Ch05 howard

Brian F. Howard is a Senior
Technologist/Engineer at GE Energy, in
Issaquah, Washington. He acts as the
corporate representative for reciprocating
machinery projects throughout the
company, supports the product development
process, and provides diagnostic support to
customers to solve problems with rotating
and reciprocating machinery. Prior to
joining GE Energy, he worked as a project
manager for Lone Star Compressor Corporation.
Mr. Howard received a B.S. degree (Mechanical Engineering,
1995) from the University of Houston. He is a member of ASME.
John A. Kocur, Jr., is a Machinery
Engineer in the Plant Engineering Division
at ExxonMobil Research & Engineering, in
Fairfax, Virginia. He has worked in the
turbomachinery field for 20 years. He
provides support to the downstream
business within ExxonMobil with expertise
in vibrations, rotor/aerodynamics, and
health monitoring of rotating equipment.
Prior to joining EMRE, he held the position
of Manager of Product Engineering and Testing at Siemens Demag
Delaval Turbomachinery. There Dr. Kocur directed the development,
research, engineering and testing of the compressor and steam
turbine product lines.
Dr. Kocur received his BSME (1978), MSME (1982), and Ph.D.
(1991) from the University of Virginia and an MBA (1981) from
Tulane University. He has authored papers on rotor instability and
bearing dynamics, lectured on hydrostatic bearings, and is a
member of ASME. Currently, he holds positions of API 617
vice-chair, API 684 co-chair, and vice-chair of the API SOME
Steering Committee.
ABSTRACT
Condition monitoring data collection capabilities have grown
over the last decade in large part due to advances in integrated
circuits (IC). These advances now permit real time thermodynamic
and process calculations on desktop computing platforms. With
these capabilities, the sophistication and amount of data available
to the users of rotating and reciprocating equipment has grown
exponentially. In particular, online reciprocating compressor
monitoring systems generate significant amounts of real time data
beyond the traditional values of pressure, temperature, vibration,
and speed.
With the growth in the amount and manipulation of data, it has
become imperative to assist the engineer or operator in quickly
locating indications of a potential problem. Many users and
condition-monitoring suppliers have designed, developed, and
tested knowledge-based tools to achieve this goal. The intent of
this paper is to describe the knowledge-based tool design and
development process for reciprocating compressors, including the
basic assumptions that must be considered and the design features
that can make an effective knowledge-based system. A discussion
of basic research into reciprocating compressor knowledge-based
tool development will be presented in an effort to help those who
undertake this development effort understand potential obstacles.
The important aspects of an effective system will be described.
Finally a series of case histories from operating process plants will
show instances where the system worked, as well as instances
where problems occurred and how they were addressed.
INTRODUCTION
Today’s petrochemical and refining plants utilize a wide variety
of equipment classes (fixed equipment, rotating machinery, and
reciprocating machinery) to produce the chemicals and fuels
required by society.Advances in design tools, information technology,
and condition monitoring have enabled reliability improvement for
all classes of equipment. Expectation of continuing operation
excellence at the plants, environmental concerns, and economic
drivers encourage plant stake holders to reduce emissions,
unplanned outages, and flare events.
Yet, reliability has not increased equally across each of these
three classes. Reciprocating equipment reliability improvement
lags behind the other two. In order to close this gap plant owners
have turned to condition-monitoring technologies to improve the
reliability of their reciprocating compressors.
Condition-monitoring vendors responded to this need by
introducing a variety of new technologies, such as the
microelectromechanical systems (MEMS) embedded in the
online cylinder pressure transducer. These new technologies,
along with advances in information systems, have increased the
amount of data available to reciprocating compressor operators.
In addition, computers can now run real-time thermodynamic,
process modeling, and calculation programs creating
additional data.
For the rotating equipment engineer tasked with operating a
unit or plant, extracting useful information from this large
collection of data presents a formidable analytic challenge.
41
KNOWLEDGE-BASED TOOL DEVELOPMENT
AND DESIGN FOR RECIPROCATING COMPRESSORS
by
Brian F. Howard
Senior Engineer/Technologist
GE Energy
Issaquah, Washington
and
John A. Kocur, Jr.
Machinery Engineer
ExxonMobil Research & Engineering
Fairfax, Virginia
Table of Contents

