9-8MV-4 Health Monitoring or Condition-Based Maintenance o a HMMWV using an Instrumented Diagnostic Cleat Copyright 9 SAE International Tiany DiPetta, David Koester, and Douglas E. Adams, Ph.D. Center or Systems Integrity, Purdue University, 5 Kepner Drive, Laayette, IN 4795 Joseph Gothamy, Paul Decker, David Lamb, Ph.D., and David Gorsich, Ph.D. Tank Automotive Research, Development and Engineering Center, Warren, MI 48397-5 ABSTRACT Operation & support costs or military weapon systems accounted or approximately 3/5 th o the $5B Department o Deense budget in 6. In an eort to ensure readiness and decrease these costs or ground vehicle leets, health monitoring technologies are being developed or Condition-Based Maintenance o individual vehicles within a leet. Dynamics-based health monitoring is used in this work because vibrations are a passive source o response data, which are global unctions o the mechanical loading and properties o the vehicle. A common way o detecting aults in mechanical equipment, such as the suspension and chassis o a ground vehicle, is to compare measured operational vibrations to a reerence (or healthy) signature to detect anomalies. The main diiculty with this approach is that many vehicles are not equipped with sensors or the acquisition systems to acquire, process, and store data; thereore, to implement health monitoring, one must overcome the economic and technical barriers associated with equipping ground vehicles to continuously monitor the response. The research in this paper explores one approach that aims to overcome this diiculty. I a vehicle cannot be equipped with sensors, then an instrumented diagnostic cleat is proposed to measure the dynamic response o the vehicle as it traverses the cleat at a ixed speed. This approach could be eective because it eliminates the need or on-vehicle sensors, but provides measurements that indicate the condition o wheels/suspensions. A simple model o a HMMWV is used to simulate the approach. Experiments are also conducted using an instrumented cleat to demonstrate the easibility o this approach. INTRODUCTION The U.S. Army is pursuing technologies that will enable Condition-Based Maintenance (CBM) o ground vehicles. Current maintenance schedules or ground vehicles are determined based on reliability predictions (e.g., mean time to ailure) o a population o vehicles under anticipated operational loads; however, vehicles that experience component damage oten lie in the tails o the reliability distribution or a given platorm. For example, a certain group o vehicles may be deployed to operate on a harsh terrain that is particularly taxing on the mechanical components in the suspensions or rames o those vehicles. Operation & support costs or military weapon systems accounted or approximately 3/5 th o the $5B Department o Deense budget in 6 (Gorsich, 7). To ensure readiness and decrease these costs or ground vehicle leets, health monitoring technologies are being developed to assess the reliability o individual vehicles within each leet. Based on a review o the open literature including Technical Note 85-3 (Thomas, 985) on ground equipment reliability issues associated with materials, it can be concluded that the most common aults occur in wheel ends (tires, brakes), suspensions, and rames. For example, Aardema (988) discussed a ball joint ailure in the HMMWV (High Mobility Multi-purpose Wheeled Vehicle). Braking systems have also experienced wear most likely due to severe operating conditions such as overheating. Reliability issues in suspensions due to wheel weights have also been reported (FORSCOM, 4). Faults in the HMMWV body chassis and rame have also been reported in reliability centered maintenance studies (Lasure, 4). Dynamics-based health monitoring is used in this work to identiy such aults because vibrations are a passive source o response data, which are global unctions o the loading and mechanical properties o the vehicle. A common way o detecting aults in mechanical equipment, such as the suspension and chassis o a ground vehicle, is to compare measured vibrations to a reerence (or healthy) signature to detect anomalies. In order to make this comparison, a library o vibration signatures must be developed and categorized according to the operational conditions o the vehicle
