vibration analysis for reciprocating compressors

48
orbit VOL. 32 | NO. 2 | APR. 2012 A Technical Publication for Advancing the Practice of Operating Asset Condition Monitoring, Diagnostics, and Performance Optimization ORBIT VOLUME 32 • NUMBER 2 • APRIL 2012 ANOMALERT* – UNDER THE HOOD MOTOR CONDITION MONITORING & DIAGNOSTICS AnomAlert* Motor Anomaly Detector – Under the Hood PG10 Vibration Data Identifies Hot Spot on Motor Rotor • PG26 NEW DEPARTMENT ITEMS Application Notes • PG25 System 1* Software Tips & Tricks • PG42

Upload: jose-prado

Post on 03-Oct-2015

200 views

Category:

Documents


23 download

DESCRIPTION

Vibration Analysis for Reciprocating Compressors

TRANSCRIPT

  • orbit VOL. 32 | NO. 2 | APR. 2012 A Technical Publication for Advancing the Practice of Operating Asset Condition Monitoring, Diagnostics, and Performance Optimization

    ORBIT

    VOLU

    ME 32 N

    UM

    BER 2 APRIL 2

    012

    AN

    OM

    ALERT* UN

    DER TH

    E HO

    OD

    MOTOR CONDITION MONITORING & DIAGNOSTICS

    AnomAlert* Motor Anomaly Detector Under the Hood PG10Vibration Data Identifies Hot Spot on Motor Rotor PG26

    NEW DEPARTMENT ITEMS

    Application Notes PG25 System 1* Software Tips & Tricks PG42

  • EDITORS NOTE

    We are also continuing with our series of ADRE* Tips, and

    for the first time in a while, have included a Recip Tips

    article as well. For the first time ever, we are including

    a System 1* Software Tips & Tricks article, and an

    Application Note summary. I anticipate sharing many

    more of these useful software tips and application

    references as we continue updating the content of Orbit.

    Speaking of sharing, we will soon be posting Orbit articles

    at a convenient public blogsite, so that our readers can ask

    questions and post comments about individual articles

    and related concepts. I look forward to learning from these

    conversations, and getting some ideas for additional

    follow-on articles to address the points that are raised.

    Finally, I couldnt help but notice that the little stop sign

    icon in our reader service card looked a bit odd for some

    reason. It was introduced in 2004, and apparently was

    not questioned until now. Gina and I fixed it for this

    issue. Can you spot the difference? I suppose these

    older back-issues will now become valuable collectors

    items like double-struck coins, or

    postage stamps that are printed

    upside down. If you are lucky enough

    to have one of these rarities, hang

    onto it as a treasured family heirloom!

    Editors Notepad

    Cheers!

    Gary

    Gary Swift

    Editor

    Orbit Magazine

    [email protected]

    Greetings, and welcome to

    Orbit! This issues cover is

    based on the graphic that

    accompanies our Feature

    article. I anticipate that

    our technical readers will

    appreciate the humorous

    analogy of a tiny V-8 engine

    symbolizing the power of

    the monitor. In keeping with

    the theme of motor condition

    monitoring, we also have

    a classic case history that

    describes how vibration

    analysis detected a problem

    with a motor that had a

    load-related thermal bow.

  • IN THIS ISSUE

    In this Issue

    orbitVolume 32 | Number 2 | April 2012

    A Technical Publication for Advancing the Practice of Operating Asset Condition Monitoring, Diagnostics, and Performance Optimization

    Publisher: GE EnergyEditor: Gary B. SwiftDesign Coordination: Eileen OConnellDesign: Gina AlteriEuropean Circulation: Estelle SjournNorth American Circulation: Karen SchanhalsPrinter: RR Donnelley

    CONTRIBUTORSGE EnergyRoengchai ChumaiCharles HatchJohn KinghamStuart RochonGaia RossiRob Winter Adrian Cobb Nate Littrell

    ArtesisCaner Kuzkaya

    CREDITSGE Measurement and ControlGlobal CommunicationsNik Noel

    Questions, suggestions, and letters to the editor may be addressed to: ORBIT Magazine 1631 Bently Parkway South Minden, Nevada USA 89423 Phone: 775.782.3611 Fax: 775.215.2855 e-mail: [email protected]

    Printed quarterly in the USA. More than 35,000 hard copies of each issue distributed worldwide

    * Denotes a trademark of Bently Nevada, Inc., a wholly owned subsidiary of General Electric Company.

    Copyright 2012 General Electric Company. All rights reserved.

    10 FEATURES

    AnomAlert* Motor Anomaly Detector Under the Hood

    NEWS

    04 Advanced Machinery Dynamics Course Invitation

    06 Celebrating Our Experience

    DEPARTMENTS

    ADRE* Tips 18 How to Display Filtered and Unfiltered Orbits Together

    Application Note 25 Resources for Managing Electrical Runout

    Case Histories 26 Vibration Data Identifies Hot Spot on Motor Rotor

    Recip Tips34 Vibration Analysis for Reciprocating Compressors Part 1

    System 1* Software Tips & Tricks42 How to Create a Machine Reference Dataset

  • Advanced Machinery Dynamics Course

    Measurement & Control, a GE Oil & Gas business invites you to extend your knowledge of machinery diagnostic techniques and rotor dynamics as applicable to rotating machinery in a 5-days Advanced Machinery Dynamics Course, from June 11th through 15th in Florence, Italy.

    This high-level course was last

    conducted in 2009 and is now fully

    updated in response to previous

    participants feedback and to meet

    the most demanding machinery

    diagnostics challenges.

    In our hands-on workshops you will

    use standard vibration diagnostic

    tools on machine simulating rotor kits.

    With us you will for sure put

    theory into practice.

    Case histories highlighting vibration

    documentation, analysis, and

    machine malfunction corrective

    techniques will be presented

    throughout the course.

    Who should attend?

    Engineers desiring to advance

    their machinery vibration

    diagnostics skills

    Engineers involved in the design,

    acceptance testing, and main-

    tenance of rotating machinery

    Post-graduate engineers

    Academic researchers and profes-

    sors involved in rotor dynamics

    Prerequisites

    Prior to this course, participants

    should have completed the Machinery

    Diagnostics course or be

    ISO category 3 certified.

    The Machinery Diagnostics course will

    be offered the week before for those

    who do not yet meet the prerequisite.

    If you wish to take part in this

    course even though you dont

    fulfill the prerequisites, your

    enrolment will be accepted but

    in this case you may not get the

    expected return on investment.

    PresentersRon Bosmans

    Global Director

    Machinery Diagnostics Services

    19952006 (Retired)

    Nicolas Peton

    MDS Technical Leader for South

    and West Europe

    GE Measurement & Control

    Rob Winter

    Senior Specialist

    Learning and Development

    GE Measurement & Control

    Arun Menon

    Global Director

    Machinery Diagnostics Services

    GE Measurement & Control

    REG

    ISTE

    RTO

    DAY

    !

    4 ORBIT Vol .32 No.2 Apr.2012

    NEWS

  • Advanced Machinery Dynamics Course1115 June 2012 | Florence, Italy | GE Learning Center

    For registration and logistics details please contact

    Marta Petruzzelli at [email protected] or call +39 0396561420

    EUR 3,500.00 (including one joint dinner)

    Apr.2012 No.2 Vol .32 ORBIT 5

  • The Bently Nevada team had a saying back in 1990:

    Duplicating our products is challenging. Duplicating our people is impossible.

    Although a lot has changed over the past 22 years,

    it is still our people who create the high-quality

    products and provide the excellent care that our

    customers depend on. In keeping with tradition,

    the employees at the home of our product line

    in Nevada, USA, pause once a year to recognize

    the dedicated work of our coworkers who have

    reached significant service milestones. The people

    listed here are only a small fraction of our total

    team, yet they represent more than 1800 years

    of combined experience! Our multinational team

    extends around the world, where similar commit-

    ment can be found in every global region.

    Celebrating Our Experience

    35Y E A R S

    BACK ROW, LEFT TO RIGHT: Ron Sanchez, Jack Howard, Al Davis. FRONT ROW: Candy Baldwin, Pam Caughron.

    Photos by Adrian Cobb

    NEWS

  • BACK ROW, LEFT TO RIGHT: Tim Walmsley, Randy Willis, Paul Blair. MIDDLE ROW: Alan Thomson, Gerry ONeill, Mike Evans, Jana Ferguson. FRONT ROW: Debbie Hartzell, Jill Evans. NOT SHOWN: Denise Clendenen, John Grant, Andrew Grimm, Doug Hoover.

    BACK ROW, LEFT TO RIGHT: Rob Rose, Dave McNeilly. MIDDLE ROW: Brenda Allmett, Jerry Pritchard, Jean Van Den Berg. FRONT ROW: Dave Whitefield, Robert Nikkels. NOT SHOWN: Sherrie Ashurst, Stan McPartland, Tim Sheets, Dave Van Den Berg.

    30Y E A R S

    25Y E A R S

    Apr.2012 No.2 Vol .32 ORBIT

    NEWS

  • BACK ROW, LEFT TO RIGHT: Thane Tahti, Mike Holcomb, Tammy Rhead. FRONT ROW: Ronnie Swan, Francie Welsh, Diana Thomas. NOT SHOWN: Rudy Capa, Ken Ceglia, Ken Forbes, Steve Kichler, Dave McElroy, Barbara Uemura.

