ROOT CAUSE ANALYSIS OF FIVE COSTLY CENTRIFUGAL PUMP FAILURES by Heinz P. Bloch Consulting Engineer Process Mac ery Const Montgomery, Texas Hin P. Bloch isa licensed Consulting Engineer with offices in Baytoren, Texas His company is Process Machinery Con- sulting Company. He retired from Exxon {nn 1956, after 24 years of service. His pro- {fessional carver included a long teri as- ‘ignment as Exxon Chemical Companys Regional Machinery Specialist for the United States. He also held machinery oriented staff and line positions with Exxon affiliates. inthe United States, Italy, Spain, England, ‘The Netherlands, and Japon. Trow Dleshooting and reliability improcement missions hace taken Him to affated companies and machinery manufacturing plants in more than a dozen additional locations worldecide, He has conducted public andior inplant courses in the United States, Canada, Kuwait, Indonesia, Pakistan, Surinam, Vew ila, and other countries. ‘The depth of Mr. Bloch’ incolvement in all aspects of machin ery reliability improcemeat is indicated in over 30 technical s0- ‘tety papers and publications, His seten comprehensive hooks ‘on practical machinery management and maintenance cost re duction hace been used for lectures and consulting assignments worldwide, bn addition, Mr. Block holds four United States a tents relating to high speed machinery improvements Mr. Bloch graduated from the New Jersey Institute of echnol- ‘ogy ith B.S. and M.S, degrees in Mechanical Engineering. He teas elected to three national honor societies, and isa member of ASME and STLE, and the Vibration Institute and is a Regis tered Professional Engineer in the States of Texas and Neve Jersey. ABSTRAC Machinery failure analysis and troubleshooting are often pur sed in somewhat haphazard, unstructured fashion. This Paper presents a novel, proven. and well focused approach ‘Wich can quickly lead to the root cause klentifiention of wrt all-any component distress, This repeatable method is hased on aly if they are subjected to e, a reactive environment, time-based exposure, or extreme temperature, Hossever, while these failure mechs fs well tll how and why a failure mode sch as brittle frac ture. melting, ete, might have aceurred in chemicalor metallur- eal terms, failure mechanisms do not define the root cause of| sing five illustrative examples, itis shown how through a process of elimination failure analysis can rapidly define which ‘of only secen possible root causes of machinery failures, desiga delicienes, materials defect, processing and manuficturing de- ciency; assembly error, off design or unintended service condi tion, maintenance deficienes, oF improper operation holds the key toa particular failure event. INTRODUCTION After afew years of operation, the overwhelming majority of ccentifngal pumps in service today will have experienced repeat failures. Most of these ate premature, of unexpected. Equip ment distess events due to component weatont, or end ie failures are quite rare. Repeat failures, often ofthe same pump component, oceur because the owner or user has either not wt ‘covered, or perhaps eleetod not to remedy, the nt cause of the problem Root cause analysis is aimed at uncovering the sometimes el sive failure sequence —anl thus, ot eause—of the events lead ing up to equipment failure. Hrreeognizes that all flares, with: ut exception, belong to ane uF more of only seren categories + Faulty desig + Material defects abrication ar processing errors ly oF installation defects + Off design or uninteuded service conditions # Asse + Maintenance deficeveies negleet, procedures) an ial + Improper operation Root cause analysis farther recognizes that without exception, the basie agents of machicery component and part failure mech. anisms are aficays force, time, temperature, ora reactive envi Fonment. One of more ofthese mechanisms may combine and hasten component degradation [1]. Contributing or interacting factors are all part ofa system; consequently, the entire system the subject tu review and scrutiny Using the above premises, a straight forward approach ean be smduced which has assisted the author in identifying the root causes of many costly falures involving centrifug process and sity services, Five such fares ate greater deta + Repeat bearing flues which were attributed to vendorde- + Several hearing failures, and finally a potentially damaging fire, caused by incorrect prelubrication of thrust bearings during shop assembly + Extreme vibration aad deterioration of gresse-lubricated sleeve bearings in hige Seawater intake pumps traceable to op= * Repeated and costly thrust bearing failures ina mining slurry pump caused by mistakes in parts documentation, fabri: ‘ation and procession. + Loss of life ina U.S. Gulf Coast plant, possibly eaused by simple maintenance oversight