gas train

8/3/2019 gas train

1/12

58 ORBIT Vol .28 No.1 2008

CASE HISTORY

8/3/2019 gas train

2/12

Vol .28 No.1 2008 ORBIT59

CASE HISTORY

n ethylene plant experienced a vibration isse after start-p of a mlti-nit chare as train

followin a relatively short mitiation otae in Ast 2006. The chare as compressor

has five staes of compression via three different compressor bodies. All three are driven by

a 35,000 hp steam trbine with a desin speed of 5,300 rpm (Fire 1). The compressors were

oriinally installed in the late 1960s and still had the oriinal floatin oil rin desin and

lbricated ear coplins. The rotors had been praded with rb-tolerant internal seals and spherically

seated tilt-pad bearins.

The short dration of the otae limited the wor scope to overhals of the A and C compressor bodies

only. The C compressor had exhibited performance problems thoht to be cased by folin, while the

A compressor was overhaled as a reslt of recent vibration isses. The B compressor was not opened p

becase it had no mechanical or performance problems. The inboard bearin on the dischare end of the

B compressor (bearin 5) was chaned becase it was felt that the clearance had increased.

The train is monitored via a Bently Nevada* 3500 series machinery protection system and all critical trains

in the ethylene nit were connected to Bently Nevada Data Manaer* 2000 (DM2000) software to provide

continos condition monitorin and machinery dianostic capabilities. For increased

data collection resoltion, an ADRE* portable data acisition system was sed to captre

shtdown and startp transient data and ament the data available

from the online software.

Dianosin

Wallace E. (Ed) Wilcox, P.E. Staff Machinery Enineer Chevron Enery Technoloy Company [email protected]

a coplin problem on a chare as trainHow vibration data and a loical process of elimination saved millions in downtime

The eThylene unIT deTaIled In ThIs case

hIsTORy has suBsequenTly Been upgRaded

TO sysTem 1* sOfTwaRe whIch addResses

many Of The daTa ResOluTIOn lImITaTIOns

ThaT weRe InheRenT In dm2000 sOfTwaRe.

8/3/2019 gas train

3/12

60 ORBIT Vol .28 No.1 2008

CASE HISTORY

5Y 5X2Y 2X1Y 1X 6Y 6X

7Y 7X4Y 4X3Y 3X 8Y 8X

CompressorBody C1 2

3 4

4

5 6

7 8Compressor

Body B

CompressorBody A

35,000 hpSteam Turbine

View as seen from turbine lookingtowards driven machines

421 1T 53

Figure 1. Machine train diagram o charge gas compressor train showing A-B-C compressor bodies,fve compression stages, and proximity probe arrangements or all eight radial bearings.

Problem

Approximately 10 mintes after the chare as

compressor train was restarted followin the otae,

radial vibration on the B compressor increased rapidly.

Althoh the speed remained basically nchaned, the

radial vibration went from 0.5 to over 3 mils, as can be

noted in Fire 2. This increase in vibration occrred

immediately after the compressor train sred. Note

the lare increase in vibration amplitde even thoh

the speed remained relatively constant on the riht sideof Fire 2. The vibration amplitde contined to rise

after startp as the load on the compressor train was

increased in response to increasin nit rates.

Overall amplitdes had reached approximately 4.5

mils by early September with a short excrsion in

late September above 5.5 mils.

Figure 2. Bode plot o bearing #6 y-probe during

start-up on 8/27/06. Note substantial change

in vibration amplitude while speed remains

essentially constant at approximately 4,300 rpm(region enclosed by the red oval). The blue cursor

position shows the frst balance resonance or

this machine occurs at about 2400 rpm.

8/3/2019 gas train

4/12


CASE HISTORY

Vibration Analysis

Analysis of the problem was essentially a process of

elimination whereby the available data was examinedto systematically determine what was not casin the

vibration. Five separate malfnctions where considered.

1. Unbalance

An initial assmption by plant operations personnel

was that the hih radial vibration on the B compressor

was cased by nbalance. This assmption was not

withot basis since olefins chare as is considered

to be a folin service. In particlar, this compression

train had experienced heavy folin several years in

the past. Two other factors spported this assmption.

