Malfunction Diagnosis:

Shaft Crack

By

Charles T. Hatch

Bently Nevada Training Development Group

Bently Nevada Corporation
 Table of Contents

Introduction to Shaft Cracks ................................................................... 1

What Is a Shaft Crack? .................................................................... 1
Crack Initiation and Growth ............................................................. 1
Effects On the Rotor System Due To a Crack ............................................ 3
Reduction of Shaft Stiffness ............................................................. 3
Asymmetric Shaft Stiffness .............................................................. 3
Shaft Crack Detection ........................................................................... 5
The First Rule of Crack Detection (1X) .............................................. 5
The Second Rule of Crack Detection (2X)........................................... 6
Other Malfunctions With Similar Symptoms ............................................. 7
1X Behavior .................................................................................. 7
2X Behavior .................................................................................. 8
Design, Operating, and Monitoring Recommendations ................................ 8
Design and Operation...................................................................... 8
Monitoring .................................................................................... 9
 Introduction to Shaft Cracks

What Is a Shaft Crack?

A shaft crack can be thought of as a slowly growing fracture of the rotor. If
undetected in an operating machine, a crack (also called a fatigue crack) will grow over
time until the remaining, reduced cross section of the rotor is unable to withstand the
static or dynamic loads that are applied to it. When this happens, the remaining rotor
section will fail in a fast brittle fracture mode. The sudden failure will release the large
amount of energy that is stored in the rotating system, and the rotor will fly apart. Shaft
fractures have caused machine parts to penetrate the machine casing and even penetrate
building walls. Damage due to this kind of failure is catastrophic and can cause serious
injury or death to anyone unfortunate enough to be standing near the machine at the
moment of failure. Obviously, shaft crack detection is a very serious matter, and
machines that are suspected of having a crack must be treated with the utmost respect.

Crack Initiation and Growth

Rotors, because of their rotation, are
subject to periodically changing, or Rotor Centered

Tension
cyclical, stresses. If a rotor orbits about
the center of the rotor system in pure 1X
0
precession, the stress at any particular
outer fiber will see no change in stress. 0 1 2
However, if the rotor is offset from the
Tension

Rotor Offset With

axis of the rotor system (typically the Elliptical Orbit

case because of a radial process load or 0

gravity), then rotor outer fibers will see a
Comp

1X variation in stress. In addition, a 1X

0 1 2
elliptical orbit (which is also typical) Shaft Revolutions
produces 2X stress cycling (see Figure 1). Figure 1. An example of outer fiber stress
Thus, even under normal, 1X operation, variation for a rotor in simple bending. When the
real rotors live in a complicated stress rotor moves about the center of the system (top)
in a 1X circular orbit, the stress is constant. When
environment that contains a mixture of the rotor is displaced from the system center in a
1X and 2X stress cycling. Any sub- or 1X elliptical orbit (bottom), the rotor sees
supersynchronous vibration that may be variable stress with a mixture of 1X (from the
present will produce an additional displacement) and 2X (from the ellipticity).
complicated pattern of cyclic stresses in
the rotor.
Cracks are initiated on the shaft in regions of high local stress. Shafts are subjected to
large-scale stresses due to static or dynamic bending and torsional twisting, static radial
loads, or residual stresses from heat treatment, welding, or machining operations. These
larger-scale stresses can be concentrated by geometric factors such as step changes in
shaft diameter, shrink fits, keyways, drilled holes, or other discontinuities. Further
concentration can occur at the microstructure level where surface machining
imperfections, chemical surface damage, or material discontinuities (such as produced by

