Balancing tutorial

ABSTRACT
The purpose of this tutorial is to describe details related to shop and field balancing of turbomachinery. The target
machinery is API type turbomachinery such as steam turbines, compressors, and other rotating machinery that is
common in refining and petrochemical plants.
An introduction to shop balancing includes a review of current practice as it relates to shop balancing of API rotors.
Included in the description and discussion is a review of API shop balancing methods as well as a review of ISO
standards related to shop and field balancing that are referenced by API standards.
Greater emphasis is presented in this tutorial on field balancing, which applies to balance correction in situ on
rotating machinery and similarly applies to methods and techniques used when conducting high speed shop
balancing.
INTRODUCTION
Rotor unbalance is a common cause of synchronous rotor vibration that is detected using non-contacting proximity
probes or with bearing housing vibration. The causes of unbalance can be varied with the actual causes depending
on manufacturing methods and procedures, repair practices, as well as balance condition changes during operation.
Some operational causes of unbalance include rotor fouling (dirt or other deposits on the rotor), bowing of rotors
due to uneven heating or shaft damage, loss of rotor material possibly from rubs, or other causes.
Balance correction is most effective when it is applied at or on the component that actually has the unbalance. In
most cases the rotor is manufactured from a number of components (impellers, balance disks, thrust collars, etc.)
that will each have some level of unbalance during assembly of the rotor. During operation, the unbalance state of
each mounted component could change due to reasons stated above. As a result, the actual balance condition of an
assembled rotor is never fully known prior to, during, or after a successful balance procedure is executed either in
the shop or field.
DEFINITION OF UNBALANCE
Unbalance will occur any time a rotor or a component mounted on a rotor has a mass center (or center of gravity,
i.e. cg.) that is not coincident with the axis of rotation. When this occurs, a force is generated due to rotation of the
shaft that is defined by the following equation and shown graphically in Figure 1:

𝐹𝑏𝑎𝑙𝑎𝑛𝑐𝑒= 𝑚 × 𝑒 × 𝜔2

𝐹𝑏𝑎𝑙𝑎𝑛𝑐𝑒= 𝐹𝑜𝑟𝑐𝑒 𝑐𝑎𝑢𝑠𝑒𝑑 𝑏𝑦 𝑡ℎ𝑒 𝑖𝑚𝑏𝑎𝑙𝑎𝑛𝑐𝑒

Where

𝑚 = 𝑀𝑎𝑠𝑠 𝑜𝑓 𝑡ℎ𝑒 𝑒𝑐𝑐𝑒𝑛𝑡𝑟𝑖𝑐 𝑝𝑎𝑟𝑡 𝑜𝑟 𝑟𝑜𝑡𝑜𝑟

𝑒 = 𝐸𝑐𝑐𝑒𝑛𝑡𝑟𝑖𝑐𝑖𝑡𝑦 𝑜𝑓 𝑡ℎ𝑒 𝑝𝑎𝑟𝑡 𝑜𝑟 𝑟𝑜𝑡𝑜𝑟
𝜔 = 𝑆ℎ𝑎𝑓𝑡 𝑠𝑝𝑒𝑒𝑑, 𝑟𝑎𝑑/ sec =
2 πN
60
𝑁 = 𝑆ℎ𝑎𝑓𝑡 𝑠𝑝𝑒𝑒𝑑, 𝑅𝑃𝑀
 Figure 1 - Unbalance Force

This force rotates about the shaft that is phased to the shaft which results in vibration at 1xRPM. It is possible to
measure the dynamic force in a balance machine, but not possible when operating the machine on the rotor support
bearings as is the case for either high speed shop balancing or field balancing. As the machine operates, the balance
condition of the rotor or components on the rotor can be evaluated at least in part by measuring and assessing the
vibration characteristics of the machine. Many other machine faults also produce vibration at 1xRPM such as
bearing or shaft misalignment, loose components, rubs, or a host of other sources.
There are two general conventions for defining an amount of unbalance for a component or a shaft including
eccentricity and unbalance magnitude. Both conventions can be used to define or describe the balance state of a
component or a shaft and are often used interchangeably.
The term balance eccentricity, or permissible residual specific unbalance as detailed in ISO 1940 (see References),
is defined by the amount of unbalance present divided by the mass of the rotor or component. When this term is
used, it can be physically related to the runout of a previously balanced component as follows:

U TIR
e= =
M 2

𝑈 = 𝑈𝑛𝑏𝑎𝑙𝑎𝑛𝑐𝑒 𝑜𝑓 𝑡ℎ𝑒 𝑝𝑎𝑟𝑡

𝑀 = 𝑀𝑎𝑠𝑠 𝑜𝑓 𝑐𝑜𝑚𝑝𝑜𝑛𝑒𝑛𝑡
𝑒 = 𝑏𝑎𝑙𝑎𝑛𝑐𝑒 𝑒𝑐𝑐𝑒𝑛𝑡𝑟𝑖𝑐𝑖𝑡𝑦
𝑇𝐼𝑅 = 𝑇𝑜𝑡𝑎𝑙 𝐼𝑛𝑑𝑖𝑐𝑎𝑡𝑒𝑑 𝑅𝑢𝑛𝑜𝑢𝑡
In more detail, the balance of a component can be defined either by the mass of the component and the amount of
unbalance (M and U), or by the mass and the eccentricity (M and e). This distinction is helpful for the purpose of
understanding and evaluating the significance of how components are attached to a shaft. In particular, API 617
defines that individual components should be individually dynamically balanced before assembly to ISO 1940 G1
levels. For a component that is installed on a 5000 RPM shaft, the allowable eccentricity (e) at G1 is 0.0019 mm
(0.000075”), or 0.15 mils TIR. Consistently maintaining less than 0.15 mils TIR for mounted components on most
shafts for industrial machinery will be impractical, making the mounting process a likely larger contributor to the
resulting unbalance of the mounted component than the balance condition of the part itself. This very issue is the
driving factor behind the use of the incremental balance procedure defined in ISO 11342 and generally specified by
API standards (i.e. API 617 and others).
ISO 1940 defines the balance quality of rotors for a variety of services by defining a host of balance quality grades
for different types of rotors. The residual unbalance for a rotor is defined in this standard using velocity magnitudes
starting at 0.4 mm/sec and increasing in factors of 2.5 (0.4, 1.0, 2.5, 6.3, etc.). The velocity magnitude is defined
using the eccentricity concept from above combined with operating speed of the rotor as follows:

𝐺 = 𝑒 × ω , 𝑚𝑚/ sec 𝑝𝑘
𝐺 = 𝐵𝑎𝑙𝑎𝑛𝑐𝑒 𝐺𝑟𝑎𝑑𝑒 (𝑖. 𝑒. 𝐺1 = 1 𝑚𝑚/ sec 𝑝𝑘)
ω = 𝑆ℎ𝑎𝑓𝑡 𝑠𝑝𝑒𝑒𝑑, 𝑟𝑎𝑑/ sec ¿
2 πRPM
60

This definition specifies the eccentricity that would result in the rotor vibrating at 1 mm/sec assuming there is no
dynamic amplification due to natural frequencies and the rotor is operating well above a critical speed. In reality, the
balance grade is much less related to observed vibration on operating machinery due to various natural frequencies.
API standards, such as API 617, specify unbalance tolerances generally by specifying U as follows:
 𝑈 = 6350(𝑊⁄𝑁), 𝑔 − 𝑚𝑚 (𝑆𝐼
𝑢𝑛𝑖𝑡𝑠)
𝑊
𝑈 = 4( ⁄𝑁), 𝑂𝑧 − 𝑖𝑛 (𝑈𝑆
𝑐𝑢𝑠𝑡𝑜𝑚𝑎𝑟𝑦 𝑢𝑛𝑖𝑡𝑠)
𝑊 = 𝐽𝑜𝑢𝑟𝑛𝑎𝑙 𝑠𝑡𝑎𝑡𝑖𝑐
𝑙𝑜𝑎𝑑, 𝑔 𝑜𝑟 Lb.
𝑁 = 𝑆ℎ𝑎𝑓𝑡 𝑠𝑝𝑒𝑒𝑑, 𝑅𝑃𝑀

It should be noted that for API standards U is specified referenced to the journal reaction force (due to static weight)
and is generally assumed to be one correction plane per bearing (two for most machines). For comparison, the
allowable unbalance for the rotor (commonly split between two planes) can be calculated from the equation above
using “W” as the total rotor weight.
Although the API standards generally specify the rotor unbalance using U, API 617 has a minimum limit on
eccentricity that is invoked for rotor speeds in excess of 25,000 RPM where the balance tolerance is limited at
250μm or 10μ inch. This limit is established in general by the capabilities of shop balance machines.

