Assessing Bearing Health for Helicopter Power Train Systems

Transcription

1 Assessing Bearing Health or Helicopter Power Train Systems Harrison H. Chin Applied Concept Research, Inc. Bedord, MA Eric Mayhew and avid L. Green Goodrich Corporation Fuel & Utility Systems Vergennes, VT Abstract Bearings are critical components in helicopter power train systems. A ailing bearing not only could jeopardize the saety o light but also induce collateral damages, adding additional maintenance costs. etecting bearing aults remains a challenge because o eeble signatures generated by bearings that are oten masked by other noise in the power train system. HUMS data collected to date rom 30 U.S. Army BlackHawk helicopters enables us to learn more about bearing ault behaviors in an operational environment. The lessons learned in this program will allow us to urther advance bearing health assessment methodology and to achieve Condition-Based Maintenance. Introduction The U.S. Army is in the process o transitioning rom current Time-Based Maintenance (TBM) to Condition- Based Maintenance (CBM) or their UH-60L and other helicopters. Unlike TBM where maintenance or retirement o components is perormed based on a ixed operating hour schedule, CBM requires timely replacement or repair o components beore they ail (i.e., unscheduled maintenance). It can be argued that CBM o a helicopter power train system requires automated diagnostics and prognostics to provide current component status, and predict the time remaining beore the useul lie is consumed. This new maintenance philosophy promises to reduce the overall maintenance costs while improving aircrat readiness. Bearings are one o the power train components that will be subject to CBM. Early detection o bearing degradations allows the maintainer to plan corrective actions so as to minimize the impact to readiness, and in many cases minimize collateral damages. Planning lead time allows the replacement assembly to be ordered and delivered beore the aircrat becomes unserviceable, and allows the replacement eort to be accomplished when the technicians, space, and special tools are available. Similarly, minimizing collateral damages has the potential o minimizing the cost and oreshortening the time required to overhaul the parent assembly. Presented at the American Helicopter Society 61 st Annual Forum, Grapevine, TX, June 1-3, Copyright 2005 by the American Helicopter Society International, Inc. All Rights Reserved. The Army has been taking several concrete steps toward CBM, including a battalion-level demonstration o the Goodrich Integrated Mechanical iagnostics Health and Usage Monitoring System (IM HUMS) installed on 30 UH-60L BlackHawk helicopters [1]. The Goodrich IM HUMS provides a mix o industry standard and privately developed diagnostics and management techniques. Much o the privately developed technology is centered in the areas o rotor track and balance (RTB) and automated mechanical diagnostics (AM). This technology has been developed through years o research and development in concert with government personnel in test stand environments and on aircrat data collection. Techniques and concepts have been developed or measuring mechanical state and assessing mechanical health. These techniques have been incorporated into the IM HUMS and are now gaining large scale application. Exposing the general rules, developed on a small sample set and test cells, to a large sample size has uncovered areas that need additional interpretation. Reining the algorithms and techniques has always been an expected stage o the pre-planned deployment or this technology. Automated Mechanical iagnostics (AM) is a unique challenge or several reasons. In contrast to main rotor out o balance conditions, mechanical aults are rare in normal operation. It is a challenge to provide evidence, o a reliable diagnostic system, in the absence o mechanical aults. Without operational evidence, theory and more importantly test cell data is all that can be provided. Navy and Goodrich eorts have led to a compilation library o various ault signatures. The eorts in understanding theoretical responses to simple