    BACK ROW, LEFT TO RIGHT: Carol Brennaman, Kyle Hoffman, Tim Gross, Landon Boyer. NEXT ROW FORWARD: Pamela Greek, Deana Cormier, Paul Lindsay, Ben Willis. SECOND ROW FROM FRONT: Leslie Yered, Beth Ferrara, Ray Jensen, Scott Williams. FRONT ROW: Enrique Corcostegui, Larry Mcdonald, Steve Schmid. NOT SHOWN: Daniel Abawi, Matt Anderson, Alex Beitel, Dale Bradley, Chien Cheng, Mitch Cohen, Doran Cushing, Mike Hanifan, Mike Rokusek, Bryan Shadel.

    20Y E A R S

    15Y E A R S

    8 ORBIT Vol .32 No.2 Apr.2012

    NEWS

  • BACK ROW, LEFT TO RIGHT: Ron Robbins, Daniel Jenkins, Todd Balcon, Paul Carrion. NEXT ROW FORWARD: Stephen Lau, Kris Wickstead, Becky Cawthorne, Donna Barber. NEXT ROW FORWARD: Jay Brown, Bev McMahon, Lisa Akins, Kelly Kondo, Sandi Bachstein, Tina Ku, Christina Caldwell. SECOND ROW FROM FRONT: Ray Murphy, Brian Steinkraus, Richard Fraser, Laura Love, Ruby Ecobisag, Lynne Towle. FRONT ROW: Manuel Lara, Violeta Della Pella, Jack Riley, Joe Jenks. NOT SHOWN: Jennifer Carlson, Ken Crosby, Michael Gaynor, Paul Gonzi, Dustin Hess, Brad Kelly, Rick Lohroff, Lelana Moralez, Paul Parisien, Jean Untereiner.

    10Y E A R S

    Apr.2012 No.2 Vol .32 ORBIT

    NEWS

  • Anom Alert Charles T. Hatch

    Principal Engineer

    [email protected]

    Caner Kuzkaya

    Vice President, Artesis A.S.

    [email protected]

    The AnomAlert Motor Anomaly Detector is a

    system of software and networked hardware that

    continuously identifies faults on electric motors

    and their driven equipment. AnomAlert utilizes

    an intelligent, model-based approach to provide

    anomaly detection by measuring the current

    and voltage signals from the electrical supply to

    the motor. It is permanently mounted, generally

    in the motor control center and is applicable

    to 3-phase AC, induction or synchronous, fixed

    or variable speed motors. AnomAlert models

    are also available for monitoring generators.

    FEATURES

    10 ORBIT Vol .32 No.2 Apr.2012

  • Anom Alert under the hood

    Apr.2012 No.2 Vol .32 ORBIT 11

    FEATURES

  • 12 ORBIT Vol .32 No.2 Apr.2012

    FEATURES

    The AnomAlert diagnostic solution can be used together with

    a vibration monitoring system as a complementary tool for

    detecting electrical faults. Alternatively, it can be used where

    dedicated vibration monitoring is not practical, economical, or

    comprehensive enough. It can detect changes in the load the

    motor is experiencing due to anomalies in the driven equipment or

    process such as cavitation or plugged filters and screens. Since it

    doesnt require any sensor installation on the motor itself or on the

    associated load, AnomAlert is especially attractive for inaccessible

    driven equipment and is applicable to most types of pumps,

    compressors, and similar loads. It is also well suited to the moni-

    toring of submersible, borehole, downhole, and canned pumps.

    The AnomAlert monitor uses a

    combination of voltage and current

    dynamic waveforms, together

    with learned models, to detect

    motor or driven equipment faults.

    Active learning is backed up by

    an additional fleet model in case

    the monitor has been installed on

    an already defective motor. The

    monitor detects differences between

    observed current characteristics

    and learned characteristics and

    relates these differences to faults.

    Motor fault detection is based on a

    learned, physics-based motor model,

    where constants in the model are cal-

    culated from real-time data and com-

    pared to previously learned values.

    Mechanical fault detection is based

    on power spectral density amplitudes

    in particular frequency bands, in

    relation to learned values. This infor-

    mation is combined automatically

    with expert diagnostic knowledge.

    Because of this spectral band

    approach, mechanical fault detection

    is not precise, but provides guidance

    toward a class of possible faults. The

    sensitivity to some faults (for example

    rolling-element bearing faults) will

    decrease with distance from the

    fault. On the other hand, faults that

    increase motor load are independent

    of the distance from the motor.

    The spectrum-based mechanical

    fault detection in the AnomAlert

    monitor seems similar to Motor

    Current Signature Analysis (MCSA),

    but several important differences

    set it apart from typical MCSA:

    The AnomAlert monitor uses cause-

    effect (voltage-current) relation-

    ships, while MCSA uses the current

    only. Changes in input voltage will

    cause changes in the current that

    could lead to false alarms in MCSA.

    The cause-effect relationship in

    the AnomAlert processing helps

    protect against these false alarms.

    The AnomAlert monitor uses a

    stable reference data set that is

    obtained from ten days of motor

    operation, and it calculates

    alarm threshold levels specific

    to the equipment itself.

    Detected anomalies are subjected

    to a sophisticated change

    persistence algorithm to guard

    against false alarms, making the

    AnomAlert monitor less sensitive to

    random fluctuations in the signals.

    We will now delve more deeply

    into the operating principles of the

    AnomAlert monitor. We will not dis-

    cuss current and voltage transformer

    selection or installation, or operating

    modes and programming; these

    aspects are covered elsewhere.1

    Data AcquisitionVoltage and current signals from

    all three phases (6 total signals) are

    sent to the monitor where they are

    digitized for further signal processing.

    Voltages less than 480 V can be

    input directly, while higher voltages

    require a potential transformer.

    Depending on the application,

    current transformers or Hall-effect

    current sensors are used to sense

    and step down the motor currents.

    AnomAlert processing operates on

    a 90 second iteration cycle. At the

    beginning of every 90 second itera-

    tion, the monitor samples voltage and

    current waveforms. The remainder of

    the period is used for post processing

    analysis and front panel update.

  • Apr.2012 No.2 Vol .32 ORBIT 13

    FEATURES

    All six waveforms can be exported to

    a text file for further post process-

    ing. The text file has no headers

    and six columns, corresponding

    to paired voltage and current

    waveforms V1, I1, V2, I2, and V3, I3.

    Modeling And Fault DetectionThe AnomAlert monitor uses four

    different approaches to fault detec-

    tion. One is based on internal motor

    characteristics; another is based on

    frequency analysis of the residual cur-

    rent spectrum; a third analyzes actual

    line voltages and currents to check for

    certain types of line and current faults;

    finally, the fourth uses fleet data from

    similar motors to provide an inde-

    pendent diagnostic reference. We will

    discuss how all of these work in turn.

    The Internal Motor ModelFor an ideal motor, voltage and cur-

    rent waveforms are sinusoidal at line

    frequency. The changing line voltage

    creates magnetic forces that cause

    the rotor to turn, and the amplitude

    and phase of the motor currents are

    related to the input voltages through

    the internal mechanical and electrical

    workings of the motor. We can think

    of the line voltage waveforms as

    inputs to the motor, and the current

    waveforms as outputs. The motor

    electrical and mechanical internals

    can be thought of as a transfer func-

    tion that converts the input voltage

    waveform into the output current

    waveform (Figure 1). This is the key

    to understanding the internal motor

    model in the AnomAlert monitor.

    The monitor uses a linear model

    for the electrical and mechanical

    internals of the motor. This physics-

    based model is derived from a set

    of differential equations, and it can

    be expressed as a transfer function.

    During the learning process, the

    monitor determines the coefficients

    of this model. For a normal motor,

    the model transfer function is a

    close approximation to the real

    physical transfer function of the

    motor. We will discuss later the

    special case of what happens when

    the AnomAlert monitor models a

    motor that already has a defect.

    While monitoring, the AnomAlert

    monitor takes the input voltage

    waveform and passes it through the

    model transfer function to obtain

    a theoretical current waveform.

    Meanwhile, the real motor transfer

    function converts the input volt-

    age waveform into the observed

    (measured) current waveform. The

    theoretical current waveform is sub-

    tracted from the measured current

    waveform to produce a residual cur-

    rent waveform (Figure 2). The residual

    waveform contains the errors

    between theory and reality, and the

    monitor uses this residual waveform

    for mechanical fault analysis.

    FIGURE 1: The motor as a transfer function. A voltage waveform is converted to a current waveform by the motor.

    Input Voltage

    OutputCurrent

    Motor

    FIGURE 2: A source voltage waveform passes through the real motor transfer function, producing a current waveform with harmonic distortion, IMotor. The same voltage waveform is passed through the learned model transfer function, producing a theoretical current waveform, IModel. The two waveforms are subtracted, producing a residual current waveform. The residual waveform represents the error between theory and reality.

    VResidualCurrent

    Motor

    Learned Model

    -1

    IMotor

    IModel

  • Motor Electrical Fault DetectionChanges in the internal character-

    istics of the motor (for example, a

    shorted winding) will cause the real

    motor transfer function to change.

    While monitoring, the AnomAlert

    unit takes the measured voltage and

    current waveforms and calculates

    a new set of observed coefficients

    for the internal motor model. The

    original model coefficients are

    subtracted from the observed

    coefficients to yield residuals.

    These residuals are used to detect

    internal electrical motor problems.

    Mechanical Fault Detection

    In an ideal motor, the rotor would

    be perfectly centered in the stator

    clearance, turn smoothly, and have no

    unbalance. In real motors, the rotor is

    never perfectly centered in the stator,

    bearings and driven equipment create

    disturbances, and the rotor always

    has some unbalance.