Condition-monitoring systems for reciprocating compressors,
which have many more moving parts than an equivalent
centrifugal compressor, generate a large amount of raw data. For
a critical, large reciprocating compressor, the condition-monitoring
measurements include those shown in Table 1. (Less critical
or spared reciprocating compressors utilize a subset of
these measurements.)
Table 1. Reciprocating Compressor Monitored Points.
Although some of the measurements change slowly (i.e., static
data), other measurements require high-speed data collection
during the revolution of the crankshaft (i.e., dynamic data).
Condition-monitoring systems typically collect frame velocity,
crosshead acceleration, cylinder acceleration, valve-cover acceleration,
cylinder pressure, and rod position data as dynamic data.
Assuming the measurements in Table 1 are applied to a typical four
throw reciprocating compressor, the condition-monitoring system
manages over 200 static points and 44 dynamic points (assumes 2
frame velocity [3 static, 2 dynamic values], 4 crosshead accelerome-
ters [6 static, 2 dynamic], 8 rod position [5 static, 2 dynamic], and 8
cylinder pressure [19 static, 2 dynamic]). The analytical tools in most
reciprocating compressor condition-monitoring systems provide
numerous plotting options for each point, representing the potential
for scores, or even hundreds, of plots to review.
Recognizing that in today’s economic climate it is not
practical to expect an engineer to audit the data presented by this
system at regular intervals, many companies and organizations
have turned to knowledge-based rules to flag suspect data and
potential problems. This paper describes the knowledge-based
tool development and design process for reciprocating
compressors, including the basic assumptions that must be
considered, and the design features that can make an effective
knowledge-based system.
TOOL DEFINITION AND DESCRIPTION
The use of outside resources to store and document knowledge
has been with the human race for thousands of years, from the
ancient Sumerian cuneiform marks on clay to the paperback books
commonplace today. Computers brought changes in not only the
way we store information, but in the type of information that could
be stored.
Writing and drawing enabled the storage of only descriptive
information. The content could only say, “go do this” or
“check that” but could not perform any of the operations.
Computers, programming languages, and digital media enabled
storage of both descriptive and operational information. The
handheld calculator presents a great example of this. Unlike
a math book, which describes how to perform addition,
subtraction, and other basic math operations, the calculator
actually takes data, performs the operations, and returns a
result. By pressing the “PI” button, the calculator also provides
descriptive information.
The term “knowledge-based tools” refers to a collection of
computational procedures containing both descriptive and operational
information. The knowledge base may include first principles or
model-based structures, empirically determined structures, or any
combination of the two. Obviously, a reciprocating compressor
condition monitoring knowledge base includes a large amount of
both descriptive and operational information.
Knowledge-based tools do not represent the only mechanism for
analysis of the data generated by condition-monitoring systems.
Other approaches, such as statistical analysis or, in some cases,
neural net analysis can be used. These approaches do not require
knowledge of how a particular machine should behave. Instead
these approaches detect anomalous behavior in the data set based
on past performance. These paradigms have the advantage of never
missing an event, but the disadvantage of requiring the local
knowledge to disposition each event and develop a baseline for
each unique compressor configuration and operational state. As
such, these anomaly detection tools fall outside the scope of
this paper.
In general, knowledge-based tools can include a wide variety of
analytical toolsets and databases. This paper focuses on a subset of
those tools, fixed-threshold alarms, fault dictionaries, model-based
diagnosis and decision trees, and the ways in which these tools can
be combined to define rules.
Fixed-Threshold Alarms
Fixed-threshold alarms, by far, find the greatest application in
reciprocating compressor monitoring. When a measured value
exceeds a preset limit, the monitoring system triggers an alarm.
Threshold alarms represent the most primitive knowledge-based
tool as they contain only two items: a single descriptive piece of
information (the set point) and a single piece of operational
information (trigger an alarm). This tool requires so little
computational effort that it has been implemented in mechanical,
pneumatic, and electronic systems.
Configuring parallel preset limits on the same measurement
enables multiple severity alarms. Although very simple to set,
process upsets or transient events may trigger multiple alarms
requiring the operator to quickly sift through the triggered alarm
list to discover the cause of the root alarm event.
PROCEEDINGS OF THE THIRTY-EIGHTH TURBOMACHINERY SYMPOSIUM • 200942
Table of Contents

