MEASURING AND MODELING VERTICAL CROSSHEAD VIBRATION IN API-618 RECIPROCATING

COMPRESSORS

Brian Howard Charles Hatch

Signal Processing Engineer Principal Engineer
GE Bently Nevada GE Bently Nevada
Minden, NV, USA Minden, NV, USA

Gregory Yonker Kevin Philips

Development Engineer Reliability Engineer
Dresser-Rand BP West Coast Products
Painted Post, NY, USA Blaine, WA, USA

Brian F. Howard, PE, is a Senior Engineer Kevin Phillips, PE is a Rotating Equipment

at GE Oil & Gas in Minden, Nevada. In this Reliability Engineer at BP Cherry Point
role he focuses on diagnostic algorithm Refinery located in Blaine, Washington.
development and research. Prior to coming His core responsibilities include
to work with the technology team Brian was monitoring, evaluation, maintenance,
the Global Technical Leader for repair, selection, and upgrading rotating
Reciprocating Machinery Diagnostics equipment to maximize equipment
where he supported the product availability and performance. Kevin has
development process and provided diagnostic support to been a Reliability Engineer at BP for 15 years. Prior to that he
customers to solve problems with rotating and reciprocating worked for a local engineering contractor for the oil industry
machinery. Mr. Howard received a B.S. in Mechanical for 5 years. Mr. Phillips received a B.S. in Mechanical
Engineering from the University of Houston and an M.S. in Engineering from the University of Washington.
Electrical Engineering from the University of Washington. He
is a member of ASME and IEEE. ABSTRACT

Charlie is a Principal Engineer with GE Reciprocating compressors constructed to API-618

Oil & Gas, where he helps develop standards find application in refineries, petrochemical plants,
diagnostic solutions and new technology for and pipelines around the world. These compressors are
Bently Nevada products. He has extensive constructed with a force-fed lubricated crosshead that drives a
experience in rotordynamics modeling and double-acting piston via a piston rod. Looseness in the running
rotating machinery diagnostics. He has also gear (piston assembly, connecting rod assembly, and crosshead
developed diagnostic training materials and assembly) results in knocks (impulse events) that are usually
has instructed at many international detected at the crosshead guide with accelerometers and, in
training seminars. Charlie has authored a book on rotating some cases, with proximity probes measuring the piston rod
machinery diagnostics and is the author of many technical vibration at the pressure packing case. Crosshead guide
papers. He holds B.S. and M.S. degrees in Mechanical accelerometers have proven successful in detecting and
Engineering from the University of California at Berkeley. preventing catastrophic failure; however, the alarm set points
for the vibration levels have been derived heuristically because
Gregory Yonker has a B.S. in Mechanical no model exists that relates forces and impacts within the
Engineering from Pennsylvania State running gear to the vibration of the crosshead guide. This paper
University. He is a Research and presents a model of the crosshead system that estimates the
Development Engineer at Dresser-Rand’s vibration response of the crosshead and crosshead guide in
Painted Post, New York operations. His reaction to the forces and impulses in the running gear. The
work centers around the design and model incorporates the oil-film behavior including non-linear,
application of reciprocating  position-dependent stiffness and damping. The paper also
compressors.     examines the system response in the time domain and the
complex plane (eigenvalue migration). The numerical model
results are validated by comparison with data acquired on
operating machines in known conditions of distress. The results
show good agreement between the model and crosshead and
piston rod vibration data, both in the timing and amplitude of
vibration. Using the results of the model and historical data
collected from operating reciprocating compressors the paper
concludes with guidelines for filtering of crosshead
accelerometer signals and alarm thresholds.

INTRODUCTION

Reciprocating compressors provide compression service in

many applications throughout industry. This paper focuses on
the large, slow speed units found in refining and petrochemical
plants. These units are typically built to the American
Petroleum Institute (API) Standard 618 and include some level Figure 2 - Reciprocating compressor crank-slider nomenclature.
of condition monitoring instrumentation. Crosshead guide
acceleration and piston rod vibration are two commonly used
measurements for condition monitoring (Schultheis and
Howard (2000) and Schultheis et. Al. (2007)). Events in the
cylinder can result in impulse events in the crosshead guide
acceleration signal while events in the crosshead can change the
piston rod vibration signal at the pressure packing case. In
some cases, this cross coupling makes it difficult to determine
the source of an impulse event or vibration signal.

Figure 3 - Schematic of crosshead support system for downrunning

condition.

Figure 1. Sketch showing location of crosshead guide accelerometer

and proximity probes (“rod drop” or “rod position” probes) at the
pressure packing case.

Detection of malfunctions could be improved if the

influence of the crosshead could be quantified. Prior work
seems to have focused on pistons in internal combustion
engines (Wilson and Fawcett, 1974). In this paper a model is
developed for the crosshead motion as a function of rod load
and crank-slider kinematics. This model is combined with a Figure 4 - Schematic of crosshead support system for uprunning
non-linear system of springs and dampers to simulate the condition.
response of the oil film and crosshead guide. The output of the
model includes the system natural frequencies, crosshead on the connecting rod, which is also related to the measured
displacement, and crosshead guide velocity. This latter output cylinder pressures of the machine.
can be numerically differentiated and filtered to arrive at an When the vertical force, , is less than the static weight,
estimated crosshead guide acceleration signal which can be the crosshead accelerates towards the lower guide. Figure 3
compared to the measured crosshead guide acceleration signal shows the schematic for this condition. (It is assumed that the
to validate the model. piston rod provides sufficient stiffness to neglect the rotation of
the crosshead).
MODEL DEVELOPMENT When the vertical force exceeds the static weight the
crosshead accelerates towards the top guide. When the
Figure 2 shows the nomenclature for the crank-slider crosshead rises above the midpoint the model switches to the
mechanism in a typical API-618 process compressor. The schematic shown in Figure 4. The simulation switches between
crosshead lubrication results in an oil film between the the two models depending upon the location of the crosshead
crosshead shoes and the crosshead guide. The forces acting on within the guides.
the crosshead include gravity and the reaction force to the load The oil film has little influence on the crosshead motion