Report Documentation Page Form Approved OMB No. 74-88 Public reporting burden or the collection o inormation is estimated to average hour per response, including the time or reviewing instructions, searching existing data sources, gathering and maintaining the data needed, and completing and reviewing the collection o inormation. Send comments regarding this burden estimate or any other aspect o this collection o inormation, including suggestions or reducing this burden, to Washington Headquarters Services, Directorate or Inormation Operations and Reports, 5 Jeerson Davis Highway, Suite 4, Arlington VA -43. Respondents should be aware that notwithstanding any other provision o law, no person shall be subject to a penalty or ailing to comply with a collection o inormation i it does not display a currently valid OMB control number.. REPORT DATE 5 OCT 8. REPORT TYPE N/A 3. DATES COVERED - 4. TITLE AND SUBTITLE Health Monitoring or Condition-Based Maintenance o a HMMWV using an Instrumented Diagnostic Cleat 6. AUTHOR(S) Tiany Dipetta; Nathanael Yoder; Douglas E. Adams; Joseph Gothamy; David Lamb; David Gorsich 7. PERFORMING ORGANIZATION NAME(S) AND ADDRESS(ES) US Army RDECOM-TARDEC 65 E Mile Rd Warren, MI 48397-5 5a. CONTRACT NUMBER 5b. GRANT NUMBER 5c. PROGRAM ELEMENT NUMBER 5d. PROJECT NUMBER 5e. TASK NUMBER 5. WORK UNIT NUMBER 8. PERFORMING ORGANIZATION REPORT NUMBER 9335RC 9. SPONSORING/MONITORING AGENCY NAME(S) AND ADDRESS(ES). SPONSOR/MONITOR S ACRONYM(S) TACOM/TARDEC. DISTRIBUTION/AVAILABILITY STATEMENT Approved or public release, distribution unlimited. SPONSOR/MONITOR S REPORT NUMBER(S) 9335RC 3. SUPPLEMENTARY NOTES Presented at 9 SAE World Congress, April 9, Detroit, Michigan, USA, The original document contains color images. 4. ABSTRACT 5. SUBJECT TERMS 6. SECURITY CLASSIFICATION OF: 7. LIMITATION OF ABSTRACT SAR a. REPORT unclassiied b. ABSTRACT unclassiied c. THIS PAGE unclassiied 8. NUMBER OF PAGES 9 9a. NAME OF RESPONSIBLE PERSON Standard Form 98 (Rev. 8-98) Prescribed by ANSI Std Z39-8
(speed, terrain, turning radius, etc.). Figure illustrates this common approach to ault identiication. There are two principle diiculties with this approach. First, the number o datasets required to develop a library o possible healthy signatures extracted rom an N-dimensional sensor suite on a vehicle given M terrains on which that vehicle can operate is o order M N (Bishop, 99). For example, 6 sensors over terrains would require that one million datasets be used to establish a ully populated reerence set or ault detection. I 4 datasets are acquired each day on average, then it would take years to develop this library o healthy signatures or each individual ground vehicle. This large number o datasets would be needed to characterize the normal operational response o the vehicle due to the non-stationary nature o the loading and the inability to control these loads in operation. Second, many vehicles are not equipped with sensors nor the acquisition systems to acquire, process, and store data; thereore, to implement health monitoring or condition-based maintenance, one needs to overcome the economic and technical barriers associated with equipping ground vehicles to continuously monitor their responses. because it eliminates the need or on-vehicle sensors. () I a vehicle can be equipped with sensors, then a reerence-ree approach to data analysis is used to compare similar response pathways on the vehicle to identiy mechanical anomalies. For example, the vertical and tracking responses o the let wheels can be compared to the same responses o the right wheels to determine i the ront/rear wheels exhibit anomalies. This approach could be eective because it diminishes the need or reerence signatures to identiy aults; however, the approach is still somewhat sensitive to operational loading and variability. Figure : Illustration o concept or an instrumented cleat that diagnoses aults in wheel ends and suspensions o ground vehicle. This paper investigates Approach (). Some o the merits o this approach are as ollows: Figure : Illustration o common approach or diagnosing aults in ground vehicles using dynamic response and operational data. To overcome these diiculties, it is useul to consider how rotating machinery diagnostic systems unction. In these machines, the repetitiveness o the operating load or a machine operating at constant speed makes it relatively easy to identiy aults in the bearings, shat, etc. In