    Mechanical faults disturb the rotor

    position and create disturbances and

    distortions in the current waveforms.

    As faults develop in the machine train,

    they will cause the output current to

    deviate further from the theoretical.

    For example, an unbalanced rotor

    will move in a 1X orbit that causes a

    rotating rotor/stator gap change. This

    change causes amplitude modulation

    of the current signals and causes

    sidebands to appear around the line

    frequency in the spectrum. In another

    example, a race fault in a rolling

    element bearing will cause a periodic

    disturbance in the rotor position;

    this disturbance in rotor position will

    create a corresponding disturbance

    in rotor/stator gap and amplitude

    modulation of the motor current.

    The modulation produces sidebands

    around the line frequency in the

    residual current spectrum, and the

    distance of the sidebands from the

    line frequency will correspond to the

    bearing defect frequency. Other kinds

    of faults can produce a wide variety

    of additional frequency content in

    the current waveforms. AnomAlert

    processing (and in general, MCSA)

    looks for this additional frequency

    content and uses it to diagnose differ-

    ent classes of mechanical problems.

    AnomAlert analysis is different

    from MCSA. MCSA involves spectral

    analysis of the observed current

    waveform (sometimes demodulated),

    while AnomAlert processing

    produces a Power Spectral Density

    (PSD) plot from the residual current

    waveform (the difference between

    the theoretical current waveform and

    the measured current waveform).

    The AnomAlert residual current

    waveform is based on a learned

    model, so the PSD is a spectrum of the

    difference between theory and reality.

    Thus, AnomAlert methodology first

    detects change in the motor current,

    and then classifies the spectral

    characteristics of that change into

    fault classes. The monitor classifies

    PSD energy into 12 typical spectral

    frequency ranges that are associated

    with particular fault classes.

    Line and Current Faults

    During the learning period, the

    monitor learns typical behavior for

    that motor. Deviations of voltage or

    current from normal behavior can

    signal a problem. The monitor checks

    for significant changes in power

    factor, voltage, and current imbal-

    ance. Because an increase in driven

    load will cause an increase in motor

    current, AnomAlert methodology uses

    abnormal current as an indicator

    of a load problem. For example,

    decreasing flow through a fan or

    blower would cause a decrease in

    fan load and motor current, and this

    could signal an obstruction in flow.

    The Fleet Model

    What happens if the monitor is

    installed on a motor that has an

    existing fault? Will it learn the fault

    and fail to detect that something is

    wrong? No. This is where the fleet

    model comes in. The monitor has a

    database of residual waveform signal

    characteristics that are representa-

    tive of a large fleet of similar motors.

    This is used as a backup to guard

    against missed alarms in case the

    AnomAlert monitor has learned

    a bad motor. When a measured

    value exceeds the High value in

    the database for that frequency

    range (Figure 3), the monitor will

    alarm assuming that the alarm

    level has passed the persistence

    test. We will discuss this test later.

    14 ORBIT Vol .32 No.2 Apr.2012

    FEATURES

  • Learning

    When first installed, the monitor

    learns the behavior of the motor it is

    hooked up to. It spends some time

    learning before starting to monitor the

    motor. Some motors drive equipment

    that operates at a constant speed and

    load. This is the simplest operating

    mode to learn and monitor because

    any change in operating character-

    istics is probably indicative of a fault.

    Many other machine trains operate

    at variable speed or variable load. In

    this case, what is normal for one load

    range may be abnormal for another.

    In this situation, the monitor learns

    and creates a separate internal motor

    model for each operating mode.

    Then, later, as conditions change, it

    will shift from one model to the next.

    The AnomAlert learning period takes

    about 10 days (Figure 4), whether

    the motor is fixed or variable

    speed. During learning, the monitor

    iterates by collecting waveforms,

    performing analysis, then repeating

    the process. During each 90 second

    iteration, it simultaneously collects

    voltage and current waveforms

    for each phase, and then performs

    numerical analysis of the data. During

    the initial, 3 day Learn phase, the

    AnomAlert unit will not monitor. It is

    busy building a preliminary internal

    motor model and spectral statistics.

    After the initial Learn phase is

    complete, the AnomAlert unit will

    begin to monitor the motor. While it

    does this, it will continue to improve

    the model for another 7 days (the

    Improve phase). For variable speed

    motors, these iterations are spread

    over as many operating modes as

    necessary. During the Learn and

    Improve phases, if motor operation

    shifts from one operating mode to

    another, the monitor will save the

    previous data and start learning

    the new operating mode. When

    the motor returns to a partially

    completed mode, the monitor will

    continue learning from the last point.

    Once the entire learning process has

    been completed, the monitor stops

    model refinement and continuously

    monitors the motor using the

    completed internal motor model

    and PSD spectral characteristics.

    If, after model completion, the motor

    enters a new operating mode that

    hasnt been seen before, the monitor

    may go into alarm if the current

    waveforms are significantly different

    from what has been modeled. At

    that time, the user can manually

    direct the AnomAlert unit to learn

    the new mode using the Update

    command. It will then learn the new

    operating mode. It will not monitor

    the new mode until the update

    learning process is completed.

    During all learning, if either motor

    power or AnomAlert power is

    interrupted, the monitor will

    automatically recover and continue

    learning from the last point.

    FIGURE 3: Residual current PSD plot showing the motor spectrum (blue) and the fleet High curve (red). If a motor frequency persistently exceeds a fleet High value, the monitor will alarm.

    FIGURE 4: The AnomAlert learning period. After installation, AnomAlert spends about 10 days learning the motor behavior. It will start to monitor after the initial 3 day Learn period is complete.

    Learning Period (10 days)

    ImproveLearn

    7 days3 days

    Monitor

    Apr.2012 No.2 Vol .32 ORBIT 15

    FEATURES

  • Change Detection, Persistence, and AlarmingBecause of noise and small changes

    in operating characteristics, there

    is always some variation between

    successively observed model and

    spectrum parameters. During the

    learning phase, the AnomAlert

    monitor builds statistics that describe

    the variation that occurs. When

    learning is complete, the monitor has

    a set of statistics for every model

    coefficient (electrical faults) and

    spectral band2 (mechanical faults).

    The AnomAlert unit operates by

    detecting differences between

    observed and previously learned

    parameters; either internal model

    coefficients or spectral band

    amplitudes. These differences must

    pass a statistical test before being

    considered significantly different.

    These tests define minimum alarm

    thresholds. Check Line alarms are

    generated based on voltage imbal-

    ance variations and voltage fluctua-

    tions from the range encountered

    during the Learn phase. A similar

    alarm method is used for power

    factor, total harmonic distortion,

    voltage and current rms values, and

    voltage and current imbalance values.

    Even large deviations could be

    expected to occur in a normal

    machine once in a while. To guard

    against false alarms, AnomAlert

    processing requires that the detected

    change be persistent over time.

    The monitor uses a sophisticated

    algorithm that compares the amount

    by which a parameter exceeds the

    threshold value and the number of

    times this has occurred in a window

    of time. This sliding window varies

    depending on the amount the

    measured parameter exceeds the

    statistical threshold. Large threshold

    exceedance will require only a short

    time window, while mild exceedance

    will require a long window. The moni-

    tor will alarm only when the persis-

    tence requirement has been satisfied.

    DiagnosticsFor the most part, the AnomAlert

    monitor does not provide precise

    diagnoses of particular faults. Instead,

    it reports categories of faults that

    act as indications and point to areas

    that should be further investigated.

    It uses four independent fault

    detection methods that cover two

    categories, electrical and mechanical.

    Electrical faults are associated with

    either motor internal problems or

    external power supply issues. The

    AnomAlert unit monitors both using

    two independent methods. Internal

    motor faults are detected using the

    learned internal motor model as a

    reference. During each monitoring

    iteration, the monitor calculates a set

    of 8 internal motor model parameters

    based on the observed voltage and

    current. These observed parameters

    are compared against the param-

    eters that were obtained during the

    learning phase, and significant and

    persistent changes are detected and

    reported as electrical faults. These

    faults include the following examples:

    Loose windings

    Stator problem

    Short circuit

    External supply is directly checked

    for voltage or current imbalance,

    voltage range, maximum current,

    and low voltage or current.

    Mechanical fault categories are

    detected and diagnosed using the

    PSD of the residual current waveform.

    The residual current represents the

    difference between the observed

    current and the theoretical current

    produced by the internal motor

    model using the same observed

    voltage. The PSD is divided into 12

    frequency ranges that are typically

    associated with certain mechanical

    problems (listed below). Analysis of

    these frequency ranges produces

    fault classes for further investigation.

    Loose Foundation/Components

    Unbalance/Misalignment/

    Coupling/Bearing

    Belt/Transmission Element/

    Driven Equipment

    Bearing

    Rotor

    Note that the Check Load alarm,

    caused by abnormally high or low

    current, is usually caused by a

    change in the driven machines load;

    machine load can change for two

    reasons, fault or process change. If

    the machine is running in a different

    condition which is not seen during

    the learn period, the user has to set

    16 ORBIT Vol .32 No.2 Apr.2012

    FEATURES

  • the AnomAlert unit to update mode

    to learn this new condition. If the

    load is changed due to a fault, the

    problem should be investigated,

    and the user needs to make sure

    the alarm is cleared in the monitor.