Copyright© 2015 by Turbomachinery Laboratory, Texas A&M Engineering Experiment Station

Figure 5 - Pressure transducer arrangement on typical API-618
cylinder.

until it nears either the upper or lower guide. In between neither

the upper or lower oil film provides much support. In addition,
the horizontal speed of the crosshead results in some variation
Figure 6 - Cylinder pressure, crosshead acceleration, and piston rod
in stiffness as the crosshead speed varies from maximum to vibration.
minimum throughout the stroke. These characteristics mean
that both the stiffness and damping values of the oil film
depend on the horizontal speed of the crosshead as well as
relative vertical displacement. Appendix A provides the details
on the system model, including the equations of motion for the
crosshead.

MODEL EVALUATION – COMPRESSOR FR66

Prior to model evaluation, data must be collected on the

reciprocating compressor. Typically API-618 compressors have
double-acting cylinders, meaning that compression occurs on
both faces of the piston. Pressure transducers have to be
installed on both ends of the cylinder, as shown in Figure 5. All
transducers (pressure transducers, crosshead guide
accelerometers, proximity transducers, etc.) are connected to an
on-line protection and monitoring system, allowing for
simultaneous data collection across all sensors. The monitoring
system stores the waveforms if there is an alarm, if a defined
amount of time has passed, or if the user requests it.
In addition to the measurements, the reciprocating masses, Figure 7 - Indicated pressure and rod load curves, full load and full
connecting rod weight, stroke, and other kinematic parameters pressure for compressor FR66, throw #4.
must be recorded from the compressor manual, API data sheets,
or measured in the field. Table C1 in Appendix C shows these event did change crank angle position with load suggesting a
values, along with the final non-linear stiffness and damping cause in the running gear, possibly in the crosshead transition
parameters. from one guide to another since the throw was uprunning.
With the data collection complete, the model evaluation Figure 7 shows the indicated pressure curves in the top
can begin. The motivation for the model began with an pane and the rod load curves in the bottom pane. The indicated
observation of an impulse event in the filtered crosshead pressure curves were exported from the condition monitoring
acceleration signal of a hydrogen make-up compressor system. The inertia load curve (red line) was generated using
(designated as FR66, throw #4). The frame rotates CCW and Equation B9, the gas load curve (blue line) was calculated by
throw #4 is on the right side, as looking driver to driven. This multiplying the piston area times the pressure for both ends,
throw is nominally an uprunning crosshead. Figure 6 shows a and the combined load curve (green line) was calculated using
sample of the waveform data. The impulse event of interest is Equation B10.
highlighted by a red rectangle and it appears in both the The combined rod load and running gear kinematics can be
unfiltered (gray) and filtered (purple) crosshead acceleration used to calculate the vertical force at the crosshead pin due to
signal. rod load using Equation B13. This is the black line in the top
The impact event had a high amplitude level, over 2g’s pk pane of Figure 8. The red line in the same pane represents the
in some cases, and did not align with combined rod load vertical force at the crosshead pin due to the connecting rod
crossings, gas load crossings, or valve open/close events. The inertia and is calculated using Equation B29. The bottom pane

Copyright© 2015 by Turbomachinery Laboratory, Texas A&M Engineering Experiment Station

Figure 8 - Vertical forces at the crosshead pin for compressor FR66, Figure 10 - Estimated crosshead guide vibration (top) and measured
throw #4. crosshead acceleration (bottom) for compressor FR66, throw #4.

Figure 9 - Crosshead displacement, crosshead guide displacement, Figure 11 - Calculated crosshead motion (top) and measured piston
crosshead velocity, and crosshead guide velocity outputs from the rod vibration in the vertical plane (bottom), full load and full
model (compressor FR66, throw #4). pressure (compressor FR66, throw #4).

shows the summation of these two forces calculated via transducer and monitoring system. This estimated crosshead
Equation B30. This curve in the bottom pane is the input force acceleration signal can be compared directly to the measured
to the model, . crosshead guide signal. The non-linear stiffness and damping
Figure 9 shows selected outputs from the model. The top parameters must be tuned so that the estimated crosshead guide
pane shows , the crosshead vertical position, as the black acceleration matches the measured crosshead guide signal at
line. The red line shows the upper crosshead guide position, . varying loads. Figure C1 shows the stiffness and damping
The bottom pane shows ′ and ′, the crosshead and curves from this tuning.
crosshead guide velocities. With the model parameters determined, the model output
The model output is compared to measurements made on (top pane of Figure 10) can be compared with the measured
the compressor. The crosshead guide accelerometer is closest to crosshead acceleration signal (bottom pane of Figure 10). The
the crosshead and is less influenced by events in the cylinder so large impulse event, at about 240° after top dead center, can be
it can serve as the primary validation measurement. The seen in the measured crosshead guide acceleration signal and is
estimated crosshead guide acceleration signal is calculated by highlighted by a blue box. The estimated crosshead guide
numerically differentiating the crosshead guide velocity, then acceleration signal shows a corresponding event at the same
bandpass filtering the signal to match the characteristics of the crank angle location, highlighted by a red box. The estimated
signal also shows two other low amplitude events (~0.5 g’s pk)