wheeled ground vehicles, loading varies signiicantly as mentioned above. I loads acting on the vehicle could be ully measured or controlled in terms o the terrain input motions and/or spindle orces/moments, ault identiication in wheeled vehicles at the component level would be more straightorward. Mechanical properties that determine the vehicle condition could be extracted rom data i loads could be controlled. There are two approaches being undertaken in this research that attempt to overcome the diiculties mentioned above: () I a vehicle cannot be equipped with sensors, then an instrumented diagnostic cleat is proposed as illustrated in Figure to measure the dynamic response o the vehicle as it traverses the cleat at a ixed speed. This approach could be eective - Cleat is portable making it practical or ield use. - Cleat can be engineered to control the amplitude and requency o the input imparted to the vehicle wheels allowing or more targeted diagnostic results. - Vehicle speed traversing the cleat can be controlled. - Coniguration o cleats can be designed to develop speciic tests or certain subsystems. - Sensors are installed within the cleat rather than the vehicle providing greater reliability. - Algorithms or analyzing response data rom the cleat are less complex than or on-vehicle diagnostic algorithms, which must address non-stationary data. First, a simpliied hal-car our degree o reedom model o a typical HMMWV (see Figure ) is used to examine the sensitivity o measured orces in the tire to aults in the wheel and suspension. Second, a proo o concept experiment is conducted using an instrumented cleat to attempt to diagnose a ault in the strut coil spring. LITERATURE REVIEW The response o the HMMWV to a cleat excitation has been studied by Faller, Hillegass, and Docimo (3). The response o the center o mass, driver, and let and right wheel o the HMMWV was experimentally determined with accelerometers during a road test over a 4 inch high semicircle cleat. The road test was conducted at vehicle speeds o 5 and 4 mph. The speed was ound to aect the response o the vehicle. Health monitoring systems usually place all measurement instrumentation on the vehicle itsel to
measure vehicle responses. However, Champoux, Richard, and Drouet (7) have used an instrumented cleat to study the wheel response o a bicycle. The cleat was instrumented with biaxial orce transducers. The rest o the bicycle was instrumented with strain gages and accelerometers to measure the cyclist s comort. MODELING APPROACH Dimensional and material data was obtained in the open literature regarding American General s standard HMMWV. Figure shows a photo o the vehicle, which has been represented using a our degree o reedom lumped parameter model. It has a length o 4.6 m, width o. m, height o.8 m, and mass o 34 kg. The rame is modeled as a rigid body with three lumped masses, M j with j=,, and 3, representing the ront, rear, and center o mass payloads carried by the vehicle. The mass moment o inertia about the center o mass is I cm3. Dimensions a and b describe the location o the center o mass. The tire stiness properties are denoted by K and K r or the ront and rear wheels, respectively. K and K denote the ront and rear suspension rate properties, respectively. Although not indicated in the schematic, proportional viscous damping is assumed in the model. The vertical base motions o the ront and rear tires are denoted by x and x. The vertical and pitch motions o M 3 and I cm3 are denoted by x 3 and θ, respectively. The nominal parameter values that were used in the model are listed in Table. Parameter Value M, M, M 3 95, 8, kg M, M r, kg I cm3 kg m a, b, 5 t K, K 5, 4 N/m K, K r 5, 4 N/m Table : Nominal parameter values in our degree o reedom model o HMMWV. Figure 3: American General HMMWV or which model has been developed (Photo is a US Army work). The lumped parameter set o dierential equations corresponding to this model was derived using Newton- Euler methods and is given below: M + M + K K ( a + c) + K + K K K K M 3 K + K r + K ( b c) I cm3 M K ( a + c) + K K x3 θ = x K x x K x r r && x3 && θ + && x M && x r r K ( a + c) ( b c) ( b c) where c=(b M a M )/ (M + M +M 3 ) and an in the stiness matrix indicates a symmetric entry in the matrix with respect to the diagonal. A viscous proportional damping model o the orm, [ ] = α[ M ] + β[ K], α =, β =. () C () is also used in Eq. () to describe the dissipative (nonconservative) eects. The unctions x and x were used to model the proile o the cleat, which provides a base excitation to each wheel at dierent times. x and x were expressed using a Hanning unction o the orm: h πt t T x t cos T or ( ) = c or t > T x ( t) = x ( t T ) b c c (3) where h is the height o the cleat, T c is the time during which a wheel is in contact with the cleat, and T b is the time it takes or the rear wheel to come into contact with the cleat ater the ront wheel has reached the cleat. This approach was used because experimentally the measurement would be triggered at t= sec once the ront wheel reaches the cleat. T c can be calculated using the length o the cleat L and the speed o the vehicle v, T c = L/v. Likewise, T b can be calculated using the distance rom wheel to wheel (wheelbase) w and the speed, T b = w/v. x and x are plotted in Figure 5 or a 5 t wheelbase, in wide cleat, and speed o 5.8 mph. Part o the instrumented cleat design is associated with the requency range over which these cleats excite the vehicle. Thereore, the requency spectra o these base excitation time histories are also plotted in Figure 6. Both inputs produce the same spectral eatures because
they are identical in amplitude but dierent in phase. The bandwidth o these excitations is 94 rad/s. vector rom Eq. () and its derivative. The state variable model is given by, d dt { x} {} x& + M = [ ] 4 4 [ I ] 4 4 [ M ] [ K] [ M ] [ C] {} x {} x& [] [] M r 6 K M K r M r 6 βk x( t) x ( t) x & ( t) βk r x& ( ) t (4) Figure 4: Simpliied our degree o reedom model or HMMWV showing wheels and suspension on ront and rear o vehicle. The desired outputs o this model are the orces inside the ront and rear tires because the goal o the instrumented cleat is to measure orces in the tire to identiy aults in the tires and suspension. Thereore, the output equation used in this state variable model is given by: x,x [m].6.5.4.3. = K K K K r r βk βk βk r x( t) x ( t) βk x& r ( t) x& ( ) t { x} {} x& +..5.5.5 3 Time [sec] Figure 5: x and x cleat inputs acting on ront ( ) and rear ( ) tires. ANALYSIS UNDAMAGED SYSTEM (MODAL PROPERTIES) (5) x,x [m] x -4.9.8.7.6.5.4.3.. 5 5 5 3 35 4 45 5 Frequency [Hz] Figure 6: X () and X () cleat inputs acting on ront ( ( ) tires. ) and rear The input-output model in Eq. () was then rewritten in state variable orm in preparation or conducting timedomain simulations. The state vector in this state space representation o the model consisted o the response The modal properties associated with the ree response o the vehicle model were calculated by solving the corresponding eigenvalue problem using the state matrix in Eq. (4). The eigenvalue ormulation takes the ollowing orm: [ ] 4 4 [ I ] 4 4 [ M ] [ K] [ M ] [ C] { x} {} x& = λ { x} { x& } (6) where {x} is the modal delection shape and λ is the corresponding modal requency (eigenvalue). For the mechanical properties chosen in Table, the eigenvalue problem in Eq. (6) was solved and the modal properties obtained are listed in Table. The irst two modes o vibration are associated with the sprung mass (pitch and bounce) and the second two modes are associated with the wheel hop resonances o the ront and rear. The modal delection shapes are only indicated to two signiicant digits to highlight the dominant degrees o reedom in each mode shape. The our undamped
natural requencies are at.63,.88, 7.9, and 7.9 Hz. Consequently, when the base excitation unctions shown in Figure 6 are applied to the vehicle moving at 5.8 mph, all our modes o vibration will be excited because the bandwidth o the primary lobes in each o the input requency spectra spans the requency range rom to 5 Hz (94 rad/s). I the vehicle is traveling more slowly, it is possible that all modes o vibration will not be excited in the orces that are measured in the tires. Undamped Freq. (rad/s) and Damping Ratio Modal Vector (Two signiicant digits) 4.,.4 [.87. -.4 -.7] T 5.5,.6 [. -.9..7] T 49.6,.89 [-. -. -..] T 49.7,. [-....] T Table : Modal parameters o HMMWV our degree o reedom model. UNDAMAGED SYSTEM (BODE DIAGRAMS) To examine the orces that are produced in the tires o the vehicle as the ront and rear wheels traverse the cleat, the Bode diagrams relating the input displacements to the wheels (x and x ) and the orces in the tires ( and, see Eq. (5)) were constructed. These