Analysis of Nonlinear Vibration-Interaction Using Higher Order Spectra to Diagnose Aerospace System Faults Mohammed A. Hassan, David Coats, Kareem Gouda*, Yong-June Shin, and Abdel Bayoumi* Department of Electrical Engineering/ Department of Mechanical Engineering* University of South Carolina 301 Main Street Columbia, SC 803-454-9461 shinjune@cec.sc.edu, bayoumi@cec.sc.edu ponents [1], [2]. In contrast, traditional time-based maintenance (TBM) involves replacing existing parts after a certain time period or a certain number of operational hours. Implementation of the CBM helps to avoid failures in critical parts due to unexpected wear which may cause operational downtime and/or potential safety hazards [3], [4], [5]. We have sought to improve the accuracy of condition indicators (CIs) used in the Vibration Management Enhancement Program. VMEP implementation resulted in on-board Modern Signal Processing Unit (a vibration data acquisition and signal-processing equipment for the health monitoring of critical mechanical components) for AH-64 (Apache), UH60 (Blackhawk) and CH-47 (Chinook) fleets [6] and aims to provide rotorcraft maintainers with a collection of diagnostic and progressively prognostic vibration-based indicators summarized by CIs or Health Indicators (HIs), which collect several CI metrics. Pre-established and baseline measurements of these typically one-dimensional CI and HI values from existing historical data and testbed verifications under extreme conditions provide rankings for the status of individual aerospace and rotorcraft components with ratings such as “Good,” “Caution,” and “Exceeded,” which in turn provide maintainers of these fleets proactive time-independent condition based maintenance decision making [7]. Abstract— For efficient maintenance of a diverse fleet of aging air- and rotorcraft, effective condition based maintenance (CBM) must be established based on rotating components monitored vibration signals. Traditional linear spectral analysis techniques of the vibration signals, based on auto-power spectrum, are used as common tools of rotating components diagnoses. Unfortunately, linear spectral analysis techniques are of limited value when various spectral components interact with one another due to nonlinear or parametric process. In such a case, higher order spectral (HOS) techniques are recommended to accurately and completely characterize the vibration signals. Since the nonlinearities result in new spectral components being formed with coherency in phase, the detection of such phase coherence may be carried out with the aid of higher order spectra. In this paper, we use the bispectrum as a higher order spectral analysis tool to investigate nonlinear wave-wave interaction in vibration signals. Accelerometer data has been collected from baseline tests of accelerated conditioning in tail rotor drive-train components of an AH-64 helicopter drive-train research test bed simulating drive-train conditions. Through bispectrum analysis, we compare the harmonics interaction patterns contained in vibration signals from different physical setting of helicopter drive train and compare that with classical power spectral density plots. The analysis advances the development of higher order statistics and two dimensional frequency health indicators in order to qualify health conditions in mechanical systems. TABLE OF Our aim is to improve the effectiveness of the MSPU by developing new general methods for fault analysis that could be used in existing or new CIs. Unfortunately, some of the existing CIs based on conventional time or spectral analysis have limited diagnostic capabilities and are used to detect more than one fault associated with the same rotating component. For example, SP2 (Spectral Peak 2) is currently employed in the MSPU to detect unbalanced and/or misaligned shafts in a tail rotor drive-train of a rotorcraft [8]. However, this CI does not specify whether the fault is unbalance, misalignment or a combination of those faults. The maintainers are told to check for more than one source that might cause that CI to exceed its limit. We use the concept of higher order spectra in vibration analysis toward the development of more robust diagnostic CI metrics, as well as the furthered understanding of underlying physical and electromechanical interactions. C ONTENTS 1 I NTRODUCTION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 S TATISTICAL RELATIONSHIPS BETWEEN VI BRATION SIGNALS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 E XPERIMENT SETUP AND DATA DESCRIPTION . . 4 A PPLICATION OF BISPECTRUM TO ANALYZE VIBRATION INTERACTIONS . . . . . . . . . . . . . . . . . . . . . . 5 N UMBER OF N ONLINEAR H ARMONIC I NTER ACTION AS A CI . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6 CONCLUSION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ACKNOWLEDGMENTS . . . . . . . . . . . . . . . . . . . . . . . . . . . R EFERENCES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . B IOGRAPHY . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 2 3 4 6 7 7 7 8 In order to obtain better understanding of the fault sources, statistical relationship (or, dependence) between vibration signals can be investigated using various orders of the correlation function. The Fourier transform of the auto-correlation is the classical auto-power spectrum which is one of the most commonly used tools in spectrum analysis [9], [10]. Similarly, higher-order correlations, and their Fourier transforms describe higher order statistical relationships. Higher order spectral densities (HOS), such as bispectrum and trispectrum, 1. I NTRODUCTION Condition Based Maintenance (CBM) is a practice that recommends maintenance actions for machines, or systems, based on the continuous condition-monitoring of their comc 978-1-4577-0557-1/12/$26.00 ⃝2012 IEEE. 1 IEEEAC Paper #1345, Version 4, 04/01/2012. 1 Sxx (f ) is shortened to S(f ) and is referred to as the power spectrum. are the Fourier transforms of third order and the fourth order correlation functions respectively. The advantage of HOS over linear power spectral analysis is its ability to characterize nonlinearities in monitored systems [11], [12]. When various spectral components interact with one another due to nonlinear or parametric processes, new spectral components are formed which are phase coupled with the permanent interacted frequencies [13]. The bispectrum describes this correlation between the source and the result of interaction process in two-dimensional frequency space. Since faulted components are often related to nonlinear phenomena, bispectrum can provide more diagnostics information than the classical power spectrum which carry no phase information. Higher order correlation and higher order spectra Higher order spectral (HOS) analysis is a powerful signal processing technique in detecting nonlinearities. When various frequency components inside the vibration signal interact with one another due to nonlinear physical phenomena, new combinations of frequencies are generated at the sum and difference of the interacting frequencies. Those frequency components are phase coupled to the primary interacted frequencies. HOS uses this phase coupling signature between frequency components to detect nonlinearities [14]. The third order auto-correlation function is called auto-bicorrelation Rxxx (τ1 , τ2 ) , and its two-dimensional Fourier transform is called auto-bispectrum Sxxx (f1 , f2 ). In common practice, when bispectrum is mentioned, it is meant to be the autobispectrum. Rxxx (τ1 , τ2 ) and Sxxx (f1 , f2 ) for zero-mean strongly stationary random signal x(t) are defined in (4) and (5) respectively [11]. In this paper, we use the bispectrum as a signal processing tool to analyze vibration data to distinguish between different seeded faults in AH-64 drive-train namely (1) misaligned shafts, (2) unbalanced shafts and (3) combination of both misaligned and unbalanced as will be described in section 3. In the following section, we will start with reviewing general concepts of the linear correlation and linear spectra and then extend those concepts to higher order spectra. In section 4, analysis of vibration signals using bispectrum will be discussed. Based on this analysis, a CI is proposed in section 5 based on nonlinear harmonic interaction Rxxx (τ1 , τ2 ) = E{x∗ (t)x(t + τ1 )x(t + τ2 )} Sxxx (f1 , f2 ) = E{X(f1 )X(f2 )X ∗ (f3 = f1 + f2 )} 2. S TATISTICAL RELATIONSHIPS BETWEEN VIBRATION SIGNALS Rxx (τ ) = x(t) ⋆ x(t) = ∞ x∗ (t)x(t + τ )dt (1) −∞ If waves present at f1 , f2 , and f1 + f2 are spontaneously excited independent waves, each wave will be characterized by statistical independent random phase. Thus, the sum phase of the three spectral components will be randomly distributed over (−π, π). When a statistical averaging denoted by the expectation operator is carried out, the bispectrum will vanish due to the random phase mixing effect. On the other hand, if the three spectral components are nonlinearly coupled to each other, the total phase of three waves will not be random at all, although phases of each wave are randomly changing for each realization. Consequently, the statistical averaging will not lead to a zero value of the bispectrum. where the superscript asterisk (∗) denotes complex conjugate and the operation of correlation is indicated by a five-pointed star (⋆). Auto-correlation Rxx (τ ) is a measure of similarity (statistical dependence) between a signal x(t) and timeshifted version x(t + τ ). For vibration signals collected from physical systems, it is not possible (from experimental point of view) to have access to all possible realizations of x(t). Therefore, the autocorrelation in this case is statistically estimated based on a finite number of realizations as Rxx (τ ) = E{x∗ (t)x(t + τ )} The auto-bispectrum Sxxx (f1 , f2 ) is a true spectral density function indicates how the mean cube value of x(t) is distributed over a two-dimensional frequency plane. Using the symmetry property of Fourier transform, X(−f ) = X ∗ (f ), in (5), one can easily prove the four symmetry properties of the bispectrum [11] which make it enough to compute the bispectrum only for the octant labeled A in Figure 1. (2) where E{.