1、07FTM11Helicopter Accessory Gear Failure AnalysisInvolving Wear and Bending Fatigueby: G. Blake and D. Schwerin,Rolls-Royce Corporation - Transmission and StructuresTECHNICAL PAPERAmerican Gear Manufacturers AssociationHelicopter Accessory Gear Failure Analysis InvolvingWear and Bending FatigueGrego
2、ry Blake and Doug Schwerin, Rolls-Royce Corporation - Transmissionsand StructuresThe statements and opinions contained herein are those of the author and should not be construed as anofficial action or opinion of the American Gear Manufacturers Association.AbstractGear tooth wear is a very difficult
3、 phenomenon to predict analytically. The failure mode of wear is closelycorrelated to the lambda ratio. Wear can be the limiting design parameter for long-term durability. Also, thefailure mode of wear can manifest into more sever failure modes such as bending. Presented is a failureanalysis in whic
4、h this occurred.Alegacyaerospacegearmeshexperiencedninefailureswithinatwoyeartimeperiod. Thefailuresoccurredaftermorethaneightyears inserviceand withintight rangeof cycles to oneanother. Eachfailure resultedinthelossofallgearteethwithoriginsconsistentwithclassicbendingfatigue.Noneofthefailureswerede
5、tectedprior to tooth loss.Non-failed gears, with slightly lower time than the failed gears, were removed from service and inspected.Gear metrology measurements quantified a significant amount of wear. The flank form of these worn gearswas measured and the measured data used to analytically predict t
6、he new dynamic load distribution andbending stress. To predict if the failure mode of wear was expected for this gear mesh, an empiricalrelationship of wear to lambda ratio was created using Rolls-Royce field data from multiple gear meshes inmultiple applications.The empirical understanding of wear
7、coupled with the analysis was used in an analytical design ofexperiments (ADOE). The results of the ADOE were then used to guide design changes. Presented are themetallurgicalfailureanalysisfindings,dynamicgearmeshanalysis,theempiricalwearratecurvedeveloped,results of the ADOE, and design changes.Co
8、pyright 2007American Gear Manufacturers Association500 Montgomery Street, Suite 350Alexandria, Virginia, 22314October, 2007ISBN: 978-1-55589-915-81Helicopter Accessory Gear Failure Analysis Involving Wear and Bending FatigueGregory Blake and Doug Schwerin,Rolls-Royce Corporation Transmissions and St
9、ructuresIntroductionGear tooth wear is a very difficult phenomenon topredict analytically. The failure mode of wear isclosely correlated to the lambda ratio 12. Wearcan be the limiting design parameter for long-termdurability. The failure mode of wear can manifestinto more severe failure modes such
10、as bending.Presented is a failure analysis in which this oc-curred.Eight events occurred in a legacy aerospace gearmesh within a three-year time period. The failuresoccurred after more than eight years in serviceandwithin a tight range of cycles to one another. Eachfailure resulted in the loss of al
11、l gear teeth with ori-gins consistent with classic bending fatigue. Thefailures were not detected prior to tooth loss.Non-failed gears, with slightly lower time than thefailed gears, were removed from service and in-spected. Gearmetrologymeasurementsquantifiedasignificantamountofwear.Theflankformoft
12、heseworn gears was measured and the measured dataused to analytically predict the new dynamic loaddistributionandbendingstress. Topredictifthefail-uremodeof wear was expected for this gear mesh,an empirical relationship of wear to lambda ratiowascreatedusingRolls-Roycefielddatafrommul-tiple gear mes
13、hes in multiple applications.The empirical understanding of wear coupled withthe analysis was used in an analyticaldesign of ex-periments(DOE). TheresultsoftheanalyticalDOEwere then used to guide design changes. Present-edarethemetallurgicalfailureanalysisfindings,dy-namic gear mesh analysis, the em
14、pirical wear ratecurvedeveloped, resultsof theanalyticalDOE,anddesign changes.Nomenclature Specific film thickness, the oil filmthickness divided by the compos-ite roughness 3.MOS Margin of safety, the allowablevaluedividedbythepredictedval-ue. MOS should be greater than1.0ChipdetectorA detection de
15、vice used in aero-space gearboxes to detect earlyfailure involving metal generationDOE Design of experimentsRaAverage roughness of involuteprofile minSuperfinishingAn isotropic chemical processperformed to achieve improvedsurface finishTiOil inlet temperature FWrWearrateexpressedininchespercycle x 1
16、015Overview of gear systemThe failed gear mesh was used to drive an electricgenerator installed on a Rolls-Royce gas turbineengine in a helicopter.A schematic of the gear train is shown in Figure 1.The gear meshes are numbered one through four.Gear meshfour is thesparedrive gear mesh that isthesubje
17、ct of thispaper. Thefailedpinionandmat-inggear werecarburized, ground, and shot peenedAMS6265. The engine torque was input throughgear mesh number one.Background of failuresThe initial indication to the pilot was generator out-put failure. Additionally, two aircraft had chip lightindications just pr