    The Fleet Model provides an

    independent analysis in the event

    that the AnomAlert unit has learned

    a faulty system. The Fleet Model

    consists of Normal and High values

    for each of the 12 PSD ranges based

    on experience with a large number

    of similar motors. If a residual

    current PSD range value exceeds

    the fleet High value, then, after

    persistence checking, the monitor

    will warn that something is wrong.

    LimitationsThe AnomAlert Motor Anomaly

    Detector is a powerful motor

    monitoring system. However,

    there are some limitations on

    its use and interpretation.

    It cannot be used for DC or

    single-phase motors.

    For variable frequency drives,

    the inverter chopping frequency

    should be higher than 2 kHz.

    Mechanical diagnostics are based

    on energy in 12 spectral frequency

    ranges. This is, by nature, an approxi-

    mate analysis, and diagnostic indica-

    tions usually only represent broad

    classes of problems. The customer will

    have to follow up using other methods

    to determine the actual fault. The PSD

    spectrum produced by the AnomAlert

    unit can be helpful, but may not be

    sufficient for problem identification.

    The AnomAlert unit cannot be

    used on motors that have rapidly

    varying voltage or power. Voltage,

    frequency and current amplitude

    must not change by more than 15%

    in six seconds. This is not a serious

    restriction for most applications, but

    some applications, like crushers, will

    not fit this requirement. Note that

    if a sudden change of load occurs,

    the monitor will reject that sample;

    however, the same machine could run

    steadily at some load, and this would

    allow the unit to monitor the machine.

    The AnomAlert unit will work very

    well on applications where the

    motor is located some distance

    from the current or potential

    transformers. However, the line at

    the current measurement point

    must be dedicated to a single motor;

    multiple motors downstream from

    a single CT cannot be monitored. On

    the other hand, one set of PTs can be

    used for all motors that are supplied

    from the same voltage source. The

    current measurement restriction is a

    consideration for subsea applications

    where power may be delivered to the

    sea floor only to branch off to multiple

    motors. In this case, an AnomAlert

    unit could not be used on the main

    delivery power line. However, it could

    be used if CTs could be installed on

    each branch (CT burden limits apply3).

    Summary

    The AnomAlert Motor Anomaly

    Detector is a powerful motor monitor-

    ing system. Its power comes from

    both sophisticated signal processing

    and analysis algorithms and from

    built-in redundancy. Its ability to learn

    makes it sensitive and flexible, and

    a fleet reference database protects

    against missed alarms caused by

    learning an already defective motor.

    Alarming is clever and uses statistical

    analysis combined with an adaptive

    persistence test. These features

    produce a product that is a significant

    improvement over conventional Motor

    Current Signature Analysis, and it has

    a proven track record documented

    by many case histories.

    * Denotes a trademark of Bently Nevada, Inc., a wholly owned subsidiary of General Electric Company.

    Copyright 2012 General Electric Company. All rights reserved.

    1 For sensor selection and installation, see Bently Nevada Guide 286752, Selection of CTs, CSs, and PTs for AnomAlert. For general ordering information, see 286754-01, Specifications and Ordering Information.

    2 Note that the AnomAlert monitor identifies the largest amplitude spectral line in a particular frequency range and uses that lines amplitude for the value in that range. It does not add up all the spectral energy in a range.

    3 The burden of a current transformer is the maximum resistance that the secondary of the CT (the part hooked up to the AnomAlert monitor) can drive and meet specification. Long wires from the CT will have more resistance that will limit the allowable distance from the CT to the monitor. See Bently Nevada Guide 286752, Selection of CTs, CSs, and PTs for AnomAlert.

    Apr.2012 No.2 Vol .32 ORBIT 1

    FEATURES

  • John W. Kingham

    Field Application Engineer

    [email protected]

    How to Display Direct and Filtered Orbits Together, Synchronized by Sample

    When I used to be a Machinery Diagnostic Services (MDS) engineer, and travelled the world to diagnose machinery problems,

    I had several plot formats that I used all of the time. One of

    my favorites was the Orbit plot. Ive always described the

    orbit plot as what the shaft would draw if there were a

    pencil lead at its centerline, and you held a piece of paper

    up to it . Seeing what the shaft is doing graphically allows

    you to interpret what it is doing mechanically, and from

    there a diagnosis may be made.

    Typically, most people look at orbit plots for the unfiltered,

    or direct data and the 1X filtered data. I particularly like to

    look at these two orbits together on the same page. A quick

    glance at the unfiltered orbit can show problems such as

    glitch (electrical & mechanical runout noise), unbalance,

    misalignment, oil whirl (instabilities), looseness and rubs.

    The 1X filtered plot gives you some insight into these

    malfunctions as well, and is especially good for observing

    shaft precession.

    The plots in Figure 1 are of a steam turbine exciting its first

    natural rotor vibration frequency due to a rub. The plots

    have been scaled using the Auto, All Plots function, which

    allows you to see at a quick glance that there is significant

    NOT 1X activity.

    18 ORBIT Vol .32 No.2 Apr.2012

    DEPARTMENTS

    ADRE* TIPS

  • FIGURE 1: Orbit and timebase waveform plots for vibration of a steam turbine rotor within the clearances of its fluid film bearings. The upper plots show Direct (unfiltered) data, while the lower plots show 1X-filtered data.

    While we are on the subject, it is easy to determine that

    the frequency is locked in to X by looking closely at

    the signal from the X (horizontal) probe. First, notice that

    there is only 1 positive peak seen for every 2 Keyphasor*

    marks. Second, to determine it is exactly 1/2X, note that

    you can draw two straight lines through the Keyphasor

    marks as shown by the red dashed lines (Figure 2).

    FIGURE 2: Magnified view of the unfiltered timebase waveform

    signal from the X (horizontal) probe.

    Overview

    For those who are fairly experienced in configuring

    ADRE plot sessions, this brief overview summarizes the

    process. A more complete description with step by

    step instructions is included after this summary.

    Create your Orbit/Timebase plot group.

    Make sure that you are using the

    Synchronous waveform.

    Copy and paste the new plot group below the original,

    and reconfigure it for 1X (or 2X or nX). While you are

    configuring, make sure that the Use Static Samples

    for Filtered Waveforms check box is cleared.

    Drag the 1X plots up into the original plot group.

    Do this at the New Orbit/Timebase Plot level,

    not at the variable (waveform) level.

    Once the plots are organized correctly, delete

    the 1X plot group it is no longer needed.

    If you are using a 2x1 plot layout with paging by

    sample, the direct orbit will be on top with the filtered

    orbit on the bottom. If you have a 2x2 plot layout,

    the filtered orbits will be on the right hand side. This

    can be changed by changing the order (dragging

    and dropping) of the plots at the plot group level.

    For steady-state machine conditions, it is recommended

    that you set scaling to auto all plots. This way, it is

    easy to see which bearing is giving you a problem. If

    you are observing a startup or shutdown transient,

    setting manual scaling may be better, as the auto all

    plots will scale for the amplitude extremes (such as

    when the machine passes through a resonance).

    Detailed Description

    1. CREATE A BASIC ORBIT TIMEBASE PLOT

    Select the channel(s) that you want to show orbit

    plots for from the Configuration hierarchy on the

    left side of the screen. It is only necessary to select

    one channel from each channel pair (Figure 3):

    Apr.2012 No.2 Vol .32 ORBIT 1

    DEPARTMENTS

    ADRE* TIPS

  • FIGURE 3: In this example, two channels have been selected for

    display. We will be viewing data for bearings 1 and 2.

    Right-click on the selected channel(s) and select the

    Orbit/Timebase plot from the shortcut menu (Figure 4):

    FIGURE 4: Selecting Orbit/Timebase plot option.

    When you do this, the Orbit/Timebase plot

    will open. Close the plot for now, since we

    will not be looking at it for a while.

    Direct Orbit/Timebase plots are typically made from

    the Synchronous waveform sample. The 1X filtered

    and Slow Roll Compensated data is always taken

    from the Synchronous waveform sample. Therefore,

    we need to make sure that the Orbit/Timebase plot is

    configured to use the Synchronous waveform data.

    Right-click on the new plot group in the plot hierarchy

    and select Configure: from the menu (Figure 5). This

    will open the plot group configuration dialog.

    FIGURE 5: Opening the plot group configuration dialog.

    In the Orbit Timebase plot group configuration dialog,

    first verify that the Sync Waveform variables are

    selected for each channel. If they are not, select the

    correct variables from the drop-down boxes (Figure 6):

    FIGURE 6: Checking which variables are selected.

    Click the arrow button to expand the drop-down

    list. Select Sync Waveform for display (Figure 7).

    FIGURE 7: Ensure the Sync Waveform is selected.

    Repeat the process for the channel pair variable (Figure 8):

    20 ORBIT Vol .32 No.2 Apr.2012

    DEPARTMENTS

    ADRE* TIPS

  • FIGURE 8: Selecting Sync Waveform for the paired variables to be displayed.

    2. DUPLICATE THE PLOT

    In the Plot Session hierarchy (right hand tree), make

    a copy of the plot group that you just created, and

    place it just below the original plot group. There are

    several ways to accomplish this. The easiest way is to

    right-click on the plot group and select Copy (Figure

    9), then Paste it into the plot session (Figure 10).

    FIGURE 9: Copying the new plot group.

    FIGURE 10: Pasting the copied plot group into the plot session.

    The result should look something like the example in

    Figure 11.

    FIGURE 11: Observe that the selected variables are now shown underneath the associated Orbit/Timebase plot groups.