} denotes the expected value operator. The auto-power spectrum Sxx (f ) is the Fourier transform of the auto-correlation Rxx (τ ) and is given by Sxx (f ) = E{X ∗ (f )X(f )} = E{|X(f )|2 } (5) The definition of the bispectrum in (5) shows how the bispectrum measures the statistical dependence between three waves. That is, Sxxx (f1 , f2 ) will be zero unless the following two conditions are met: 1. Waves must be present at the frequencies f1 , f2 , and f1 + f2 . That is, X(f1 ), X(f2 ), and X(f1 + f2 ) must be non-zero. 2. A phase coherence must be present between the three frequencies f1 , f2 , and f1 + f2 . Linear correlation and linear spectra In signal processing, the auto-correlation function Rxx (τ ) for a stationary signal x(t) is defined as ∫ (4) In fact, bispectrum will be evaluated digitally. Thus, the sampling theory implies that all f1 , f2 and f3 = f1 + f2 must be less than or equal to f2s , where fs is the sampling frequency. Thus, when auto-bispectrum is digitally computed, it will be plotted within the triangular region defined by the lines f2 = 0, f2 = f1 , and f2 = f2s − f1 . The auto-bispectrum can be estimated directly from the discrete Fourier transform of M realization of sampled version of x(t) as follows [11]; (3) where X(f ) is the Fourier transform of x(t). The auto-power spectrum, Sxx (f ), has the dimensions of mean square values/Hz and it indicates how the mean square value is distributed over frequency. In common practice, 2 Figure 2. Experimental helicopter tail rotor drive-train assembly with labeled shaft and supporting components. precedes the forward hanger bearing shown in Figure 2 while the #4 driveshaft connects the two vital bearings. Segment #5 precedes the intermediate gearbox (IGB) while the #6 driveshaft serves as link between the intermediate and tail rotor gearboxes (TRGB). The high speed shafts numbered 3-5 operate at a rotation speed of 4863 RPM (81.05Hz) while the gearboxes step the rotational speed down to 3660 and 1414 RPM respectively at increased torque, each corresponding to full-speed of shaft rotation on the fielded rotorcraft. Vibration and temperature data along with speed and torque measurements are all taken with the focus of this study centered on accelerometer vibration data at each hanger bearing and each of the two gear boxes (the intermediate and tail rotor gearboxes). Figure 1. Symmetry regions and region of computation for the auto-bispectrum Ŝxxx (l, k) = M 1 ∑ Xi (l)Xi (k)Xi∗ (l + k)} M i=1 (6) The magnitude of the bispectrum at coordinate point (l, k) measures the degree of phase coherence between the three frequency components l, k,and l + k. However, this magnitude is also dependent on the magnitude of the relevant Fourier coefficients. Therefore, a normalized version of the bispectrum estimator in (6) is adopted throughout this research as indicated in (7). Ŝxxx (l, k) S̄xxx (l, k) = √∑ ∑ 2 ∀i ∀j |Ŝxxx (i, j)| Signals denoted as FHB and AHB in Figure 2, measured at forward and aft hanger bearings vibrations respectively, are gathered at two minutes intervals at a sampling rate of 48 kHz over the course of thirty minute baseline test runs. Measurements are taken for each set of new bearings under test which include baseline shaft and bearing configuration, unbalance in different shafts configuration, and shaft misalignment, all common issues on AH-64 drivetrains. Misalignment between drive-shafts #3, #4 and #5 is introduced by moving shaft #4 distance 2.5 inches parallel to the center centerline to simulate the maximum severe condition in the actual aircraft. This causes different angles of misalignment between shafts #3 and #4, and shafts #4 and #5 where their lengths are 30, 130, and 130 inches for drive-shafts #3, #4, and #5 respectively as illustrated in Figure 3. (7) The magnitude of the normalized bispectrum S̄xxx (l, k) is bounded between “0” and “1” and it evaluates the share of that magnitude to the norm of the overall bispectrum matrix. 