bode diagrams relate the amplitudes and phases o the input displacements to the amplitudes and phases o the orces measured within the instrumented cleat, which is proposed or use in diagnosing vehicle aults. Figure 7 shows the Bode diagrams or the our requency response unctions relating the tire input displacements to the tire output orces. The modal requencies given above or the sprung vehicle mass are evident in the peaks o the Bode magnitude plots. The two wheel hop requencies are also evident but are much more heavily damped than the bounce and pitch modes as expected rom Table. DAMAGED SYSTEM (BODE DIAGRAMS) Damage due to ractured suspension tie bolts or aulty struts and tires that are underinlated or contain separated plies are analyzed in this work. First, a 5% reduction in K (see Figure 4) is used to model damage in the ront suspension. Figure 8 shows the resulting Bode diagram relating the input displacement at the ront wheel to the orce in the ront tire in the undamaged ( ) and damaged ( ) states. The requency range most sensitive to this damage is the mid-requency range in the vicinity o the resonances o the sprung mass. This result is consistent with the location o the damage in the system relative to the delection mode shapes listed in Table. It is evident rom the bounce motion at 4 rad/s (and to a lesser extent in the pitch motion at 5 rad/s) that there is more delection and velocity across the suspension than in the tire hop delections. Thereore, these motions o the sprung mass are most sensitive to the suspension damage in K. In contrast, the response in the requency range above 4 rad/s is most sensitive to changes in the ront tire rate, K. Mag H,x [db] Phase H,x [deg] Mag H,x [db] Phase H,x [deg] Mag H,x [db] Phase H,x [deg] Mag H,x [db] Phase H,x [deg] 6 4-3 5-5 - 6 3 4-3 5-5 - 3 Frequency [rad/s] 6 4-3 5-5 - 6 3 4-3 5-5 - 3 Frequency [rad/s] Figure 7: Bode diagrams (magnitude and phase) or the ollowing output/input requency response unctions: (a) F /X, (b) F /X, (c) F /X, and (d) F /X. DAMAGED SYSTEM (FORCED RESPONSE) The damage mechanisms in the ront suspension and tire that were simulated in the previous section were again introduced in this section. The orced response in the time and requency domains or the excitation unctions shown in Figure 5 was then calculated. Figure 9 shows the time and requency domain orces in the ront tires or the ault scenario involving a 5% reduction in the ront suspension system. In Figure 9(a,b), two sets o orces in the time and requency domains in the tire are plotted. The solid lines correspond to tire orces in the undamaged ( ) and
damaged ( ) vehicle assuming the orce can only be measured while the tire is traversing the cleat. The dotted lines correspond to the same scenario assuming the orce can be measured throughout the entire time period shown. It is evident that there are subtle changes in the time history due to a ault and more pronounced changes in the requency spectrum. It is interesting to note that the primary changes in the spectrum occur in the requency range dominated by the pitch and bounce degrees o reedom due to the previous conclusion about the sensitivity o the orce in the tire to aults in the vehicle (see Figure 8). increases throughout the entire requency range. A wider cleat places more o the excitation in the lower requency range resulting in larger amplitudes o displacement across the wheels and struts, which increases the sensitivity to aults in the tire. For the suspension ault, the increase in sensitivity is also noticeable or wider cleats but only in the low requency range below 5 rad/s. These results suggest that or a given height, changes in the width o the cleat aect the sensitivity o the measured orce in the cleat to tire aults more than to suspension aults. 