2 mechanical anomalies and measuring the response on a controlled test acility have driven the implementation as seen today on the IM HUMS AM unction. The second major challenge is to properly model the relationship o measured responses to component state based on visual representations o the data. The planned approach Goodrich has taken to mitigate the eects o development on a inite sample size is to incrementally update thresholds and data blending techniques as the leet data is acquired. In the absence o aulted data, deviation rom leet normal is being reported. Provided that enough evidence has been gathered to conidently say that the measurements are sensitive to degradation modes, deviation rom statistical leet normal is interpreted as approaching or achieving ault detections. In addition to HUMS ground station sotware provided to the Army personnel, Goodrich has developed a versatile engineering sotware utility called Mechanical iagnostics Analysis Toolkit (MAT) that enables detail analysis o HUMS data Since 2002, the MAT has been used to conduct detailed data analysis and accomplish rapid prototyping needed to respond to the unique challenges associated with operations o the UH- 60L power train in a variety o operational environments. This data has given the researchers the means to reine and optimize HUMS unctionalities, including bearing diagnostic algorithms. Furthermore, the data also serves as the basis or multi-aircrat statistical analysis, as well as or the development o advanced prognostic algorithms. Given the number o aircrat involved and the amount o data generated, a health evaluation tool has been developed to perorm health assessment or every power train component automatically. With the aid o this tool, aircrat with the worst component(s) can be readily identiied or special analysis, the results o which have been used to validate automated detections. In this paper, issues related to bearing health assessment are discussed, which reveal the similarities and dissimilarities between theoretical understandings and actual observations o bearing degradation. Several important indings related to IM HUMS bearing data are described. The Bearing Challenge etecting bearing aults remains a challenge. A aulted bearing generates minute vibrations that are oten masked by other noise sources in the power train. The bearing signatures are also subject to a certain amount o scatter due to variability in installation and operating environment. This urther complicates the eorts to properly detect and quantiy bearing aults. Many bearing monitoring techniques such as vibration analysis, acoustic emission, and oil debris analysis have been developed over the past decades to address this problem. Among them, vibration monitoring has been proven to be a reliable and cost-eective technique or on-board implementation. Onboard implementation allows data to be downloaded upon landing and in some cases allows onboard alerts o impending in-light ailures. This last eature requires a detection system that is comprehensive, extremely accurate and highly reliable. Bearing Fault Progression Bearing ault progression typically can be described by 4 distinct stages [2]. Symptoms oten start at high requency excitation and move to lower requencies as damage progresses, as shown in Figure 1. In Stage I, there are some ultra high requency activities in the ultrasonic region (Zone 4), which represent the spike energy produced by bearing micro-deects. It requires a sensor speciically designed to detect such spikes in this region. Visual inspection o the bearing at this stage may not show any identiiable deects. In Stage II, the aulty bearing begins to generate signals associated with natural requencies o the bearing parts as they are being excited by the deects. It produces a notable increase o signals in Zones 3 and 4 in this stage. Beginning signs o deects will be ound upon inspection. Fundamental bearing deect requencies start to appear in Stage III o bearing damage. There may be higher harmonics o these requencies, depending upon the quantity o deects and their distribution around the bearing races. These requencies are sometimes modulated by the shat requency as the deects interact with the shat rotation. Bearing deect requencies and their harmonics are observed in Zone 2. Signals in Zones 3 and 4 continue to grow throughout this stage. Stage IV is the last condition beore catastrophic ailure o the bearing. This stage is associated with numerous modulated undamental requencies and harmonics, indicating the deects are distributed around the bearing races. ue to the increased degradation o the bearing the internal clearances become larger, which then allow the shat to vibrate more reely. Consequently an increase in shat undamental requency and harmonics is observed, which can be associated with imbalance, misalignment, and/or looseness o the shat. Toward the