Fault Dictionary
When discrete events, such as threshold alarms or measurements,
are grouped together in such a way that a particular pattern, or set
of patterns, of alarms or events drives an output this system is
referred to as a fault dictionary. For example, a simple fault
dictionary might consist of two measurement points: a velocity
transducer at the drive end of the compressor and a second velocity
transducer at the nondrive end of the compressor. If the direct
vibration amplitude of either transducer crosses a threshold alarm,
a contact closes turning on a light behind a panel that reads “HI
FRAME VIB.” When both measurements cross, or have crossed,
the threshold a contact closes turning on a light behind the “HI HI
FRAME VIB” panel. Diverse and complex fault dictionaries can
be compiled using computer software programmed with Boolean
logic steps.
Model-Based Diagnosis
In model-based diagnosis, a model of the process runs
concurrently with the physical process. Input values originating in
the field drive the model, and the output of the model is compared
to the measured output of the physical process. Differences in the
observed values and model’s predicated values can be evaluated.
For example, the compression process inside a reciprocating
compressor cylinder closely matches an isentropic compression
process. From this observation, a mathematical model can be
constructed where discharge temperature is calculated as a function
of gas composition, suction temperature, and compression ratio.
Using a simple threshold alarm, the discharge temperature of the
model can be compared to the measured discharge temperature at
the cylinder. In contrast to a fixed set point, which could be crossed
due to process upset or other transient conditions, this threshold
changes as conditions inside the cylinder change. Multiple
combinations of threshold alarms and fault dictionaries can be
used with model-based diagnosis.
Decision Trees
Decision trees provide a method for mapping diagnostic
methodology. Figure 1 shows a simple decision tree describing the
problem solving methodology for selecting a generic soda over a
name-brand soda when the potential for a name brand coupon
exists. Threshold alarms, fault dictionaries, and model-based
diagnosis can all be combined within decision trees to provide
flexible, robust knowledge-based tools.
Figure 1. Sample Decision Tree.
Figure 2 shows a decision tree incorporating threshold alarms
and fault dictionaries. The decision tree consumes temperature data
from “LP Stg3 Suct W” and “LP Stg 3 Suct Temp,” performs basic
mathematical operations and compares the result to the constant
“Suction Valve Coefficient” in the first threshold alarm, labeled
“Temperature Threshold Alarm.” In a separate branch of the tree,
“Leak – Cylinder to…” value is compared to a severity constant.
Based on the result of this combination of threshold alarms and
fault dictionary, the rule delivers different severities for the
“Suction Valve Leak” alert.
Figure 2. Knowledge-Based Tool Showing Decision Tree
Incorporating Fault Dictionary and Threshold Alarms. Red Ellipse
Indicates Temperature Threshold and Blue Ellipse Indicates Leak
Threshold Alarms.
ROBUST TOOL DESIGN
Threshold alarms, fault dictionaries, model-based diagnosis,
and decision trees all seem to be useful tools, but what is the best
way to synthesize a useful set of tools for reciprocating
compressor condition monitoring? The following describes the
design process and lessons learned from knowledge-based tool
development for heavy-duty, slow-speed, API-618 (2007)
reciprocating compressors.
Characteristics of Successful Tools
A successful rule has only two functions: where no malfunction
indicator or malfunction exists, the rule should be silent; where an
indicator or malfunction exists, the tool must provide an alert.
Moving from this accurate, but simple, characterization to a
robust knowledge-based tool requires that many implicit needs be
addressed, some of which have a direct impact on the design
process. Here are some of the concerns that arose during the rule
development process for reciprocating compressors:
• The rule and rule infrastructure design need to address the
human-interface needs. For example, what happens to an
unacknowledged alert when the severity increases? How does the
display prioritize and drive focus to high-priority alerts during
multiple events?
• The rule must perform these functions at all operating and load
conditions, including the presence of other malfunctions.
Reciprocating compressors have a variety of capacity-control
devices including valve unloaders, pocket unloaders, and hydraulically
actuated “stepless” unloaders making this a real challenge.
• Condition-monitoring system infrastructure must be pristine.
Transducer installation, wiring, system configuration, and information
technology (IT) infrastructure must be in excellent condition prior
to deployment of the rule to avoid false/missed alerts.
• The knowledge-based tool must be able to differentiate between
harmless transient conditions (such as debris passing through a
valve) and serious changes in machine condition.
• The tool must be sufficiently flexible to fit into the cultural
DNA of the organization. For example, some plants change every
compressor cylinder valve during an outage; some change only the
43KNOWLEDGE-BASED TOOL DEVELOPMENT
Table of Contents

one in distress. To the first plant a knowledge-based tool that
indicates exactly which valve is in distress has no value; to the
second, it is essential.
• To every rule for reciprocating compressors, there exists an
exception. The rule and development environment must be flexible
enough to allow deployment over a large fleet and accommodate
these exceptions.
Beyond the technical needs, the development process also
highlighted infrastructure needs at the plant. Some of those include:
• Involve information technology at the earliest opportunity.
Condition-monitoring systems often require data from the
distributed control system (DCS) and, therefore, the plant IT
locates them in the control room and only connects them to the
control network. In this case, information does not get from
the rule to the people that need it. Directing this information
appropriately requires that the condition-monitoring system
be connected to the business network as well as to the
control network.
• Establish condition-monitoring system ownership. The way in
which a plant designates ownership of the system has a direct
impact on the value rule alerts return to the organization. For
example, consider a broken wire. In a plant where the rotating
equipment engineer owns the machine, but instrumentation and
controls (I&C) own the transducer, the rotating equipment engineer
detects the problem and must issue a work order describing the
problem to the I&C shop. In contrast, if the rotating equipment
engineer owns the transducer and wiring, the engineer can make
the repair directly.
• Who should get the alerts? The unit operators are in the best
position to use the data immediately to take corrective action. But
if every alert goes to the operators, they quickly become resistant
and ignore the rule alerts. The machine stakeholders need to
decide who gets what information. The rule must be flexible
enough to allow the alerts to flow to a variety of levels within
the organization.
• Rule infrastructure has as much to do with the success of the
rules as does the technical content. Data presentation, alarm
management, data storage, and other factors all provide important
value and features to the end users. Even something as relatively
minor as a color scheme can create confusion if the user needs’ and
expectations are not understood.
Tool Design Process
Figure 3 shows an overview of the design process used today to
develop rules for reciprocating compressors. The process looked
quite different in the first development phase so the chart in Figure
3 includes a variety of lessons learned.
For example, the initial rule design process began with the
effort to map the failure effect on measurements and then
proceeded directly to the assembly of the decision tree. During
the proof-of-concept deployment, questions from the plant
personnel about operation and delivery of alert messages raised
several concerns about how to best manage the results
generated by the rule. This resulted in the second step shown
in Figure 3.
Another gap in the initial rule process became apparent as
the initial results of the proof of concept deployment were
integrated into the rule design. In adding or modifying decision
tree components within the rule, changes could inadvertently result
in unexpected operation during regimes where the tool had
previously been stable. To ensure that changes to the rule do not
impact stability, a test program must be designed and implemented
prior to rule development. This test program should incorporate
several features to ensure proper rule behavior.
First, the test suite must cover a broad range of machine
operation. In practice, this usually reduces to a collection of
databases describing normal operation and malfunctions. Data
from these sources flows through the rule in a controlled
environment. Figure 4 shows such a development tool in which the
various components of a decision tree can be monitored during
testing. Next, the test suite must be able to execute quickly.
Generally, the test suite should take less than 10 minutes to
execute. In addition to speed, the test suite should be automated as
this improves error trapping and feedback to the developers. Lack
of speed and feedback causes developers and rule designers to
avoid testing, resulting in poor rule operation. Finally, the test suite
should be viewed as a living document, readily modified as lessons
learned accumulate from deployment of the rule.
Figure 3. Reciprocating Compressor KBT Design Process.
Table of Contents