First, as previosly mentioned, the C compressor was

overhaled drin the otae and some folin was

fond. Second, the B compressor was the only case

not opened drin the otae, so its cleanliness was

nnown. While most of the vibration was indeed

occrrin at the rnnin speed (1X) of the compressor,

nbalance was not the major contribtin case of

the problem. This was ascertained by notin that the

lare increase in amplitde on the riht side of Fire 2

had occrred at relatively constant speed. While it was

possible that some folin material cold have been

thrown off one of the impellers resltin in a sdden

chane in balance, this wold have been accompanied

by a sdden phase shift. However, Fire 2 shows that

no sch phase shift occrredthe phase remained

constant at approximately 215, eliminatin nbalance

as the case of the lare observed increase in vibration

amplitde. To lend additional credence to this concl-

sion, an acceptance reion plot (i.e., trend of amplitde

and phase sin polar coordinates) of the bearin #6

y-probe was examined for a data rane encompassin

the two wees after startp. It showed that the synchro-

nos phase anle was almost constant (Fire 3).

Figure 3. Bearing #6 y-probe acceptance region plot or 2-week period between 8/25/06 and 9/7/06 showing

an essentially constant phase angle o 215.

8/3/2019 gas train

5/12

62 ORBIT Vol .28 No.1 2008

CASE HISTORY

2. Locked oil seals

The possibility also existed that the floatin rin oil

seals cold have loced in an eccentric position, actin

as additional rotor spports instead of jst the jornal

bearins. However, this type of problem wold have

ndobtedly exhibited some sb-synchronos vibration

peas in the spectrm, becase of the oil whirl that

wold be present in the seals at this speed. Examination

of the spectral data showed that the vibration was

almost entirely at 1X and had no sb-synchronos

components. Conseently, loced seals were extremely

nliely. Additionally, loced seals typically affect the

rotors natral freencies becase they effectively

redce the rotors span between spports and ths

increase its stiffness. Had a chane in the compressors

1st balance resonance been noted (particlarly an

increase), this wold have pointed toward the oil seals.

However, comparison of the Bode plot collected prior

to the shtdown (Fire 4) with that taen drin the

post-otae startp (Fire 2) showed that the rotors

first balance resonance had not chaned. Combined,

the lac of sb-synchronos freency components

and no observed chane in the first balance resonance

eliminated loced oil seals from frther considerationas a liely root case.

3. Cracked shat

A craced shaft will normally be indicated by a decrease

in shaft stiffness that reslts in a noticeable (and

sometimes sbstantial) redction in the rotors critical

speed. As noted previosly by comparin Fires 2 and

4, the rotors critical speed measred prior to the otae

was essentially nchaned from that measred after

the otae, main a craced shaft very nliely.

4. Internal rub

In contrast to a craced shaft, a hard rb on internal

seals will normally increase the critical speed becase

of the redced span between rotor spports, since the

seals are now actin as asi-bearins. For the same

reasons listed above, the lac of chane in the first

balance resonance between the pre-otae coastdown

and post-otae startp data did not spport a rb

hypothesis. Additionally, a rb wold normally be

observable as flat spots or other anomalies in the

shaft orbits. However, the shaft orbits (Fire 5)

exhibited no sch evidence. Combined, this made

a rb extremely nliely.

Figure 4. Bode plot o bearing #6 y-probe on shutdown, 8/3/06, prior to the outage. The blue cursor positionshows the frst balance resonance or this machine occurs at about 2400 rpm, essentially unchanged rom

the post-outage resonance o Figure 2.

8/3/2019 gas train

6/12


CASE HISTORY

Figure 5. Bearing #6 orbit ater start-up

showing no evidence o seal rub.

5. Locked Coupling

Althoh the chare compressor train had a history

of loced coplin problems, these had not manifest

for several years. After a maintenance trnarond

several years prior, the radial vibration on the A and B

compressors increased to approximately 3.5 mils, and

at the time, it was assmed the problem was cased by

coplin loc-p. However, the vibration had trended

down to normal levels over time. Additionally, the

coplins removed drin the Ast 2006 otae did

show indication of wear between the end of the hbs

and spacer, and for this reason were replaced with

identical new components. The fact that the coplins

were brand new seemed to rle ot the lielihood

of coplin loc-p. Liewise, the alinment of the

compressor had been checed and fond to be in ood

condition. Measrements had been taen in the past

that showed the thermal rowth assmptions in the

cold alinment tarets were ood.

For all of these reasons, the coplin was not

immediately sspect.

IN THIS SPECIFIC INCIDENT,

THE ABILITY TO ACCuRATELY

DIAgNOSE THE ROOT CAuSE

ALLOWED MACHINERY

PERSONNEL TO RECOMMEND

A REDuCED WORk SCOPE.

THIS SAVED APPROXIMATELY

1011 DAYS OF LOST

PRODuCTION WHEN

COMPARED WITH A COMPLETE

OVERHAuL THAT WOuLD

HAVE BEEN REquIRED IN THE

ABSENCE OF SuCH DATA.