1
slag inclusions or chemical impurities) can produce high, local stress concentrations. All
of these stresses combine to produce a local stress field that changes with time. The end
result can produce a small local region where stresses exceed the maximum that the
material can withstand, and a microcrack will form in the material. Because rotor bending
tends to produce the highest stresses at the outer surface, rotor cracks usually, but not
always, start at or near the outer surface. Sometimes, because of chemical or other
processing problems in the rotor billet, a
microcrack may exist inside the rotor Torsion
before it is put into service. Crack
Transverse
Once initiated, and if sufficiently high
Crack
cyclic, or alternating, stresses are present,
the crack tip will slowly propagate in a
direction perpendicular to the orientation of
the local maximum tensile stress at the
crack tip. The orientation of this stress field
is affected by the type of stress (bending or
torsional) and by any geometric factors. If
a rotor is subjected only to simple bending Figure 2. A transverse crack results from pure
stresses, then the stress field will be bending stress in the shaft and propagates directly
oriented along the long axis of the rotor, into the shaft. A crack resulting from pure
torsional stress forms a spiral at 45° to the long
and the crack will propagate directly into axis of the shaft. Most shafts contain a mixture of
and circumferentially across the rotor bending and torsion stress. The local stress field
section (Figure 2). Pure torsional stress will at the crack tip, which can be influenced by local
produce a crack that is oriented at 45° geometry, determines crack propagation
relative to the long axis. The crack will direction. The crack tip propagates perpendicular
to the direction of the maximum local tensile
propagate into the rotor, but the crack will stress.
tend to form a spiral on the shaft surface.
In rotor systems, the stress field usually contains a mixture of bending and torsional
stress. Bending stress is usually the dominant component; thus, the crack will usually
propagate into the rotor more or less as a transverse crack. However, other crack
geometries are possible.
As the crack propagates, less and less material is available to transmit loads in the
rotor shaft, and the local stress across the remaining shaft material becomes higher and
higher. At some point, the section will become so small that, during the next load
application, the local stress intensity will exceed the fracture toughness of the material.
The fracture toughness is a measure of the material’s resistance to fast fracture and is a
function of the alloy, heat treatment, the material temperature, and the rate of loading of
the shaft. When the fracture toughness is exceeded, the remaining section will undergo a
fast brittle fracture, and the rotor will break in half.
Rotor shafts are usually manufactured out of materials with high fracture toughness.
Rotor shaft cracks have exceeded 90% of the shaft cross sectional area before final
fracture, although certainly one should not depend on this. It is not an easy matter to
determine crack size in a rotating machine, and any machine suspected of having a shaft
crack should be shut down as soon as possible.

2
 Effects On the Rotor System Due To a Crack

Reduction of Shaft Stiffness

Rotor shaft bending stiffness is proportional to the cross section area moment of
inertia. The larger this term, the higher the stiffness. As the crack propagates across the
shaft, the area moment of inertia becomes smaller, and the bending stiffness of the shaft
decreases. For a given radial force on the rotor, if the stiffness decreases, the rotor
deflection will become larger.
This reduction in shaft stiffness allows the rotor to bow more in response to a load
such as a rotating unbalance. Depending on the orientation of the fracture surface, the
original heavy spot location, and the relationship of rotor running speed relative to
resonance, the rotor could bow in virtually any direction. Thus, it is likely that crack
propagation will produce a change in both 1X amplitude and phase over time.
The 1X amplitude and/or phase are likely to change slowly in the early stages of
crack growth. But, as the crack propagates across the shaft section, further reducing the
shaft stiffness, 1X amplitude is likely to increase at an ever-faster rate. 1X phase may not
change so dramatically in the later stages of crack growth.
Rotor bow will have a significant influence on measured 1X slow roll vectors. As the
crack propagates across the rotor section, the slow roll vector is likely to change
dramatically from shutdown to shutdown. Vibration amplification while passing through
a resonance may yield the rotor at the weak point and cause very dramatic changes in 1X
response.
Balance resonance frequencies can
be affected by changes in rotor shaft
stiffness. Thus, as a crack propagates High
Phase

across the rotor, balance resonances Low Stiffness

may be shifted downward to a lower Stiffness
speed (Figure 3). Depending on the
mode shape and the location of the
crack, some resonances (or modes) of a
Amp

rotor could be affected more than

others. Also, rotor mode shapes may
change, depending on the location of the 1 2 3 4

crack and how it effects the stiffness Speed

distribution of the shaft. Figure 3. As a crack propagates, rotor shaft
stiffness will decrease, and resonant speeds may
move downward. Here, the changes in 1X
Asymmetric Shaft Stiffness amplitude and phase that could occur are shown
As a transverse crack propagates for four different operating speeds.
across the rotor shaft section, the shape
of the remaining, uncracked material will become increasingly asymmetric in shape
(Figure 4). This asymmetric shape produces an asymmetric stiffness in the rotor.
This can be demonstrated with a flexible ruler. The ruler behaves like a beam, and the
stiffness in one direction is very much larger than the stiffness in the perpendicular
direction (Figure 5). If the ruler is simply supported horizontally, and a steady,
unidirectional force is applied at midspan, the resulting deflection will be very different