Copyright© 2015 by Turbomachinery Laboratory, Texas A&M Engineering Experiment Station

Figure 12 - Indicated pressure and rod load curves, reduced pressure Figure 13 - Calculated crosshead guide vibration (top) and measured
(compressor FR66, throw #4). crosshead acceleration (bottom), full load and reduced pressure
(compressor FR66, throw #4).
indicating minor impact events, not much higher than the noise
floor of the measured accelerometer signal.
The model also provides an estimate of the crosshead
position within the crosshead guide. The top pane of Figure 11
shows this position as a function of crank angle. At about 240°
the model shows the crosshead energetically contacting the
upper guide. This coincides with the timing of the impulse
event in the measured crosshead acceleration signal.
In addition to checking the crosshead acceleration, the
crosshead displacement can also be compared to the piston rod
vibration signals collected by the proximity probes at the
pressure packing case. Since these probes are farther away from
the crosshead (and influenced by the piston motion inside the
cylinder) the correlation between piston rod vibration and
crosshead motion will not be as close as between the estimated
and measured crosshead acceleration signal. The bottom pane
of Figure 11 shows this measured piston rod vibration. It has
large upward movement when the crosshead transitions from
the lower to upper guide. This is good evidence that the model
is accurately approximating the response of the crosshead
system.
In the preceding section the compressor was fully loaded. Figure 14 – Calculated crosshead motion (top) and measured piston
Next, results will be examined at full load, but lower discharge rod vibration (bottom), full load and reduced pressure (compressor
pressures. Figure 12 shows the cylinder pressure and rod load FR66, throw #4).
curves. In this operating mode the cylinder discharge pressure
estimated and measured signal indicates the model
is only 1700 PSIg [117 bar] compared to 1800 PSIg [124 bar]
approximates the physical system.
in the previous data set.
The calculated crosshead displacement, shown in the upper
The pressure change is not large in terms of percentage;
pane of Figure 14, still indicates energetic contact between the
however, it has an effect on the model output and the measured
crosshead and upper guide, but also shifted left by 3°. This
crosshead acceleration. Figure 13 shows the estimated and
correlation provides more evidence that the impulse event in
measured crosshead acceleration signal for this load. The first
the measured crosshead acceleration signal is related to the
small event in the estimated crosshead acceleration signal
crosshead contacting the upper guide. The estimated crosshead
appears similar in both Figure 13 and Figure 10. The second
position and the piston rod vibration, shown in the lower pane
event has nearly disappeared from the estimated crosshead
of Figure 14, appear similar. From 60° to 120° both signals
acceleration signal. The third, and largest, event has shifted left
show an upward movement. From 210° to 240° both signals
by 3° in both the estimated and measured crosshead
acceleration signals. This close tracking of events in both the

Copyright© 2015 by Turbomachinery Laboratory, Texas A&M Engineering Experiment Station

Figure 15 – Indicated pressure and rod load curves at 66.8% load Figure 16 - Estimated crosshead guide vibration (top) and measured
(compressor FR315, throw #2). crosshead acceleration (bottom), 66.8% load (compressor FR315,
throw #2.
show a similar pattern of movement as well, although the piston
rod vibration signal is lower in amplitude.
Similar results were obtained at other load steps on this
compressor. Based on the results of the model, the customer
decided to continue to operate the compressor. Over the next
few years, the vibration gradually increased rising to nearly 3
g’s pk. During a scheduled outage the crosshead to crosshead
guide clearance was checked and found to be at the outer limits.
The crosshead was re-shimmed back to recommended
clearances and the compressor placed back in service. The
clearance adjustment reduced the amplitude of the impulse
event back to the levels shown in Figure 6 confirming that the
impulse event results from the crosshead impacting the upper
crosshead guide.

MODEL EVALUATION – COMPRESSOR FR315

The next compressor frame (FR315, throw #2) has a

history of piston hop that makes the analysis more interesting.
In contrast to FR66, this compressor has a higher rod load
rating and runs slightly faster (277 RPM versus 257 RPM).
Throw #2 is downrunning. The design includes suction bottle Figure 17 - Calculated crosshead motion (top) and measured piston
supports connected at the upper crosshead guide, resulting in a rod vibration (bottom), 66.8% load (compressor FR315, throw #2).
much larger upper guide mass ( ) value than on the previous
machine. Appendix C has the specifics of the frame as well as blue box). The levels are low, but there seems to be reasonable
the non-linear stiffness and damping curves. correlation between the estimated and measured signal.
This compressor has eleven different load steps. The 66.8% Comparing the calculated crosshead motion to the
load step had many alarm events so this step was selected to be measured piston rod vibration signal, Figure 17, does not show
modelled. Figure 15 shows the indicated pressure and rod load good overall agreement. Given the reasonable agreement in the
curves for the machine loaded at 66.8%. Both ends are loaded; crosshead accelerometer signals this is unexpected. In fact, as
however some pockets in the head end have been opened. the crosshead transitions down at about 65° the piston rod
Figure 16 shows the estimated crosshead guide actually shows a small rise.
acceleration signal and the measured crosshead guide Although the piston motion does not correlate well with
acceleration signal. The estimated signal shows three events, the estimated crosshead position, the measured vibration
but only one is far enough above the noise floor to be of interest amplitudes are all at reasonable levels; however, seven seconds
(highlighted by a red box). There is a corresponding event in later the measured responses of both the crosshead
the measured crosshead acceleration signal (highlighted by a accelerometer and piston rod vibration probes have increased
significantly. The indicated pressures and rod loads, shown in

Copyright© 2015 by Turbomachinery Laboratory, Texas A&M Engineering Experiment Station