8/3/2019 gas train

7/12

64 ORBIT Vol .28 No.1 2008

CASE HISTORY

The first stron indication that the coplin miht be the

problem afterall came approximately one wee after

the startp. Becase of a overnor control oil problem,

the trbine speed dropped rapidly (from approximately

5000 to 3500 rpm in a few seconds). When this occrred,

the overall radial vibration at bearin #6 dropped from

approximately 4.5 mils to 0.7 mils. However, the speed

retrned almost as icly as it had dropped, and the

hih vibration retrned (Fire 6). Initially, this event was

not that enlihtenin becase the 1X trend (see Fire 7)

seemed to indicate that the synchronos amplitde and

phase anle had not chaned maredly, in star con-

trast to the chanes observed in the overall nfiltered

vibration of Fire 6. However, after contactin gEs

Bently Nevada Machinery Dianostic Services (MDS) per-

sonnel, it was learned that the overall vale displayed in

the DM2000 software is received directly from the 3500

series machinery protection system and merely passed

throh, while the 1X vale is calclated in the software

from the raw vibration waveform. Since the chane in

speed and vibration happened so rapidly, the DM2000

softwares tracin filter was not fast enoh to captre

this rapid excrsion. Ths, it seemed liely that the 1X

amplitde and phase had in fact chaned alon with

the overall nfiltered amplitde drin this excrsion.

Figure 6. Rapid drop in bearing #6 y-probe unfltered overall vibration amplitude during speed upset.

Note that ollowing the event, the amplitude returns to nearly the same level as prior to the speed upset.

Figure 7. Same data as in Figure 6, but fltered to compressor running speed (1X).

8/3/2019 gas train

8/12


CASE HISTORY

In an attempt to re-create the drop in vibration

mentioned above, approximately one month after the

otae, the compressor speed and load were decreased

drin a redction in nit rates. Drin this test, an

ADRE portable data acisition system was connected

to the 3500 machinery protection rac as it provided

hiher data resoltion than the DM2000 software.

Becase the speed decrease was radal, there was

no rapid decrease in vibration as seen earlier. However,

drin the redction in nit rates, the compressor was

accidentally sred and becase the ADRE system

was connected, hih resoltion data was acired.

Examination of this data showed that the synchronos

amplitde had dropped rapidly drin the two sre

events. Liewise, the synchronos phase anle had

chaned as well (Fire 8). These facts pointed stronly

to the lielihood that the coplin between the B and

C compressors was loced in a riid condition, bt

had momentarily broen free drin the sre event

becase of the rapid flow reversal and sbseent

relaxation of torsional loadin on the coplin.

Figure 8. Bearing #6 y-probe 1X fltered amplitude and phase data collected during surge events on 9/28/06.

Note high-resolution time span on horizontal axis (1 second per division) indicating that the back-to-back

surge events occurred within approximately 10 seconds o one another.

8/3/2019 gas train

9/12

66 ORBIT Vol .28 No.1 2008

CASE HISTORY

Examination of the shaft centerline plots of the #6

bearin revealed that the jornal had dropped approxi-

mately 2.0 mils drin the coast-down before the

otae, bt had raised 4.0 mils drin the sbseent

start-p (see Fires 9 and 10 respectively). This meant

that the sction end jornal was ridin in the center of

the bearin, instead of the lower half, since the desin

bearin clearance was approximately 8 mils. The indica-

tion was that somethin was liftin the B compressor

sction end jornal off of its bearin. This was frther

confirmed by examination of the bearin metal tem-

peratres for bearin #6 (see Fire 11). As can be seen,

the bearin #6 temperatre is approximately 15 coolerafter the mitiation otae, even thoh the vibration

was obviosly hiher. In contrast, the bearin #7 tem-

peratre was approximately the same before and after

the otae, indicatin that no other external factors,

sch as oil temperatre or ambient temperatres, were

affectin the reslts.

Figure 9. Shat centerline plot or bearing #6during pre-outage shutdown on 8/3/2006.

Figure 10. Shat centerline plot or bearing #6during post-outage startup on 8/27/2006.

THIS TEST WAS A FuRTHER

CONFIRMATION THAT THE B

COMPRESSOR ROTOR WAS

BEINg LIFTED OFF BEARINg

#6 BY A RIgID COuPLINg

BETWEEN THE B AND C

COMPRESSORS.

8/3/2019 gas train

10/12


CASE HISTORY

Figure 11. Trend rom the plants process historian showing bearing #6 (red) and bearing #7 (blue)

temperatures beore and ater outage. Note that the temperature o bearing #6 drops approximately 15

degrees ater the outage, while the temperature o bearing #7 stays approximately the same. This helped

corroborate that bearing #6 was more lightly loaded and riding more towards the center o its clearance

(i.e., lower eccentricity ratio) ater the outage.