    3. CONFIGURE THE FILTERED ORBIT

    For clarity, I renamed the bottom Orbit plot group 1X

    Orbit/Timebase Plot Group. You can too, but it isnt neces-

    sary (Right click on the plot group and select Rename).

    When we are done, we will delete this plot group, and

    rename the reconfigured plot group Direct & 1X Orbit/

    Timebase Plot Group again, this will be optional, but it

    is a nice thing to do if you are going to use this plot group

    as a template. I do this as a matter of bookkeeping. If I

    didnt delete the plot group, at the end of the day, Id have

    two plot groups the one that I want, with both direct and

    1X data and also a plot group that only has 1X plots.

    To establish the plots to be 1X filtered, right-click on the

    appropriate plot group and select Configure. In the top

    half of the configuration grid (Figure 12), select 1X from

    the drop down menu under the Filtering column. Select

    this for all channels that you want filtered orbits for.

    FIGURE 12: Selecting 1X filtering option from plot General properties.

    At this same time, make sure that the check boxes

    under Use Static Samples for Filtered Waveforms

    are cleared (Figure 13). If this isnt done, the filtered

    and unfiltered samples will be indexed differently and

    become unsynchronized, which would be confusing.

    Apr.2012 No.2 Vol .32 ORBIT 21

    DEPARTMENTS

    ADRE* TIPS

  • FIGURE 13: Clear the Use Static Samples check boxes.

    Now expand the 1X Orbit/Timebase Plot Group by

    right clicking and choosing Expand All (Figure 14).

    FIGURE 14: Example of expanded plot groups.

    Drag the first Orbit/Timebase Plot from the 1X plot group

    (Figure 15) up into the top plot group. Do this by clicking

    on the highlighted plot, and dragging it with your mouse

    up to just below the first Orbit/Timebase plot in the top

    group of the plot session. As you are completing this, your

    result will look something like the example in Figure 16.

    FIGURE 15: Selecting the first Orbit/Timebase Plot for dragging.

    Observe that as you drag the plot up, the pointer

    will change from a circle with a slash through it to a

    horizontal straight line. When the insertion point it is

    where you want it to be in the hierarchy, drop it in.

    FIGURE 16: The first Orbit/Timebase plot will be dropped at the horizontal line insertion point.

    After dropping the plot at the insertion point, your

    result should be similar to the example in Figure 17:

    FIGURE 17: Example of Plot Session Manager Hierarchy after the plot has been dragged to the insertion point.

    Repeat this procedure with the remaining channels,

    to drag their associated plots up into the appropriate

    plot group. My results are shown in Figure 18.

    FIGURE 18: Example of Plot Session Manager Hierarchy after all required plots have been dragged up into the appropriate plot group.

    22 ORBIT Vol .32 No.2 Apr.2012

    DEPARTMENTS

    ADRE* TIPS

  • Just to confirm you have put the plots where you want

    them, use the Expand All command on the top Orbit/

    Timebase plot group in the hierarchy (Figure 19):

    FIGURE 19: Example showing that all four needed plot sessions (outlined in red) have been dragged into the top plot group in the Plot Session Hierarchy.

    Since this looks good, we can go ahead and delete

    the unneeded 1X Orbit/Timebase Plot Group (Right

    click on the group and select Delete (Figure 20).

    Note: The reason we dragged these new plots into

    the new plots is to enact plot overlays in ADRE. We

    started with two distinctly separate plot groups one

    is for Direct data, and one is for 1X data. By drag-

    ging and dropping the 1X plot into the Direct plot,

    we can see both sets of data in the same plot.

    FIGURE 20: Selecting the unneeded 1X Orbit/Timebase Plot Group for deletion.

    This step is optional, but helps to keep things clearly

    labeled: Rename the New Orbit/Timebase Plot Group to

    Direct and 1X Orbit/Timebase Plot Group (Figure 21).

    FIGURE 21: In this example, we renamed the new plot group (in red outline box) with a descriptive name.

    IVE ALWAYS DESCRIBED THE ORBIT PLOT AS WHAT THE SHAFT WOULD

    DRAW IF THERE WERE A PENCIL LEAD AT ITS CENTERLINE, AND YOU

    HELD A PIECE OF PAPER UP TO IT. SEEING WHAT THE SHAFT IS DOING

    GRAPHICALLY ALLOWS YOU TO INTERPRET WHAT IT IS DOING

    MECHANICALLY, AND FROM THERE A DIAGNOSIS MAY BE MADE.

    Apr.2012 No.2 Vol .32 ORBIT 23

    DEPARTMENTS

    ADRE* TIPS

  • 4. HOUSEKEEPING, OR MAKING IT LOOK GOOD

    In the plot group configuration (bottom part of

    configuration window), there are a few things that we

    can do to make the plots tidy, and simpler to navigate:

    Change the Paging Mode to By Sample (Figure 22).

    If you leave it set for paging mode By Channel, your

    plots will only show data from one channel you will

    scroll through all of the data for that channel, and

    then you will scroll through all of the data for the next

    channel, and so forth. Typically, I want to see what is

    happening on all of the channels at the same instant

    in time. This is much more meaningful to me.

    FIGURE 22: Selecting By Sample from the drop-down list.

    Change the plot layout to display the plots

    appropriately. Either show 2 plots per page

    (Figure 23) or 4 plots per page (Figure 24):

    FIGURE 23: The 2 x 1 option will show two plots per page.

    FIGURE 24: The 2 x 2 option will show four plots per page.

    Click OK on the plot configuration to close it when you

    are finished.

    5. OPEN THE PLOT

    If you used a plot layout of two plots per page, the 1X

    Filtered plot will be located below the unfiltered (direct)

    plot (Figure 25):

    FIGURE 25: Example with two plots per page.

    If you used a layout with four plots per page mode, they

    are displayed in a two-by-two arrangement (Figure 26).

    FIGURE 26: Example with four plots per page.

    Re-ordering the plots in the plot tree will change

    the arrangement of the four plots on the page:

    For instance, by changing the order, you could

    make the top plots show unfiltered data, with the

    bottom two plots showing 1X filtered data.

    For analyzing data that was collected from steady state

    (constant speed) conditions, it is recommended that you

    set scaling to auto all plots. This way, it is easy to see

    which bearing is giving you a problem. If you are observing

    a startup or shutdown, setting manual scaling may be

    better, as the auto all plots will scale for the extremes

    (such as when the machine passes through a resonance).

    Hopefully, this tip will make you more productive and help

    you diagnose machinery problems a little more easily.

    See you the next time the Keyphasor* comes around!

    * denotes a trademark of Bently Nevada, Inc., a wholly owned subsidiary of General Electric Company. Copyright 2012 General Electric Company. All rights reserved.

    24 ORBIT Vol .32 No.2 Apr.2012

    DEPARTMENTS

    ADRE* TIPS

  • [This is the first installment in a continuing series of Application Notes on important topics regarding the effective use of Bently Nevada* products. Watch for additional Application Notes topics in future issuesEditor]

    Resources for Managing Electrical Runout

    Electrical Runout is the term used to describe the unwanted signal from an eddy current probe due to variations in material properties. Many machines are required to meet specifications limiting the baseline vibration. Electrical runout can cause problems with acceptance of new

    machinery, or with diagnostics of machinery with low levels of vibration.

    The first step in obtaining an accurate measurement is ensuring that the eddy current probe is installed

    optimally. Once there is confidence in the probe installation, we are ready to evaluate the causes and take appropriate corrective actions for electrical runout problems.

    The Orbit article at the following link was written by Nate Littrell. It includes useful guidance for probe installation, as well as for evaluating the causes of electrical runout, and planning effective corrective actions:

    http://www.ge-mcs.com/download/orbit-archives/2001-2005/3q2005_runout.pdf

    * Denotes a trademark of Bently Nevada, Inc., a wholly owned subsidiary of General Electric Company. Copyright 2012 General Electric Company. All rights reserved.

    DEPARTMENTS

    APPLICATION NOTE

    Apr.2012 No.2 Vol .32 ORBIT 25

  • Vibration Data Identifies Hot Spot on Motor Rotor

    Roengchai Chumai

    Technical Leader

    Bently Nevada Machinery

    Diagnostics Services

    [email protected]

    Executive Summary

    This case history describes how vibration analysis

    identified a thermally-sensitive rotor in an induction

    motor driving a condensate pump at a newly com-

    missioned power plant in Thailand. Induction motors

    have characteristic vibration behavior based on the

    electrical, magnetic and mechanical effects that they

    experience. In order to capture the significant data for

    analysis, it is important to perform test procedures

    and data collection very carefully. In some situations,

    problems with machine casings, mounting foundations

    and associated structures and even the configuration

    of the driven machine can influence motor vibration

    behavior. All of these factors should be taken in account.

    In this particular case, the analysis was concerned

    with increasing vibration amplitude that occurred

    when the pump was running at loaded conditions.

    The phase angle of 1X filtered vibration kept changing

    over the running period and a thermal vector was

    identified. It was suspected that a hot spot was

    causing the rotor to bow, so a repeatability test was

    performed to check for consistency of its location.

    The motor was run uncoupled from its pump to

    reduce load and therefore operating temperature of

    the rotor. Testing verified that the rotor did indeed

    straighten out when it was run solo, which indicated

    that the thermally-induced bow had gone away, and

    with it, the previously-observed high 1X vibration.

    Shop inspection showed clear evidence of burnt rotor

    insulation resin on the rotor surface, which verified the

    location of the hot spot that caused the thermal bow.