Figure 18 - Indicated pressure and rod load curves at 66.8% load Figure 20 - Calculated crosshead motion (top) and measured piston
(compressor FR315, throw #2). Data taken seven seconds after that rod vibration (bottom), 66.8% load (compressor FR315, throw #2).
shown in Figure 14.
Additionally, the calculated crosshead position (top pane of
Figure 20) shows little resemblance to the measured piston rod
vibration (bottom pane of Figure 20). In fact, the piston rod
vibration shows a high amplitude event beginning at about 65°
and peaking at about 90°. The large impulse event in the
measured crosshead acceleration signal at 115° coincides with a
sharp valley in the measured piston rod vibration signal, likely
when the crosshead contacted the lower crosshead guide.
The large piston rod vibration is caused by piston hop.
Liquids accumulate in the cylinder and, as the piston velocity
increases, the liquid forms a wedge that lifts the piston
assembly from the bottom of the cylinder. The lift and drag
forces are largest when the piston velocity is highest. For this
compressor the peak piston velocity can be found by
substituting the machine parameters into Equation B8, setting it
equal to zero, and solving for the crank angles. This results in
velocity peaks at 79° and 281° after TDC. These are the points
where the inertia curve (red line in the lower pane of Figure 18)
crosses zero. If the force created by piston hop is large enough
and rises fast enough, it can act as an impulsive force change to
the loading on the crosshead and alter the crosshead position
Figure 19 - Calculated crosshead guide vibration (top) and measured and the measured crosshead guide vibration. To understand
crosshead acceleration (bottom), 66.8% load (compressor FR315,
throw #2).
why piston hop can have such an influence on the crosshead
motion, the analysis has to shift from the time domain to the
Figure 18, appear virtually identical to that in Figure 15. That complex plane (eigenvalue analysis).
makes a change in load unlikely to be the cause of the vibration Appendix D provides an overview of system analysis in
change. the complex plane. The poles of the system are the roots of the
Figure 19 shows the estimated crosshead guide characteristic equation. For non-linear models the roots move,
acceleration and measured crosshead guide acceleration. The or migrate, in the complex plane. It is desirable to stay in the
estimated crosshead guide acceleration appears identical to the left hand plant and away from the imaginary axis (those
previous case (top pane of Figure 16); however the measured systems turn into oscillators) and to be close to the real axis to
crosshead guide acceleration differs greatly. In this case, there have a well-damped system response. Since the poles are
is a high amplitude event at 115°, highlighted by the blue box. complex conjugates only the positive roots are shown in Figure
This event does not appear to relate to the events in the 21. The plot only has labels for those eigenvalues with a non-
estimated crosshead guide acceleration, but does occur at the zero imaginary component. All other eigenvalues without
same time the piston rod vibration signal in Figure 20 shows labels fall on the real axis (overdamped).
high frequency vibration (also highlighted by a blue box).

Copyright© 2015 by Turbomachinery Laboratory, Texas A&M Engineering Experiment Station

 Figure 23 - Calculated crosshead motion (top) and measured piston
Figure 21 - Eigenvalue migration plot for the crosshead, 66.8%
rod vibration (bottom), 66.8% load (compressor FR315, throw #2).
load (compressor FR315, throw #2).

Figure 24 - Average eigenvalues in the 60° to 70° band as

compressor load changes (compressor FR315, throw #2).

crosshead system. The correlation in timing between the peak

Figure 22 – Crosshead damped natural frequency and piston damped natural frequency and the peak piston velocity suggests
velocity, 66.8% load (compressor FR315, throw #2). that as the lift and drag forces in the fluid wedge drive the
piston upwards, it also imparts an impulsive force to the
The eigenvalues migration plot shows that the system has
crosshead head system that results in vibration of the crosshead
at least 4 crank angle bands in which the system has an
system and piston assembly.
underdamped response. In the complex plane it is hard to
The model also indicates a high damped natural frequency
differentiate the different peaks. It would be helpful to correlate
near 225°. This point is not as close to the maximum piston
these results from the model with piston speed since piston hop
velocity; however, in some cases the condition monitoring
is most likely to occur when the velocity of the piston is
system has collected data with a second piston hop beginning at
highest. To provide this correlation the damped natural
this point. Figure 23 shows an example of this data. This data
frequencies (the projection of the eigenvalues on the imaginary
was acquired at the same load step as that Figure 20, but six
axis) are plotted along with piston velocity in Figure 22.
months later. There is the first piston hop event starting at 60°
The positive peak in the piston speed is very nearly at the
and a second one starting at 225°. This provides more
some point in the revolution at which the system response has a
correlating evidence suggesting the vibration induced by piston
large underdamped response. An impulse force to the crosshead
hop is influenced by the crosshead system response.
at this point in the revolution would result in vibration of the
Although the data has focused on the 66.8% load step, the