3
for the two ruler orientations.
Remaining
Furthermore, if the ruler is at an angle, a Shaft Section Crack
force produces a response that has a
quadrature component. If the ruler rotates
about its long axis, then this weak/strong
transition will produce a “snapping”
action two times per revolution. Thus, the
asymmetrically stiff, rotating ruler which
is subjected to a steady, unidirectional,
radial force will produce a 2X vibration Lower Higher
component. Stiffness Stiffness
In a rotor shaft, the crack produces a
similar asymmetric stiffness. If the rotor Figure 4. As a crack grows, the remaining shaft
is subjected to a unidirectional radial section shape becomes asymmetric. This
produces an asymmetric rotor shaft stiffness that
force, then a crack will tend to produce a rotates with the rotor.
2X component in the rotor response. The
radial force acting on the rotor could be
due to gravity in a horizontal machine,
fluid reaction forces in a horizontal or
vertical machine, unequal steam
admission in a steam turbine, or
misalignment. The 2X rotor precession
produced by crack stiffness asymmetry
tends to be predominately forward in
direction.
The 2X snapping action due to the a b c
crack also produces a strong 2X torque
impulse in the rotor system that is Figure 5. The effect of asymmetric rotor stiffness
available to excite any torsional can be demonstrated with a simply supported
resonance at that frequency. Because ruler. In the figure, identical downward loads (red
torsional vibration can easily cross couple arrows) are applied to a ruler in three positions.
into lateral vibration, this is likely to The responses are shown as blue arrows. The
ruler has minimum stiffness and largest
increase the 2X lateral vibration further. deflection at position (a). It has maximum
The 2X component of lateral vibration stiffness and smallest deflection at position (b).
can be greatly amplified if a rotor At intermediate position (c), the ruler has a
operates at half the speed of a lateral or perpendicular (quadrature) component of
(because of cross coupling) half the speed deflection. If the ruler is rotating with a similar
unidirectional load, a snapping motion will be
of a torsional resonance. As we will see, seen two times per revolution (2X).
when it is present, this behavior can
provide a useful, additional piece of information when diagnosing a possible shaft crack.
But remember, for a crack to produce 2X vibration, a radial unidirectional load must be
present in the rotor system, and this may not be the case for a particular machine.
When it occurs, the 2X forward vibration due to a crack is likely cause an increase in
the amplitude of the 1X alternating stress in the rotor near the crack. Note that +2X
forward precession with +1X forward rotation produces a net +1X forward stress cycling.
(2X – 1X = 1X). Any 2X reverse vibration components will also increase the frequency

4
of the alternating stress (-2X precession – 1X rotation = -3X stress cycling). The
combination of these effects is very likely to promote more rapid crack growth and hasten
the failure of the rotor.

Shaft Crack Detection

The First Rule of Crack Detection (1X)

The first rule has to do with 1X filtered vibration: If a rotor is cracked, it is very likely
to be bowed. And that bow is likely to change over time. A change in rotor bow will
change the effective location and magnitude of the heavy spot, which will produce a
change in 1X rotor response. Thus, continuous changes in 1X amplitude and/or phase are
the best primary indicator of a shaft crack. As the crack grows and the associated bow
develops, 1X amplitude and phase will change in such a way as to produce a non-
repeating pattern on a Bode or polar plot over time (Figure 6). The time scale of this
change can range from months to weeks in the early stages of crack growth, to weeks to
days as the rotor begins to seriously weaken, and to hours as the rotor nears catastrophic
failure.
As failure nears, 1X vibration
amplitude will usually increase rapidly. At Generator Outboard X 1800 RPM 1X Uncomp
this point, 1X vibration is likely to be the
Phase Lag

dominant source of vibration in the system,

so overall direct vibration will also
increase rapidly. Thus, steady increases in
unfiltered peak-to-peak vibration over time
Amplitude