3. E XPERIMENT SETUP AND DATA DESCRIPTION The CBM center at The University of South Carolina has an AH-64 helicopter tail rotor driveshaft apparatus for on-site data collection and analysis from which data has previously been gathered in [15]. The apparatus is a dynamometric configuration which includes a pair of 400 horsepower motors for drive and regenerative tail end loading of a multi-shaft drive train. These shafts support tests of a pair of gearboxes connected to the absorption motor to simulate the torque loads that would be applied by the tail rotor blade as well as a pair of hanger bearings tested typically under faulted conditions. Figure 3. Misalignment conditions comparing (a) an aligned drive-train diagram focused on #4 shaft and associated bearings (b) configuration for misalignment of shafts #3-#5. Four drive train shaft segments (actual helicopter components designed to military specifications) are of interest in this study and are numbered as segments #3-#6. Segment #3 3 third harmonic of the shafts (243 Hz) dominates all other harmonics due to the loading effect in both the AHB and the FHB spectra. Moreover, other different spectral components are affected by either attenuation or amplification. Unbalance of the shafts was also applied in the test conditions as a comparison point for baseline operation. The unbalance in the shafts of the drivetrain is introduced by adding thin, weighted foil lining segments to the interior wall of the driveshafts #4 and #5. Different combination misalignment and unbalance are tested with Table 1 summarizing these test conditions and their designations. Table 1. Tail Rotor Driveshaft Experimental Settings Shaft Status Aligned Misaligned Balanced 00321 20321 Unbalanced 10321 30321 4. A PPLICATION OF BISPECTRUM TO ANALYZE VIBRATION INTERACTIONS As described in the previous section, for each test setting described in Table 1, 17 acquisitions (realizations) have been gathered at sampling rate 48 kHz with 65536 points each. However, the bigger the number of realizations we use, the more accurate bispectrum estimation we get. In order to increase the number of realizations, we split each acquisition file into two realizations to have 34 total realizations for each test (32768 sample points in each realization). Then, the stationarity assumption of the vibration realizations is validated using the nonparametric runs test with 0.05 level of significance [16]. Figure 5. Cascaded plot of FHB power spectrum showing change of load conditions for unbalanced case (10321) The average power spectra of AHB during loaded condition of all tested faults are summarized in Figures 6 and 7 with the 3rd harmonic dominating all other harmonics in all tested cases. It is not an easy task to distinguish between different cases using only the traditional linear power spectrum. For example, the power spectra of the baseline (00321) and the misaligned (20321) cases in Figure 6 have the same dominating spectral peaks with very slight changes in the minor peaks. A similar situation occurs when we compare the unbalanced (10321) and the misaligned-unbalanced (20321) cases in Figure 7. Figure 4. Cascaded plot of AHB power spectrum showing change of load conditions for unbalanced case (10321) The vibration signals from the forward (FHB) and aft (AHB) hanger bearings have been investigated under different loading conditions by controlling the torque applied by the absorbtion motor. It was found that the power spectral content of the vibration signals vary dramatically considering the normal loading and the free loading operations. Figures 4 and 5 show this effect during the test of unbalanced shafts for the AHB and the FHB signals respectively. During this test, the speed of shafts #3-#5 is maintained constant at 4863 RPM (81.05Hz). A total 18 runs of the power density spectra (PDS) are plotted during the no load setting, then 16 runs are plotted with output torque loading. It is clear that the Figure 6. Average power spectrum of the AHB in loading conditions: baseline (00321) in (a), and misaligned (20321) in (b) 4 Figure 9 shows the bispectrum of the misaligned case. More interactions among harmonics appear along 1R frequency axis. Along this axis, interaction with the 3R and 2R are the most obvious. Other interactions exist with very low magnitudes. The increased magnitude of frequency interactions along the 1R axis can be used as an indicator of the misalignment between shafts. Figure 7. Average power spectrum of the AHB in loading conditions: unbalanced (10321) in (a), and misalignedunbalanced (30321) in (b) The vibration signals under loading conditions are then analyzed using the bispectrum as shown in Figures 8-11 for the AHB. The vibration interaction in the baseline case shown in Figure 8 shows the least nonlinearity with only the 3rd harmonic interacting almost exclusively with itself. In fact, this coordinate point representing this 3rd harmonic interaction shows up in all the other faulted cases. However, unlike