CHECKLIST APPROACHES GENERALLY AVAILABLE It would be difficult te think of machinery troubleshooting tasks that would not benefit fom a structured approach, Time Js saved, accuracy impnoved and the risk of encountering repeat failures is reduced whenever the troubleshooter makes use of a16 PROCEEDINGS OF THE SEVENTH INTERNATIONAL PUMP USERS SYMPOSIUM. comprchensive checklist such asthe one compiled by Karassk, ‘Tables 1 and 2 [2]. An internationally recognized authority on ppumps and their application, Karassik believes that, while no listof pump troubles can ever be complete, it make sense to use checklists in diagnosing centrifugal pump troubles. The ‘checklist approach shown in Tables 1 and 2 correlates observed symptoms with possible causes of trouble. Of course, cheeklsts could be further expanded by observing the symptoms of bear ing distress with corresponding possible causes, or mechanical seal distress could be tabulated together with possible contribut- ing causes, Similarly, vibration symptoms could be contrasted with causes, or stuffing bor packing deterioration diagnosed from a symptom es eause comparison mattis, Table 1. Check Chart for Centrifugal Pump Problems. eat pty der rt pu de 2 A ane deh ans 20 Pricer ueoy orig 80,18 9, 9857,68,0,00. 7.72, 21, Peed fo dee sed no 22. Nos alse $5.56 55, 5 9862.75.18 ign al md as se é Table 2. Possible Causes of Trouble, 2 Pump cn pe compe id with bd 5 Ice be Roepe tma faar Aen tne 10 ie foe ge bed Haren sti he 2 Saaanl pp mt Secon Spel pe ee Hen rooms ‘pose sry ea ec iconmeemace “ Tol ea orem berth a fo te ye oes ds mtn a Eeaeueureeupayperr A. Fin mer per 8 Sh ronnie rere st ana £3 meres up pig ping aan feet Le ee a a Remar ton ns 5 eer al ofa eas ah a a dug ion. 1S Motor ttt it 5. Eewnecotgt eer regeROOT CAUSE ANALYSIS OF FIVE COSTLY CENTRIFUGAL PUMP FAILURES " FAILURE STATISTICS CAN BE HELPFUL Another highlighted approach [1] attempts to give a statistical indication ofthe most frequently encountered centrifugal pump troubles (Table 3). The numbers listed in columns A through H Indicate the probability ranking of the sariows causes which could result in a particular symptom manifesting itself Table. Troubleshooting Guide for Centrifugal Process Pumps. omar | Psat i ; fae BEET TTT Pc Soke — it il licence i i | Fee rae oe i 0 Dame bari Thus, looking at Table to determine the most probable cause for insufficient pressure generation (Symptom D) will deter- mine that investigators siould look for possible causes in this sequence: + Noncondensibles( + Pump speed too low + Wrong direction of rotation + Tota system head lover than design head of pump—pump is “running out + Viscosity too high + Tivo oF more pumps ia parallel operation but having unsult- able head es flow characteristics + Internal wear, ie., wear ring clearances, excessive, in liquid However, while the use of cheekliss and/or probability rank ings is steongly recommerded, the person engaged in pump fale ture analysis may do well to remember that all problems ean be assigned to one or more afthe seven eanse categories mentioned tarler. In addition, the tmubleshooter should keep in mind the basic agents of machinery component and part faire mecha- isms, ie, free, time, temperature, and a reactive environment. Itis doubtful whether statisties have been compiled to show the overall distribution cf failures as they relate to the seven ‘eause categories given in the INTRODUCTION, At hest, the reviewer might expect tone fille eause and failure mode dis- tributions for critical components [1] or entire machine centsfigal pumps 3) and gears [1]. The later1s PROCEEDINGS OF THE SEVENTH INTERNATIONAL PUMP USE! reference subdivides gear failure eauses into vendor problems {G6 percent) opersting problems (47 percent) and extrancons influences (IF percent). tis interesting to note that mnpublished statistics from alarze petrochemical plant tend to shows problem distributions for several machine eategories to he in the same: overall range SYSTEMATIC APPROACHES ALWAYS VALUABLE Considerable involvement with pump maintenance and re- pir would lead usta estimate a fare ease distribution forcen= trifzal pumps in US. process plants as xiven in Table 4. This failure analysis and troubleshooting approach attempts to focus con this estimated cause distribution, In other words, an ap proach which seeks to first find the rot causes of failures i the Categories with the highest probability ranking might he xener ally endorsed, This approach daes not, however, overlook the need to + Start a the beginning by reviewsing the pump cross-section drawing ‘thinking throngh” how the individual parts finetion or snafanction. ‘understanding the process loop and process + Take a systems approach, Never lose sight of the fact that + the pup is only part ofthe overall loop, + the part