SHOP BALANCING METHODS

The methods employed in shop balancing can have a profound impact on the resulting balance condition of the
rotor. The impact of shop balance technique is most important when the rotor is relatively flexible and/or long as is
common with most turbomachinery.
To improve the balance condition of most high-speed flexible rotors, the following procedure is generally followed:
1. Balance the bare shaft without added components
a. Assure that any keyways are fitted with half keys in accordance with ISO 8821 unless two keys are
located at the same axial position and are 180⁰ apart
2. Balance the attached components separately to ISO 1940 grade G1 or better
a. Balance should be accomplished normally using shop mandrels or other balance hardware.
Mandrels should be precision balanced and have eccentricity <0.0001 in and a mass of <25% of
the component to be balanced.
b. Concentricity of mounting hardware such as mandrels during the shop balancing should be
adequate to prevent mounting eccentricity that can result in component balance error that
exceeds the intended balance tolerance
3. Mount no more than 2 components to the shaft at a time and re-check balance, and if corrections are
required only correct on the added components
4. Perform a check balance on the fully assembled rotor after the component assembly procedure above,
with final corrections normally on two correction planes near the ends of the rotor (near bearings)
a. Check balance process will normally allow limited balance correction, since major corrections
would have been completed during assembly
b. “Large” corrections to an assembled rotor pose a nontrivial risk of vibration above the first critical
speed due to modal balance, since large unbalance (that could be located away from balance correction
planes) can be corrected to minimize shop balance machine output to comply with the 4W/N rule at
the two planes defined but not correct the modal balance condition for higher flexible rotor modes
The motivation for following this incremental balance procedure is to minimize the unbalance of the rotor in
general, but to specifically reduce the modal unbalance that can result if this method is not followed. If the rotor is
fully assembled and balanced after being fully assembled (opposed to the incremental balance), unbalance of
components or more specifically the mounting eccentricity of the components can result in very large modal
unbalance even though a low speed balance machine may indicate that the rotor is successfully balanced.
API Residual Unbalance Verification
API Describes a method for verifying that the residual unbalance of a rotor falls within the standard 4W/N
tolerance. This test is conducted after final low-speed balance of a rotor assembly is completed in a shop. A trial
weight equal to twice the 4W/N tolerance for hard bearing balance machines (four times 4W/N for soft bearing
machines) is applied at six equally spaced angular positions at each balance planes with the rotor operated at the
balance speed between each application of the trial weight. The vibration amplitude vs. trial weight location is
plotted on polar graph paper. The plot should approximate a circle with the center of the circle indicating the
residual unbalance in the rotor. The API procedure, outlined in API 684 and other API standards, provides
additional details.

UNBALANCE DISTRIBUTIONFor most turbomachine rotors, the shaft is fabricated from a number of
parts (impellers, blade rows, etc.). As the rotor is assembled, an accumulation of unbalance vectors that
result from each component being added can produce an unbalance state that will excite different rotor
modes at different levels.
For a rigid rotor, the unbalance state can be fully defined by a static unbalance and one balance couple as shown in
Figure 2. Based on the rigid rotor theory, the static correction can be made at any location along the shaft although it
is generally preferred to correct the static near the center of the rotor. The dynamic correction is applied with equal
and opposite (180⁰ out of phase) balance correction at two separated planes. When a larger separation between the
two dynamic planes is used it results in a lower required dynamic correction weight.
This balance approach is generally acceptable for “rigid” rotors, or rotors that do not demonstrate critical speeds or
significant flexibility in operation. This is also the practical limit of most low speed shop balancing machines in that
they can correct for the static correction for the rotor and for dynamic couple in two planes.
When a rotor is balanced on a low speed shop balancing machine, the actual unbalance distribution is not known. If
the rotor was component balanced per normal API procedures, the modal unbalance will hopefully be under
reasonable control for most rotors. However, for rotors that are repaired or refurbished and sent to a shop balance,
the actual balance state is not known, and shop balancing may or may not improve the balance condition of the rotor
at normal operating speed.
One common practice during shop balancing is to make static corrections near the center of a rotor using the low
speed shop balance method by correcting for about 50% of the static near the center of the rotor. Final corrections
(remaining 50% of static and the required dynamic) are made at the balance correction planes on the end of the
rotor. Since the low speed balance machine cannot distinguish the actual axial position of the unbalance, the
selection of the amount of static to correct at the rotor center is based on experience and the intuition of the balance
machine operator. If the rotor is repetitively shop balanced using this method, it is possible that each balance
procedure will progressively move more static correction either to the center of the rotor or away from the center
depending on what the actual balance distribution is for the rotor. Therefore, making large balance corrections to
assembled rotors is a risky proposition.

Any actual rotor

unbalance distribution = Static unbalance
amplitude + Dynamic unbalance
amplitude
Figure 2 - Static and Dynamic Balance Concept

MODAL UNBALANCE CONCEPT

Since many turbomachinery rotors are in fact flexible from a rotordynamics and balancing perspective, it is helpful
to discuss balancing using a modal balance concept. This is important since rotors can generally only be balanced
using the rigid rotor method (static and one dynamic couple) using a shop balance machine. This becomes a
challenge since it is common for most turbomachinery rotors to operate above a first critical speed.
As rotors pass through critical speeds, the sensitivity of the rotor to unbalance depends on the deflected mode shape
at each shaft speed. For a first critical speed of a between-bearing machine (center of gravity between the bearings),
the rotor is generally most sensitive to unbalance near the center of the rotor, or unbalance on the ends that are in
phase. As the rotor approaches a 2nd critical speed, the rotor will then have a node point near the middle of the
rotor. Unbalance near a node point for any mode will not impact the vibration response of that mode.
The concept of modal unbalance is to take the actual unbalance in the rotor (or more commonly used when selecting
trial weight locations and phasing for field balancing) and separate it into balance magnitudes for each rotor mode
shape that may exist (i.e. 1st critical and 2nd critical speeds). For example, a rotor with a 1st and 2nd critical speed
will have rotor mode shapes as shown in Figure 3.
In general, an appropriate use of shop balancing with the incremental balance method can reduce the resulting
unbalance for both a 1st and 2nd critical speed for most rotors. However, once a rotor has been in service and/or
repaired, the use of a low speed balance method, although frequently successful, cannot assure that high modal
unbalance is prevented. This is a possible cause of elevated vibration after repair, and particularly likely when the
first critical speed vibration is acceptable and vibration increases dramatically above the first critical speed. Use of a
modal balancing concept is therefore essential for the balance of flexible rotors.
Field balance corrections, although intended to reduce vibration at all speeds due to unbalance, can often reduce the
response to one mode (i.e. 1st critical speed) while making another (i.e. 2nd critical speed) far worse, or vice versa.
Limited access to correction planes in the field can also complicate the process since ISO 11342 details that at least
N planes and usually N+2 planes are required for proper balance of a flexible rotor where N is the number of modes
either passing through or approaching. Since most turbomachine rotors operate above the first and approach the
2nd, as many as 4 balance correction planes may be warranted for proper balance at all shaft speeds.

2nd mode sensitive to

1st mode sensitive to weight out of phase on
weight in phase on ends and insensitive at
ends and at center center