    Background and Sequence of EventsOne block of the newly-commissioned combined-cycle

    power plant includes three vertical motor-driven

    condensate pumps. The drive motors are of 4-pole

    induction design, with synchronous speed of 1500

    Hz (50 Hz power supply). The drive motors operate at

    constant speed, and condensate flow is controlled

    26 ORBIT Vol .32 No.2 Apr.2012

    DEPARTMENTS

    CASE HISTORIES

  • by throttle valves. The multistage centrifugal

    water pumps are of typical canned design.

    Motor Nameplate Data Power: 250 hp (186 kW)

    Speed: 1485 rpm

    Power Supply: 380 vac, 3 phase, 50 Hz

    Current: 347 amp

    Service Factor 1.15

    Time Rating: Continuous

    Power Factor: 0.855

    During initial plant startup activities, all three

    condensate pumps exhibited high vibration at the

    motor Non-Drive End (NDE). Since the steam plant

    could not be started without all three of these pumps

    running, the project construction contractor called

    Bently Nevada* Machinery Diagnostic Services

    (MDS) to assist with vibration testing and analysis

    to determine the root cause and to provide on-site

    advisory for resolution of the high vibration conditions.

    The MDS Engineer arrived at the plant site and

    discussed the history of the problem with the customer.

    He set up portable data acquisition instruments

    and temporarily-installed vibration transducers for

    individual testing of all three condensate pumps.

    InstrumentationAn orthogonal (perpendicular) pair of radial velocity

    transducers was installed at both the Non-Drive End

    (NDE) and the Drive End (DE) of each motor while it

    was being tested. The temporary installation also

    included an optical Keyphasor* sensor for providing

    reference phase angle and supplementary machine

    speed measurement (Figure 1). All vibration signals

    were sent to an ADRE* Data Acquisition Interface

    Unit (DAIU) using coaxial cables. A laptop computer

    running ADRE for Windows software was connected

    to the DAIU for capturing, digitizing and presenting

    vibration data in a variety of plot formats for analysis.

    BANGKOK SPARKLES LIKE A JEWEL AT NIGHT, SYMBOLIZING THE IMPORTANCE OF ELECTRICAL POWER TO THAILANDS ECONOMY.

    Apr.2012 No.2 Vol .32 ORBIT 2

    DEPARTMENTS

    CASE HISTORIES

  • FIGURE 1: An orthogonal pair of velocity sensors is temporarily installed at the DE and NDE of one of the condensate pump drive motors. An optical sensor (not visible in this photo) was also installed for direct observation of a one event per turn feature on the rotating shaft which was provided by a strip of reflective tape.

    FIGURE 2: Trim balance correction weight installed at motor rotor NDE location. The motor dust cover has been removed to reveal the weight plane. Observe that one bolt of appropriate mass has been threaded into the required location to provide trim balance.

  • Initial Investigation & Actions

    Based on available vibration data, it was discovered

    that all three of the pumps (units A, B & C) were

    experiencing higher than acceptable vibration due

    to unbalance. The MDS Engineer trim balanced the

    pumps to acceptable vibration levels based on ISO

    10816-3 Standard, by adding appropriate correction

    weights at the NDE of each motor rotor (Figure 2).

    Additional Pump A TestingAfter trim balancing, Pumps B & C behaved normally.

    However Pump A exhibited an unusual change in the

    vibration amplitude and phase over the observed

    running period (Figures 3 and 4). These trend plots show

    the change of vibration amplitude over a time period

    of just over 4 hours at constant speed and load.

    This kind of behavior often indicates a thermally-sensitive

    rotor that bows during operation due to a hot spot caused

    by a local fault in the rotor. With such a fault, the amount

    of heating and thermal bow typically depends on motor

    load, and the associated current flow in the rotor iron.

    With the Pump A motor coupled with its pump, plant

    personnel performed alignment checks and then started

    the pump and ran it at a steady state operating condition.

    Vibration data was captured throughout the running

    period. Motor soft foot and pump baseplate rocking

    effects were checked using phase angle relationships

    of the vibration timebase waveforms. These evaluations

    verified there was no sign of these possible problems.

    FIGURE 3: First test run: Four hour trend plot of vibration phase (upper plot) and amplitude (lower plot) from the 1XV sensor (oriented to the 0 degree North reference) at the NDE location of the Pump A drive motor. Sensor names in these vibration plots correspond to labeling in the photo of Figure 1.

    FIGURE 4: First test run: Four hour trend of vibration phase and amplitude from the 1YV sensor (oriented 90 degrees to the right of the North reference when viewed from the driver toward the driven load.

    Apr.2012 No.2 Vol .32 ORBIT 2

    DEPARTMENTS

    CASE HISTORIES

  • FIGURE 5: Second test run: Four hour trend of vibration phase and amplitude from the 1XV sensor. Results are consistent with the first test run.

    FIGURE 6: Second test run: Four hour trend of vibration phase and amplitude from the 1YV sensor. Results are consistent with the first test run.

    FIGURE 7: First test run shows a 1X vibration vector that changed significantly in both amplitude and phase lag angle over the period of the test.

    FIGURE 9: Vibration data from the 1XV sensor for a 2-hour solo run. As the rotor cooled, the Direct and 1X vibration levels dropped significantly.

    FIGURE 8: Second test run shows almost exactly the same results as the first run.

    FIGURE 10: Vibration data from the 1YV sensor for a 2-hour solo run. As the rotor cooled, the Direct and 1X vibration levels dropped significantly.

    30 ORBIT Vol .32 No.2 Apr.2012

    DEPARTMENTS

    CASE HISTORIES

  • Testing Under Load

    In order to check for repeatability of the temperature

    dependent vibration effects, a second test was run

    the following day, after shutting down the motor and

    allowing it to cool. Similar results were obtained (Figures

    5 and 6). This validation meant that the location of the

    suspected rotor hot spot was indeed at a fixed position.

    The vibration data for these two test runs was also

    plotted in a polar format. Again, the results of the

    two runs were consistent, and the effects of the

    thermal vector are quite apparent (Figures 7 & 8).

    No-Load Solo Testing

    Immediately after the pump was shut down after the

    second series of loaded testing, the motor was uncoupled

    and run solo. Since the rotor was still hot and thermally

    bowed, the vibration amplitude was relatively high at

    the beginning of the test. However, as shown in Figures

    9 and 10, vibration amplitude dropped as the rotor

    cooled and straightened out during the uncoupled run.

    Note: The observed cyclic variation in vibration amplitude can be a classic symptom of broken or cracked rotor bars, or high-resistance joints between the bars and the rotor end rings. In electrical diagnostics

    FIGURE 11: Polar plots show the 1X vibration vector for the motor NDE and DE during its solo run. This data shows that rotor was bowed when hot, as it had high 1X vibration amplitudes with the same phase angle at both ends of the machine. As the amplitude dropped, the data points can be seen moving closer to the center (zero amplitude) of the plot. Finally, the classic phase shift can be seen as the motor is tripped and coasts down (as indicated by rpm labels).

    (not performed in this particular case), the motor current often shows corresponding fluctuations, that correspond to the pole-pass frequency of the motor. In fact, this slow modulation of motor vibration often produces an audible low-frequency beating sound. We will look more closely at this symptom near the end of this article.

    Polar plots of 1X vibration amplitude and phase (Figure 11)

    verifies that the rotor was straightening out as it cooled,

    causing reductions in amplitude and changes in phase.

    RecommendationsBased on review of the vibration data, the fol-

    lowing recommendations were made:

    The rotor iron should be inspected for any evidence

    of lamination smear region, which would indicate

    local heating and generate a hot spot on the rotor.

    Check for proper function of the rotor cool-

    ing air system. It is possible that non-uniform

    airflow or a plugged flow path within the motor

    can contribute to abnormal rotor heating.

    Closely monitor the vibration amplitude of the subject

    unit to ensure that it does not increase with time. The

    existing rotor should be replaced for a permanent repair.

    Inspection ResultsThe rotor inspection showed clear evidence of local

    overheating, as indicated by a small discolored spot

    where rotor insulating resin had seeped to the surface

    of the iron laminations and charred (Figure 12).

    THE ENTIRE PROJECT COST WAS

    APPROXIMATELY 250K US DOLLARS,

    SO THE NEW ONLINE SYSTEM

    MORE THAN PAID FOR ITSELF

    IMMEDIATELY AFTER INSTALLATION."

    Apr.2012 No.2 Vol .32 ORBIT 31

    DEPARTMENTS

    CASE HISTORIES

  • A hot spot in the laminated core of the

    Pump A induction motor rotor caused

    uneven thermal expansion along the

    rotor (more expansion on the side of

    the rotor with the hot spot, and less

    expansion on the undamaged side of

    the rotor). This thermal bow resulted in

    increasing synchronous (1X) vibration

    amplitude under loaded conditions,

    with in-phase vibration measure-

    ments at both ends of the rotor. Since

    thermal bowing is related to rotor

    current, the motor was also tested

    under unloaded solo conditions to

    check whether the bow would relax as

    the rotor cooled from hot conditions,

    and straightened out. This effect was

    observed, validating the diagnosis.

    Pump A Corrective Actions

    The customer ordered a new rotor

    to replace the damaged rotor.

    However, the lead time for a new

    rotor was approximately 3 months.

    So in the interim, the existing rotor

    was rebuilt for temporary use during

    the plant commissioning period.

    After rebuilding, it was balanced at

    the repair shop and then returned to

    the generating site for installation.