Copyright© 2015 by Turbomachinery Laboratory, Texas A&M Engineering Experiment Station

machine operates with multiple load steps. To simplify the
presentation, average eigenvalues will be calculated between
60° and 70° for each load step. Figure 24 shows the imaginary
part of these average eigenvalues. The model shows that load
steps in this range of 66.8% to 80.1% do have eigenvalues with
damped natural frequencies close to, or just above, the running
speed of the compressor. These are the load steps where it
would be expected to see piston hop. Although the model
predicts piston hop at 0%, in this load step no gas is flowing
through the compressor so no liquids can enter from process.
Cylinder lubrication could accumulate, but the plant rarely
operates the compressor unloaded for more than a few hours at
a time.
Review of the historical condition monitoring data
spanning twelve months of operation (~8,400 operating hours)
found many instances of piston hop on this throw. Table 1
summarizes the results. The first column shows the load step,
the second column shows the number of waveform samples that
indicated a piston hop event had occurred, and the third column
shows the number of operating hours at this load step. Since the
compressor spends different hours at each load step, the last Figure 25- Calculated crosshead motion (top) and measured piston
column shows a normalized value of events per thousand hours. rod vibration (bottom), 55% load (compressor FR315, throw #2).
Except for the 55% load step, the normalized event count
generally increases the closer the average eigenvalue in Figure The difference in motion is likely due to the vertical
24 is to the 1X running speed. loading on the crosshead at this step, shown in Figure 26. The
throw is located on the right side of the machine and the
Table 1- Summary of piston hop events on FR315, throw #2 crankshaft rotates CW, as viewed from the motor to the
compressor. As the crank web rotates below the throw
Normalized centerline, the connecting rod rotates about its center of mass
event count and exerts an upward force on the crosshead pin. At the same
Samples with Operating (events per time, pressure builds in the crank end chamber placing the
piston hop hours at this thousand piston rod in tension and exerting a downward force on the
Load Step event load hours)
crosshead pin. As shown in the top pane of Figure 26, these two
0.00% 0 8 0
forces nearly cancel each other out resulting in a very lightly
49.8% 9 497 18 loaded crosshead. This is why the crosshead vertical force in
55.0% 173 950 182 the bottom pane of Figure 26 is almost zero. With such a lightly
loaded crosshead the system has low stiffness and any forces
68.1% 2 187 11
acting on the piston assembly excite a larger response in the
66.8% 114 284 402 piston rod and crosshead. Looking at the bottom pane of Figure
30 108 277 25 it can be seen that the larger peaks occur as piston velocity
69.8%
reaches high values. This could be explained by liquid in the
71.5% 45 167 270
cylinder. Since the crosshead loading is low and the system
78.6% 27 1496 18 stiffness is low the interaction of the piston and liquid results in
35 1278 27 high piston rod vibration, but not because a resonant response is
80.1%
excited as may be the case with the other load steps.
90.6% 16 1820 9
100% 50 1579 32

It is worth taking a look at the data associated with the

55% load step. All but five of these events occurred in a single
six-hour span of time. Figure 25 shows a sample of the
estimated crosshead motion and measured piston rod vibration
at this load. The measured piston rod vibration does have high
vibration amplitude; however, it does not have the same shape
as the piston hop event captured in Figure 20. The motion is
more erratic and does not have the fast rise time and oscillatory
response seen in Figure 20.

Copyright© 2015 by Turbomachinery Laboratory, Texas A&M Engineering Experiment Station

Figure 26 - Vertical forces at the crosshead pin (compressor FR315,
throw #2). Figure 27 – Compressor FR66, throw #4 estimated upper crosshead
guide acceleration (top pane) and estimated lower crosshead guide
APPLICATION TO PROTECTION SYSTEM acceleration (bottom pane).
INSTALLATION AND CONFIGURATION
The model can also be used to validate configurations for
With accurate results in the time and complex domains, the machinery protection. Experience with the crosshead guide
model seems to do a reasonable job of approximating the acceleration signal has shown that bandpass filtering the signal
behavior of the crosshead and crosshead guide in response to with a highpass corner of 3 hertz and a lowpass corner of 2000
loads and impacts. One of the advantages of a model, over a hertz gives a signal that contains mechanical knock and impact
physical system, is that transducers can be placed anywhere on information, but excludes the valve and flow noise information.
the model. This can be helpful in understanding how to locate Figure 28 shows the spectral content from the unfiltered
the crosshead guide accelerometer. model output of Figure 13. The spectrum has most of the
Some OEMs and condition monitoring suppliers significant frequency content below 1000 hertz, although some
recommend that the crosshead guide accelerometer be mounted does extend up to 1500 hertz. The spectra were similar for the
to the upper guide for uprunning crossheads and to the lower other estimated crosshead guide acceleration signals suggesting
guide for downrunning crossheads where possible. One of the that the empirically derived filter settings are reasonable
interesting results of the model is that very few reciprocating starting points for detection of mechanical knocks and impacts.
compressor throws operate as either uprunning or downrunning
over all of the load steps and operating modes. That makes
selection of the proper guide more difficult.
Since the model allows the transducers to be placed
anywhere it is possible to plot the results with transducers
located on the upper guide and lower guide. Figure 27 shows
the estimated upper and lower crosshead guide acceleration
signals for FR66, throw #4. Even though this was the throw
with a large impulse event due to the crosshead contacting the
upper guide, the model shows the lower guide had a higher
amplitude response than did the upper guide. The difference
comes about because the model has more damping between the
guides than between the lower guide and the foundation so the
lower guide has a higher damped natural frequency. The
difference is not always so dramatic. In most of the cases the Figure 28 - Spectrum of estimated crosshead guide acceleration
upper and lower crosshead guide acceleration signals have (compressor FR66, throw #4).
similar amplitudes. Thess model results suggest that placement
of the accelerometer on either the upper or lower guide will The model also proves useful in assessing the amplitude of
provide good results, regardless of whether the crosshead is crosshead acceleration. From the preceding discussion it can be
nominally uprunning or downrunning. This finding needs to be seen that estimated crosshead guide acceleration values below
validated by checking the operating deflection shape (ODS) of 0.5 g’s pk typically do not show up in the measured crosshead
crosshead guides on running machines. guide acceleration signal. This suggests that the noise floor for