Figure 4. Knowledge-Based Tool Test Session.
Another important change to the process includes the cyclical
nature of the development processes. Just as Antony van
Leeuwenhok discovered new worlds of microbiology when he
applied the tool of the microscope to everyday items, so too will
users of rules on reciprocating compressors discover many new and
unexpected details about the operation of their compressors. In
many cases, this new learning will impact the failure effect
mapping and, therefore, trigger the process over again. As the
organization deploys rules, this cycle enables corporate learning to
be leveraged by the entire enterprise.
The remainder of the paper describes some of the findings from
the application of rules to reciprocating compressors.
CASE HISTORIES
Effect of Buffer Settings and
Hysteresis on Rule Operations
Easily one of the most frustrating experiences with deployment
of rules has been designing the necessary infrastructure for
managing data, alarms, and alarm set/reset events.
Anyone tasked with auditing data coming from a reciprocating
compressor has likely seen spurious transient events in the data.
Problems such as debris passing through a valve, process changes,
or gas composition changes can cause short-term deviations that
typically do not warrant an alert. How does one manage the rules
so that these problems do not trigger an alert, but real machine
problems do?
Part of the answer has been buffering, in which the rule averages
the data coming in over a fixed amount of time. To date, no real
first-principles approach has emerged to determine appropriate
values for these buffers. The values in Table 2 have been
determined empirically, based on results of installation on API-618
(2007) style compressors.
Table 2. Buffer Settings.
Using buffers provides improved rule operation, but the rules
require additional infrastructure in order to operate robustly. In order
to understand this infrastructure, recall that the opening of this paper
included a definition of fixed-threshold alarms and mentioned that
such systems had been implemented in many different mechanical,
pneumatic, and electronic systems. A simple pressure switch, as
shown in Figure 5, provides an example of a “simple” mechanical
system that provides fixed threshold alarm capability.
Figure 5. Mechanical Pressure Switch.
Consider what happens to the switch as pressure on the piston
face increases: a voltage potential exists at one contact and a
voltmeter connects the other contact to the potential source as
shown in Figure 5. As the pressure force begins to overcome the
spring force, the piston moves toward the contacts. At Point 1 in
Figure 6 the piston has experienced sufficient displacement to
enable the conducting stem to touch the contacts. This completes
the circuit, allowing current to flow from one contact to the other.
As a result the voltmeter indicates a value. Point 2 in Figure 6
represents this state.
Figure 6. Voltage Potential and Pressure.
Now consider the switch reaction to a decrease in pressure.
Reduction of pressure does not immediately change the value of
the voltage at the contacts. Friction exists between the piston and
the walls, and the piston, stem, and spring have inertia resulting in
resistance to movement. The line from Point 2 to Point 3 represents
the switch state for this condition. Eventually, the pressure drops
low enough that the spring force overcomes friction and inertia
allowing the stem to move away from the contacts, breaking the
circuit. Point 4 represents this state.
The difference between the pressure that causes the switch
contacts to close and the pressure that causes them to open again is
referred to as hysteresis. All mechanical switches have some
degree of hysteresis and designers for these systems strive to keep
the values bounded as excessive hysteresis has a strongly negative
influence on switch operation.
Software systems suffer from no such limitations. In fact,
software alarms have no intrinsic hysteresis at all. This leads to
some difficulties. Figure 7 shows a typical repetitive advisory
Table of Contents