18、ior to landing. Two others hadchips discovered on the magnetic plug when trou-bleshootingthecauseofageneratorproblem. Oneaircraft had a chip light during the mission (chip de-tectors had large and numerous chips when ex-amined after flight the engine was removed andreplaced). The sixthevent occurred
19、before takeoffwith mission equipment on and blades in flat pitchposition. In all cases, the aircraft generator driveshaft and spare drive gearshaft in the accessorygearbox (AGB) turned freely. Each failure resultedinallgearteethbeingfatiguedormilledoffthesparedrive gear (Figure 4).2Figure 1. Schemat
20、ic of gear trainEight events occurred over a three-year period.Fiveoftheeightwerefromthesamefleet. Theeightfailures are detailed in Table 1. A lognormal proba-bility plot was createdtoshowthescatter inthefail-ure data. The bending fatigue failures occurred be-yond ten million cycles and within a sma
21、ll range ofcycles as shown in Figure 2.Failure investigationAnengineeringteamconductedadetailedfailurein-vestigation. Dynamics, geometry, manufacturingquality, gear design, application, lubrication quality,operationinthefield,andmaterialswerecategoriesinvestigated in detail. None of these categories
22、were identified as root cause during the investiga-tion. Thefindingsfromeachofthesecategoriesarebeyond the scope of this paper and are notpresented.Table 1. Cycle and hours of failures by dateDate of failure Time (hrs) CyclesNov 2003 1297.7 9.37E+08Feb 2004 1347.0 9.72E+08May 2004 1577.2 1.14E+09May
23、 2004 1705.7 1.23E+09Aug 2004 1738.6 1.26E+09Sep 2004 1687.2 1.22E+09Oct 2004 1214.0 8.76E+08Apr 2005 1253.9 9.05E+08Figure 2. Probability of failure vs. spare drive gear life3Photos of sample post failures of the pinion andspare drive gear are shown in Figures 3 and 4.These photos are typical of al
24、l the failures that oc-curred. A fatigue origin in the root radius area isshown in Figure 5 and thedetached teethcollectedpost failure are shown in Figure 6.Thecasehardness,casedepth,andcorehardnessweremeasuredineachof thefailedgears. Samplefindings are listed in Tables 2. The microstructure,chemist
25、ry, and hardness of the spare drive gearsand the pinion gears were evaluated anddetermined to meet all drawing requirements.Figure 3. Failed pinionFigure 4. Failed spare drive gearTable 2. Sample measurements of sparedrive gearsPinion GearCase hardness HR15N 91 HR15N 90Case depth at root 0.018 in. 0
26、.020 in.Case depth at 1/2 tooth 0.020 in. 0.022Core hardness HRC 40 HRC 40Figure 5. Sample SEM images of fatigueorigin area and crack propagationFigure 6. Condition of detached spare drivegear teethA nonfailed spare drive gear and the mating piniongearwereremovedbyfieldmechanicsandreturnedfor invest
27、igation. The gear set had visible wear ontheloadedtoothsurfaces. ThepedigreeofthisgearsetislistedinTable3.These two gears were manufactured during thesame period as the bending fatigue failures. Fur-ther, the gears had accumulated near the samenumber of cycles as the bending fatigue failures.4Thegea
28、rsetofferedanintactviewofthegearspriorto the bending fatigue failure.Thetoothgeometryofthetwogears wasmeasuredtoquantify thevisiblewear. A largeamount ofwearcouldbeseenintheinvoluteandleadtraces ofbothgears as shown in Figures 7 and 8.Table 3. Pedigree of sample worn gear setPinion GearUser Fleet AT
29、otal time hrs 1147.7Total cycles 1.07E+09 8.29E+08Figure 7. Analytical inspection of worn pinionFigure 8. Analytical inspection of worn sparedrive gearAnalysisThe gear set was originally designed using theAGMAindexmethodforbending,contact,andflashtemperature. The margin of safety (MOS) for1/10,000 f
30、ailure rate is shown in Table 4.Table 4. Baseline AGMA index MOSMOSBending 2.06Contact 1.10Flash temperature 1.085Afineranalysismethodwasusedtoevaluatetheef-fects of the tooth flank wear. The actual involute,lead, and index measurement data from the worngear mesh were used to predict the static and
31、dy-namic mesh conditions compared to nominal blueprint. The analysis was performed at maximumcontinuouspower.Auniversitywrittenloaddistribu-tion program was used to perform the gear meshanalysis. Figure 9 shows the increased range ofpeak to peak transmission errors over one full gearrevolution.Thewo
32、rntoothgeometrysignificantlyincreasedthestatic transmission errors of the gear mesh. Therotationspeedandinertiaswerethenusedtocalcu-latethedynamicpeakroottensilestressofthegear.ThedynamicroottensileMOSforonemeshcycleisshown in Figure 10. The results were calculated at100% and 10% rotational speed fo
33、r blueprint andworn conditions.Figure 9. Static transmission error, one gear revolution, worn vs. blue printFigure 10. MOS dynamic gear root tensile stress over one mesh cycle6Thepeakdynamicrootstressfortheworngeometrywasgreaterthan3Xthatoftheblueprintnominal. Itisassumedthatthegeartoothprofileswoul