should be taken very seriously and

investigated.
Because of the changing bow of the 1 Mar 1 Apr 1 May 1 Jun 1 Jul 1 Aug
Time
rotor, the amplitude and/or phase of the 1X
filtered slow roll vector is also likely to
change as the crack propagates. Slow roll Figure 6. Example 1X APHT plot showing
vectors should be compared to historical possible 1X vibration changes due to a crack. The
machine runs at a constant speed, and amplitude
slow roll data. and phase change slowly over time. Near the end,
Occasionally, a diagnostician may shaft stiffness will drop rapidly as the crack
encounter a machine with a “balance weakens the section (red arrow), and vibration
problem.” Perhaps the machine had no amplitude will increase rapidly. The details of
history of such a problem before. Changing each machine’s behavior are different.
rotor bow due to growth of a shaft crack will change the location and magnitude of the
effective heavy spot of the rotor. If this happens, a previous balance correction may soon
be rendered ineffective, and the 1X vibration will increase again. If the root problem is a
shaft crack, repeated attempts to balance the machine will not solve the problem.
Changes in 1X rotor behavior in resonances are an indication that something has
changed in the rotor system. A significant downward shift in a balance resonance speed is
a clear indication that the stiffness of the rotor system has decreased. One then has to ask
why this has happened. A weakening shaft due to a crack is a possibility.

5
 1X vibration is usually very sensitive to the presence of a shaft crack because of the
relationship between crack growth and rotor bow, and it can provide significant early
warning of a crack.

The Second Rule of Crack Detection (2X)

The second rule has
to do with 2X filtered
vibration: If a rotor Vertical
with a crack has a Frequency
steady, unidirectional Relationship

Rotor Speed (rpm)

radial load, then a
strong 2X response may
appear when the rotor
is turning at half of a
balance resonance
speed. In addition, this
2X component is likely
to be predominately
forward (although it
may be elliptical). The
2X snapping action of
the rotor produces
lateral and torsional Precession Frequency (kcpm)
impulses in the system.
Because these impulses Figure 7. Full spectrum cascade startup data from a rotor with a shaft
occur twice per cross-section asymmetry and a unidirectional radial load. A significant
2X component appears when the running speed (left axis) reaches ½ of
revolution, the rotor the balance resonance speed near 3600 cpm. Note that a reverse 2X
will respond at the 2X component is also visible but is smaller than the forward component.
frequency. If a Thus, the 2X component is forward and elliptical.
resonance exists at
twice running speed, then the 2X vibration will be amplified. This forms a vertical
relationship on a half or full spectrum cascade plot (Figure 7).
Note that a rotor could pass through such a 2X speed relationship during startup or
shutdown, or a rotor could normally operate at half of a resonance speed. Obviously, it is
less likely that such a relationship would exist at normal operating speed. For that reason,
and because of the additional requirement that a unidirectional radial load be present, a
crack may not produce significant 2X vibration at running speed.
In fact, in about 75% of cases, 2X vibration does not appear when operating at design
speed. For this reason, 2X vibration should never be used as the only tool for crack
detection. 2X vibration data should be used together with 1X data to ensure the earliest
and most reliable detection of a shaft crack.
Like 1X vibration, 2X vibration amplitude and/or phase is likely to change as the
crack propagates through the rotor shaft. Startup and shutdown 2X Bode and polar plots
should be examined for any evidence of change (Figure 8). Also, 2X amplitude and phase
should be trended during steady state operation. In one case, a reactor coolant pump
developed a crack while the pump was operating at a constant speed. As the crack

6
 2X Filtered Data Note Phase Change in Leading Direction
90° 90°

TN
19 June 28 June
RO

1853
* * 1756
5011 525
5541 * * * 1715
* 1908 425 * 2170 * 875
6034 * * 4840 *
* * 3835 * 323 * 1676
6998 * 4359 * 6776 * * * 5008
2667 * * 3231 5961 * 4868
180° * 0° 180° * 6947 0° * 1657
7101 7193 * 2934
* * * 6987 * 1039
* * 1638
3031 * 7029
7146 *
* 1059 1599
* 7068 7194 * * 1200 * 1560
7171 * * 1257
*
* * 7149 1440
7107 *
7127
270° 270°
4 mils pp Full Scale
Figure 8. 2X filtered polar plots showing data from two startups of a machine with a rotor crack. The
startup on 28 June shows drastically different behavior than the earlier startup on 19 June. On the later
startup 2X amplitude is larger, and some unusual, leading phase behavior is visible.