Copyright© 2015 by Turbomachinery Laboratory, Texas A&M Engineering Experiment Station

most API-618 machines is about 0.5 g’s pk. This can be tested Table 2 - Summary of estimated response from the model and
using historical data. measured response from the crosshead guide accelerometer.
One of the authors (Howard) maintains a fault database
that stores waveforms from a variety of condition monitoring Data Meas. Est. Meas.
Set Est. Vib. Vib. Location Location Reference
platforms, both portable and on-line. Each waveform is (g’s pk) (g’s pk) (°) (°) Figure
reviewed and assigned a malfunction label. The database also FR66,
includes waveforms from machines known to be in good Full 1.81 2.00 237 236 10
condition. Figure 29 shows a histogram of the peak value for Load
these waveforms. As was expected from the model, the center FR66,
of mass of the histogram is near 0.5 g’s pk. The historical data Low 1.75 1.25 234 233 13
implies that for the typical healthy machine 0.5 g’s is a Press.
reasonable noise floor for crosshead accelerometer event FR315,
detection on API-618 machines. No 0.53 0.41 70 71 16
Hop
FR315,
Piston 0.53 1.12 71 110 19
Hop

Application of this model to operating cylinders indicates

that it provides useful information about both the vibration
response at the crosshead guide as well as insight into
interaction between the events in the cylinder and crosshead
assembly. Future development and application of such models
should result in an improved understanding of physical
parameters (crosshead to crosshead guide clearance, oil flow,
oil pressure, etc.) that can be adjusted to reduce vibration
response and stress on reciprocating compressors.
Figure 29 - Histogram of filtered crosshead acceleration peak values
on API-618 machines in confirmed good condition. NOMENCLATURE

When the estimated crosshead acceleration amplitude r Stroke/2

levels exceed 1.0 to 1.5 g’s, the events amplitudes of the same l Connecting rod length
level can almost always be seen in the measured crosshead mcp Connecting rod crank pin mass
acceleration signal. This suggests that amplitudes above this mcr Connecting rod crosshead pin mass
level for the filtered crosshead acceleration signal are related to m Mass of crosshead assembly, rod
mechanical impacts in the running gear. The impact may be an mrec Reciprocating mass
indication of machine distress (loose piston) or an operating m2 Lower guide mass
characteristic of the machine (crosshead motion), but the root m3 Upper guide mass
cause needs to be identified when the level is this high. This Fy1 Vertical force from connecting rod
suggests that the alert thresholds in the range of about 1.5 g’s Fg Gravitational force
pk are reasonable starting points for machines that have no bp Instantaneous oil film damping
OEM provided thresholds and that have no operating history. bn Nominal oil film damping
These are a few examples of how the model and historical be Damping exponent
data can be used to validate settings on protection and condition bo Damping offset
monitoring systems. b2 Damping, lower guide to foundation
b3 Damping, upper guide to lower guide
CONCLUSIONS kl,ku Instantaneous oil film stiffness
kn Nominal stiffness exponent
This paper shows how a model was developed for the ke Stiffness exponent
crosshead and crosshead guide system of API-618 reciprocating ko Stiffness offset
compressors. So long as forces acting at the piston come from k2 Stiffness, lower guide to foundation
the gas and inertia loads the model does a reasonable job of k3 Stiffness, upper guide to lower guide
estimating the vibration response at the crosshead guide. ω Rotational speed, radians/second
Impulse forces acting on the piston assembly, such as piston
hop, can alter the crosshead guide vibration response. Table 2
summarizes these findings.

Copyright© 2015 by Turbomachinery Laboratory, Texas A&M Engineering Experiment Station

APPENDIX A – SYSTEM MODEL 0 0 0 1 0 0
0 0 0 0 1 0 0
The resultant vertical force on the crosshead arises from 0 0 0 0 0 1 0
the connecting rod load combined with the angle of the 0 0 0 0 0
connecting rod. For example, when the connecting rod is F
exactly parallel to the piston rod the resulting vertical force 0 q3 0 q4
must be zero. Likewise, when the connecting rod is not parallel 0
0
to the piston rod, there must be a resulting force in the vertical q5 q6
direction. Appendix B has more details on how this vertical
force, ,, is calculated. Eq. 2A
For the downrunning condition, the following equations Where:
can be derived (LaGrange notation is used for the derivatives
and, unless otherwise stated, the independent variable is time, q1
i.e. / ):

q2
'' ′ F
q3
'' ′ ′
′ ′ ′
q4
'' ′ ′

Similarly, for the uprunning condition the following q5

equations of motion can be derived:
q6
'' ′ F

'' ′ ′ ′ These equations were integrated numerically within

MATLAB using ODE45. The state-space matrix was also used
'' ′ ′ ′ to extract the eigenvalues.
′
Where stiffness and damping have the following non-linear
functions (where t=0 when the piston is at TDC):
0.5 2

These equations of motion are order-reduced and placed in
state-space form. Equation 1A shows the state-space form for
the downrunning case and Equation 2A shows the state-space
form for the uprunning case.