that failed is very Ren not the root cause ofthe vblem and unless we find the root cause, repeat failures are Tikely to occur, + Collect all the pieces, The missing part may contain elves which must be examined andl which may have had an influence ‘on fue cause or fine progress + Use a caleulation approach while not, or course, neglecting the intuitive or prior esperience-utilization approach, ‘With thisin muind, the frst of the five pump prublems can now be examined Table 4, Failure Cause Distribution Estimate For Centrifugal Pumps tn US Process Plants % Probability Incidence Ranking Maintenance Deficiencies (Nealeet, Procedures) 30% 1 Assembly Or Installation Defects 25% 2 Off-Design Or Unintended Service Candlitions Io 3 proper Operation 12% 1 Fabrication Or Processing Ervurs 8% 5 Faulty Design 6% 6 Material Defects 4% FAULTY DESIGN CAUSES PREMATURE BEARING FAILURES Not too long ago, 125 hp, 3560 rpm, 310 gpm, 670 ft head single-stage overhung impeller centrifugal pump in hydrvear- hhon service experienced frequent bearing failures. With" Faulty Design” ranking nest to last in the Failure Cause Distesbution Tisting of Table 4, i was certainly ot logical to immediately ss SYMPOSIUM pect a fundamental design error or vendorrelated engineering problem. Becaise ofthe probability ranking, maintenance-t pe fuses were pursied first. Table 2 was consulted (tems through $5), snd a supplementary 53-tem bearing problem checklist [1] sas used to ascertain that falty assembly o ‘maintenance could aso be riled out. Nest, the failure analy sis review fensed on “OT-Design Conditions” and “Improper Op- tration.” When no pmblems were found in any of these ares, and it ss further established that there were no material de Fects inthe ring clement bearinas, the investigation hewn to ‘concentrate heavily ou the possibility of « vendor ermor, he Faulty Design Punip owner and pump manufacturer agyeed to perform a field test on this falue-prone pamp. ° special test ri, Figure 1, was designed and fabricated by the pump manufacturer. It consisted of means tallow the pimp rotorbearing assembly to move in the axial Grapeller thrust) dieetion. The total xia ‘movement 64s limited so as not to exceed permissible impeller travel. Also, the axial thrist value was measured by three load cell (Figure 1), whose connecting cables are visible in the eld test setup shoe in Figure J. Design Drawingof Centrifugal Pump Avia Thrust Test Rig Figure 2. Field Instabation of Special Bearing Housing Which Verified Gentrifizal Fump Axial Thrust to be Escessice Test results were phtted ane compared tothe mannfaturers caleulated and origialy antieipated thrust sales for this pup. As indicated in Figur 3, the experimentally verified thrust at shutof? was 2.6 times greater than anticipated. Since ball bear ing life varies as the cube of the load changes, the life of the pump bearing would thus be reduced by factor of IrROOT CAUSE ANALYSIS OF FIVE COSTLY CENTRIFUGAL POMP 2MLURES 9 \CEGaine na Fn so caPaciTe mW) Figure 3, Experimentally Verified (Actual) Rotor Thrust Ex ceeds Manufacturer’ Anticipated Values by a Factor of 26. ‘The test results convinced hth the operating company and the pump manufacturer that the pump internals had to be rede~ signed to Kinit hydraolially induced thrust values to more reasonable linits. Obvionsl, the haste agent ofthe bearing i tire mechanism was excessive force ASSEMBLY PROCEDURES MUST BE JUST RIGHT How a seemingly minor assembly oversight or procedural shortent ean have adisprop e impact on equipment pe formance and plant safety i ilstrated in the next example During a two week period, several process pumps its large ctliylene plant developed problems soon after they had been = tured to the field after having underaone seal and bear placements. fire resulted shen & 250 hp, 3560 rpm tae pun failed within eight hours after one such repair The pumnp in Figure 4 is identified as one of the MP-IT 8 or Bets feeding 320 gpm of taraltemperatre the primary fractionator tar boot and coke filter M tar evolers and on to downstream process equipment. Pump de sign inlet and discharge pressure conditions were 10 psig and SIN! psig, respectively: pant M01 PRIMARY FRACTIONATOR To cooueRS ro ) | e178 tT) Mp.t70 Figure 4. Flow Schematic Showing Tar Pump (MP-IA)Incotved in Fire Event Because the filed MP-I7A and its eo companion pumps ad previons given satisfactory service, wt was decided initially nt to pursue "Fanlty Design” asthe most probable cause. Similarly, Fabrication or Processing Errors” were jaded somewhat likely on rolling element hearings failing in succession. How ‘exer, since the MP-17 pimps operate in hot sersice and take Suction from a reactor with « varying liquid level, operations related causes were reviewed with control wm personnel. Pr ‘cess technicians reported that