scenario. The top pane shows the process variables, in this case rod
load tension, rod load compression, and degrees of reversal. The
middle plot shows the intermediate buffered values, averaged over
three hours. As expected, the buffering reduces the appearance of
the transient events at 16:21 and 16:35 in the output of the rod load
and reversal functions. Yet, as shown in the lower plot in Figure 7,
the rule still changed state. The cause of this change in rule state
requires close examination of the arithmetic sum of both the rod
load and the reversal function (used as input to the rule). As Figure
8 shows, the buffered values move above and below the severity
threshold several times a day. Note the small changes with respect
to the scale on the left vertical axis.
Figure 7. Trend Plots Showing Repetitive Severity. Top Pane Shows
Inputs to Rule, Middle Pane Shows Buffered Intermediate Values
Within Rule, and Bottom Pane Shows Rule Alert Status.
Figure 8. Arithmetic Sum of Rod Load and Rod Reversal Functions.
Clearly the small changes in values, and subsequent rule
advisories, do not represent actual changes in machine
condition. In order to introduce a hysteretic effect in a software
tool, a dedicated operand, shown in Figure 9, needs to be
included in the toolset to control the actuation of the alerts.
When the “Set” flag is true the hysteresis operand passes the
value through after the “Set Duration” time has elapsed. The
value remains set until the “Reset Duration” time has elapsed
and the flag at the “Reset” input equals true. The operand then
remains in the reset condition until the “Set” is true and the set
duration time has elapsed.
Figure 9. Threshold Operand.
Hysteresis settings for reciprocating compressors have
been determined empirically, as shown in Table 3. With these
buffer settings and hysteresis values, repetitive alerts have
been eliminated.
Table 3. Threshold Settings.
Suction Valve Failure
In this case history, a collection of rules is designed to
assess the condition of the cylinder trim components: valves,
piston rings, and packing rings. This collection includes rules
to detect:
1. Pressure packing leak
2. Frame loading
3. Crosshead pin loading
4. Leak—cylinder to low pressure
5. Leak—high pressure to cylinder
6. Suction valve leak
7. Discharge valve leak
Likely, with the exception of the fourth and fifth items, most
operators and owners of reciprocating compressors understand the
malfunctions. The fourth and fifth items refer to distortion of the
pressure versus volume (PV) curve caused by the leaks. In the
case of the fourth item, a leak from the cylinder chamber to a
low-pressure reservoir (such as the suction valve manifold,
distance piece, or atmosphere) occurs and the cylinder end in
distress can no longer build up pressure fast enough during the
compression process and pressure falls as piston velocity slows
near top dead center (TDC)/bottom dead center (BDC). The top
two graphs in Figure 10 show this case. The fifth item addresses
the opposite case in which gas from a high-pressure source leaks
back into the cylinder. In this case, the pressure builds up too
quickly during the compression process and cannot fall quickly
enough as piston velocity slows near TDC/BDC. The bottom two
plots in Figure 10 show this case.
Table of Contents

Figure 10. Chamber Leak Effects on Pressure-Volume Curves.
In the case of a leaking suction valve, the indicated discharge
pressure drops with respect to line pressure, while, in the case of a
leaking discharge valve, the indicated suction pressure rises with
respect to line pressure. The leak detection algorithms use these
pressure data points from the PV curve to determine if a leak exists
on the cylinder and whether the leak originates in a high-pressure
or low-pressure reservoir.
In order to pinpoint the component in distress, the rules
combine the above cylinder pressure PV information with
suction temperature, discharge temperature, and valve cover
temperature information.
In this example, a four-throw, balanced-opposed, horizontal
reciprocating compressor provides compressed hydrogen service
at a large refinery. A head end (HE), suction valve plug-type
unloader on each cylinder and a variable pocket clearance unloader
on the head end of the first stage provide capacity control. Each
compressor provides three stages of hydrogen make-up and one
stage of recycle service. Each compression stage consists of one
double-acting compressor cylinder. Shell and tube heat exchangers
provide interstage cooling between each compression stage.
Compressors have been instrumented as per Table 1 plus valve
cover acceleration.
The condition-monitoring system, which includes rules, first
provided an alert to a suction valve problem on 13:59:47 on 22
February 2008, as shown in Figure 11. Review of waveform
data, shown in Figure 12, shows the cylinder pressure curves
associated with this event. The plot shows a significant
difference between the indicated pressure curve (orange) and the
adiabatic pressure curve (green) on the head end. (The adiabatic
pressure curve assumes an isentropic thermodynamic process
and uses the indicated pressures at TDC and BDC to generate
the curve.) The difference appears most pronounced during the
compression process, 100 percent to 75 percent of cylinder
displacement, when the indicated pressure curve rises slower
than the theoretical pressure curve. Looking at the same lines
just after TDC, it can be seen that the indicated pressure falls
faster than the adiabatic curve. This pattern of difference
between the indicated and adiabatic pressure curves signifies a
leak from the cylinder to a low-pressure area, such as is found
with a suction valve leak.
Figure 11. Rule Suction Valve Alert Advisory.
Figure 12. First Stage Cylinder Pressure Versus Volume.
Typically, a valve temperature measurement provides confirming
evidence to determine which valve has failed with multivalve
cylinders. For the cylinder in question, valve temperature data
taken 30 minutes prior to the PV diagram (Figure 13) does indicate
that the 1HS1 valve temperature had risen far above the suction
temperature: 171ЊF (77.2ЊC) valve cover temperature versus 70ЊF
(21.1ЊC) suction gas temperature. This high temperature difference
confirms a leaking valve, as reported by the alert.
Figure 13. 1HS1Valve Temperature and Suction Temperature Trend.
Table of Contents