the case of linear power spectrum, the bispectrum shows different harmonic interaction patterns between the different cases which might be used to design a more accurate diagnoses model and condition metrics. Since our discussion in the following part will be based on different harmonic interactions, we will use “1R, 2R, 3R, ...” for easier notation of “first, second, third, ...” harmonics of the rotating shafts frequency. Figure 9. Bispectrum for the misaligned case (20321) Pure unbalanced shafts act on the 3R axis of interaction as shown in Figure 10. Along this axis, 3R interacts with 6R and 9R (notice that 6R and 9R are the second and third harmonics of 3R itself). Although high magnitudes exist along the 1R axis, we can easily distinguish between the misaligned (20321) and the unbalanced (10321) cases by those 6R and 9R interaction with 3R. Figure 8. Bispectrum for the baseline case (00321) Figure 10. Bispectrum for the unbalanced case (10321) 5 Table 2. Comparison with baseline case in terms of SP1, SP2, and SP3 (dB) Combining the misalignment and unbalance faults, one may expect a combined signature of interactions discussed in each separate case. This conjecture can be justified by the bispectrum shown in Figure 11. In this case (misalignedunbalanced 30321), increased harmonic interactions along both 1R and 3R are obvious. Moreover, minor magnitudes along 2R and 4R axes exist in this case. Although careful inspection of the two dimensional bispectra in Figures 8-11 gives better insight to all wave-wave interaction inside the signal spectrum, it would be useful to follow an automated procedure similar to the spectral peak comparison discussed above to diagnose and isolate a certain fault signature in the frequency domain. As discussed in the previous section, increased number of harmonic interactions along 1R axis alone indicates pure misalignment. Unbalance adds more interaction along 3R axis, and a combination of the two faults results in increased number of harmonic interactions in both 1R and 3R axes. Therefore, Number of Nonlinear Harmonic Interaction (NNLHI) in the bispcetrum of the vibration signal can be used as a condition indicator to diagnose shaft faults. NNLHI(1R)/NNLHI(3R) is a number that indicates how many harmonic interactions exceed the threshold limit along 1R/3R axis. Table 3 shows how NNLHI(1R) and NNLHI(3R) indices are calculated for each faulted case. Values of the bispectral peaks at shaft harmonic interaction points (with 1R or 3R) are compared to their counterparts from the baseline case in logarithmic scale. When a value exceeds the threshold limit (we used 6 dB in our case), it counts toward the total NNLHI at the corresponding axis (1R or 3R) as summarized in the last row of Table 3. In both the unbalanced (10321) and the misaligned (20321) cases, 1R interacts with 1R, 2R, 3R, 4R, and 9R and the value of each bispectral peak at those points exceed 6 dB compared to the baseline case, NNLHI(1R)=5. However, no interaction with 3R exceeds the 6 dB threshold, NNLHI(3R)=0, in the misaligned (20321) case compared to NNLHI(3R)=2 in the unbalanced (10321) case. The misaligned-unbalanced case (30321) shows increased number of both 1R and 3R interactions, NNLHI(1R)=8 and NNLHI(3R)=3. Figure 11. Bispectrum for the misaligned-unbalanced case (30321) 5. N UMBER OF N ONLINEAR H ARMONIC I NTERACTION AS A CI In order to develop a CI based on the bispectrum analysis discussed in the previous section, we will start with reviewing the current CI being used to detect the fault cases under study. As mentioned in section 1, SP2 is currently used in the MSPU to indicate shafts misalignment and/or unbalance. This CI calculates the magnitude of the spectral peak at the second harmonic of the shaft rotating frequency and compares it with reference baseline case (taken under well defined normal operating conditions with the rotorcraft in known good condition). When the SP2 exceeds certain predefined limits compared to the baseline, the “Caution” or “Exceeded” status of this CI are designated. However, this CI can not specify whether the fault is misalignment or unbalance. Table 3. Comparison with baseline case in terms of NNLHI (dB) For the sake of discussion, we recall that the comparison with the baseline is usually done on a logarithmic amplitude scale with increases of 6-8 dB considered to be significant and changes greater than 20 dB from the baseline considered serious [17]. Table 2 summarizes the results of the spectral peak comparison of the three faulted cases (10321, 20321, and 30321) with the baseline case (00321) in terms of SP1, SP2, and SP3. Values of spectral peaks at the first three harmonics of the shaft rotating speed (1R, 2R and 3R) are extracted from the averaged PSD plots in Figures 6 and 7, and