at end of stage 2. Spike energy grows (for example .25 to .50 gSE).

Stage 3stage3.gif Stage 3 : Bearing defect frequencies and harmonics appear when wear progresses. More defect frequency harmonics appear and a number of sidebands grow, both around these and around bearing natural frequencies (see Vibration Case History Number 3 for actual example). Spike energy continues to increase (for example, from .5 to over 1 gSE). Wear is now usually visible and may extend throughout periphery of bearing, particularly when well formed sidebands accompany any bearing defect frequency harmonics. replace the bearings now.

Stage 4stage4.gif Stage 4 :Towards the end, the amplitude of the 1x RPM is even effected. It grows, and normally causes growth of many running speed harmonics. Discrete bearing defect and component natural frequencies actually begin to “disappear” and are replaced by random, broadband high frequency “noise floor”. In addition, amplitudes of both high frequency noise floor and spike energy may in fact decrease, but just prior to failure, spike energy will usually grow to excessive amplitudes.

Formulae to Calculate Specific Bearing defect Types.




Wear / Clearance Problems

spacer20v.gif (51 bytes)Typical Spectrum

Latter stages of sleeve bearing wear are normally evidenced by the presence of whole series of running speed harmonics (up to 10 or 20). Wiped sleeve bearings often allow high vertical amplitudes compared to horizontal. Sleeve bearings with excessive clearance may allow a minor unbalance and/or misalignment to cause high vibration which would be much lower if bearing clearances were to specification.

Oil Whirl Instability
Typical Spectrum Shaft Diagram
oilwhirlspec.gif shaftdraw.gif
Oil Whirl instability occurs at 0.42 – 0.48 x RPM and is often quite severe. Considered excessive when amplitude exceeds 50% of bearing clearances. Oil whirl is an oil film excited vibration where deviations in normal operating conditions (attitude angle and eccentricity ratio) cause oil wedge to “push” the shaft around within the bearing. Destabilising force in the direction of rotation results in a whirl (or precession). Whirl is inherently unstable since it increases centrifugal forces which increase whirl forces. Can cause oil to no longer support the shaft, or can become unstable when whirl frequency coincides with a rotor natural frequency. Changes in oil viscosity, lube pressure and external pre-loads can affect oil whirl. 

Oil Whip Instability
Typical Spectrum
A Spectral Map showing Oil Whirl becoming Oil Whip Instability as shaft speed reaches twice critical. 


Oil Whip may occur if a machine is operated at or above 2x rotor critical frequency. When the rotor is brought up to twice critical speed, whirl will be very close to rotor critical and may cause excessive vibration that the oil film may no longer be capable of supporting. Whirl speed will actually “lock onto” rotor critical and this peak will not pass through it even if the machine is brought up to higher and higher speeds.


Business Practice Area Predictive Maintenance

Business Practice Area Reliability Improvement
1. Set up database

Setting up database Pdm ( equipment & technology matric ) berdasarkan SERP & FMEA
2. Schedul

Jadwal bulanan pelaksanaan Pdm termasuk didalamnya resource manhours dan peralatannya
3. Persiapan tehnis lapangan

Identifikasi  dan persiapan pelaksanaan pekerjaan : orang, alat, metode, link bagian lain
4. Pengukuran ( Monitoring )

Pengamatan kondisi peralatan dilakukan dengan mengukur level vibrasi, kondisi pelumasan, panas, impurities dll menggunakan peralatan vibration monitoring, tribology, infra red dll
5. Data management

Penanganan data – data kondisi peralatan secara computerized dari data pengukuran dan data lainnya, termasuk didalamnya membuat trend data, warning sistem dsb
6. Anilsa & Rekomendasi

Analisa dari data terkumpul dan seluruh kondisi yang mempengaruhi operasi peralatan pembangkit dan memberikan rekomendasi kepada O/M
7. Tindak lanjut

Pelaksanaan, pengamatan atau perubahan schedul dan pekerjaan dari hasil analisa dan rekomendasi
8. Cost Benefit Analysis

Kalkulasi biaya pelaksanaan Pdm dan hasil rekomendasinya dbanding dengan biaya yang akan timbul jika pemeliharaan tidak terencana