Interestingly, the customer’s inclusion of a measurement not typically
found in online reciprocating compressor condition-monitoring
systems (valve cover vibration) provides an opportunity for
additional analysis. The vibration from this transducer can be
plotted with the cylinder pressure data, as shown in Figure 14.
Figure 14. Cylinder Pressure, Head End Suction Valve Acceleration
and Crosshead Acceleration, 22 Feb 2008, 7:17:16.
In the areas highlighted by the red ellipses in Figure 14, gas
begins to leak past the valve into the cylinder suction manifold.
This flow of gas creates an acoustic noise detected by the
accelerometer. The red ellipses in Figure 14 highlight this noise
The good agreement between the theoretical and indicated
pressure curves in Figure 14 illustrates that this incipient leak has
not yet significantly affected the cylinder performance and would
not be detected using a PV diagnostic tool.
The presence of these signatures at 7:16 in the morning of 22
February 2008—nearly seven hours prior to the suction valve
alert—suggests valve cover vibration may provide the earliest
indication of valve distress.
The valve condition continued to deteriorate until the plant took
the unit out of service on 27 February.
Machine outage and overhaul commenced on 27 February 2008.
Referring to Figure 15, the 1HS1 suction valve experienced both
concentric ring and valve seat failure. The plant replaced the valve
assembly and returned the unit to service.
Figure 15. Typical Suction Valve Assembly.
As Figure 16 shows, the good agreement between the theoretical
and indicated pressure curves and low amplitude valve cover
accelerometer signals confirm the new valve to be in good health
and operating correctly.
Figure 16. Stage 1 Cylinder Pressure, Suction Valve Cover, and
Unfiltered Crosshead Accelerometer Signal, 28 Feb 2008, 12:05:14.
Capacity and Process Effects on
Combined Rod Loading and Reversal
In this example a six-throw, balanced-opposed horizontal
reciprocating compressor provides make-up hydrogen service at a
large refinery. An HE suction valve plug-type unloader on each
cylinder and a variable pocket clearance unloader on the head end
of the first stage provide capacity control. Each compressor has two
distinct services, based on process stream requirements; each of
these services has three compression stages; and each compression
stage consists of one double-acting compressor cylinder. Shell and
tube heat exchangers provide interstage cooling between each
compression stage. The compressor is instrumented as per Table 1.
Shortly after the commissioning of the collection of rules, the
plant engineer observed that one of the throws indicated an alert.
Opening the event viewer, the plant engineer discovered that an
alert had been issued (Figure 17).
Figure 17. Crosshead Pin Loading Condition Alert.
As the rules had been recently installed and not completely
commissioned, the plant engineer initially believed the alert to be
false. In addition, plant operations had begun the process of
Table of Contents

shutting down the compressor, which lowers the pressure across the
compressor. It seemed unlikely that a reduction in pressure would
have caused loading problems on a crosshead pin.
Evaluation of the rule configuration confirmed correct configuration.
The supporting evidence indicated that a change in rod load and
reversal had occurred and driven the alarm. To confirm that the data
consumed by the rule accurately reflected machine condition, the rod
load curves were reviewed. (Refer to APPENDIX A for definition of
terms and how the condition-monitoring system generates these
curves.) Figure 18 shows these curves. At first glance, the two curves
appear quite similar; however, closer inspection revealed that a
critical change in forces at the crosshead pin had occurred.
Figure 18. Rod Load Curves Before (Top) and After (Bottom) Alert.
Figure 19 shows a crosshead assembly and nomenclature. The
crosshead pin connects the connecting rod to the crosshead. The
crosshead pin does not turn by any large amount. In order to achieve
complete lubrication, the crosshead pin forces must alternate so that
the pin moves from one side of the bushing to the other. If the forces
at the crosshead pin do not alternate sufficiently, metal-to-metal
contact occurs between the pin and the bushing. Unchecked, the wear
leads to excessive clearance and high-amplitude shock events to the
connecting rod, which can cause the connecting rod to fail.
Figure 19. Typical Direct Rod Connection Crosshead Assembly.
The plots in Figure 18 revealed that the alert had correctly
identified a shift in the rod load forces at the crosshead pin. The
blue lines represent the force exerted by the gas acting on the piston.
The red line represents the inertia load imposed upon the pin by the
reciprocating motion of the crosshead assembly, crosshead nut,
piston rod, and piston assembly. Summing the inertia and gas forces
results in the net force experienced by the crosshead pin. The green
line in Figure 18 shows this combined rod load force.
In each plot, the combined rod load crosses the neutral axis at
two points, shown by black dots. The minimum distance between
crossings, referred to as degrees of reversal, serves as a good
indicator of lubrication between the pin and crosshead bushing. As
the degrees of reversal decrease, the pin has less time to flush out
old lubricant and allow new lubricant to flow in.
In addition to degrees of reversal, the ratio of maximum tensile
load to maximum compressive load also impacts the lubrication
condition at the crosshead pin. Ideally, the magnitude of the
compressive and tensile forces would be equal. The more unequal the
forces, the more difficulty the pin has in achieving good lubrication.
As can be seen in Figure 18, both degrees of reversal and
symmetry in tensile/compressive forces decreased. The rule alert
had correctly identified a problem that could have led to substantial
machine damage.
As a result of this analysis, the plant technician consulted with
operations to review the process for machine shutdown. It was
discovered that prior to lowering discharge pressure, the shutdown
procedure decreased the head end clearance volume of the first
stage cylinder.
Figure 20 shows the cylinder pressure curves before and after the
volume pocket change on the first stage cylinder. Prior to the
volume change, the third stage cylinder had the highest compression
ratio. The change in clearance volume on the first stage cylinder
caused this cylinder to operate with a higher compression ratio, but
at the expense of the compression ratio on the third stage.
Figure 20. First Stage, Second Stage, and Third Stage Cylinder
Pressure Curves Before (Top) and After (Bottom) Change in First
Stage Head End Volume Clearance.
As the third stage is a high-pressure cylinder, the rod diameter is
not so different from the diameter of the piston. The resulting
inequality in crank-end and head-end piston area makes the cylinder
sensitive to changes in load. Contrary to “common sense” thinking,
an increase in the compression ratio of the third stage cylinder in
fact results in improved degrees of reversal and symmetry of force
Table of Contents