level, flows, and pressures had been quite normal until the etual failure event. With adequate NPSH critically important tothe safe, eavitation-free operation ‘of centrifugal pumps, the availablity of a strip chart recorder tape sheacing sulicent level in the suction vessel sas consid cred a particular advantaze Satishetory operation‘ graphically represented in Figure 5, the strip chart obtained frm the trend recorder for fractionator boot and main vessel level. At 17:30, or 5:30 p.m., the chart shows the boot level toe 100 percent, About 10 minutes later, the hoot level was reduced to 35 percent far approimately 5 to Wainntes hefore finally eturningto 1OO percent. The ehart ver ies tha the boot level never deupped below 35 percent on the day of the failure incident. Therefore, pump operation errors.ar tlf-design sersice conditions did not seem at fault andl were OY ae ~ Figure 5. Strip Chart Showing That Adequate Liquid Level Exivted in Suction Dru at all Times, The faire analysis and troubleshooting effort now shifted to the areas "Maintenance Deficiencies” and “Assembly or lnstalla- tion Defects.” Debriefing the shop and field mechanical work forces shed some light on maintenance techniquesand assembly ‘quality-control procedures employed during repairs, which pre ceded the final failure event Duving a peri of three days, the pump had been removed from its field location and taken to the shop three times for seal repairs, bearing replacement. and adjustments of one type or ‘other. After final overhaul the pun was started up around 41:30 p.m. Hews reported on five at 730 pam ‘When the pump was dismantled, the Seal area was found i clean and undamaged condition, Some solids were found in the impeller. Impeller wear rings and inboard hearing appeared satisfactory. The duplex thrust (outboard) hearings were totally destroved., Severe metal loss was noted on virtually every bear= ing ball. Many balls were deeply embedded in the inner race; the ball separators had sisintegrated. The shaft was severely bent in the region adjacent to the duplex bearing lock nut Fig ture 6). Two ofthe four seal gland nuts had loosened, athird one had fallen off completels. 4 pedestal support bracket a point A had not heen connected to the pump easing. The ductile iron Dearing bracket was fractured at point B. It was then theorized that failure of the duplex thrust (out board) bearing set in motion the chain of events leading to the2 PROCEEDINGS OF THE SEVENTH INTERNATIONAL PUMP USERS SYMPOSIUM a= 1 Figure 6, Schematic Cross-Sectional View of Centrifugal Pump. fire. Excessive friction resulting from severe and nearinstan- taneous hearing failure could have caused the shaft to bend. Ex. treme vibration was certain to be generated and could have ceused the less-than-optimally installed seal to release pum= pause, By this time, the outboard bearing area was thought to have been red hot, causing spilled pumpage to ignite Twas noted that near-instantaneous massive fulure of rolling contact bearings is most frequently caused by deficient lbrica tion and overheating. The failure was too massive and tow far progressed to allow determination ofthe origin of overheating Installation method, housing bore dimensions, shaft dimen- sons, driver to pump alignment accuracy, clas of bearing (ve. rullingelement tolerance) and sparkingaction between auxiliary land and shaft sleeve could have played a mole in the event. However, the primary cause of hearing overheating upon initial bperation of new bearings at this plant was not related to any of the above. lustead, the most probable canse was the practice af prelubricating with a penetrating oil which was never meant to be applied in this manner. Its extreme low viscosity (75 SUS @ 100°F) makes it suitable only for bieyele and door lock type of Tubrication duties. This oi was found very volatile and would evaporate at temperatures well helow those antiipated for new antiviction bearings operating at relatively high speeds ‘The chart in Figure 7 was used to determine the viscosity re ‘quired to adequately lubricate the MP-I7 hearings. Ata mean diameter of 87.5 mm and a speed of 3570 rpm, a mininan Ie bricant viscosity of §.3 eSt is required, As shown in Figure 8, ‘maintaining this minimum viscosity s possible only operating temperatures do not exceed 10°F. Newly installed duples and double ro thrust hearings will, however, experience temper tues well in excess of IPF. Although a superior grade oil mist Tubricant was supplied to these particular pump bearings, the nist habe could not overcome the dilating effect ofthe inferior low viscosity ol which was present ina “tough” formed by the Dearing outer race at the sik oelock position. The existence of this trough,” or minisumnp explains why vl mist ubricated bear ings in horizontally arranged pumps and