Figure 3 – Modal Unbalance Distribution

It should be noted that high speed shop balance (“at speed balance”) is normally accomplished by
making balance corrections to more than 2 correction planes for the purpose of reducing unbalance
response at or approaching multiple critical speeds. Therefore, 4 shop correction planes for high speed
rotor balancing may be required for appropriate balance for the entire speed range. Additional information
on high speed rotor balancing may be found in the reference by Ehrich (1980).
FIELD BALANCING
Once a rotor is installed in the rotating machine, the unbalance condition cannot be directly determined. The only
way to assess the unbalance state is to add balance correction weights at various locations and using some
calculation tool such as the influence coefficient method, and if the vibration is greatly reduced at most (or all)
speeds, then a reasonable estimate of residual unbalance can be calculated based on observed vibration response to
multiple trial weights added to the shaft.
When this is done, the balance options include several different methods including:
 Single plane balancing
o Corrections are only added at a single correction plane (or equal weights are installed on both ends
of an impeller or rotor at the same phase angle)
 Two plane balancing
o Correction weights are added to two correction planes normally with one plane near each support
 bearing
 Multiplane and/or multispeed balancing
o Calculation technique where more than two planes and or multiple speeds are considered in the
calculations
The selection of the field balance technique is dependent on the dynamic characteristics of the rotor combined
with available balance correction planes and knowledge of the machine operating requirements.
SINGLE BALANCE METHODS
Graphical Method
The most basic method for rotor balancing is the single plane balance using a graphical approach. The biggest
advantages to being proficient in applying this method are
that it requires no specialized software, it can be
successfully applied to the majority of balance- of-plant
equipment, and it is the basis for and can be used as a
sanity check of more advanced multiplane methods or
software tools. The general approach is to collect reference
synchronous vibration amplitude and phase data, apply a
trial weight, collect response data, plot the initial and trial
vibration vectors, and determine a correction weight
graphically. The only tools needed for successfully
applying the single plane graphical balancing method are a
vibration transducer, a tachometer, a scale and some polar
plotting paper.
The machine to be balanced is operated at full or reduced
speed and the synchronous vibration amplitude and phase
are measured and recorded. This trial run vibration vector,
normally called the O vector (original), is plotted on polar
graph paper. It is helpful to define a scale on the graph
paper appropriate for the vibration amplitudes considered.
Since the graphical method relies on an accurate plot, the
larger the vectors on the plot, the easier it will be to determine
Figure 4 - Single Plane Balance Vector Diagram
the amplitude and angle for correction weights.
A trial weight is applied to the rotor and a new vibration vector is recorded from operation at the same speed as the
first run. This vector, called the O+T (original + trial) vector, is plotted on the polar graph paper. Now a line is
drawn from the tip of the O vector to the tip of the O+T vector. This vector will be called the T vector (very
creative!). The T vector is measured and then translated so that it starts from the center of the plot instead of the tip
of the O vector.
Since the T vector represents the change in synchronous vibration amplitude and phase caused by the application of
the trial weight, it may also be referred to as the Effect vector. Since the field balancing goal is to minimize
vibration, the plot can be inspected to determine the direction and angle needed to rotate the trial weight so that the
T or Effect vector is 180° opposed to the O vector. The amount of the trial weight is scaled up or down by the ratio
of the length of the O vector to the Effect vector. Appendix A contains additional discussion on phase conventions
for balancing that have been applied in this procedure.
As an example of single plane balancing using the graphical method, we will review balance information from a
large centrifugal induced draft fan in a coal fired power plant. We ran the fan to full speed and measure our
synchronous vibration amplitude and phase. Our analyzer showed us that the vibration is 5.6 mils pk-pk at an
angle of 135°. We then plotted this on our polar plot paper (shown in Figure 4) and labeled it as the O vector
using an appropriate scale. Our next step was to apply
a trial weight to the rotor and re-run the fan to get our trial vibration vector. In this case, we chose to install a 74 oz-
in weight at an angle of 315° from our phase reference mark as measured in the direction opposite rotation and mark
this location on our polar plot. Since we were installing our trial weight at exactly 180° from the high spot (our O
vector), we were assuming that this rotor ran well below the first critical speed and that our lag angle was 0°. In the
majority of cases (including this one) we will see this was a poor assumption. Once the machine was run up to the
same speed, we measured the synchronous vibration amplitude and phase to be 3.3 mils pk-pk at 238° and we
plotted this on our polar graph and labeled it as our O+T vector. We drew a line from the tip of our O vector to the
tip of the O+T vector and then translated this from the tip of the O vector to the origin on our plot giving us our T
vector. This represented the vector change of the vibration with the application of our trial weight. Since our goal in
balancing is to cancel out our original vibration (O vector) we want to shift the angle of our trial weight so that our
T vector is oriented 180° from our O vector or, in this example, we rotated our trial weight by 26° in the positive
direction (opposite rotation). We also want to make our T vector equal in length to our O vector. Since our T vector
was longer than our O vector in this example, we knew we need to reduce the amount of weight. We divided the
length of the O vector by the length of the T vector and multiplied by the amount of the trial weight to get 58 oz-in.
So, from our vector diagram we graphically determined our correction weight for this fan. Once the correction
weight was applied to the rotor, data was collected and evaluated against the vibration acceptance criterion for the
subject rotor. In this example, the measured vibration was acceptable and no further corrections were required. If the
vibration was still above acceptable levels, it may be necessary to apply an additional correction weight. This is
typically done by drawing a new vector diagram considering the T vector from the first trial weight as the original or
reference run and considering the first correction weight to be a trial weight. It is recommended that this be plotted
on a new polar plot so an appropriate scale can be used for the (hopefully) much lower vibration amplitudes. This
method is often referred to as “taking a new O”.

Influence Coefficient Method

Single plane balancing using the influence
coefficient method takes the graphical
method and applies math to get the same
results. Since the synchronous (1x) vibration is
described by a vector with magnitude and
phase, vector math is used to manipulate the
results. The general procedure for
measurement is identical to the graphical
method. The reference vibration is measured
and recorded (O vector). A trial weight
(TW) is applied to the rotor and the response
vector is measured (O+T vector). Vector
math is then used to calculate the T vector
and the influence coefficient. Using the
influence coefficient, a correction weight is
calculated and applied to the rotor.
Looking at the graphical example, the O
vector was 5.6 mils pk-pk at 135° and our
O+T vector was 3.3 mils pk-pk at 238° after
applying a 74 oz-in trial weight at 315°. To
calculate the T vector, we subtract the O
vector from the O+T vector. This is done by
converting the vectors from polar
coordinates (magnitude and phase) to
Cartesian coordinates (real and imaginary),
subtracting the components, and then
converting them back to polar coordinates
(magnitude and phase). An influence
coefficient vector is calculated from the trial
weight and the T vector. The influence
coefficient is a system property that
describes how a rotor reacts to a balance
correction weight. Multiplying the O vector by the influence coefficient vector reveals to heavy spot on the rotor. In
this example, balance corrections are being made by adding weights to the rotor so the correction weight is
calculated by adding 180° to the heavy spot vector to determine to correction location for the correction weight. The
influence coefficient method gives the same correction weight location as the graphical method, as expected.

PRACTICAL SINGLE PLANE CONSIDERATIONS

Trial Weight Selection
One of the keys to a successful balance job is selecting an appropriate trial weight in both magnitude and location.
An appropriate trial weight selection means that the effect of the trial weight provides at least a 10% vector shift
from the reference vibration. In the polar plotting method, the T vector should be at least 10% of the length of the O
vector. This will provide sufficient resolution on the polar plot to accurately calculate the influence coefficient. It is
not a requirement that the trial weight result in a reduction in vibration amplitude.
To select an appropriate trial weight magnitude, several methods have been suggested by others. Jackson (1991)
recommends that the trial weight magnitude be selected to produce a dynamic force equal to 10% of the rotor static
journal weight or:
56.33∗W
TW = 2
oz−¿
N
Other trial weight magnitude selection methods include using residual balance limits from API standards or ISO
1940. The typical residual unbalance limit per API standards is:
4∗W
U= oz−¿
N

ISO 1940 residual unbalance limits are specified using:

6.015∗G∗W
U= oz −¿
N
The ISO 1940 G level varies depending on the class of the machine to be balanced. Common levels are G0.67
(equivalent to the API limit), G2.5 (typical for large motors), and G6.3 (typical for pumps and fans). At low speeds,
the 10% static weight method results in a much larger trial weight magnitude.
At moderate shaft speeds, the 10% rule becomes more conservative and is a good rule of thumb to follow for trial
weight selection.

Figure 5 - Comparison of Trial Weight Selection Methods vs. Rotor Speed. API and ISO 1940 and 10% Rule

The angle of the trial weight should also be determined with care. As shown in Figure 6, at low speeds well below
the first critical speed, the vibration response (high spot) will be in phase with the unbalance vector (heavy spot) and
the lag angle will be zero. As the speed approaches a rotor resonance, the high spot will lag the heavy spot by some
angle and at resonance the lag angle will be 90°. Rotors that can be balanced using single plane methods will
operate below their first critical speed will have a lag angle between 0-90°. The rate of change of the lag angle as
the critical speed is traversed has an inverse relationship with the amount of system damping. A sample
synchronous response amplitude and phase plot (Bode plot), shown in Figure 6, shows this lag angle as the critical
speed is traversed. Selecting a moderate lag angle of 45-60° for a trial weight for a rotor operating below the first
critical speed will generally result in a reduction in vibration amplitude as long as the trial weight magnitude is not
excessive.

Angle Corrections
Locating the heavy spot is the key to balancing. There are some practical issues that must be accounted for to find
this correctly. The first thing is the lag angle that has already been discussed. The sensor angle must also be
considered. This is the physical angle between where the vibration sensor is mounted and where the phase reference
is measuring (for a laser tachometer, this is where the red dot shows up on the shaft). Per convention, this angle is
positive in the direction opposite rotation. An example is shown in Figure 7. Additional discussion on
recommended phase angle conventions is included in Appendix A of this tutorial.
 Figure 7 - Sensor Angle Diagram

A word of caution here on what we are calling sensor angle is appropriate. Many resources available for balancing
procedures define something called “sensor lag” that must be accounted for. This refers to the lag introduced by a
sensor such as a velocity pickup with a spring and mass inside it and does not really refer to the physical angle
where the sensor is mounted relative to your phase sensor. Since it is fairly uncommon to find a reason to use these
for balancing these days, it is not relevant to include any additional consideration for this lag.
The integration angle must also be considered. Many times,
balancing will be done using displacement measurements in units of
mils pk-pk. Other times, it may make sense to use velocity
measurements in units of in/sec-pk (or IPS-pk) or even G’s-pk. This
may make sense when the acceptance criterion is in units of velocity
or acceleration. It must be noted though that the lag angle previously
discussed is the angle between the high spot and the heavy spot and
the high spot is defined in terms of displacement. All this means in
practice is that you have to consider what we’ll call the integration
angle. For displacement measurements this is 0°, for velocity
measurements this is 90° and for acceleration measurements this is
180°. Figure 6 - Vibration Amplitude and Phase Lag

Now that the terms have been defined, a standard equation can used to determine where the heavy spot is from a
measurement. The formula for calculating the angle of the heavy spot is:
(Heavy Spot) ° = (Vibration Phase) ° - (Lag Angle) ° + (Sensor Angle) ° + (Integration Angle) °

As long as the sign conventions are followed as defined above, this formula will always give the physical location
of the heavy spot in degrees relative to the phase reference mark on the shaft in the direction opposite shaft rotation.
As an example of applying this, consider a machine monitored via proximity probes. It is being balanced with both
these and the addition of magnet-mount accelerometers that are single-integrated to velocity (IPS-pk) and assume a
portable tachometer reference. The setup is shown in Figure 8.