magnitudes. Lower compression ratio, as in this instance, actually
reduces degrees of reversal and symmetry of force magnitudes.
As a follow-up action based on the advisory, the plant altered
shutdown and operational procedures to avoid loading the first
stage cylinder.
In this case, the alert provided a timely notification that
allowed the plant to both update the operating instructions and to
continue monitoring to ensure that the machine remained below
design conditions.
CONCLUSION
The previous case history illustrates the importance of getting
timely, meaningful alerts to the right people. Designing and
deploying knowledge-based tools has challenges.Yet, organizations
that successfully deploy and manage knowledge-based tools to
leverage information provided by condition-monitoring systems
recognize advantages in their plant operations.
APPENDIX A—
ROD LOAD TERMINOLOGY
AND CALCULATION
Introduction
The term “rod load” has been used for decades to describe the
maximum forces a reciprocating compressor assembly can
withstand. The term has some ambiguity, but recent papers have
clarified some of the key definitions and terms.
With the improved reciprocating compressor condition monitoring
understanding beginning to permeate the industry, customers have
begun to ask questions about how condition-monitoring systems
calculate these values. The purpose of this Appendix is to answer
those questions.
The terminology and context for this application note are
American Petroleum Institute (API) Standard 618, Fourth Edition
and Fifth Edition (2007). Previous editions of API 618 used other
descriptions and definitions for rating values of reciprocating
compressors. For ease of reference, the terms and definitions used
herein appear at the end of this application note.
Calculation Methodology
Figure A-1 shows the rod load curves and data generated by a
reciprocating compressor condition-monitoring system.
Figure A-1. Rod Load Curves Generated by Condition-Monitoring
System.
Inertial Force
The red line in Figure 21 represents the forces due to inertia.
The inertia mass for nearly all reciprocating compressor
condition-monitoring installations includes the crosshead
assembly, crosshead nut, piston rod, and piston assembly. This
collection of mass for inertia is consistent with the definition
provided in API-618 Fifth Edition (Data Sheets, Page 7, line
31). The condition-monitoring system does allow the user to
exclude the crosshead mass from the inertia force calculations
(Figure A-4); however, the configuration is rarely encountered
in the refining/petrochemical segments and is not consistent
with API-618 Fourth or Fifth edition terminology. For these
reasons, this definition falls outside the scope of this
application note.
Gas Force
The blue line in Figure A-1 represents the gas forces acting on
the compressor’s static components and running gear. This force
is the gas load referenced in API-618 (paragraph 6.6.2). To obtain
this force, the indicated cylinder pressure on the head end is
multiplied by the head-end area of the piston. This is shown in
green in Figure 22 for a typical double-acting cylinder. The
resultant force is then subtracted from the indicated crank-end
cylinder pressure times the crank-end piston area (shown in brown
in Figure A-2).
Figure A-2. Piston Areas Used for Gas Rod Load Calculation.
This summation represents the net force (sign convention for
forces in rod load show compressive forces as negatives and
tensile forces as positive) acting on the piston rod and can be
written as:
The cylinder pressure varies continuously throughout the
revolution of the crankshaft so the calculations must be
performed multiple times throughout the revolution to obtain
a curve. For each 360 degrees of crankshaft rotation, a
condition-monitoring system collects 720 indicated cylinder
pressure data points simultaneously for both the head and crank
end. The gas load calculation is thus performed 720 times for
each revolution.
The gas pressures in the chamber act not only on the piston, but
also on the heads of each cylinder. The combined gas force on the
crank- and head-end heads has the same absolute value as the gas
load calculated above, but acts in the opposite direction (i.e., has
the opposite sign) (Figure A-3).
(A-1)
Table of Contents