5 Mag H,x [db] Phase H,x [deg] 6 5 4 3 3 5 5 3 Freq [rad/s] Figure 8: Magnitude and phase o Bode diagram or input at ront wheel and output orce in ront tire or undamaged case ( ), damage in ront suspension ( ), and damage in ront wheel ( ) showing requency ranges sensitive to damage. The same orced response simulation was perormed or a scenario involving a 5% reduction in the ront tire stiness. Then the resulting orced response or this ault in addition to the orced response or the suspension ault were both subtracted rom the undamaged orced response. The spectral magnitudes o these dierences due to the two distinct aults were plotted as shown in Figure out to rad/s. The eects o the suspension ault ( ) and tire ault ( ) aect dierent requency ranges as explained in Figure 8. Moreover, the suspension ault exhibits larger changes in the low requency range whereas the tire ault exhibits larger changes in the high requency range. When the entire orce time history is measured throughout the vehicle motion, the dierences due to aults are more apparent. However, the dierences are also apparent in the case when only the short segment o orce data is available as the tires traverse the cleat. To examine the eects o a change in the aspect ratio o the cleat, the width was increased by a actor o (4 in) and 3 (36 in), and the change in orce was again calculated or the scenario involving only a ault in the ront tire. The percentage change in orce spectrum was then plotted in Figure or the case when the orce is measured in the tire throughout the entire vehicle travel. The igure shows that as the width o the cleat becomes larger or a ixed height, the sensitivity to the tire ault Force [N] -5.5.5 4 Time [sec] Force [N] Fault - 4 Freq [rad/s] Figure 9: Forced response in the (a) time and (b) requency domains with ( ) and without ( ) a ault introduced in the ront suspension using the complete ( ) and partial ( ) orce time histories. Change in Force [N] 3 - - -3 5 5 Freq [rad/s] Figure : Magnitude o change in orce or a suspension ( ) and tire ( ) ault using the complete ( ) and partial ( ) orce time histories. % Change in Force - - No ault Suspension Tire 3 in in -3 5 5 5 Freq [rad/s]
Figure : Percentage change in magnitude o change in orce or a ront tire ault or a in ( ), 4 in ( ), and 3 in ( ) wide cleat using the complete orce time histories. EXPERIMENTAL VALIDATION To validate the analytical indings, a rubberized cleat was instrumented with two PCB 356B8 tri-axial accelerometers and a small truck was used as the test vehicle. These accelerometers were used to measure the responses on the let and right side o the cleat. These responses are indicative o the orcing unction that acts through the tire as the vehicle traverses the cleat. The accelerometers were positioned in the center plane o the cleat using metal plugs and cables were run out to the data acquisition system through the base o the cleat. The plugs were installed so that they were not touching the ground to provide measurements that would be sensitive to the orces acting through the tire. The instrumented cleat used in the experiment is shown in Figure (a) with a close up o one o the accelerometers and plugs in Figure (b). Figure : (a) Instrumented cleat used in experiments, and (b) tri-axial accelerometer installed inside cleat. The experiment consisted o six tests: a irst baseline, a simulated suspension ault, three simulated tire aults, and a second baseline. The baseline vehicle had no aults and the pressure in all our tires was 35psi. The ault in the vehicle suspension was simulated by inserting a metal spacer into the ront right coil spring o the vehicle as shown in Figure 3. The three dierent tire aults were simulated by reducing the pressure o the ront right tire to 3psi, 5psi, and psi. Figure 3: Metal spacer inserted into coil to simulate suspension ault. Each test consisted o the vehicle being driven over the instrumented cleat at 5 mph ive times and the average accelerations were calculated rom the measure data. The data was initially sampled at 6,384 Hz and then down sampled to 89. Hz to highlight the lower requency content that is more indicative o the wheel end and suspension response. Figure 4 shows the (a) right and (b) let cleat responses in the vertical, lateral, and tracking directions or the irst baseline measurement as the ront tire traversed the cleat. The time histories observed when the back wheels traversed the cleat were similar. Note that the let cleat measurement was slightly delayed by 7 msec relative to the right cleat measurement. The reason or this delay is that the two tires strike the cleat at slightly dierent times. The response amplitudes in the three directions were dierent with a peak acceleration o.5 g. First, the suspension ault simulated as shown in Figure 3 was considered. Figure 5 shows the vertical acceleration spectra or the (a) right and (b) let wheels..5.5 -.5 - -.5.9....3.4.5 Time [sec].5.5 -.5 - -.5.9....3.4 Time [sec] Figure 4: Acceleration responses on (a) right and (b) let sides o instrumented cleat with ( ) vertical, lateral ( ), and tracking ( ) directional responses with 35 psi tire pressure and 5 mph. These plots correspond to the data acquired as the ront wheels traversed the cleat. The solid dark ( ) and dotted dark ( ) lines correspond to the two baseline datasets. The lighter solid line ( ) corresponds to the suspension ault dataset. Note that on the top plot or the right wheel in Figure 5(a), the suspension ault data exhibits two strong peaks at 7.5 and 5 Hz, respectively. The peak at 7.5 Hz is associated with one o the suspension modes probably at Hz in the other two datasets. The modal peak when the metal spacer is inserted is lower in requency because by splitting the coil spring o stiness k into two shorter coil springs o stiness k, the resultant eective stiness o the spring is lower, e.g., k/. The peak at 5 Hz is a second harmonic o 7.5 Hz due to the nonlinear response o the suspension as the spring coils compress on the metal spacer. This behavior was not modeled in the simpliied model o Figure 4; however, nonlinear behavior is expected in the suspension or this type o ault. In contrast, the data in Figure 5(b) or the let wheel does not exhibit signiicant dierences between the two
baseline datasets and the aulty dataset. To quantiy these dierences, the dierence between the second baseline dataset and the irst baseline dataset and the dierence between the aulty dataset and the irst baseline dataset were calculated as a unction o requency. Then the area underneath these two unctions were calculated and plotted as a unction o requency. Figure 6 shows this ault index. Note that the aulty dataset exhibits a larger dierence rom the irst baseline dataset than the second baseline dataset. An appropriate threshold would need to be chosen in order to detect the suspension ault using this result. To veriy that this approach is eective at isolating the ault, the data was also analyzed as the rear wheel traversed the cleat. Figure 7 shows the comparison o the spectra. Note that now there is no indication that the aulty dataset is signiicantly dierent rom the baseline datasets. This result veriies that the ault is indeed in the ront suspension as opposed to the rear suspension. Figure 6: Comparison o ault index or second baseline dataset ( ) and aulty dataset ( ) indicating larger dierences due to the ault than due to measurement variability..5 x -3 No ault evident 5 5 5 3 Freq [Hz].8 x -3 No ault evident.6.4. 5 5 5 3 Freq [Hz] x -3 Fault.5 5 5 5 3 Freq [Hz].8 x -3 No ault evident.6.4. 5 5 5 3 Freq [Hz] Figure 5: Front vertical acceleration responses on (a) right and (b) let sides o instrumented cleat with ( ) irst baseline, ( ) second baseline, and ( ) aulty datasets indicating ault near 7.5 and 5 Hz. Figure 7: Rear vertical acceleration responses on (a) right and (b) let sides o instrumented cleat with ( ) irst baseline, ( ) second baseline, and ( ) aulty datasets indicating no ault. CONCLUSION A simpliied our degree o reedom model o a HMMWV was developed to study changes in the orces in the tires as a unction o aults in the wheels and suspensions. Simulations showed that tire aults were more readily detected than suspension aults at lower requencies using measured orces in a roadway cleat. Longer cleats were shown to produce data that better separated healthy and aulty wheel end and suspension responses. Tests on a small truck showed that a simulated suspension ault could be detected and isolated to the ront right corner o the suspension using an instrumented rubberized cleat to measure tire orces. REFERENCES Fault Index 5 x -3 4.5 4 3.5 3.5 Fault.5.5 Second baseline 5 5 5 3 Freq [Hz]. Gorsich, D., Reliability Centered Maintenance, 7, Condition-Based Maintenance Workshop, Tank Automotive Research Development Engineering Center, Warren, MI.. Thomas, M., Major Ground Equipment System Accidents Caused by Materiel Failure, 985, TN 85-3. 3. Aardema, J., Failure Analysis o the Lower Rear Ball Joint on the HMMWV, 988, AD- 894. 4. Lasure, Maj. K., Pilot RCM I HMMWV Analysis, 4. 5. FORSCOM Saety Alert Message, TACOM SOUM 4-7, 4. 6. Bishop, C. M., Neural Networks or Pattern Recognition, 995, Clarendon Press, Oxord. DEFINITIONS, ACRONYMS, ABBREVIATIONS
CBM: Condition Based Maintenance HMMWV: High Mobility Multipurpose Wheeled Vehicle