Figure 8 - Measurement Setup for Balancing

Based on the diagram, it can be determined that the rotation direction is CW and the tachometer is oriented at 90°
Left. This means it is rotated 90° to the left of vertical when looking from driver to driven. The drive-end (DE) or
inboard (IB) and non-drive end (NDE) or outboard (OB) accelerometers are also located at 90° Left. There are also
two proximity probes oriented at 45° Left.
For this example, assume that this machine runs below the first critical speed. Based on the previous discussion,
assume a lag angle of 45°. For the proximity probes, the sensor angle would be 315° or -45° since they are oriented
45° away from our tachometer in the direction of rotation or 315° from the tachometer in the direction of positive
phase (opposite rotation). The integration angle for the proximity probes is 0° since they measure displacement. The
integration angle for our accelerometers is 90° since the have already been single integrated to get velocity readings
(IPS-pk).
Now assume that a reference set of data was taken where this machine was operated up to the normal operating
speed and 1X amplitude and phase data from all of the sensors has been recorded and are shown in Table 1.

Table 1 - Example Trial Measurement Data

Measurement Location 1X Amplitude 1X Phase
DE-Prox. Probe 2.74 mils p-p 227°
DE-Accelerometer 0.251 in/s-pk 91
NDE-Prox. Probe 2.59 mils p-p 234°
NDE-Accelerometer 0.223 in/s-pk 93

Now the rules discussed above can be applied to calculate the location of the heavy spot on the rotor using the
assumed lag angle. This is shown in Table 2.

Measurement Location Vibration

- Assumed Lag
+ Sensor Angle
Table 2 - Heavy Spot Calculations for Example Data
Phase Angle + Integration
Angle = Heavy Spot

DE-Prox. Probe 227° 45° -45° 0° 137°

DE-Accelerometer 91° 45° 0° 90° 136°
NDE-Prox. Probe 234° 45° -45° 0° 144°
NDE-Accelerometer 93° 45° 0° 90° 138°

The results in Table 2 show that all four of our measurement locations show that heavy spot is around 139°.
This means that the appropriate location for a trial weight added to the rotor 180° from where the heavy spot is with
the assumed lag angle of 45° would be 319°.
In practice, not all measurement locations will suggest that the heavy spot is at the same angle on the rotor. If the
rotor has just static imbalance (there is a heavy spot at the same angle on both ends of the rotor), all the calculated
heavy spot angles should be close to the same. If there is one outlier, verify that the sign conventions were applied
correctly and the measurement is good. If different angles for the heavy spot are calculated on opposite ends of the
machine, the unbalance is not pure static and it is more of a couple imbalance (there are heavy spots on both ends of
the machine that are at different angles on the rotor).
Several commercial tools are available for carrying out single plane balancing including free or low-cost
applications for portable devices. The authors have not tested any of the available applications and prefer to use in-
house software for single and multi-plane balancing. Readers are encouraged to use this tutorial as a guide for
developing balancing tools using spreadsheets or other suitable programming languages. At a minimum, readers
should validate any commercial tools using hand or spreadsheet calculations or by testing the tools by balancing a
rotor kit.
Runout Corrections
Any field balance effort using proximity probes for response measurement should include corrections for rotor
runout. The runout measured at low speed by proximity probes will include any residual rotor bow as well as any
eccentricity or non-circular surface and any electrical runout at the probe target area. While it is certainly possibly to
leave the rotor runout in the vibration data when completing a field or shop balance and getting acceptable
amplitudes at the proximity probes, the consequence is that the unbalance forces transmitted to the bearings will be
much higher than if the runout is subtracted. This will be obvious is bearing housing vibration is also measured
since the displacement measured by proximity probes may be minimized but the bearing housing vibration will
increase due to the transmitted force.
Most, if not all, vibration data acquisition systems will allow users to designate a particular sample as a slow-roll or
runout reference sample. The software saves the 1X amplitude and phase at this sample and allows users to correct
1X vibration amplitude and phase at other speeds by vector subtraction of the runout from the vibration. The slow-
roll or runout compensated data should be used in the balance calculations. The slow-roll reference data should be
selected at the lowest speed where the amplitude and phase is stable and well below any speed where the measured
vibration is increasing due to unbalance forces or amplified by the first critical speed.
SINGLE PLANE BALANCING USING MULTIPLE MEASUREMENT POINTS
The use of multiple measurement points described in the previous section shows that an analytical approach may be
required to best minimize the vibration at all points simultaneously. The graphical technique can be used, but it is
common that different points (i.e. vertical vs. horizontal) will have different reaction to a balance weight. Since the
different points will have different effects and initial vibration, each measurement point will individually calculate a
different correction weight. With more measurement points than balance planes, the calculation for correction
weight is over specified. That is, there is not necessarily a correction weight that can be calculated to drive all the
vibration to exactly zero. Instead a numerical method must be used to minimize the vibration.
The most common method used to calculated balance correction weights is the influence coefficient method. This
method generates an influence coefficient matrix using measured vibration data (amplitude and phase) as well as a
known weight change. Once the influence coefficients are known (vibration per unbalance), the residual vibration
can be minimized by selecting a balance weight.
The most common minimization routine uses the least squares numerical method.
Before the numerical process can be described, the concept of vibration and weight changes must be well
understood. With a single plane method using one sensor, the original vibration is measured (O), a trial weight is
added (TW), and the response with the trial weight is measured (O+T). The original vibration run will normally
have no trial weight, so the weight change equals the trial weight. The vibration change is called the trial response
(T), and is found by determining the amplitude change from O to O+T from a vector plot.
When this is shifted over to a numerical process, terms are defined for each of the values described above but by
using subscripts to help accommodate conversion to a computer program:

Point i - Measurement points will be used with i = 1,2,3,… based on the number of measurement points used (no
limit to the number of points)
Run j - Runs include the reference run (Run 1), as well as the trial run (Run 2), followed by any additional
sets of vibration data (Run 3, Run 4, ….)
Vi,j - Vibration measurement (amplitude and phase) measured at point i during run j
Wj - Weight vector (amplitude and phase) installed when Run j was recorded.
o Note that W1 should normally be zero (no weights added on
initial reference)
o Weight addition can become a complicated accounting process with resulting confusion between
trial/final weights added as well as trim weights. Recommendation is to document weights using clear
notes regarding weights that have been added and removed throughout the process to prevent
confusion.
ΔVi,j-k - Vibration vector change at point i between run j and run k.
 Vi,j-k = Vi,j - Vi,k (Using vector difference)
ΔWj-k - Weight vector change between Runs j and k.
 Wj-k = Wj – Wk (Using vector difference)

For a single balance plane, these definitions can be used to calculate the influence coefficients for each
measurement point. Two vibration runs are required (j and k) that have a known weight shift. All other influences
on the vibration are assumed to be unchanged.
Ci,j-k = Influence coefficient at measurement point i that is based on the vibration/weight shift from run j to run k.

Applying these definitions to a single measurement and single plane example illustrates the physical relevance of
the vectors.
Single Plane, Single Sensor Example

Initial weight W1 = 0 (no initial weight)

Initial vibration V1,1 = Original amplitude and phase (“O”) at point 1 for run 1
Trial weight W2 = Trial weight magnitude and phase (“TW”) installed for run 2
Trial run vibration V1,2 = Vibration with added weight (“O+T”) at point 1 for run 2

W2-1 = W2 – W1 = W2 (“TW”) is the weight change from run 1 to run 2

V1,2-1 = V1,2 – V1,1 (“T”) is the vibration change from run 1 to run 2 at point 1

Influence Coefficient = V1,2-1 / W2-1 = “T”/”TW” is the influence coefficient for point 1

Final Weight = FW (remove trial and add final weight)

The influence coefficient for each sensor can be determined by calculating the vector vibration change and the
change in the trial weight as shown above with the results configured into an influence coefficient vector as shown
below:

Where

Alternatively, the vibration and weight data can be stacked into vertical vectors and the influence coefficients can be
calculated using matrix operations.