    Post-Repair Symptoms

    The Bently Nevada MDS Engineer

    was requested to visit the site again

    when the Pump A motor (now with

    rebuilt rotor installed) was coupled

    to its condensate pump. Vibration

    testing was performed following

    FIGURE 12: The local high temperature at the hot spot showed up as a heat-damaged area on the surface of the rotor.

    CONCLUSIONS

    32 ORBIT Vol .32 No.2 Apr.2012

    DEPARTMENTS

    CASE HISTORIES

  • BACK-TO-BASICS SIDEBAR

    Motor Synchronous Speed120f = np, where

    f is the frequency of the power supply

    n is the speed of the machine in rpm (or cpm)

    p is the number of poles

    n = 120f/p = 120 (50)/4 = 1500 rpm (or cpm)

    So with a power supply frequency of 50 Hz, a 4-pole

    induction motor has a synchronous (no-slip) running

    speed of exactly 1500 rpm.

    Synchronous Vibration Frequency1500 cpm / (60 Hz/cpm) = 25 Hz.

    So with no slip, synchronous vibration frequency at

    1500 rpm is 25 Hz.

    Slip FrequencySince the motor was actually running at 1494 rpm,

    the slip frequency was:

    1500 cpm 1494 cpm = 6 cpm.

    Converting to Hz6 cpm / (60 cpm/Hz) gives a slip frequency of 0.1 Hz.

    Pole Pass Frequency (slip frequency x number of poles):

    (0.1 Hz)(4 poles) = 0.4 Hz.

    Actual 1X Frequency for Running SpeedWith the motor running at 1494 rpm, actual 1X

    frequency is found by converting to Hz:

    1494 cpm /(60 Hz/cpm) = 24.9 Hz.

    SummaryThe vibration spectrum example in Figure 13 shows

    a 1X peak at 24.9 Hz, with small sideband peaks at

    0.4 Hz.

    the same procedures that were used earlier. This time,

    no signs of thermal sensitivity were observed.

    However, high synchronous vibration was observed

    at the motor NDE, with a small amount of amplitude

    modulation at pole passing frequency (slip frequency

    times number of poles). This indicated that there was

    significant mechanical unbalance, with a small amount of

    modulation caused by rotor bars that were still cracked

    or broken or high resistance joints that still existed

    between rotor bars and shorting rings at the rotor ends.

    The vibration spectrum in Figure 13 shows a peak at 24.9

    Hz center frequency, which corresponds to synchronous

    (1X) vibration for the running speed of 1494 rpm.

    Sidebands are seen at about 0.4 Hz, which corresponded

    to the pole pass frequency (see sidebar for calculations).

    Since vibration amplitudes at the sideband frequency

    components were relatively low compared with

    synchronous vibration amplitude, it was determined

    that no immediate electrical work was required on

    the rebuilt rotor. The unit was trim balanced at solo

    run (uncoupled) to reduce synchronous vibration

    amplitude down to acceptable levels, then the motor

    was re-coupled to the pump. The unit was returned to

    service for normal operation until the new rotor could

    be delivered to site for permanent replacement.

    * denotes a trademark of Bently Nevada, Inc., a wholly-owned subsidiary of General Electric Company. Copyright 2012 General Electric Company. All rights reserved.

    FIGURE 13: Vibration spectrum measured at motor NDE of the rebuilt rotor showing predominant frequency at synchronous (1X) with sideband components at motor pole pass frequency.

    Apr.2012 No.2 Vol .32 ORBIT 33

    DEPARTMENTS

    CASE HISTORIES

  • [This is the first installment in a mini-series of Recip Tip articles that is planned by our experienced Italian Field Application Engineer (FAE), Gaia Rossi. Editor]

    Vibration Analysis for Reciprocating Compressors (Part 1)

    Gaia Rossi

    Bently Nevada Field

    Application Engineer

    [email protected]

    Vibration analysis of reciprocating machines creates some unique challenges. This article explains the reasons and gives clarity on recommended monitoring and analysis practices and tools. Years of field experience have demonstrated that techniques which may be well understood for measuring and analyzing the vibration of purely rotating machinery can produce confusing results when applied to reciprocating machinery.

    Vibration associated with rotational speed is the dominant motion for most industrial rotating machines. This synchronous (1X) behavior allows the direct application of traditional vibration analysis concepts towards addressing common machinery malfunctions such as rotor unbalance. The typical frequencies

    observed with those common rotor-related malfunctions generally occur between a quarter of running speed and twice running speed and correlate excellently with machine mechanical conditions. Consequently, principles and diagnostic methodologies for these machines are broadly accepted and harmonized within the machinery diagnostic community.

    This is not quite true for reciprocating compressors. Vibration analysis of these machines creates some unique challenges; many forcing functions produce a complex vibration signature that makes any attempt of using standard analysis techniques used for rotating equipment ineffective.

    34 ORBIT Vol .32 No.2 Apr.2012

    DEPARTMENTS

    RECIP TIPS

  • FIGURE 1: This drawing shows typical vibration monitoring locations for a reciprocating compressor. Sensors are installed at the crosshead guides (4 red hexagons) and on the frame (4 blue diamonds). [Reference 1]

    Apr.2012 No.2 Vol .32 ORBIT 35

    DEPARTMENTS

    RECIP TIPS

  • Compressor Frame Vibration

    Vibration measured at the frame

    results principally from the response

    of the mechanical system to the

    forces and moments that are

    occurring in the machine at the

    normal running conditions. These

    include the following factors:

    Gas Load Forces: These forces act

    on the piston and stationary compo-

    nents at 1X and at integer multiples

    of running speed. They are generally

    significant up to about 10X and in the

    direction of the piston rod travel. For

    large slow speed compressors (up to

    roughly 500 rpm), gas forces are typi-

    cally the largest contributor to piston

    rod and compressor frame load.

    Inertial Load Forces: These forces

    are caused by the acceleration

    of the reciprocating components

    (piston, piston rod, and crosshead).

    These components represent

    a large amount of mass to be

    accelerated back and forth with

    each stroke. Inertial loads of

    400,000 Newton (~90,000 pounds)

    of force or more are not uncommon

    with very large compressors.

    Reciprocating & Rotating Masses

    Unbalance Forces: These forces

    are predominant at 1X and 2X

    compressor speed, and are caused

    by asymmetrical crankshaft design

    and imperfect manufacturing toler-

    ances. They are usually much smaller

    than inertial and gas load forces.

    FIGURE 2: Time waveform plot of the velocity signal from a frame-mounted vibration sensor. Observe that many different frequency components are present in the signal.

    FIGURE 3: Frequency domain (spectrum) plot of velocity signal shown in Figure 2. Fast Fourier Transform (FFT) processing allows us to see the various frequency components that are included in the complex waveform.

    Gas Unbalance Forces: These are

    caused by pressure in the pulsation

    bottles and pulsation at the cylinder

    nozzle area and on piping. Allowable

    pulsation levels are defined in API-618.

    Although these pulsating forces are

    usually much smaller than the forces

    listed above, they can be destructive

    to piping and piping support systems

    if they happen to correspond to reso-

    nant frequencies for the structures.

    As a consequence of these factors,

    the extent of vibration is inherent

    with the reciprocating compressor

    design and its response to all the

    applied forces and moments. This

    causes these machines, even when

    in good condition, to vibrate much

    more than a comparable rotating

    machine. The examples in Figures

    2 and 3 show that many harmonics

    are produced by the complex shape

    of the frame velocity waveform.36 ORBIT Vol .32 No.2 Apr.2012

    DEPARTMENTS

    RECIP TIPS

  • Frame vibration frequencies typically

    include components below 10 Hz.

    For this reason, a velocity transducer

    (with extended low frequency

    response) is usually better suited than

    an accelerometer for detecting an

    increase of rotation-related forces

    (due to gas load or inertial loads,

    imbalance, foundation looseness,

    excessive rod load, etc.). The preferred

    location for the frame vibration

    transducer is on the side of the frame

    oriented in the direction of piston

    rod travel, on the centerline of the

    crankshaft and at a main bearing

    where dynamic load is transmitted

    (Figure 1). Magnitude for a filtered

    frame velocity signal is usually low

    (less than 7 mm/s); however, at low

    frequencies, even small amplitudes of

    measured velocity may correspond

    to large amounts of displacement.

    On the other hand, measuring only

    frame vibration can be insufficient

    for effective condition monitoring,

    as the increase in frame velocity

    from incipient failures developing

    at the running gear or cylinder

    assembly will be small and typically

    covered by the larger signal that

    is produced by normal machine

    movement. Experience has shown

    that by the time the malfunction has

    been detected by the frame velocity

    transducer and the compressor shut

    down, major secondary damage may

    have already occurred because of

    the malfunctions. These malfunctions

    include liquid or debris carryover,

    loose piston or piston nut, loose

    crosshead nut, or loose cylinder liner,

    and typically manifest themselves as

    impacts transmitted at the crosshead.

    Monitoring Vibration & Impact

    Vibration transducers monitoring

    rotating machinery generate station-

    ary signals; this means they have

    constant frequency content over each

    revolution of the rotor (Figure 4).

    In contrast, vibration measurements

    on reciprocating compressors present

    both stationary and non-stationary

    content. In particular, the signal gen-

    erated by an accelerometer placed

    vertically on a crosshead guide is

    characterized by different frequencies

    with different amplitudes that occur

    at specific points in the revolution.

    Figure 5 shows a typical waveform

    from a crosshead accelerometer.

    The signal shows high amplitude,

    FIGURE 4: Example of stationary vibration sample taken at an electric motor bearing. The higher frequency components are typical of the characteristic vibration produced by the interaction of the rolling elements with the bearing races.