3 end o Stage IV, the bearing undamental requencies will actually decrease and be replaced with an elevated noise loor or hay stack at higher requencies. Signals in Zone 4 will decrease ollowed by a signiicant increase just prior to catastrophic ailure [2]. Bearing Vibration Monitoring Generally speaking vibration monitoring is the most widely used technique or detecting rolling element bearing ailures. Vibration characteristics rom bearing aults are well documented, and proven techniques to extract such inormation are available [3][4]. Furthermore the accelerometers used or bearing monitoring are also suitable or monitoring shats and gears in the power train system. This makes it the most cost-eective solution without additional hardware. In a rolling element bearing, the undament shat vibration is supplemented by the mechanics inherent with the additional moving bearing parts. A healthy bearing generates very little or no vibration while a damaged bearing has a distinct vibration signature. This vibration signature is typically proportional to the bearing load. Bearing ault development typically starts with a crack or spall in the outer race. These aults can usually be detected rom several months up to hal a year beore they become critical. These aults may become smoother ater a while. Inner race aults are oten developing as a result o an outer race ault, and may be detected some weeks beore they become critical. Inner race aults oten shows side bands due to modulation. Faults on rolling elements and cages are a sign o the inal ault development o a bearing. Bearings with these aults will oten last or only a ew hours [5]. The bearing aults can be detected with envelope analysis technique which reveals the periodic impulses produced by a bearing deect. Envelope analysis demodulates the vibration signal in a narrow requency bandwidth (in Zone 2), typically centered on a structural resonance. The time period between these periodic impulses, which are enhanced by the resonance, gives an indication o the location and nature o the bearing damage because the impulse requencies can be traced back to the our bearing deect passage requencies: cage, ball/roller, outer race, and inner race. The above our bearing deect requencies can be calculated based on the measured shat speed and bearing geometric parameters: pitch and rolling element diameters, number o rolling elements, and contact angle. Figure 2 shows the geometric parameters or a ball bearings and equations or the deect requencies [5]. Bearing manuacturers normally supply the geometric data and/or the bearing deect requency ratios as part o the bearing speciications. The actual requencies may dier slightly due to loads and slippage but are expected to be close to the calculated requencies. However the calculated requencies become unreliable i the ratio between the radial and axial orces is not constant (e.g., change in contact angle) or there is excessive ball sliding (e.g., loose clearance it). In addition to vibration monitoring, the ollowing techniques are oten used or bearing diagnostics: Acoustic or ultrasound emission sensors designed to detect bearing micro-deect signatures in the ultra high requency range (above 250 khz). Thermal probes to detect any temperature changes or a bearing. Increase in measured temperatures normally indicates a possible loss o lube or excessive bearing deects that lead to increased riction between rolling elements, which can then result in bearing ailures. Oil debris monitor and chip detector are used to detect the presence o wear metals and to iner the integrity o lubricant. Oil analysis using atomic emission or atomic absorption instruments is designed to detect the presence o wear metals in extremely low concentrations by spectrometric analysis o luid samples. The IM HUMS utilizes the envelope analysis technique on vibration data acquired rom multiple high requency accelerometers to determine bearing health. Several Condition Indicators (CIs) are extracted rom the envelope data and subsequently used into a Health Indicator (HI) to relect the health o each bearing. The HI has a value between 0 (healthy) to 1 (ailure). Bearing Health Analysis One o the primary goals o the Army HUMS demonstration program is to veriy and validate the IM HUMS CIs and HIs with regard to their diagnostics perormance. Contrary to test-stand environment [6], perorming component health management in an operational environment or 30 aircrat is a complex task, due to dierent stages o component usage and aircrat-to-aircrat variability. Each aircrat s power train system contains more than 70 bearings that need to be monitored. That represents a total o over 2,100 individual bearings or 30 aircrat. To automate the health assessment process, a comparative analysis tool has been developed. The