LEAST SQUARES SOLUTION METHOD

The influence coefficients and initial vibration provide an overdetermined system when multiple vibration
measurements are used for balancing at each plane. That is, there are more known data points (vibration
measurements) than degrees of freedom (balance planes). Since the system is overdetermined, generally there is not
an exact solution. Instead of an "exact" solution, a best fit solution must be used. There are several criteria for
finding a best fit solution and they all have the objective of minimizing the residual vibration by placing a final
correction weight or weights.
A standard numerical method for solving overdetermined systems is to use Linear Least Squares. With Linear Least
Squares, the minimized value is the sum of each sensor's residual vibration squared. This minimization objective is
convenient for solving with linear algebra because it has a single unique solution. There is only one best fit solution
to minimize the objective given all the available data. Additional discussion on the application of the least squares
minimization technique as applied to rotor balancing may be found in the reference by Goodman (1964).
LEAST SQUARES WITH MULTIPLE BALANCE PLANES
Having multiple balance planes changes the procedure for calculating the Influence Coefficients, but it does not
affect the least squares method for determining the solution. With multiple balance planes, the influence coefficients
are placed in a matrix of size # channels x # balance planes, and must be calculated using linear algebra. The
vibration change and weight change data must be arranged and solved to provide the correct influence coefficient
matrix.
LEAST SQUARES WITH MULTIPLE SPEEDS
Vibration data at multiple speeds allows provides additional data points that can be used to calculate balance
correction weights. Essentially, vibration at an additional speed can be thought of as additional data at new sensors.
A simple way to configure the calculation is to stack the vibration data for each additional speed in to the same
vertical array.
WEIGHTED/SCALED LEAST SQUARES
In some scenarios it can be beneficial to give a preference to reducing the vibration at certain sensors compared to
other sensors. This can be accomplished by using what is traditionally called Weighted Least Squares. For this
paper, to avoid confusing the term Weighted with unbalance correction weight we will use the term Scaled Least
Squares. This calculation method gives a scale value for each vibration measurement. This allows for certain
measurements to be given more/less importance for determining the correction weight.
There are several examples where using Weighted/Scaled Least Squares may be beneficial. For example, if you are
balancing a rotor with bearings that have different clearances the acceptable vibration may be different at each
bearing. Also if you are balancing a machine using both proximity probes (shaft vibration) and accelerometers
(casing vibration), then the acceptable vibration for each sensor may be different. Another scenario is that the data
may show high error/uncertainty for measuring the 1x amplitude or phase for a particular sensor. If the analyst
would like to calculate the results ignoring a suspected problem channel, then the calculation can be done by
discounting that channel.
When the input data has known error or uncertainty, then the scale factors can be selected to minimize the influence
of the error. The best linear unbiased estimator uses scale factors that are equal to the inverse of the variance for the
measurement. When collecting the 1x amplitude and phase it is typical to measure many samples to ensure that the
amplitude and phase are statistically consistent. In doing this process, the measurement variance can also be
calculated and provides a direct measurement of whether certain channels should be discounted due to measurement
variance.
ALTERNATIVE OBJECTIVE FUNCTIONS
Using linear least squares has many benefits, but there are other possible numerical tools to calculate a balance
correction weight. It is possible to minimize any objective function by using Non-Linear Least Squares.
With the standard linear least squares method, the calculated balance correction weight may give a solution with a
high vibration at one location and very low vibration at all other sensors. Since the standard objective is the sum of
the squares, the method does not care whether a single channel is higher than the other channels. An alternative to
this approach would be to minimize the maximum vibration amplitude. With this objective function, the best fit
solution would be found to minimize the maximum measured amplitude. With the standard linear least squares
method, there is no penalty for adding additional weight. This is not an issue if a rotor is very out of balance.
However, if the error and noise in the vibration data has a significant influence compared to the residual vibration
then the "best fit" solution may try to add more and more weight to cancel out any noise in the measurement. This
phenomenon is only relevant when two or more balance planes are used. Measurement noise should be assessed by
the analyst in terms of phase stability and amplitude. An alternative to address this would be to use an objective
function that seeks to minimize both the residual vibration and the weight added or to set an objective for the least
squares minimization to set the maximum projected amplitude to some value less than the acceptable vibration
amplitude.
CASE STUDY 1: MULTISTAGE COMPRESSOR FIELD BALANCE
A seven-stage centrifugal compressor was overhauled including un-stacking the rotor and re-assembling. The
impellers were component balanced during the re-stacking procedure and the rotor assembly was shop balanced on
a balance stand prior to installation in the compressor case. During commissioning, overall vibration amplitudes
exceeded the acceptance limit of 1 mil pk-pk as measured on the four available proximity probes with the 1X
component exceeding 2 mils pk-pk on one of the probes. Field balancing was recommended to reduce amplitudes to
acceptable levels. The normal operating speed for the compressor is 8,610 RPM.
Measurement Selection and Reference Vibration Collection
The compressor was monitored by two proximity probes physically located at 45° left of vertical (Y Probes) and
45° right of vertical (X Probes) as viewed from the drive end of the compressor at each bearing housing. No
keyphasor probe was available so a temporary laser tachometer was located at the horizontal split-line on the left
side as viewed from the drive end of the compressor. A small piece of reflective tape was applied to an exposed

and the Y Probe sensor angle is 315° for both ends of the rotor.
shaft location to trigger the tachometer. Using the convention outlined above, the X Probe sensor angle is 225°

The rotor was operated up to full speed and data was collected from all four probes. From the data, it was clear that
the rotor operated above the first critical speed. It was also noted that there was a fair amount of runout based on the
1X vibration at relatively low speed (0.21-0.44 mils pk-pk).
Slow Roll Compensation
Since proximity probes were used for balancing, slow roll vibration vectors were recorded and all reference and
response 1X vibration amplitudes used on the balancing procedure were slow roll compensated. If uncompensated
displacement values were used in the balance process, the final uncompensated displacement values may be reduced
to amplitudes less than the rotor runout but the consequence is that the unbalance forces transmitted to the bearings
may be much greater. This will show up as higher vibration on the bearing housing from seismic measurements in
the final data compared to the reference seismic data. Using slow roll compensation for proximity probes, the
lowest theoretical final displacement data would be equal to the rotor runout and will result in no unbalance force
transmitted to the bearings. The lowest practical final displacement data will be something greater than the rotor
runout.

Trial Weight Additions and Response Vibration

The reference data showed two key pieces of information. First, the vibration was higher on the outboard end of the
rotor. Second, there was a significant phase difference across the rotor from outboard to inboard. This information
was valuable for guiding the selection of trial weight locations and also suggested that a two plane calculation
would be required. Since the highest amplitude was on the outboard end, the first trial weight was installed on this
end of the machine. The rotor had provisions for installing set screws into balance rings with 20 equally spaced
holes on either end of the machine. A trial weight of 6.8 gm was installed at 252° on the outboard end balance ring
for the first trial. The machine was returned to full speed and vibration response vectors were recorded.
Since a two plane balance procedure was being used, a second trial weight run was required with a trial weight
installed on the inboard end of the rotor. In the interest of time, the first trial weight on the outboard end of the rotor
was left in place for this run. This is normally an option in most two plane balancing programs. Since the reference
vibration phase data indicated that the couple unbalance was more significant than the static unbalance in the rotor
and the response to the first trial weight on the outboard end was favorable, the second trial weight of 6.8 gm was
installed at 72°, 180° from the first trial weight location. The 1X vibration data collected for the three runs is
summarized in Table 3. Note that the phase data in this table is not corrected using the sensor angles and must be
corrected prior to calculating influence coefficients.

Reference First Trial - OB Second Trial -

Vibration Weight OB/IB Weight
Amp Phase Amp Phase Amp Phase
OBX 0.97 310 0.80 324 0.89 312
OBY 2.20 213 2.00 210 2.00 200
IBX 0.84 102 1.05 92 0.95 84
IBY 0.58 3.5 0.53 331 0.33 315.1

Influence Coefficient Table 3 - Reference and Trial Weight Response Calculation

Vectors
The vibration data from the three runs along with the trial
weight vectors were used to calculate a set of influence coefficients using the procedure outlined above. The only
significant difference is that the first trial weight on the outboard end of the rotor was left in place for the second
trial run and this had to be accounted for in the influence coefficient calculation. If the first trial weight is removed,
the effect of the second trial weight is calculated from the vector difference of the second trial response and the
reference vibration vectors. If the first trial weight is left in place for the second trial run, the effect of the second
trial weight is calculated from the vector difference of the second trial response and the first trial response.
With four measurements and two balance planes, eight influence coefficients were calculated and are shown in
Table 4. Note that the units for the influence coefficient magnitude used were displacement per mass instead of the
typical displacement per unbalance.
Since the balance weights were installed in rings with a constant radius, there was no ability to vary the
balance weight radius so referencing the mass only was sufficient. Reviewing the influence coefficients, it
is clear that there is significant cross-coupling since the influence coefficients for the inboard
measurement response to outboard weight addition (and vice versa) are of similar magnitude to the
inboard measurement response to inboard weight addition.
 Outboard Inboard
mils pk-pk/gm degrees mils pk-pk/gm degrees
OBX 0.04 58.0 0.029 47.9
0
OBY 0.03 123.3 0.051 358.0
4
IBX 0.03 31.8 0.025 115.4
9
Table 4 - Calculated Influence Coefficients
IBY 0.04 311.5 0.034 57.0
Correction 6 Weight
Calculation
The aforementioned least squares minimization was used to calculate an optimal correction weight location from the
reference vibration data and the calculated influence coefficients. The results called for more weight on both ends of
the machine with relatively small changes to the weight locations relative to the trial weights. This shows that
appropriate trial weight locations were selected and that the trial weight magnitudes were conservative. Since the
locations for available holes for correction weights did not match the calculated correction weight locations and the
amount of weight in each hole was limited, the correction weights were installed in more than one hole with the
weight vectors added to match the calculated correction as close as possible. The trial weights were removed prior
to installing the correction weights.

Calculated Installed
gm degree gm degrees
s
OB Correction 17.5 230 18.6 233
IB Correction 30.3 0 28.2 0
Table 5 - Calculated and Installed Final Correction Weights

Vector diagrams for the reference vibration, trial responses, and final vibration response are shown in Figure 9.
Lessons Learned Figure 10 - Turbine Generator Balance/Measurement Planes
While the results of this field balance show that the procedure was successful, there were issues that could have
potentially derailed the job. During the measurement setup, the probes on one end of the machine were found to be
wired incorrectly. Fortunately, this was identified prior to making any weight corrections. If the probe orientation is
incorrect, the influence coefficient method will still result in the correct calculated balance correction angle.
However, this can cause problems when attempting to identify a trial weight location.
Another potential issue was identified with the balance rings on the rotor. The rings were identical in that they
both had 20 threaded holes and were numbered one through 20 with the numbers stamped into the balance rings.
However, numbering of the holes increased with rotation on one end and opposite rotation on the other end. Many
balance programs do not Figure
allow 9the
- Vector
user toDiagrams
specify for
thisTwo Plane Influence
convention. Coefficient Balance
This highlights the need to use care in noting
the rotation direction, phase convention, and hole numbering convention on both ends.