Validation

In an attempt to redce the radial vibration amplitde atbearin #6, and confirm that that the low jornal eccen-

tricity was associated with the problem, an online test

was condcted to chane the relative vertical alinment

between the B and C compressors. The basic concept

was to pll the B sction end jornal down frther in

its bearin, thereby increasin its eccentricity and bear-

in stiffness and redcin the radial vibration. This was

accomplished by sin an air mover with water spray

on the dischare foot of the C compressor, and a steam

hose on the sction pedestal of the B compressor. The

objective was to chane the alinment, lowerin the C

case dischare end and raisin the B case sction end

(Fire 12). Becase heatin and coolin effects were

localized to the compressor spports, it did not chane

the bearin hosin temperatre, which potentially

cold have sewed the shaft centerline plots de to the

probes rowin away from the shaft.

The chanes had exactly the effect predicted. The shaft

centerline plots of bearins #6 and #7 are shown in

B Case C Case

Cooling applied to C

case dischargewobble foot

Heat applied to

B case suctionpedestal

Figure 12. Graphical depiction o online alignment

adjustments used to confrm the locked couplingpostulate.

Fire 13. Liewise, the redction in radial vibration is

shown in Fire 14. This test was a frther confirmation

that the B compressor rotor was bein lifted off bearin

#6 by a riid coplin between the B and C compressors.

8/3/2019 gas train

11/12

68 ORBIT Vol .28 No.1 2008

CASE HISTORY

Figure 13. Steady-state shat centerline trends o bearings 6 (let) and 7 (right) showing eects o online

alignment adjustments. The rotor moves downward in its bearing clearances or bearing #6 and upward or

bearing #7.

Figure 14. Drop in bearing #6 y-probe vibration during online alignment adjustment occurred exactly

as predicted.

8/3/2019 gas train

12/12


CASE HISTORY

The hih-resoltion data (Fire 8) taen drin the

compressor sre showin the rapid synchronos

phase chane, and the drop in vibration (Fire 14)

observed drin the online alinment chane all

indicated that coplin loc-p was the problem.

Approximately two months after the otae, the overall

vibration at bearin #6 was still trendin above 4 mils

with occasional excrsions above 5 mils.

Corrective Actions

Another otae was planned in the winter months

to address the vibration isse. De to the isolation

reirements of the compressor, the coplins cold

be chaned withot completely clearin the compres-

sors. This wold allow the compressor to be ready for

maintenance approximately one day after the nit was

sht down. In contrast, if the case had to be opened,

this wold reire a minimm of 4 days jst to clear

the compressor cases before maintenance wor cold

bein. The end reslt was that a coplin chane

cold be accomplished with 3 to 4 days of interrpted

prodction, while a complete overhal wold reire at

least 2 wees, incrrin 10 or more days of additional

downtime. While this made the coplin wor loo very

attractive, the prospect of chanin only the coplin,

restartin the compressors, and then discoverin that

the hih vibration was still present was hihly ndesir-

able. For this reason, some advocated the safe rote

of a complete overhal. However, based on the clear

indication of the coplin loc-p from the test data,

the site machinery personnel were able to alleviate

manaements concerns and redce the wor scope

to that of jst replacin the coplins.

Reslts

The chare as compressor train was broht down

and the last two ear coplins in the train were

replaced with dry, flex-element style coplins. As

expected, the vibration on bearin #6 retrned to

normal levels of approximately 0.5 mils. Fire 15 shows

the ear coplin removed from between the B and C

compressor bodies. It clearly exhibits an abnormal wear

pattern considerin it had only been in operation for

approximately two months.

Conclsions

Chevron is increasinly sin condition monitorin data

as an effective method of enhancin decision-main

confidence and accracy, and the willinness of plant

manaement to rely on this data when main operat-

in and maintenance decisions is increasin as well. This

case history was merely one example of more than half

a dozen that the athor has compiled drin the last

several years.

In this specific incident, the ability to accrately dia-

nose the root case allowed machinery personnel to

recommend a redced wor scope. This saved approxi-

mately 10-11 days of lost prodction when compared

with a complete overhal that wold have been reiredin the absence of sch data. Each day of downtime typi-

cally costs in excess of 200k uSD for this plant, meanin

that several million dollars in savins were realized de

to accrate and timely machinery data.

* denotes the trademars or reistered trademars of Bently Nevada,

LLC, a general Electric company.

Figure 15. November 2006 photo showing

replaced gear coupling between B and Ccompressor bodies. Abnormal wear is clearly

evident ater only 2 months o operation.

gas train

Documents