    FIGURE 5: Timebase waveform of a crosshead acceleration signal.

    Apr.2012 No.2 Vol .32 ORBIT 3

    DEPARTMENTS

    RECIP TIPS

  • short duration impulse peaks fol-

    lowed by a ring down that occur

    at certain parts of each crankshaft

    revolution. This signal is not filtered

    so the transducer is picking up

    the widest range of frequencies

    (typically from 10 Hz to 30 kHz).

    These acceleration peaks can be

    referred as responses to impulse

    events occurring during compressor

    operation (valve opening and closing,

    gas flow turbulence, crosshead

    pin shifting at load reversal, etc.).

    Such impulses excite the structural

    resonances of the machine compo-

    nents - resulting in high frequency

    free vibration and the characteristic

    impact/ring-down profile.

    As mentioned, the main source of

    vibration on the compressor frame

    is related to periodic forces. While

    the overall frame vibration increase

    is certainly a concern, the primary

    interest of crosshead vibration

    monitoring is detecting peaks

    associated with structure response

    to impulsive events. Conditions

    that increase the excitation of such

    resonances are generated by develop-

    ing faults such as fractured or loose

    components or excess clearance.

    Loose rod nuts, loose bolts, excessive

    crosshead slipper clearance, worn

    pins as well as liquid in the process

    can be detected at early stages of

    development using crosshead impact

    monitoring, thus allowing appropriate

    countermeasures and avoiding

    potential catastrophic consequences.

    Of all vibration measurements that

    can be applied to reciprocating

    compressors, crosshead accelera-

    tion is probably the most effective

    protection measurement available,

    if appropriately employed.

    While crosshead acceleration has

    proven itself to be a sound measure-

    ment for detecting mechanical

    failures, industry has little experience

    in applying and analyzing it, resulting

    in increased risks of false or missed

    alarms, and poor diagnostic value

    from diagnostic systems. The follow-

    ing paragraphs describe some basic

    requirements for a reliable monitoring

    system and diagnostic software.

    Requirements for Monitoring Systems

    General considerations on the

    effective employment of crosshead

    acceleration for monitoring and

    protection are described here:

    Transducer SelectionAmplitude measurement units should

    be generally selected based upon the

    frequencies of interest. For crosshead

    vibration monitoring an accelerometer

    should be selected as it emphasizes

    the higher frequency components.

    The unit of measurement used should

    be the natural units of the transducer

    used (signal integration is not a

    recommended tool for this purpose).

    Transducer MountingFrequency response is sensitive

    to mounting techniques and may

    be affected by any reduction of

    the mechanical coupling between

    accelerometer and mounting surface

    such as the use of an adhesive,

    magnetic isolation base, or non-flat

    mounting surface. The transducer

    should be installed directly on the

    machine structural component to

    be measured, avoiding brackets or

    plates as a support, or mounting on

    flanges or covers. Accuracy of an

    accelerometer can also be affected

    by ground loops, base strains, and

    cable noise. These can be minimized

    by following the recommendations

    from transducers and monitoring

    systems manufacturers as well as

    applying appropriate cable tie-downs.

    Signal Processing & Alarming

    One of the concerns in applying

    crosshead vibration measurement

    for compressor shutdown is the risk

    of false alarms due to spurious peaks

    in the signal. The peak detection

    circuit in the protection system should

    be designed to manage impulsive

    vibration in order to avoid nuisance

    alarms; this can be accomplished

    by counting the number of readings

    that exceed an alarm threshold in a

    set time before triggering an alarm.

    Additionally, an appropriate time delay

    needs to be configured for the alert

    and shutdown thresholds. Careful set-

    ting of these thresholds, counts and

    alarm delays will allow us to minimize

    the possibility of false alarms. The

    recip Impact/Impulse channels

    in the Bently Nevada* 3500/70M

    monitor include these features.

    Signal FilteringAnother essential aspect to care-

    fully consider is signal filtering. As

    38 ORBIT Vol .32 No.2 Apr.2012

    DEPARTMENTS

    RECIP TIPS

  • described previously, an accelerom-

    eter can detect vibration components

    up to very high frequencies. While

    acceleration analysis in a broad

    frequency range may have diagnostic

    value, the main object of crosshead

    impact monitoring is protecting the

    machine from the consequences

    of mechanical failures. A signal

    with too high corner frequency for

    the low-pass filter may introduce

    the risk of false alarms due to the

    presence of high frequency content

    not related to mechanical malfunc-

    tions (and consequent impacts

    transmitted to the crosshead guide

    Amplitude Measurement

    Our last important note is about

    vibration measurements taken in

    either root mean square (rms), zero-

    to-peak (peak or pk), or peak-to-peak

    (pp) amplitude measurement systems.

    A few international standards

    recommend rms measurement for

    assessing machinery health based

    on overall casing vibration and this

    is traditionally adopted by many

    practitioners. Rms values provide an

    indication of the energy content of a

    signal, and for malfunctions such as

    loose foundation or load unbalance,

    this energy content relates well with

    machine condition, as well as opera-

    tor perception of machine condition.

    However, rms calculation applied to

    an impulsive frequency-rich signal

    such as crosshead vibration (Figure

    5) does a poor job in correlating with

    other critical conditions such as

    mechanical knocks, which have rela-

    tively little energy content, but prove

    vital in assessing machine condition.

    For these types of malfunctions,

    peak amplitude measurement is

    recommended as it correlates

    well with both high-energy and

    low-energy malfunctions typical of

    reciprocating compressors. Applying

    rms processing to crosshead

    vibration signals would provide

    under-predicting values.

    Crank Angle Domain Analysis

    When viewed in the time domain, the

    non-stationary crosshead vibration

    signal looks like multiple disconnected

    events (Figure 5), so diagnostic

    methodologies such as spectral

    analysis provide little value due to the

    discontinuous frequencies involved.

    The most appropriate analytic

    methodology is therefore based on

    signal timing; Bently Nevada 3500

    monitors synchronize the vibration

    signal with crankshaft rotation to

    associate peaks to a piston posi-

    tion along the stroke. Individual

    monitoring and alarming on crank

    angle bands allows association

    of peaks to the problem area.

    For example, a peak occurring when

    the piston is travelling toward the end

    of its stroke near Top Dead Center

    (TDC) can be correlated to liquid or

    debris ingression in the compression

    chamber. When the piston moves

    towards its TDC position, the impact

    with the non-compressible material

    will generate an impulse event. The

    monitoring system will then raise

    an alarm for the corresponding

    crank angle band (for example,

    starting 10 degrees before top

    dead center and ending 10 degrees

    after). Figure 6 shows case of

    liquid ingestion as detected by the

    crosshead guide accelerometer.

    YEARS OF FIELD EXPERIENCE HAVE DEMONSTRATED THAT TECHNIQUES WHICH MAY BE WELL UNDERSTOOD FOR MEASURING AND ANALYZING THE VIBRATION OF PURELY ROTATING MACHINERY CAN PRODUCE CONFUSING RESULTS WHEN APPLIED TO RECIPROCATING MACHINERY.

    Apr.2012 No.2 Vol .32 ORBIT 3

    DEPARTMENTS

    RECIP TIPS

  • FIGURE 6: Crosshead acceleration in crank angle domain, presenting a high peak at Top Dead Center (TDC). The horizontal axis represents 360 degrees of crankshaft rotation (one full revolution), where 0 indicates TDC. The System 1 plot also displays a Throw Animation (in the upper right corner of this plot) showing the piston movement synchronized with the plot cursor. In this example the cursor is set at 2.5 degrees, and the animation shows that the piston is very close to the TDC position

    FIGURE 7: The 3500/70M module returns two waveform samples to System 1 software from a single crosshead acceleration signal with two different filtering characteristics.

    Understanding Frequency ContentAdditional advanced analysis tools

    are available in System 1* diagnostic

    software. As noted before, not all

    impulse response events within the

    crosshead accelerometer signal

    contain the same frequencies.

    Mechanical knocks excite resonances

    of the reciprocating compressor

    components such as crosshead

    guides, distance pieces, etc. that

    generally lie below 2 kHz. In contrast,

    events originating in gas flow noise,

    valve opening or valve closing events

    express a much higher frequency.

    Searching for a mechanical event in

    an acceleration signal that contains

    the whole transducer frequency

    response range is practically impos-

    sible due to the high amplitude and

    frequency peaks that cover smaller,

    yet more critical, peaks related to

    mechanical events. Such overlap

    prevents early indication of an incipi-

    ent malfunction. It is for this reason

    the signal must be filtered. Figure

    7 shows crosshead acceleration in

    the crank angle domain using 3 to

    30 kHz (left plot) and 3 to 2 kHz (right

    plot) band pass filtering. The peaks

    present in the narrower pass-band

    correspond to mechanical impacts,

    which are difficult to distinguish in

    the signal with broader filter corners.

    System 1 software is integrated with

    the 3500/70M monitor to allow dual

    signal processing and both storing

    and displaying the accelerometer

    signal with two different filter settings.

    40 ORBIT Vol .32 No.2 Apr.2012

    DEPARTMENTS

    RECIP TIPS

  • Diagnostic Approach

    To wrap up this first installment, let us

    consider how we can effectively asso-

    ciate a malfunction to a specific vibra-

    tion pattern and to obtain an early

    failure diagnostic. Experience has

    shown that associating vibration with

    additional measured dynamic param-

    eters such as rod load have proven

    to be of great value in pinpoint