CASE STUDY 2: TURBINE GENERATOR FIELD BALANCE

The turbine generator set is a 5 MW back pressure turbine in a three bearing configuration. Each bearing is
equipped with two X-Y type proximity probes in the traditional API configuration with the balance/measurement
planes identified in Figure 10.
The turbine was initially unable to reach full speed of 3600 RPM due to elevated vibration primarily on the
generator end. Previous balance attempts by others included removing all the existing weights from the generator
and adding a single plane shot on the outboard end of the generator.
Excessive static and couple combination on the generator was the cause for excessive vibration.
Balance Method
The first shot included taking the weight added on the outboard end of the generator and distributing that between
the two ends of the generator based on the assumption that the added weight may have been approximately the
amount needed for static correction of the generator. A second generator shot was conducted to define the two plane
calculation for the generator using a calculation speed of 1500 RPM which was the speed slightly below the
generator 1st critical speed.
Once the generator 1st critical was reduced with the combined static/couple shot, the turbine could be run at full
speed of 3600 RPM.

Generator Balance Data

Generator balance data at 1500 RPM is as follows:
Location/Description Reference Trial 1 Trial 2 Trial 3
#1X 0.49@346° 0.54@350° 0.45@331° 0.59@357°
#1Y 0.37@108° 0.39@113° 0.33@93° 0.40@121°
#2X 0.72@105° 0.64@103° 1.42 @76° 0.03@122°
#2Y 0.45@224° 0.38@217° 0.85@193° 0.01@170°
#3X 0.47@180° 1.07 @176° 1.24 @110° 0.11@194°
#3Y 0.31@304° 0.71@297° 0.73@232° 0.10@332°
Installed weights
#2 End, grams 109@338° 386@345° 525@359° 532@279°
#3 End, grams 685@351° 386@345° 520@358° 672@12°

Once the Trial 3 run above was complete, the #2 and #3 bearings were well controlled with #3 below 1 mil through
3600 RPM. The #1 and #2 bearings reacted to the turbine critical speed near 2500-2800 RPM, so the next shot was
to reduce turbine vibration.
Turbine Balance Data
Turbine balance data at 2300 RPM is as follows:

Location/Description Trial 3 Trial 4 Trial 5

(turbine reference)
#1X 2.91@22° 1.32@348° 0.52@345°
#1Y 2.41@155° 0.83@108° 0.39@114°
#2X 1.53@44° 0.25@359° 0.26 @5°
#2Y 0.92@172° 0.15@132° 0.19@129°
 #3X 0.61@322° 0.46 @315° 0.27 @177°
#3Y 0.53@80° 0.35@51° 0.21@293°
Installed weights
#1 End, grams None 75@202° 98@184°
#2 End, grams None 75@202° 98@184°

Once the Trial 5 run above was complete, the highest vibration at 3600 RPM was at the #3 end with a maximum
amplitude of 1.09 mils. Vibration during the startup was never above 1.09 mil on any probe.

Lessons Learned
This case study describes an example where a combination of single plane and static-couple balance procedures
were required to correct the balance on a rotor. Additionally, balancing at reduced speeds may be required when
excessive vibration amplitudes at critical speeds restrict operation at full speed.

ONE SHOT BALANCING USING PREDICTED INFLUENCE COEFFICIENTS

All the previous discussion focused on methods to calculate influence coefficients from measured vibration data for
single or multi- plane balancing. The discussion showed that a reference run and trial weight runs for each balance
plane used are required to calculate influence coefficients. For large machinery with flexible rotors where multi-
plane balancing is required and multiple critical speeds are encountered below normal operating speed, the standard
influence coefficient method will require many runs to define the influence coefficient matrix. This becomes costly
when considering the lost production and resources required to start and stop machinery in many plants. This is
especially concerning when an incorrectly located or sized trial weight has an adverse effect on vibration or, even
worse, no effect. Knowledge of the rotor mode shapes or previous field balance data can be used to help guide the
location of trial weights and the trial weight sizing suggestions in this tutorial can help determine a trial weight
magnitude. Another approach that can save a very significant amount of time is to use a set of predicted influence
coefficients to determine a one shot balance correction.

Description of Method

This method requires the development of a rotordynamic model of the system. Unbalance response analysis is
included in a typical lateral rotordynamic analysis according to API. This analysis is normally used to predict the
location of critical speeds and the vibration amplitude at probe locations when the rotor model is subjected to
various theoretical unbalance distributions. However, this can also be used to analytically determine influence
coefficients. The procedure is as follows:
 Develop model of rotor-bearing system.
o Normal modelling procedures outlined in API 684 are appropriate.
o Include stations at each available balance plane and measurement location. Include substructure
(bearing housing) models where appropriate.
 Apply an unbalance (theoretical trial weight) at the first available balance plane. The size and location is
not important but should be documented.
 Calculate the synchronous vibration response at each probe location with the locations matching those
available in the field. Repeat the application of a theoretical trial weight at the second available balance
plane, removing the first theoretical trial weight and re-calculated the synchronous response. Repeat for all
available planes.
 Calculate influence coefficients from the vector of all theoretical trial weights and the calculated response
at each measurement location.
 Record actual synchronous vibration vectors from field measurements.
 Use a least squares minimization to calculate a one shot balance correction from field measured reference
vibration data and predicted influence coefficients.

Advantages
This procedure offers several advantages over a traditional multi-plane influence coefficient balance. The most
obvious is that production loss during the process of applying trial weights for calculating influence coefficients is
eliminated since this is done analytically from the rotor model.
A second significant advantage is that influence coefficients can easily be calculated for all measurement positions
and balance planes at any speed from the predicted response data. The output of the synchronous vibration response
calculation for each application of a theoretical trial weight is the predicted trial weight response vector at each
speed. The least squares minimization can be applied not only to the normal operating speed field-measured
reference data, but to a set of coast-down data from all measurement positions using the calculated influence
coefficients at all speeds. The weighting concepts outlined in this tutorial can also be applied to particular speeds (or
speed ranges) such as critical speeds or normal operating speed to optimize the correction weight move.

Challenges
There are several challenges to implementing this method. As they say, there is no free lunch. The first challenge is
developing an accurate model of the rotor bearing system. Fortunately, commercial rotordynamic software including
bearing codes are fairly advanced and following good practice in model development will often result in a
theoretical model that accurately reflects field data. For large, critical turbomachinery where such an effort is
warranted, structural dynamics are often significant and should be included in the model. Another analytical
challenge is correctly documenting phase conventions used by the selected rotordynamics software. In the same way
a single plane balance trial weight can be incorrectly located when sensor position or rotation direction is incorrectly
applied, the calculated one shot balance correction will be incorrect if the analyst does not correctly account for the
machines actual sensor positions and rotation direction in the model.
Challenges with field measurement of reference data can also impact the calculated one shot correction. As with any
balance job, incorrectly labeled/wired or non-functional probes will introduce error in the calculation. The reference
data collected from a machine coast-down will likely not be sampled at equally spaced speed intervals. When
calculating a correction weight from transient reference data, spline curve fitting the measured data can be used to
correct for this.
Machine behavior that does not match the analytical model will introduce error in the calculation. The amplitude
of response at critical speeds may be non-linear or may vary due to thermal effects or from a coast-down to a
startup. This can be addressed by weighting speed ranges and applying the least squares optimization technique to
the calculated influence coefficients with more than one set of reference data (hot and cold or coast-down and
startup).
While the challenges appear to be a deterrent to applying this method, they are really no different than more
traditional methods. Phase conventions must be correctly applied. Sensor orientation must be accounted for.
Thermal and non-linear effects may be encountered with a single plane balance. The only (non-trivial) difference
with this method is the development of an accurate rotordynamic model of the system.

Field Data

The proposed method has been applied to a six-bearing steam turbine generator. The results show that the method,
when correctly applied, gives excellent predictions of post- balancing response. In this particular application,
accessible balance planes in the field were limited so the technique was used to select a balance correction that
would reduce critical speed and operating speed vibration amplitudes using only the balance planes that were easily
accessible. Additional complexity was involved in this application since the vibration amplitude was significantly
different during startup and coast down.
Figure 11(left) shows and overlay of the measured vibration amplitude and phase vs. speed during both a startup
and coastdown as well as the predicted vibration amplitude and phase during startup and coastdown after the
application of balance corrections for one of the 12 proximity probes. Figure 11 (right) shows the measured
vibration amplitude and phase vs. speed after the application of the balance corrections. The predicted vs. measured
results show excellent agreement at all probes. Ideally, the results would have shown more reduction in the first

critical speed amplitude. However, further reduction

of the response
Figure of thisMeasured
11 - Original mode wasVibration
not possible with the accessible
Amplitude/Phase balance
during Coast downplanes in thewith
and Startup field withoutPost-Balancing
Predicted a significant
increase in theAmplitude/Phase
amplitude at operating speed.
vs. Speed (left) and Measured Post-Balancing Amplitude/Phase vs. Speed (right)

Method Summary
The one shot balance method using calculated influence coefficients can be used to reduce machine downtime when
compared to a traditional influence coefficient multiplane/multispeed balancing procedure. It is essentially a
combination of an influence coefficient and a modal balancing method with both the influence coefficients and
mode shapes being determined analytically. Like with other methods, the application of this method requires care in
development of the rotor model, careful documentation of phase conventions, and a detailed review of the
operational characteristics of the machine behavior to be successful. The time required for the analytical effort to
develop the model and predict influence coefficients is significant but is may be easily justified when compared to
the avoided lost production over other balancing methods.