drivers generally sue- vive for periods of eight oF more hours after the oil mist supple has been turned off. Unfortunately; ifthe minisump ified with ‘dilutant, the beneicial effeets of appleing highly viscous Tn brricants cannot come into play until the damage is done In conchasion, prelubrication with an inadequate lubricant was proven to be the root cause ofthis and other sevions pump fil tures which had preceded this particular event, Had the review cerallowed himself tobe distracted by additional but, in the final ‘analysis, minor deviations such asthe out-of perpendiculaity of seal fies, defective sleeve gasket, and a missing support brack: wie ae evan 848 “FH Figure 7. Temperature Viscosity Chart Mlustrates that Exces- sively Love Viscosity Lubricants Will Not Support Oil Film at ‘Temperatures Typically Encountered by Process Pumps. Figure 8, Oil Viscosity Requirements Plotted vs Mean Bearing Diameter and Shaft Speed. 4 the true rot cause might not have heen uncovered until nvch later By way ofrecap, the mot eauses ofthis particular failure event were uncovered by fist looking a the entire system, Nest, the review proceeded to examining all pieces, and plaving through 4 typical “what if” scenario ultimately provided the needed Tens + Failures ater two hour run length of MP-TA on brand new bearings, and two and eighthour runsof another punnp on brandROOTCAUSE ANALYSIS OF FIVE COS! new bearings in the same general span of four or five work days ‘were thought to fallow a pattern pointing to possible commonal- ity of failure causes. The common Tink in all failures was prelu- Lireation with « Inbricant approaching the eharacteristies of penetrating ols, + Extreme unbalance vibration originating at the impeller, due to possible coke plugging, would have been expected to cause wear ring and inboard bearing: defects. These possible events were thus ruled out as quite improbable + Seal failure preceding bearing failure was inconsistent with the surprisinaly clean appearance of the seal after the fire * Sparking action due to rubbing contact between a non- standard auxiliary gland (made of 316 $5) and seal sleeve (made ‘of 410 $5) was considered possible but should have resulted in severe galling of the softer ofthe two materials, No such galling ‘was observed. * Tar Teakage between shaft and sleeve had probably oc- ‘curred, but was not thought to have started the fire. Experience shows that pump fires brought on by seal distress must reach a very high intensity before outboard thrust bearings disintegrate tastrophicaly. Tow level fire lasting for five to 10 minutes simply did not fit this seenario It should be noted that plotting the temperature-visosity re lationship of this penetrating oil on an ASTM chart (Finure 8) ‘contd be considered the equivalent of analytical caleulations Caleulations are very. often superior to conjecture and guesswork when attempting 0 Bnd the root eauses of pump fil As to the basic agent ofthis bearing failure mechanism, it as quite cbvionsly temperature, Prelubrication with a much mote ‘iscous oi was initiated and there have not heen any sia fil FABRICATION AND PROCESSING ERRORS CAN PROVE COSTLY There isan interesting story behind along series of randomly ‘occurring theust bearing failures in one particular typeof shurry pump in service ata South American buusite mine. Apparently the thrust bearings would sometimes fal after afew das or, at other times, after a few weeks of operation. Before the mechan: ies produced a cross-sectional vies similar tothe simplified ver sion depicted in Figure 9, the author had been told that i was ‘often necessary to rebush and line bore the bearing housing, ‘The relevance, accuracy, ar importance of this verbal failure de: seription becomes evident only when the drawing is examined in dota With the impeller inverted so as to reduce the differential pressure across the shaft packing area, iis immediately shox thatthe primary thrusts from right toleft. The to an taet hearings on the extreme left are corveetly oriented to take dient Figure 9. Bearing Housing Showing Ball Thrust Bearings Which Failed Frequently. LY CENTRIFUGAL PUMP FAILURES. at up the predominant load. However, the outer rings are com pletely unsupported, because the fabricator had somehow de: cided to overhore the howsing in the vicinity ofthese two bear ings, Consequently, the entre radial load aetingon the coupling end ofthe pump had to be absorbed by the remaining third an- sgular contact thrust bearing. This bearing was thus averloaded to the point of rapid failure and was, ofcourse, prone to rotate in the housing. Using a double row spherical rller bearing at the hydraulic end of the pump would normally make for a stur ds: well designed pump. In this case, however, the spherieal ty tation or compliance feature tended to further