propagated, the rotor shaft stiffness SPEED: 1187 rpm

2X Filtered
dropped so much that a resonance that
originally existed above twice the 0
Phase Lag

90
operating speed moved down in 180
frequency and passed completely through 270
the 2X frequency before the pump was 360

finally shut down (Figure 9). 10.0

Amplitude

8.0
6.0
Other Malfunctions With Similar 4.0
2.0
Symptoms 0.0
1SEP 25SEP 19OCT 12NOV 6DEC

1X Behavior Figure 9. 2X filtered APHT plot of a Reactor

Coolant Pump with a crack. The pump operated
Many other malfunctions can produce at a constant speed of 1187 rpm. As the shaft
a change in 1X rotor response. weakened, the reduced shaft spring stiffness
A loose bearing support, or soft foot caused a resonance that was originally above
can cause a change in 1X vibration. twice running speed to move down and pass
Usually, but not always, this is through twice running speed. Note that, when the
machine was shut down, the 2X amplitude was
manifested as an increase in 1X vibration
decreasing.
amplitude. This is the kind of malfunction
that could develop over time with a slow
increase in 1X vibration. Because it mimics the behavior of a shaft crack, it can be very
difficult to determine the root cause. If casing measurements are available, an increase in
casing vibration with little or no increase in shaft relative might suggest a soft foot
problem, while an increase in shaft relative with little increase in casing might suggest a
crack. But, there are no firm rules here.

7
 Thermal growth and subsequent changes in alignment can affect the rotor bearing
stiffness and produce changes in 1X vibration. Thermal bow of a rotor such as a
generator can also produce a similar change, as could an alignment change. These
changes in vibration should stabilize once the machine reaches thermal equilibrium at
steady speed and load.
Rub can cause changes in both 1X and 2X vibration. These changes can be sudden,
occurring at operating speed, or the changes can show up as changes in transient behavior
during startup or shutdown. Rub can disappear if the part in contact wear away (this can
happen in seals). Or, if the rub is severe, rub contact may be maintained for a
considerable time. However, rub is not as likely to produce a steadily increasing 1X
vibration level over a long period of time.
A loose rotating part can produce changes in 1X response. If a part moves to a
different angular or axial position on the rotor, the resulting total unbalance of the rotor is
likely to change, and the 1X amplitude and/or phase will change accordingly. Loose parts
can shift occasionally, producing stepwise changes in 1X response, or they can shift
continuously, producing a continuously changing response. Continuously moving parts
will tend to produce a cyclic, repeating behavior on a polar or APHT plot. A loose part is
not likely to produce a steady, long-term increase in 1X vibration amplitude.
Shifting debris or liquid inside a rotor can produce significant differences in heavy
spot location in a machine. This will produce corresponding changes in 1X vibration
response and cause a machine to go out of balance frequently.
A locked gear coupling can also produce a sudden step change in 1X vibration.
The key to crack identification is to realize that a developing crack is likely to
produce a steady and accelerating increase in 1X vibration amplitude over time as the
shaft stiffness decreases. While some malfunctions will produce periodic changes in 1X
vibration amplitude and/or phase, shaft cracks will tend to produce non-repeating
patterns on polar and APHT trend plots, with the 1X amplitude trending to ever higher
levels.

2X Behavior
Nonlinearities in rotor system stiffness can cause harmonics (2X, 3X, etc.) of running
speed to appear in spectra. Nonlinear stiffness can be caused by high eccentricity ratios in
fluid-film bearings or by rub impacting. Also, coupling problems can produce 2X
vibration.
If any source of 2X vibration exists in a machine, it will be available to excite a
resonance at half of a balance resonance speed. Thus, the presence of 2X at half a
resonance, while suspicious, is not in and of itself confirmation of a crack.

Design, Operating, and Monitoring Recommendations

Design and Operation

Shaft cracks usually result from a combination of design shortcomings, improper
machine installation, and/or poor operating practice. There are several things that can
contribute to the initiation and propagation of a shaft crack.