SAFETY DISCUSSION

The discussion above about field balancing is a technical description of the methods and practices used to improve
the balance condition of rotation machinery in situ. Conducting the balancing process as described obviously
requires operation of large rotating machinery while varying the balance condition of the rotating assembly. This
process is not without risk. Therefore, it is prudent to describe some important items and concerns related to the
field balancing process.
First off, the whole proposition of field balancing implies (and virtually assures) that the machine is already
operating at vibration levels in excess of desired amplitudes. In some cases, the entire purpose of the balancing
effort is to prevent or prolong rotor repairs resulting from obvious rotor damage or other sources of unbalance such
as loss of rotating components, severe erosion or damage to the rotor, and in some unfortunate cases to mask the
cause of vibration such as misalignment by intentionally adding unbalance to the rotor to reduce the vibration (the
symptom) opposed to reducing the actual forces in the machine.
Operating machinery at elevated vibration amplitudes is inherently risky or else the vibration limits would not be
defined. Adding trial weights to rotors with elevated vibration can therefore be inherently riskier. Addition of the
trial weights is intended to be completed so that the correction weight is located opposite the actual unbalance, with
the purposeful intent of reducing the vibration after the trial weight is added.
In some cases, it is not possible to adequately predict the proper location of required trial weights, and if the weight
is added on the heavy side rather than to opposite that, the vibration may in fact increase. Causes for the improper
location of the trial weight include simple causes such as human error or complicated ones such as three
dimensional damped rotor modes resulting in non-planer mode shapes. Whatever the cause, even higher vibration
can result from addition of trial weights or even final weights.
Rotor response should be linear and predictable but often is not observed to be so. Once the rotor sensitivity to
balance is determined with influence coefficients or balance effects, weight calculation and placement is determined
by simple vector graphics or by computer calculation. However, it is possible for the expected influence to be faulty
such as can occur if too small of a trial weight is added or if the system is highly non-linear. When this occurs, it is
possible to add excessive weight so that resulting vibration can be much higher than expected.
With these considerations in mind, it is wise to consider and discuss safety around the machinery during the
process. The primary items to consider are to manage the risk by proper assessment and placement of the balance
correction weights applied and to limit personnel exposure to the machinery during the balancing process.
Controlling the risk for adding the weights can include the following:
 Verify that the initial trial weight will produce no more than 10% of journal reaction force due to the
unbalance
 Verify the balance lag angle with review of coastdown data (not always 100% effective but better than
nothing)
o Conduct a technical review of the weight type and mounting location to assure that the weight will
not damage the component it is attached to and will not come off
o Clamp on weights (C-clamp style) are normally adding to the inside of blades so that centrifugal
force pushes the weight against the blade and the locking set screw is there primarily to keep the
weight in position axially opposed to clamping on the outside where the set screw is the only
mechanism to keep the weight in position
o Weights added to holes in fan blades should only be added when the blades are designed to
accommodate the weights, and weights do not exceed the intended design. Excessive balance
weight can overstress blades and result in blade failure.
o Welded weights (common for large fans) require technical review of the amount of weld metal
used and the location of the added weldment. The force generated by the weight must be
restrained by the weld, making the design of the weld critical.
o Using engineered weights intended for the purpose of field balancing in engineered weight
mounting locations (drilled balance holes or dovetail type balance slots) is always preferred to
other methods
Limiting personnel exposure sounds simple, but there is an inherent draw for many people to approach machinery
under test (particularly to feel how “bad” it is) and also to stand around the machinery during the process and
especially during machine startup. It is always wise to use vibration sensors with long cables to allow the analyst to
position him or herself in a relatively safe location during the operation of the machine. In particular, the locations
depicted in Figure 12 include concerns for components flying off (such as trial weights), possible failure of
components (such as couplings), chemical exposure (seals), and electrical faults (such as arcing of terminal boxes).
All these items are at much higher risk during starting of large machinery, and at elevated risk due to the probability
of sequential starts for balance shots.
Vibration personnel are often called to assess high vibration amplitudes on a variety of processing equipment. In the
case of pumps and compressors, there will always be the possibility that that seals could be leaking or could occur
during the testing. Since the shaft is rotating where seals exit the casing, it is quite possible that product could be
leaking and be slung away from the shaft and not be a visible drip or spray. With compressors, gas leaks will at
times not be detectable without appropriate monitoring equipment.
In many cases where possible faults of larger, critical processing equipment and particularly during starting of high
horsepower equipment, it is common to have a number of people that choose to be in the area because of the
heightened concern for plant production impact, or simply because it appears to be an interesting experience. The
crowd seems to generally be larger when the risk is higher or the potential faults are more severe, which can easily
produce heightened excitement for the vibration analyst, and possible relaxation of concern for possible chemical
exposure.
Therefore, any pump or compressor that has known high vibration should be approached as though it has product
leaking from every seal. The thought process should be the same as used in defensive driving training where you are
always expecting the other guy to put you at risk. Some simple defensive strategies should include:

1. Always take note of the wind direction prior to approaching the machine, and approach from the upwind
side. In the case where the equipment is located on an elevated platform, climb stairs or a ladder cage on
the upwind side or as a minimum verify that egress can be made from the upwind side.
2. Take note of the nearest safety shower prior to
approaching the machine.
3. NEVER stand perpendicular to the shaft at a process
seal (see Figure 12).
4. If the machine becomes unstable or the noise/vibration becomes dramatically more intense while standing
next to it, stop vibration sampling and exit the area on the upwind side. As a minimum, go to stand on the
upwind end of the machine until transient events stop.
5. NEVER enter an area around a machine that has suspected faults and walk around to the side of the
machine where you have no route of escape from. In the event of failure, you should always have an
unobstructed route of escape.
6. Spend a minimum amount of time near the machine with the possible fault, and in cases where the risk of
chemical exposure is higher (lethal or explosive products), always mount sensors and use longer cables so
that the risk is reduced by simply placing your body further away from the source of the potential leak.
7. NEVER attempt to record vibration data on a pump that is leaking or slinging liquid due to a seal leak or
other damage unless contact with potentially hazardous materials can be prevented.
8. NEVER congregate or linger around machinery that may have mechanical damage or is being misapplied
(surging compressor, deadheaded pump, etc.). Take vibration readings and make visual observations in the
shortest time possible, then leave the area. If other personnel are in the area, recommend to them that non-
essential people leave the area.

Damage to machinery can easily occur during plant or machine startup/upset conditions. Some situations that
indicate higher risk would include:
 Any time a compressor is surging. Compressor surging can quickly cause heating and sudden failure of
aluminum impellers. In addition, suction and discharge piping will often experience significant movement
due to the high fluid loads associated with a compressor surge.
 Any time that relief devices are opened such as during flaring in plants with flammable materials. Use of
the startup and flare piping will usually produce high noise and vibration around the relief and dump
valves that feed the flare header. This noise and vibration can cause fatigue failures in rather short time.
 Operation of control valves with the actuator mostly closed and at high differential pressure. These
conditions can produce excessive vibration and pipe fatigue very quickly.
 Water hammer can occur when liquid systems are started or when check valves or other control valves are
quickly opened or closed. Water
Figurehammer events
12 - Locations to can easily
Avoid produce
During Testingmuch higher pressure surges than the
equipment is designed for, and can cause sudden leakage of flanges, distortion of pressure vessel
components, and high deflection of piping.
Vibration analysis is frequently required on machinery with known faults. In many cases the machine is kept on-
line, or is restarted for vibration analysis prior to tagging the unit out for repair. The vibration analyst will arrive on
the site and be expected to proceed directly to the machine for vibration testing followed by quick review to
determine if it should be shut down immediately or if it is believed to be safe to continue operating. When this
occurs, the risk of damage to machinery and injury to personnel is certainly higher than normal.
Many machinery areas are not equipped with adequate means of exit in the event of major equipment failure. In
some cases, the testing must be done from temporary scaffolds or from platforms with only one exit. These
situations are certainly ones that taking an extra minute to consider the exit options would be wise.
Some general recommendations should be considered when vibration is sampled on equipment with known faults:
1. Never stand next to drive couplings, or other locations where components would likely come out in the
event of failure.
2. If temporary test equipment is setup for extended monitoring, locate the equipment on the end of the
machine train, usually on the drive end.
3. Plan an escape route when approaching the machine.
4. Determine a threshold vibration level above which continued testing will not be performed. Discuss this
level with plant personnel prior to testing if necessary so that appropriate action can be quickly taken to
shut the machine off if the threshold values are exceeded.
5. Be prepared at all times to stop testing, move to a lower risk area, and possibly shut the machine down if
conditions change so that noise or vibration levels obviously increase.
6. NEVER stay around a machine that has known faults with increasing severity.
7. NEVER continue testing once the pre-determined safe vibration threshold has been identified to be
exceeded on any sample point.
8. NEVER continue operating a machine with an obvious mechanical fault such as loose hold down bolts,
coupling element progressing damage (rubber material falling under coupling), metal shavings or bolts
falling from the machine, etc.