increase the ra- dial load transfered tothe one remainingouthoard bearing. The hasie agent ofthe component flute mechanism was, of corse, force ‘An equally serious bunen was imposed on this pump by the ‘well intentioned person who, in an effort to ink the spate parts requirements of the Norh and South American plants of this, ‘major aluminum prodicer added to the drascing the parts list, partially reproduced under Figure 9. Having left off the appro priate alpha-numeric coding behind the bearing identification hhumber the bearing ilentifiation number 7312, this plant and its sister facilities would receive thrust bearings in other than rmatehed sets, A quick loocat the bearing manufscturer'sdimen- sion tables (Second insest, Figure 9) shows simple type 7312 bearings to have a width hich may difer from the next bearing by as much as 0.006 in, Mounting two such bearings in tandem may’ reent ofthe led, while the ‘other one would simultaneunusly carry 50 to (percent load. On the other hand, matchee sets intended for tandem monnting would be precision-geound for equal lol sharing and would be furnished with code lett sto indicate this design intent Did the author go through the seven cause categories to iden tify the above root cases? Frankl no, When both the fabrice tion sketch and the procurement documentation— "information, processing —show bso very obvious errors, itis reasonable that Tectification of these deviations should be a prerequisite to Further fine-tuning. This is just another way of saying thats it Tooks like a duck, walks like-a duck, and quacks ike a duck, we ‘otto call ita duck and dispense with further research into the ancestry ofthe bird OPERATIONS ERRORS CAN : FREQUENT BEARING FAILURES ‘Thenext problem involved for 2500p vertical pumpsin sea water service. Operating at 595 rpm. these 14000 mh (60000 pn), 4m 135) head pumps experienced high shaft vibration ad extreme wear of the grease-Inbricated bottom beatings, The pant operators realized that high vibration oceurted prin uring low Now operation or whenever additional parallel operating pumps were started up. An outside consultant recom tended. that the automatic pressure-fed_grease-lubriated bine hearings ofthe four pumps be coverted to continuous water lubrication ata cos of 800,00, It would appear that this, ‘consultant made the typical mistake of concentrating only on an ‘examination of the priveipal part that had failed, ‘e., the Dearing ‘The analysis strategy eensisted of a review ofthe vibration rec- fords, repair history, spere parts consumption, and physical ‘examination of bearing and smpeller wear patterns, Since the ‘pumps had been designed and manufactured by an experienced con ‘were generally running quite Well, the cause categories "Faulty Design,” “Fabrication or Processing Errors, and Assembly or Installation Defects” were not considered high fn the list of probable failure initiators. There did not appear to hheany material defeetsor either the badly worn bottom bearing or the slightly cavitation-eroded impeller vanes. Accordingly,2 PROCEEDINGS OF THE SEVENTH INTERNATIONAL PUMP USERS SYMPOSIUM ‘we concentrated on a review of the exteyories “OM-Design ot Unintended Service and “Improper Operation The pump performance curve, Figure 10, rapidly furnished the o wa . 3 g : g ; é ss o00 20 2% C2 1000 CJ m ‘ow « ‘a 0 5 0 Bo mYhi08 Figure 1. Performance Chart for Vertical Pus tn Senate intake Serie, Nite NPSH re When operatinu.at throughputs in the vieinity of 5,000 enbie meters per hou, the required NPSH (net penitive section head largely a measnee of inlet pressure) would he exceedingly high: ‘certainly twice the NPSH required at 10,000 eubie meters per Thour. With the available NPSH for each seawater intake parmp ‘essentially fied at roughly 12 meters, veduetions in throughput Mow fron the exstomary 8,008 cubie meters per hon per patap cold rapidly drive the pump performance into the ewitation range whene high vibration would be encountered. The equip: tment owner sas advised to consider installing antomatically ‘contelled los flaw: bypass loop ora combination of supervisory instrumentation and seawater eonsumerto-pamp-operator ‘communication Tink (audio-visual electronic interfice), This ‘would give the utility crew sufficient time to always have only that number of pumps on line which would be needed by the downstream processes while tthe same time satisfying the flow requirements ofeach individhal vertial pump, The principal eause category in this fulure e proper operation. The basie agent af the parts faire mechs nisms was excessive force MAINT OF LIFI One of the more trax pump fulure incidents aceurted at a Ihydlicatbon processing plant int the U.S, Gulf Coast area in 1982 Figure 11), It involved the pump shown in the foreground of Figure ANCE OMISSIONS: N CAUSE