Figure A-3. Reciprocating Compressor.
Combined Rod Load
Finally, the green line in Figure A-1 represents the combined rod
load, or crosshead pin load. This force is the combined rod load
referenced in API-618 (paragraph 6.6.1). The gas load is added to
the inertia force at each point of measurement to obtain this force.
Since, as noted in the previous section, the gas load is computed
720 times for each crankshaft revolution, the condition-monitoring
system also performs the combined rod load calculation 720 times
for each crankshaft revolution. For each of these 720 points of
measurement during the crankshaft revolution, the calculation can
be written as:
When the mass used in the inertia force calculation includes the
crosshead, the smallest distance between the points of zero force
(shown by black dots at approximately 35 degrees and 200 degrees
of crank angle in Figure A-1) represents the degrees of rod reversal
referenced in API-618 (paragraph 6.6.4).
Figure A-4 shows a typical configuration screen that allows users
to select rod load calculations at either the crosshead pin or piston
rod. Had the user configured the condition-monitoring system to
calculate rod load at the piston rod, the inertia forces would no longer
include the crosshead mass. The combination of this inertia force and
gas force results in the forces that act on the piston rod, next to the
crosshead. Note that this force no longer reflects those acting on the
crosshead pin, and therefore cannot be used to calculate rod reversal.
Figure A-4. Configuration Screen.
Terms and Definitions
The following definitions draw extensively from Atkins, et al.
(2005). The reader is encouraged to consult this reference for a
complete discussion of these terms and others, as well as a valuable
historical perspective.
• Combined rod load—The sum of actual gas load (including
valve and passage losses) plus inertia loads at the crosshead pin,
in the direction of the piston rod.
This is the force curve labeled “Combined Force” in the
condition-monitoring software’s rod load plots when the inertia
force includes the mass of the crosshead assembly. This force
varies continuously throughout the revolution. The term appeared
in both the Fourth and Fifth Edition of API-618 with the same
definition; however the Fourth Edition required that it be
calculated every 10 degrees and the Fifth Edition required that it
be calculated every 5 degrees. Reference 3.7 of API-618 Fifth
Edition for a formal definition.
• Crosshead pin load—Same definition as “combined rod load.”
Note this term is not defined within API-618, but is a commonly
encountered industry term used to clarify the component at which
the combined rod load calculations are being done.
• Gas load—The force resulting from the internal pressure
in each chamber acting on the associated piston- and cylinder-
head surfaces.
This is the curve labeled “Gas Force” in condition-monitoring
software plots. As it depends only on the pressure and cylinder
geometry, it remains the same whether the force calculation is
done at the crosshead pin or piston rod. The term appeared in both
the Fourth and Fifth Editions of API-618 with the same definition;
however, the Fourth Edition required that it be calculated every
10 degrees and the Fifth Edition required that it be calculated
every 5 degrees.
• Maximum allowable continuous combined rod load
(MACCRL)—A value determined by the original equipment
manufacturer (OEM) based on design limits of the various
components in the compressor frame and the running gear
(bearings, crankshaft, connecting rod, crosshead assembly, piston
rod, piston assembly).
With very minor exceptions (refer to 6.6.5 of API-618, Fifth
Edition), no single value of combined rod load can exceed the
manufacturer’s ratings for maximum allowable continuous
combined rod load (refer to paragraph 3.19 of API-618 Fifth
Edition for a formal definition). OEMs may have individual
limits for compressive rod load, tension rod load, and compressive
plus tension.
• Maximum allowable continuous gas load (MACGL)—A value
determined by the OEM based on the design limits of the static
components (frame, distance piece, cylinder and bolting).
With very minor exceptions (refer to 6.6.5 of API-618,
Fifth Edition), no single value in the gas load curve can exceed
the manufacturer’s ratings for maximum allowable continuous
gas load. Reference 3.20 of API-618 Fifth Edition for a
formal definition.
• Rod reversal—The shortest distance, measured in degrees of
crank revolution, between each change in sign of force in the
combined rod-loading curve.
Reference paragraph 3.49 of API-618 Fifth Edition for a formal
definition; however, note that the API 618 definition is not entirely
accurate as it references “piston rod loading” instead of “combined
rod loading.”
(A-2)
Table of Contents

APPENDIX B—
SELECT RECIPROCATING
COMPRESSOR NOMENCLATURE
Figure B-1. Select Reciprocating Compressor Nomenclature.
Figure B-2. Plug Type Loader Assembly.
Figure B-3. Variable Clearance Pocket Assembly.
REFERENCES
Atkins, K. E., Hinchliff, M., and McCain, B., 2005, “A Discussion
of the Various Loads Used to Rate Reciprocating
Compressors,” Proceedings of the Gas Machinery Conference.
API Standard 618, 2007, “Reciprocating Compressors for
Petroleum, Chemical, and Gas Industry Services,” Fifth
Edition, American Petroleum Institute, Washington, D.C.
BIBLIOGRAPHY
Sijm, A. and Desimone, G., 2004 “Improving Reliability for Recips,”
Orbit, pp. 32-35, (1).
Dvorak, D. and Kuipers, B., 1991, “Process Monitoring and
Diagnosis: a Model-based Approach.” IEEE Expert 6(3):
67-74, June.
Table of Contents

Ch05 howard

More Related Content

Similar to Ch05 howard (20)

Recently uploaded (20)

Ch05 howard