CONCLUSIONS
In this tutorial, rotor balance definitions and balance tolerances were reviewed and general shop balancing
procedures were discussed. Unbalance distribution on a rotor was reviewed and the need for modal balancing
procedures for flexible rotors was identified. A thorough discussion of field balancing concepts and procedures was
presented with examples of applying single plane graphical and influence coefficient balancing. A discussion of
trial weight magnitude and location was presented. A detailed discussion of accounting for the various sensor and
integration angles encountered in field balancing was presented with recommended conventions for successfully
documenting a measurement setup and applying rules for trial weight locations. The single plane balancing
concepts were extended to single plane balancing using multiple measurements. The least squares minimization
procedure was discussed for multiple measurement locations, multiple balance planes, and multiple speeds.
Concepts for least squares minimization using weighting (or scaling) and other optimization techniques was
discussed. The tutorial includes case studies illustrating the application of various balance techniques including
relevant data for the reader to replicate the balancing procedure. A method for one shot balancing at multiple planes
and multiple speeds using calculated influence coefficients was presented.
e = Eccentricity, in
Fbalance = Force cause by imbalance, lb
IPS = Vibration velocity, in/sec-pk
NOMENCLATURE
M = Rotor mass
m = Unbalance mass
N = Shaft speed, rot/min
T = Trial vibration effect vector
TIR = Total indicated runout, mil pk-pk
TW = Trial weight, oz-in
U = Residual unbalance, oz-in
V = Vibration vector
W = Journal static load, lb

𝜔
W = Weight vector
= Angular Velocity, rad/s
C = Influence coefficient
DE = Drive end
FW = Final correction weight
G = Balance Grade
IB = Inboard
NDE = Non-drive end
O = Original vibration vector
OB = Outboard

APPENDIX A. PHASE ANGLE CONVENTION

Background
The selection and use of a phase angle convention impacts the polar plotting methods used with balancing and can
pose challenging issues with locating the trial weights at optimal locations. In particular, the methods used to
determine the lag angle and vibration response direction (phase) as compared to measured amplitudes and phases
can be extremely confusing. This appendix is intended to describe and define a phase angle convention that can be
used consistently to determine proper locations for balance trial weights in most situations.

Phase Angle Documentation

Phase angles are documented by most vibration measurement devices using phase lag angles. If the equipment you
are using report phase lead vs. phase lag, the descriptions below do not apply. The phase lag angle defines the
relative position of the peak of a vibration signal relative to a timing reference or tachometer as shown in Figure A.
1 with phase normally reported units of degrees:

Figure A. 1 - Phase Determination

As shown, the phase is defined by the peak of the signal in a time waveform relative to the tachometer pulse with
zero degrees being at the tachometer signal, with phase angle increasing as time passes. For the example above, the
phase angle is about 55°.

Use of Vectors
Balance data is normally documented and plotted using vectors on a polar plot. The vector magnitude includes and
amplitude and phase at 1xRPM that is measured and reported with appropriate instrumentation. Magnitudes will
typically be documented with values such as 1.3 mils pk-pk @ 43°. The vibration measurement units will depend on
the job setup. Phase angles are always phase lag as stated above.
The vector amplitudes can be plotted on a polar plot such as the plot below (Figure A. 2). This plot uses a radial
distance from the center of the plot for vibration amplitude, and rotational position based on the phase associated
with the vibration. The vector shown in the plot is 0.9 mils pk-pk @ 135°.

Figure A. 2 - Polar Plot Example

Using the polar plot approach and vectors to represent the vibration amplitudes, the data gathered can be used to
determine the location for trial weights when properly documented.
Some details of the polar plot make the data display useful including:
 Shaft rotation direction should always be shown on the plot
 The polar plot should be drawn with angles increasing in the opposite direction to shaft rotation
 0° on the plot should be drawn in line with the tachometer or trigger location
The plotting of vibration vectors for balancing will typically include X-Y orthogonal probes (i.e. shaft probes for a
bearing) for each bearing location. Proper plotting of the vector amplitudes on the polar plot will help with proper
location of the initial trial weight position.

Trial Weight Location

The trial weight location is desired to be 180° from the “heavy spot” so that the vibration vector shifts opposite to
the existing vibration, and with an amplitude that results in near zero vibration. The actual relationship between
vibration response and weight placement is normally not known, but can be estimated based on knowledge of the
machine operating conditions.
The vibration response will always lag behind the location of the unbalance (the “heavy spot”) by a phase angle
between 0 and 180° when the vibration response is measured near the axial location of the unbalance. The actual
phase lag angle is determined by the relative location of the shaft critical speed to the operating speed. Operation
well below the critical speed results in near 0° phase lag. Operation at resonance results in 90° lag. Operation well
above the critical speed results in near 180° lag.
The vibration response phase can then be used to estimate the angular location of the unbalance at least within 90°
by estimating the phase lag angle and locating the trial weight relative to a properly plotted vibration vector.

Angle Plotting
There are several angles that should be defined for the proper plotting of vectors as follows:
 Vibration measurement angle – this is the phase angle in the measured vibration data
o If displacement units are used for this measurement, the phase angle is the “high spot”, or the
angular
o If velocity or acceleration units are used for the measurement, the measured phase angle
with have an integration offset described below
 “High Spot” – the phase angle where the shaft or component has its maximum movement in the direction
of the sensor = Measured Vibration phase + Integration Angle
o High spot = measurement angle when using displacement High spot = measurement angle + 90°
when using velocity
o High spot = measurement angle + 180° when using acceleration
 Corrected phase angle = High Spot + sensor angle
 Sensor angle = angle difference between the sensor and the tachometer location
The proper method for plotting the vectors on the plot is to use the corrected phase angle defined above with the
phase angle scale based on 0° being located at the tachometer position. This is functionally identical to plotting the
vibration vectors with the measured phase angle plotted as the angle from the probe location opposed to the angle
relative to the tachometer.
When this convention is used, an X-Y probe pair will have overlaying vectors on the polar plot when the probes are
located 90° apart and the vibration response shows a 90° phase angle difference as is typical with a balance
response. The method also provides a direct way to apply the phase lag angle for proper location of the first trial
weight.
Phase Angle Plotting Example
The example below shows how the vector plotting is executed for a typical balance problem. The original vibration
data and configuration are as follows and plotted in Figure A. 3:
 Shaft rotation direction is clockwise when viewing from end of driver towards driven
 Tachometer location is on the left a 90° from the top (9:00 clock position)
o Tachometer angle = 0°
 X probe is on the left at 45° (10:30 clock position)
o X Probe sensor angle = 315°
 Y probe is on the right at 45° (1:30 clock position)
o Y Probe sensor angle = 225°
 Probes are proximity probes (displacement type so integration angle = 0°)
 X probe vibration is 1.3 mils pk-pk at 47°
 Y probe vibration is 2.1 mils pk-pk at 125°

Figure A. 3 - Probe Angle Conventions

Using the method described above, the corrected angles used for plotting are as follows:
 X probe = 47° + 315° = 2°
 Y probe = 125° + 225° = 350°
Based on an estimated phase lag angle of 120° (operating above first critical speed), the heavy spot leads the
corrected phase angles above by about 120°. That puts the heavy spot near 116°. The proper location of the first trial
weight would then be 180° from there, or 296° relative to the tachometer location.

REFERENCES
API Standard 617, 2009, “Axial and Centrifugal Compressors and Expander-compressors for Petroleum, Chemical
and Gas Industry Services,” American Petroleum Institute, Washington, D.C.

API Standard 684, 2005, “Tutorial on the API Standard Paragraphs Covering Rotor Dynamics and Balancing: An
Introduction to Lateral Critical and Train Torsional Analysis and Rotor Balancing,” American Petroleum Institute,
Washington, D.C.

Ehrich, R., 1980, “High Speed Balance Procedure,” Proceedings of the Ninth Turbomachinery Symposium,
Turbomachinery Laboratory, Texas A&M University, Collection Station, Texas, pp. 25-31.

Eshleman, R.L., 2004, Basic Machinery Vibrations: An Introduction to Machine Testing, Analysis, and Monitoring,
VIPress, Inc.

Eshleman, R.L., 2004, Rotor Dynamics and Balancing, Vibration Institute.

Goodman, T. P., 1964, “A Least-Squares Method for Computing Balance Corrections,” J. Engrg. Indus., Trans.
ASME, pp 273-279.

ISO Standard 1940, 2003, “Balance Quality Requirements for Rotor in Constant (Rigid) State,” International
Organization for Standardization, Geneva, Switzerland.

ISO Standard 11342, 1998, “Mechanical Vibration – Methods and Criteria for the Mechanical Balancing of Flexible
Rotors,” International Organization for Standardization, Geneva, Switzerland.

ISO Standard 21940-32, 2012, “Mechanical Vibration – Rotor Balancing – Part 32: Shaft and Fitment Key
Convention,” International Organization for Standardization, Geneva, Switzerland.

Jackson, C., 1991, “Single Plane Balancing,” Proceedings of the Eighth International Pump Users Symposium,
Turbomachinery Laboratory, Texas A&M University, Collection Station, Texas, pp. 105-127.

Kelm, R.D., 2008, “Advanced Field Balancing,” Vibration Institute.

Vance, J. M., 1988, Rotordynamics of Turbomachinery, New York, New York: John Wiley and Sons.