8
 At the design level, any stress concentration can lead to trouble if it is not properly
accounted for in the fatigue life of the design. Keyways, notches, drilled holes, and step
changes in shaft diameter are all capable of producing large stress concentrations in the
shaft. Shrink fits produce high tensile stresses at the surface of the rotor in the vicinity of
the fit. Retaining-pin holes and grooves should not be used near or under shrink fits
because the combination of shrink fit stresses and geometric stress concentrations can
lead to very high local stresses.
Low vibration will help minimize the chance of a shaft crack. High dynamic vibration
in operating machines can produce higher cyclic stresses than the designer may have
anticipated in his or her design. High dynamically induced stresses have combined with
geometric factors to produce catastrophic cracks.
Predicted vibration mode shapes can be helpful in the prediction of deflections and
stresses. But the dynamic behavior of individual machines can change when several
machines are coupled together. For this reason, the dynamic characteristics of the entire
coupled machine train should be taken into account when estimating mode shapes,
deflections, and stresses.
Good machine train alignment will help minimize the chance of a shaft crack. Severe
misalignment between adjacent machines in a train can cause bending of the rotor
between machines. This can result in very high 1X cyclic stresses that may exceed the
allowable limit of the design.
Operation of machines on or near resonances should be avoided. It can produce high
sensitivity to unbalance and high vibration with the potential for high stresses. Operation
of a machine at half of any lateral or torsional resonance should also be avoided. If 2X
vibration is present, it is likely to be amplified by the resonance at twice running speed.
The 2X vibration will produce cycling stresses and increase the risk of a shaft crack. 2X
Bode or polar plots should be examined for any evidence of a resonance at or near twice
operating speed.

Monitoring
In one example, during a startup a steam turbine encountered high 1X vibration and
was shut down. The problem was assumed to be an unbalance problem. Successive
attempts to restart the unit encountered increasing 1X vibration levels and the appearance
of some 2X vibration. A decision was made to disassemble the unit, where a partial shaft
crack was found in the rotor. Subsequent examination of all of the vibration data showed
that, in addition to the high 1X vibration amplitude, a significant 1X phase shift of nearly
40° had taken place during the attempted startup. But the phase had not been monitored
even though the equipment had been available to do so. Remember: if you don’t look for
signs of a shaft crack, you may not detect one in time to prevent a catastrophe.
It is not possible to look at a single piece of data and positively diagnose a shaft
crack. Many times it is difficult to make a diagnosis with all the data at hand. However, if
you apply the following monitoring recommendations, you will have a much better
chance of detecting a shaft crack in a timely manner.
At operating speed, monitor direct (unfiltered) vibration levels and 1X and 2X
amplitude and phase. All these data should be trended over time. This data trending can
be as simple as a list of hand logged data (1X and 2X amplitude and phase) or, better, the
same data can displayed in the form of APHT and polar plots. In addition, 1X and 2X

9
acceptance regions (Figure 10) should be
defined in order to detect significant 0°
changes in vibration when they occur.
These acceptance regions may need to be
adjusted for different load conditions.
Steady state monitoring of this kind has
provided warning as early as a 25% crack. 270° 90°
Every time a machine is shut down and
restarted, shutdown and startup data (in the
form of 1X and 2X Bode and polar plots)
should be compared to earlier data. Look
for significant changes in 1X and 2X
180°
behavior through resonances, a decrease in
one or more resonance speeds, or other Figure 10. Polar plot showing Acceptance
abnormal behavior (for example, phase Regions in green. Acceptance Region boundaries
changing in a leading direction through a can depend on load or other operating conditions
and they can be constructed for either 1X or 2X
resonance, see Figure 8). filtered vibration vectors. Operation in the red
As a crack propagates, the bow of the area outside of the Acceptance Regions is cause
rotor is likely to change, and the amplitude for alarm.
and/or phase of the 1X filtered slow roll
vectors will also change. 1X slow roll vectors should be logged and compared to earlier
data.
Significant changes in transient machine response should be a cause for concern and
warrant further investigation. Transient monitoring of this kind has provided warning as
early as a 20% crack.
While there are no guarantees, with careful and thoughtful machine monitoring you
have a very good chance of detecting a crack before a catastrophic failure occurs.

10