Loss. Figure 1. Broader View of Pumps in 1982 Hydrocarbon Process ing Plant Accident. When a pump malfinetion was detected by control mom per: sonnel at this plant, two operators went to the area and realized fiom the dimensions ofa propane vapor clowd thatthe pusap had the shut doven. As they approached the equipment, the spor jquited causing both men to sufer extreme burns. One af the "uve operators later die Ass stl in stich cases, an effort as later male by a aca expert to reeonstruct the event and determine the case ofthe Fire. His teport noted that the mechanical seals had received flush liquid via APL Pn 31, 4e., recirculation fiom the pup cease throngh aeyelone separator deliveringclewn propane to the seal and, in his own words, certain amounts of entrained water to the pimp suetion. Alter esamining the pump internals (Pi ture 13) the loeal expert determinedROOT CAUSE ANALYSIS OF FIVE COSTLY CENTRIFUGAL PUMP FAIL that a failure of the pomp occurred and that the dynamic forces distorted the pump in order to thro the rotating se tion between the two beavings out and away from the center line of the shaft, causing i to wear on exactly the same side throughout the length of the pump inside the easing Teis the opinion of [the local expert] that a pump failure of this type with these results cold not be anticipated by the ‘operating personnel af the owner’ plant]and that theve was ho Failure of adequacy of instruction ta operating personnel at the phunt. Tes the further opinion of the [local expert] that the two leaks that were present and that iznited alter the pump fire cecurred were the type of leaks that are nora the openition ofa phant of this mature and that they, in and themselves, are not indicative ofan falureof ether ond en sneering practices or proper maintenance Authorized by the owner, the local expert supervised the re wal of seals, bearings, and impellers (Figures 14, 15, 16, and fn eflorts ta find a crack or cracks in the shall material which, ated the eatastrophie failure of the r he reasoned, might have in pap, ture 19. Pumps Internals Immediately After the Fire Together with five oF sis other equipment and component rmunufacturets, a repair shop which had worked om the pump Fon years pris to this incident had to defend itself eonart, The attorney representing thispump repairshope dnd requested reviews of depositions and ph pare defense arguments Although the plaintiffs expert had already: gone on yeeord with the statement that pump failure originated with a eruck sonesshere in the pump shaft, our review effort was agin aimed at eliminating at least some ofthe seven principal falure ced the author tographs to pre ‘eateries hy assemblingcas much pertinent data as was possible Figure 15. Questionable Procedure for Remocing Rolling Ele- ment Bearings (Pump Inboard Side). at this late stage in the investigation. The eategory “OM Design ‘or Unintendeal Service Conditions” was ruled ont on the basis that similar pumps had heen installed elsewhere and had oper ated well under similar conditions. “Improper Operation” did hot appear likely since tre unit was running normally and the pump in question was wet ina startup or shutdown phase a the time of the incident, “Baulty Design” was not judged likely in view ofthe age and experience record of this pup model. "Fab- Fication or Processing Errors” andAssembly or Installation De fects” were ranked somewhat more likely, and "Maintenance Deficiencies” and” Material Defects” tentatively, and somewhat arbitrarily, putat the top of the list Nest, asite visit was arranged. As is appropriate when using a conscieitons systems cpproach which includes a review ofall relevant component pars, a bos of broken parts was examinedu PROCEEDINGS OF THE SEVENTH INTERNATIONAL PUMP USERS SYMPOSIUM. Figure 16, Double-Row Thrust Bearing Showing Adequate Lab- ication at Time of Pump Failure. Figure 17. Impeller Destroyed at Disassembly. at the plant site (Figure 18), but no coupling components were found. Since a motorto-pamp couplings, of course, part of the system, i was judged important to review its condition. Porta nately the eoupling was fond on the wrowad in close proxinity tothe pu 19) and serious wear was immediately evident (Figure 20), Other pertinent observations rapidly fol Tosved and led to a rather concise summary repo of the most hls seayenee of events a this ts Nate the “pont ination of misalignment and lack of lubrication inthe robably led to excessive vibration ear coupling very Points of evidence: severe ridges were visible inthe softer of the tio mating gears; no traces of lubricant were found in the drive-side coupling onthe failed pump set; also, no traces were Found of lubricant in the eouplings of both adjacent identical pump sets “High levels of vibration and severe misalignment probably ‘caused crack