0 0 0 1 0 0
0 0 0 0 1 0 0
0 0 0 0 0 1 0
0 0 0 0 0
F
q1 q2
0
0
0 0

Eq. 1A

Copyright© 2015 by Turbomachinery Laboratory, Texas A&M Engineering Experiment Station

APPENDIX B – VERTICAL FORCES IN sin
RECIPROCATING COMPRESSORS asin Eq. B4

After some manipulation the distance from TDC to the

Vertical forces on the crosshead arise as a result of the
current piston position can be written as:
kinematics of the running gear, the inertia forces, and the gas
load forces. Figure B1 shows the nomenclature associated with sin
the running gear. 1 cos 1 1 Eq. B5

To find the expression for velocity, return to Equation B1

and differentiate each element with respect to time:
0 Eq. B6
Acceleration can be found by differentiating Equation B6
with respect to time:

2
Eq. B7
Figure B1 - Nomenclature used in kinematic analysis of reciprocating 0
compressor crank slider assembly.
After some simplification the acceleration can be
The kinematic analysis can be better understood if the expressed as:
crank-slider is reduced to a loop diagram in which vectors are
substituted for the essential dimensions of the actual cos 1 cos 2
mechanical components. Figure B2 shows the nomenclature cos ⁄
sin
associated with the loop diagram and the relationship between 1
the vectors and mechanical components.
Eq.B8

With the linear acceleration known, the scalar form of

Newton’s second law can be applied to find the reciprocating
inertia load:
′′ Eq. B9
Combining the inertia load with the gas load results in a
combined rod load expression of:
Figure B2 - Equivalent loop diagram of crank slider arrangement.
Eq. B10
This system has the following nomenclature: Since the connecting rod rotates at each end, the combined
r Crank radius (1/2 of the stroke) rod load results in a vertical force at the crosshead pin. Figure
θ=θ2 Crank angle (0° at TDC), equals ω*t B3 shows how these forces interact. In this figure the crosshead
ω Crankshaft speed in radians/second pull to the right (tensile loading) and the connecting rod reacts
l Connecting rod length upward and to the left. This connecting rod force, , can be
ϕ=θ3 Angle of the connecting rod resolved into a horizontal component (equal to and opposite of
d Crankshaft centerline to crosshead centerline the combined rod load force, ) and a vertical component.
θ4 Angle of vector d, fixed at 180°
xp Distance from top dead center (TDC)

Begin the derivation by writing the loop equation from

Figure B2 as:
0 Eq. B1

Since is fixed at 180°, Equation B1 can be simplified

and separated into real and imaginary parts:
Figure B3 - Forces at the crosshead pin (point B in Figure B1 and
cos cos 0 Eq. B2 B2).

0 Eq. B3 From Figure B3 it can be observed that if the horizontal

Equation B3 can be solved for the connecting rod angle, : forces are summed and set to zero, the connecting rod force,

Copyright© 2015 by Turbomachinery Laboratory, Texas A&M Engineering Experiment Station

 , can be expressed as a function of the connecting rod angle, end, , from Equation B17 and then solve for the mass at the
360° , and the combined rod load, : crankpin end, :
0 → cos 360° 0 Eq. B18

Equation B18 can be substituted directly into Equation B16

Eq. B11 to find the mass at the crosshead end, :
cos
The vertical force, , , can readily be expressed as: Eq. B19
, sin 360° Derivation of the connecting rod reaction begins with a
few assumptions to keep the analysis simple. These
, sin assumptions include:
cos 1. No reciprocating mass (Equation B10 accounts for
this)
, Eq. B12
2. Frictionless operation (only a vertical force is allowed
Since it is usually the crankshaft angle, , that is known, at point B, the crosshead pin)
the vertical force is usually written by combining Equations B4 3. The crank turns counter-clockwise.
and B9 as: Figure B5 shows a sketch of the crank-slider, along with
sin the associated nomenclature.
, tan asin
sin
,
Eq. B13
1 sin
The crosshead pin not only experiences vertical force due
to the combined rod load, but it also experiences a vertical
force as the connecting rod rotates about its center of mass.
Most API-618 style connecting rods can be thought of as a
slender rod joining a substantial bearing at each end. For this
arrangement the center of gravity, , can be approximately
located as:
1 Figure B5 - Schematic of crank-slider with connecting rod inertia.
Eq. B14
3
2 For this case of plane motion, the scalar form of Newton’s
Eq. B15 second law can be used to sum the forces and the moments
3
about the center of gravity, , of the connecting rod.

, ′′ Eq. B20

, , ′′ Eq. B21

, , ,
Eq. B22
′′
In Equation B22, indicates the rotational moment of
inertia about the center of mass, . The next step is to define
the path of travel for the connecting rod center of mass. It
travels in an elliptical path (Den Hartog, 1985) with the
following expressions for displacement and acceleration:
Figure B4 - Connecting rod center of gravity. Eq. B23
The crank pin end mass, , and the crosshead pin end
mass, , can be related to the center of gravity and the total Eq. B24
connecting rod weight, , by using Newton’s second law.
Eq. B25
Eq. B16

Eq. B26
0⇒ 0 Eq. B17

Use Equation B16 to eliminate the mass at the crosshead

Copyright© 2015 by Turbomachinery Laboratory, Texas A&M Engineering Experiment Station

 With three equations and three unknowns, each force can
readily be found.

, Eq. B27

tan tan
,
Eq. B28
Cos
tan tan
,
Eq. B29
Cos
The total vertical force at the crosshead pin can be found
by combining Equations B13 and B29:
F , , Eq. B30

Copyright© 2015 by Turbomachinery Laboratory, Texas A&M Engineering Experiment Station

APPENDIX C – MODEL PARAMETERS

Table C1 – Model parameters for reciprocating compressor FR66, Table C2 – Model parameters for reciprocating compressor FR315,
throw #4. throw #2.