4 result o this analysis is a list o components that show elevated and/or signiicantly increasing HIs. The HIs in the list are the result o blending a set o selected bearing CIs by employing both rule-based and datadriven statistical approaches. The list then serves as the basis or urther engineering investigations on those components. With the aid o this comparative analysis, several bearings have been identiied as having potential aults. One such bearing is the Intermediate Gear Box (IGB) Input Thrust Bearing on one o the aircrat. The health o this bearing rom the 30 aircrat is shown in Figure 3, ranked according to their HIs. It can be seen that among the 30 aircrat, the HI o AC # 11 has the highest value (nearly 0.9) that is more than 4 times higher than the second highest aircrat. This elevated HI prompted a more comprehensive engineering analysis on the subject bearing. The CIs and, in some acquisitions, raw/intermediate data can be viewed and extracted rom the MAT tool. Figure 4 shows a screen shot o the MAT tool displaying the CIs or the subject bearing, where an increasing trend can be observed on several CIs. To conirm a bearing ault, the CIs must be traced back to one or more o the bearing deect requencies mentioned earlier. For this, envelope spectrums rom several acquisitions were extracted and examined. Figure 5 shows the topographical plot o 147 envelop spectrum waterall, sorted by the averaged engine torque. Several distinct tones associated with bearing requencies are seen throughout the plot, including the cage, outer race (OR), inner race (IR), and 2X OR. In addition to the bearing requencies there is also the gear mesh (GM) requency which appears more consistently at higher torque. Figure 6 shows the comparison between one o AC # 11 envelope spectrums and one rom AC # 6, representing a healthy bearing. The OR and 2X OR tones can be clearly identiied in the AC # 11 spectrum, accompanied by a higher noise loor. The above analysis has demonstrated the IM HUMS bearing diagnostics capability in identiying a aulted bearing. The next step is to minimize the scatter in CIs so a more consistent HI can be obtained as oppose to the one shown in Figure 7. Upon close inspection o the calculated OR requencies vs. the actual requencies in the envelope spectrums, it was ound that the actual requency is on average 1.5 Hz lower than the calculated (see Figure 8). This suggests that the bearing geometric parameters used to calculate the bearing requencies are not exact, or may be due to excessive slippage in the bearing. The only possible parameter to change is the contact angle that also aects the other three bearing requencies. The only other visible requency in the spectrum is the IR requency which showed the actual requency is on average 1.87 Hz higher than the calculated. By reducing the contact angle by a ew degrees in the equations, a much better it was achieved or both OR and IR requencies. Such ine-tuning o the bearing deect requencies will enable us to obtain more consistent CIs and HIs. Another bearing example involves the shiting o a bearing deect requency. It is generally assumed that the our bearing deect requencies are approximately constant or a given shat speed. This is because each o those requencies is expressed as a unction o shat speed and bearing geometric parameters only, as shown in the equations in Figure 2. Recent observations on one aircrat (AC # 7) reveal several elevated ball energy CI values o Right Input Spiral Bevel Pinion Gear Ball Bearing (see Figure 9). These high data points ar exceed the nominal data scatter observed on other aircrat. An in-depth analysis was conducted to determine the source o these high data points. Ater examining the envelope spectrums associated with this bearing, the ball requency and its shat modulation requencies were ound to vary with torque, that is, higher torques produce higher requencies, as shown in Figure 10. It also shows that there is very little vibration when the engine torque alls below 40%. Figure 11 shows the ball requency as a unction o the right engine torque. The ball requency changes rom low to high with increasing torque, with a value between 1,720 Hz (40% Q) and 1,920 Hz (80% Q). The expected requency is around 1,820 Hz, which corresponds to roughly 50% Q. It is postulated that there is a change o ratio between the axial and the radial load when dierent power is applied to the bearing rom the shat. In addition, since the deect has compromised the internal clearance (i.e., larger gap has developed) and thereore it promotes more excessive ball sliding that can also aect the deect passage requency. In addition to the two cases described above, this comparative analysis has also uncovered several instances where tail drive shat bearing HIs are elevated but no actual bearing deect tone was ound. Instead, the HIs are driven by high level o broad band noise.