propagation in each of the four pump support legs. Teas noted that a similar crack had been repair welded on one support leg of an adjacent identiel pump, Figure 19, Worn Gear Coupling + At this time, a combination of shaft misalignment and cou- pling iaflexbilityisthight to have eaused amplified vibration ‘which Ted to shaft bow and internal rubbing + Severe intemal misalignment now caused the hall separator ‘on the radial hearing neurdrive end) todisintesrate Figure 13), + Vibration next caused Fatigue failure of a pipe apple eon nected to the eyelone separator (Figure 1). This caused a mas sive spill of pumpage and also deprived the mechanical seal of flush TigiSE ANALYSIS OF FIVE COSTLY CENTRIFUGAL PUMP PULURES. ey Figure 20. Closeup of Worn Gear Coupling «The bearing now reached a temperature of approximately SUPP; at about the same time, the mechanical sa fees began twheat-check and secondary leakage started to develop, + Hydrocarbon vapors and/or liquids with an auto ignition temperature ranging between 450°F and 650°F ignited and a flash five resulted Additionally, the author noted that the entire pump install tion had to be considered vulnerable, due tothe lack of thermal apabilty of pump suction pipins, lackofseal weld ipples filled with flammable liquid, and lack of cow Tnbrication. Exen visual abservation allowed the observer tocomclade thatthe shalt was bent. This bend was hated near the hot bearing and followed the classic pa machinery engineers on the vast majority of pumps involved in thistype of failure progression. Had the shaft heen bent tobexin with, this3,600 spin pup set wold have exhibited abnormally high vibration from the tine ofcommissioning. Al internal parts rubs and also the cracking pattern on shaft sleeves were com pletely as anticipated in this particular event and were judged the consequence of the sequence indicated above. ‘Aniumber of valuable lessons ae contained in this story. First very few fare events are the result ofa single ereor or omis observed by sion, What ifthe pump suction piping had been designed more flexibly and would not equently have pushed the equip ‘out of alignment? What someone would ha the relatively hes ported differently of, better yet, would have challenged its highly questionable usefulness in the frst place? What if some had decided that highly flexible nonlubricated or elas tomerie couplings should he used on these pumps? Or, what if someone had simply greased the gear couplings twice a year? ly would have vastly reduced the probability the sng as the basic agent of the failure mechanism ea ‘coupling distress “And, finaly, fom a lure analysis and toubleshooting point ‘of view how much more quickly would the most probable flre Sequence and its root eaises have been imeovered if someone had used a more reasona te and wellstructared flare analysis approach? MAKING THE CASE FOR FAILURE PREVENTION AHEAD OF FAILURE ANALYSIS. Itis worth noting that whife pump problems and failure ine dents eat often be trace! hack to a iven root cause oF origin catastrophic failures are rarely the result of only a single viola tion, Many times, a seres of omissions, eversights, or errors ‘combine and lead to the inevitable flare Puinp users can do mich to reduce the risk of experiencing uipient failures, Wel thonght-ont specifications tand docnment reviews, compliance with uneompro Stallation provedsres, oprator and mechanival work foree tain ing, and a gove! combination of preventative and predictive maintenance (periodic cndition monitoring) are just a Few of the proven wass that corte to mind ‘When flutes do occur, there is no substitute for applying, properly: fueused and repeatable approaches to failure analysis And troubleshooting. Tho use of checklists is encouraged, as is the “systems approach,” and collecting all the pieces before a tempting to determine what happened, “The next step is forthe failure analyst o machinery tom bileshooter to remember the seven prineipal causes eatexories aud to rank them in loge order. Ustngga process ofelimination, the most prububle cause categories, or perhaps the ones that are most easly and rapidly sereened, are investigated first, The four basic agents of machine's eomponent and part failure mechs nism also have to be kept in mind, The final and most important review will then almost naturally focus on the one area sshich fontalns the root cause fa failure event REFERENCES 1. Bloch, H. P.,and Geitner, F. K.; Machinery Failure Analysis and Troubleshooting, Houston, Tesas: Gulf Publishing Com pany (983) Karassik, 1 J, cessing, September 158, pp. ‘Centrifugal Pump Clinic,” Chemieal Prov 123-135, 3. Bluch, H. Ps Improcing Machinery Reliability, Revised 2nd Eadition, Houston, Texas: Gulf Publishing Company (1958),PROCEEDINGS OF THE SEVENTH INTERNATIONAL PUMP USERS SYMPOSIUM