34、dcontin-ue to wear with use and the dynamic root stresswould continue to increase with wear.The worn gear tooth geometry provided a dynamicsource of increased bending stress. It is assumedthat the dynamic bending stress would continue toincrease with additional wear.Gear wearReference 3 provides gui
35、dance for predicting theprobabilityofwearasafunctionofpitchlinevelocityand . However, a method was needed to deter-mineifthemeasuredamountofwear wasexpectedgiven the original design assumptions.Severegearandpinionflankwearwasfoundonthebending fatigue failures. Analysis of the sampleworn gears conclu
36、ded a 3X or greater increase indynamic bending stress due to wear.Thirtydatapointswereobtainedandanexponentialcurve fit to the data. The regression coefficientshown in Figure 11 of the curve was lower than de-sired.Thefailuremodeof wearis closelycorrelatedtothelambdaratio12. Fieldexperiencewasexamin
37、edto test the magnitude of this relationship. All of thegears evaluated were carburized and groundAMS6265materialusedwithMIL-L-23699lubrica-tion. The field experience was collected from hightime gears (1k hours) with known hours of opera-tion. Data were collected from high power turboprop gear meshe
38、s and accessory gear meshes.Wearwasquantifiedusinganalyticalinspectionandassumed that the unloaded side of the tooth repre-sented the unworn tooth geometry. Specific oil filmthickness () was calculatedandawear rateestab-lished for each gear. Figure 12 is a plot of this dataand the empirical wear lif
39、e curve created.The empirical curve was used to evaluate the wornpinionandsparedrivegearshownpreviouslyinFig-ures7and8. Table5comparesthemeasuredwearversusthepredictedwearusingtheempiricalcurve.The predicted flank wear for this gear as a functionof cycles is plotted in Figure 12. The actual wearand
40、cycles completed were also plotted.Table 5. Spare drive gear wearmeasured vs. predicted calculated0.181Spare gear wear cycles 8.29E+08Gear flank wear (inches)Predicted 0.0015Measured 0.0019Figure 11. Wear rate vs. 7Figure 12. Estimated flank wear of spare drive gear using empirical wear rate curveTh
41、e predicted wear was minimal up to 108cyclesand then dramatically increased.The bending fatigue failures occurred between9X108and 1.3X109cycles. The empirical curvewouldpredictasignificantamountofflankwearandthe dynamic analysis would predict a 3X or greaterincrease in bending stress after the wear
42、reachedthis point.Wear in the fieldAfter identifying the link between bending fatigueandwear,thenextstepwastofurtherisolatetheex-tent of the wear problem for the spare-drive gearmesh.ConditionreportswerepulledfromteardowninspectionsinFleetA. Fromthesereports,anythingresembling surface distress on th
43、e gear teeth wasdocumented. The data showed that surface dis-tress was consistent year to year averaging about14%of thetotal fleet and averagetime of indicationwas identifiedas1160hoursas showninFigure13.A statistical difference was identified between sur-face distresses in gearboxes operating fleet
44、 Acompared to the rest of the entire fleet as shown inFigures14and15. Thisprovedtobeusefultoaidinroot cause identification.As shown in Figures 14 and 15, the mean time be-tween removal for surface distress was higher forFleet A applications compared to the remainingfleet.Design of experimentsA desig
45、n of experiments was conducted usingana-lytic methods to bothidentify theprimary drivers forgear tooth wear and also quantify the designspacefor thesignificant factorswithrespecttotoothwear.Analyticmethodswerechosenoveramoreconven-tionalDOE duetocosts associatedwithtestingandacquiringhardware. DOEru
46、ns aremuchcheaperifthe code exists and the error associated with thetest is at an acceptable level. In the case of thespare drive, a program consistent with AGMA 925,was used to capture the inputs and relate that to alambda value. The empirical curve seen in Figure11was thenusedtoconvert lambdato an
47、associat-edwearrate. Thewearratethenbecametheoutputof the DOE.8Figure 13. Spare drive gear reported field wearFigure 14. Fleet A vs. rest of fleet run chartFigure 15. Fleet A vs. rest of fleet box chart9InanefforttofullycapturethesystemandultimatelythecorrectinputstotestduringtheDOE,aParame-terDiagr
48、am(P-Diagram)wasused(seeFigure16).The P-Diagram was also used to capture the link-age between the inputs, design variables, noise,andtheoutput. This wasimportant beforecomplet-ing the DOE because it was an aid to ensure thatfactors and points outside of normal operationwouldnotbetestedandpotentially
49、skewtheresults(see Figure 17).The experiments conducted were as follows: ascreeningdesign,optimizedesign,andfinallyasur-face response. The screening design was a 2-lev-el,12factorResolutionIVdesign,intendedasafirstpass used to eliminate the insignificant factors andfocusonthesignificantmaineffects. AsseenintheFigure 18 the most significant factors were surfacefinish followed by oil temperature. The pareto(Fig-ure 18) chart described the standardized effect foreachfactor as themodelwas reduced. For this ap-plication,factorswithstandardizedeffectsles