5 They later were identiied by ield engineers as improper shimming that produced excessive end-load on the bearing. Other installation problems can also produce high shat vibration which in turn produces looseness and higher load on the bearing. The above installation problems are believed to have contributed to accelerate bearing wear and premature ailures. The ability o IM HUMS to identiy installation problems is key to prolonging bearing lie. Conclusions The decision to build the Army IM HUMS program around 30 aircrat that have accumulated an average o 2,000 light hours, and were involved in a variety o typical tactical operations has been validated by the results accumulated to date. The ability to make onaircrat detections o bearing degradation is a current capability o the IM HUMS as installed on the UH- 60L aircrat. Several analysis tools/techniques such as MAT and comparative analysis have been shown to be eective in HUMS data analysis and interpretation. MAT has been proven to be a valuable tool in the control o data scatter, and provides an integrated development environment or rapid prototyping trending techniques that are essential to advanced leet wide diagnostics. Several important indings related to bearing aults have been identiied by the above program, which were not observed during earlier controlled test cells or single aircrat testing. These indings give us an insight into bearing ault behaviors in a complex helicopter power train system under real-lie operations. They also allow us to reine/develop more eicient and reliable diagnostic/prognostic models, which is a crucial step in the transition to CBM. Signiicant indings o IM HUMS bearing data and their implications are summarized below: The onboard detection o the aulted IGB input thrust bearing and post-light analytical validation conirms the ability o the IM HUMS to automatically detect progressive bearing aults. The tail rotor drive shat bearing detection reported above validates the IM HUMS ability to automatically detect improperly installed drive train bearings. The improvement o bearing algorithms to assure reliability o ault degradation detection in the presence o broad band noise demonstrates the ability o the AM bearing algorithms to unction correctly in a high noise environment. Airworthiness is assured by the AM because o early detection so as to avoid light with a bearing in a near ailure status, as well as automatic in-light detection and reporting o extremely high HIs that indicate the probability o an imminent bearing ailure. Battalion level CBM is acilitated by the ability to trend HIs so as to predict the service lie remaining. Fleet management CB logistics is acilitated by early detection, tracking, and reporting the status o bearings showing progressive degradation. Reerences [1] ora, Ray, Wright, J., Hess, R., and Boydstun, B., Utility o the IM HUMS in an Operational Setting on the UH-60L Blackhawk, AHS International Forum 60, Baltimore, M, [2] Rolling Element Bearings, STI Field Application Note, Sales Technology, Inc., League City TX. [3] Randall, R. R., Antoni, J., and Bonnardot, F., Recent Advances in Bearing iagnostic Techniques or Helicopter Gearboxes, Proceedings o the 58th Mechanical Failures Prevention Technology Meeting, Virginia Beach, VA, [4] Chin, H., Automated Bearing Fault etection with Unsupervised Neural Nets, Proceedings o the 49th Mechanical Failures Prevention Technology Meeting, Virginia Beach, VA, [5] Eisenmann, R. C and Eisemann R. C., Jr., Machinery Malunction iagnosis and Correction, ISBN , Prentice Hall PTR, Upper Saddle River, NJ 07458, [6] Hess, A., Hardman, W., Chin, H., and Gill, J., The US Navy s Helicopter Integrated iagnostics System (HIS) Program: Power rive Train Crack etection iagnostics and Prognostics, Lie Usage Monitoring, and amage Tolerance; Techniques, Methodologies, and Experiences, NATO Conerence, 1999.

6 ZONE 1 ZONE 2 ZONE 3 ZONE 4 Normal Stage I Stage II Stage III Stage IV Figure 1. Typical bearing ailure progression which can be described by 4 stages with signatures in 4 zones. Number o balls Shat req. Outer race req. Inner race req. Ball req. Cage req. or ir N s ball = 1 cos β 2 pitch N s ball = 1 + cos β 2 pitch s b = 2 pitch ball ball pitch s ball c = 1 cosβ 2 pitch 2 (cos β ) Figure 2. Rolling element bearing geometry and deect requencies (courtesy o [5]).

7 IGB Input Thrust Bearing HI Comparison HI Aircrat # Figure 3. IGB Input Thrust Bearing HI comparison or the 30 aircrat. Aircrat # 11 has an elevated HI among 30 aircrat. Figure 4. Goodrich MAT Mechanical iagnostics engineering tool showing CIs or the IGB Input Thrust Bearing or Aircrat # 11.

8 CAGE OR IR GM 2X OR 70% 60% 50% 40% 20% Figure 5. Topographical view o AC # 11 IGB Input Accelerometer envelope spectrums rom 147 acquisitions. Outer Race 2X Outer Race Figure 6. Envelope spectrum comparison brtween AC # 11 (with bearing deect tones) and AC # 6 (without bearing deect tones).

9 Figure 7. AC # 11 IGB Input Thrust Bearing HI time history showing an increasing trend. Figure 8. Comparison between calculated and actual IGB Input Thrust Bearing outer race requencies at various shat speeds. An oset between the actual and calculated requencies is observed.

10 Figure 9. MAT shows several elevated Ball Energy CI values (inside red ellipse) or AC # 7 Right Module Input Spiral Bevel Pinion Gear Ball Bearing. 70% Ball Frequency 60% 50% 40% Engine Output Shat Shat Modulation Figure 10. Topographical view o AC # 7 Right Module Input Flange Accelerometer envelope spectrums rom 53 acquisitions.

11 Ball Frequency vs. Torque Ball requency (Hz) Torque (%) Figure 11. Ball requency vs. engine torque or AC # 7 Right Module Input Spiral Bevel Pinion Gear Ball Bearing.