Parameter Value Units Parameter Value Units

r 254 mm r 229 mm
l 1276 mm l 1219 mm
CE rod dia 102 mm CE rod dia 127 mm
HE rod dia 0 mm HE rod dia 0 mm
Bore dia 206 mm Bore dia 622 mm
Isent. exp. 1.37 - Isent. exp. 1.46 -
mcp 204 kg mcp 318 kg
mcr 45 kg mcr 114 kg
m 868 kg m 680 kg
mrec 887 kg mrec 1126 kg
m2 4000 kg m2 8000 kg
m3 4000 kg m3 24000 kg
Fg -9.8 N Fg -9.8 N
bn 1 N-s/m bn 1 N-s/m
be 70500 - be 70500 -
bo 75000 - bo 75000 -
b2 5.5E+5 N-s/m b2 1.5E+6 N-s/m
b3 1.5E+6 N-s/m b3 2.4E+7 N-s/m
kn 1 N/m kn 1 N/m
ke 8.6E+5 - ke 8.6E+5 -
ko 3.8E+6 - ko 3.8E+6 -
k2 4.8E+10 N/m k2 1.8E+11 N/m
k3 3.1E+10 N/m k3 1.5E+11 N/m
ω 26.9 rad/s ω 29.0 rad/s

Figure C1 - FR66 throw #4 nonlinear stiffness and damping curves. Figure C2 – FR315 throw #2 nonlinear stiffness and damping curves.

Copyright© 2015 by Turbomachinery Laboratory, Texas A&M Engineering Experiment Station

APPENDIX D – SYSTEM DYNAMICS IN THE
COMPLEX PLANE Eq. D2
2
The and terms are determined by the initial conditions
To show how the poles in the complex plane relate to
of the system. The denominator of Equation D2 is called the
mechanical behavior it is helpful to examine a second
characteristic equation. The roots of this equation describe the
order system, such as the mass-spring-damper system of
dynamic behavior of the system and can be written as:
Figure D1. With no external forces, other than gravity, the
mass is at rest. Now, someone comes along and pulls the 1
mass downward, as indicated by the outline in Figure D1,
and releases it. The spring will exert a force on the mass These roots determine the transient and steady state
pulling it back to the equilibrium position. As the mass response of the system. To understand how, it can be observed
travels, its velocity increases. In fact, at the point of that the roots have real and imaginary parts. These parts can be
equilibrium the velocity will be at a maximum and plotted in the complex plane, as shown in Figure D2.
momentum will carry the mass a distance of Δx above the
equilibrium, where it stops, and begins to travel downward.
It will again pass through the equilibrium point, but will
not quite reach the starting point because of energy
dissipated by the damper. The cycle with the mass travels a
little less each cycle until it eventually comes to rest.

Figure D2 – Eigenvalues of the mass-spring-damper system plotted in

the complex plane.

The roots, or eigenvalues, have a magnitude equal to the

system’s undamped natural frequency. Regardless of the
damping ratio, if mass and stiffness remain constant, the system
will always have the same undamped natural frequency. The
damping ratio describes the angle between the undamped
natural frequency and the real axis. The projection of the pole
onto the imaginary axis gives the damped natural frequency.

For this simple model, the eigenvalues occur as complex

Figure D1 - Mass-spring-damper system conjugate pairs so displaying both pairs does not add much
information to the plot. The convention when plotting the
For this system, the forces in the vertical direction can be eigenvalues is to show only the positive value. Figure D3
summed as: shows the eigenvalue positions in the complex plane along with
the system response. As the damping approaches zero, the poles
0⟹
move toward the imaginary axis. If the damping is zero, the
This can be written to normalize the coefficients by the poles will be on the imaginary axis and the system will behave
mass: as an oscillator. In contrast, as the damping ratio approaches
unity, the poles move toward the real axis. If the damping ratio
2 0 Eq. D1 is unity, then the poles sit on the imaginary axis and the system
is described as critically damped. This means that for free
vibration the system converges to the equilibrium position as
, fast as possible without oscillation.
2√
One way to solve this equation is to use the Laplace
transform. This transform allows Equation D1 to be written as a
polynomial (capital letters are used to designate the variables in
the Laplace transforms):

Copyright© 2015 by Turbomachinery Laboratory, Texas A&M Engineering Experiment Station

 REFERENCES

API Standard 618, 2007, “Reciprocating Compressors for

Petroleum, Chemical, and Gas Industry Services”, Fifth
Edition, American Petroleum Institute, Washington, D.C.

API Standard 670, 2014, “Machinery Protection Systems”,

Fifth Edition, American Petroleum Institute, Washington,
D.C.

Schultheis, S. M., and Howard, B. F., “Rod Drop Monitoring,

Does It Really Work?”, Proceedings of the Twenty-ninth
Turbomachinery Symposium, Texas A&M University,
College Station, pp 11-20.
Figure D3 – System response in the complex plane.
S. M. Schultheis, C. A. Lickteig, R. Parchewsky,
The model of the crosshead system is more complex than “Reciprocating Compressor Condition Monitoring”,
the simple mass-spring-damper system used in this appendix; Proceeding of the Thirty-sixth Turbomachinery
however, the basic ideas still apply. Because the model used for Symposium, Texas A&M University, College Station, pp.
the crosshead system is non-linear, the methodology for 107-113, 2007.
calculating the eigenvalues is slightly different. At each step of
the integration, the state-space matrices are saved. Once the R Wilson, J.N Fawcett, “Dynamics of the Slider-Crank
simulation is compete, the eigenvalues are found for the state Mechanism with Clearance in the Sliding Bearing”,
space matrices at each step of the integration. This is called a Mechanism and Machine Theory, Volume 9, Issue 1,
frozen-time approach and results in the eigenvalues as a Spring 1974, Pages 61-80.
function of crank angle. Plotting these eigenvalues results in the
plot seen in Figure 21 of the main text of the paper. The Den, Hartog J. P. Mechanical Vibrations, New York: Dover
imaginary parts (damped natural frequency) of these Publications, 1985.
eigenvalues are plotted in Figure 22.

Copyright© 2015 by Turbomachinery Laboratory, Texas A&M Engineering Experiment Station