compared with the baseline case in logarithmic scale. It is clear that values of the SP2 for all faulted cases exceed the 6 dB threshold compared to the baseline and therefore it provides a good Indicator for all of the three faulted cases. However, SP2 has limited diagnostic capability to distinguish between those health conditions of the shafts. It is worthwhile to mention here that although bispectral analysis provides a more accurate tool to diagnose different faults compared to the conventional power spectral analysis, this comes on the cost of computational resources and time. 6 6. nance (CBM) Component Inspection and Maintenance Manual Using the Modernized Signal Processor Unit (MSPU) or VMU (Vibration Management Unit),” Aviation Engineering Directorate Apache Systems, Alabama, Tech. Rep., Oct. 2010. CONCLUSION Bispectral analysis has been used to study vibration signals from an AH-64 helicopter tail rotor drive-train. A normalized version of the bispectrum has been proposed and then used to compare between different seeded fault cases in the shafts of the drive-train using hanger bearing vibrations. It has been shown that using the bispectral analysis provides more details about the spectral content of the vibration signal and how different harmonics nonlinearly interact with one another. Different faults have shown different harmonic interaction patterns which have been used to design a diagnostic algorithm based on the number of nonlinear harmonic interaction (NNLHI) in the vibration signal. The proposed indicator has shown more diagnostic capability to differentiate between shafts misalignment/unbalace than the currently used CI based on conventional spectral peak comparison. The tradeoff of the proposed technique is its higher complexity and computational cost. Current work is focusing on the application of bispectral analysis to study vibrations from the intermediate and tail rotor gearboxes in the AH-64 helicopter tail rotor drive-train. [7] N. Goodman, and A. Bayoumi, “Fault Class Identification Through Applied Data Mining Of Ah-64 Condition Indicators,” Proceedings of AHS 66th Annual Forum and Technology Display, May 2010. [8] P. Grabill, J. Seale, D. Wroblewski, and T. Brotherton, “iTEDS: the intelligent Turbine Engine Diagnostic System,” Proceedings of 48 International Instrumentaion Symposium, May 2002. [9] A. S. Sait, and Y. I. Sharaf-Eldeen, “A Review of Gearbox Condition Monitoring Based on vibration Analysis Techniques Diagnostics and Prognostics,” in Rotating Machinery, Structural Health Monitoring, Shock and Vibration, Vol. 8, T. Proulx, Ed. New York: Springer, 2011, pp. 307-324. ACKNOWLEDGMENTS [10] P. D. Samuel, and D. J. Pines, “A review of vibrationbased techniques for helicopter transmission diagnostics,” Journal of Sound and Vibration, vol. 282, no. 1-2, pp. 475-508, Apr. 2005. This research is funded by the South Carolina Army National Guard and United States Army Aviation and Missile Command via the Conditioned-Based Maintenance (CBM) Research Center at the University of South Carolina-Columbia. Also, this research is partially supported by the Egyptian government via the Government Mission Program for Mohammed Hassan, and National Science Foundation Faculty Early Career Development (CAREER) Program as well as a National Science Foundation Graduate Research Fellowship for David Coats. [11] Y. C. Kim, and E. J. Powers, “Digital bispectral analysis and its application to nonlinear wave interactions,” IEEE Transactions on Plasma Science, vol. 7, no. 2, pp. 120131, July 1979. [12] S. Elgar, and R.T. Guza, “Statistics of bicoherence,” IEEE transaction on Acoustics Speech and Signal Processing, vol. 36, no. 10, pp. 1667-1668, Oct. 1988. R EFERENCES [1] A. Bayoumi, N. Goodman, R. Shah, L. Eisner, L. Grant, and J. Keller, “Conditioned-Based Maintenance at USC - Part I: Integration of Maintenance Management Systems and Health Monitoring Systems through Historical Data Investigation,” Proceedings of AHS International Specialists’ Meeting on Condition Based Maintenance, Huntsville, AL, Feb. 2008. [2] A. K.S. Jardine, D. Lin, and D. Banjevic, “A review on machinery diagnostics and prognostics implementing condition-based maintenance,” Mechanical Systems and Signal Processing, vol. 20, no. 7, pp. 1483-1510, Oct. 2006. [3] A. Bayoumi, W. Ranson, L. Eisner, and L.E. Grant, “Cost and effectiveness analysis of the AH-64 and UH-60 onboard vibrations monitoring system,” IEEE Aerospace Conference, pp. 3921-3940, Mar. 2005. [4] A. Bayoumi, and L. Eisner, “Transforming the US Army through the Implementation of Condition-Based Maintenance,” Journal of Army Aviation, May 2007. [5] V. Blechertas, A. Bayoumi, N. Goodman, R. Shah, and Yong-June Shin, “CBM Fundamental Research at the University of South Carolina: A Systematic Approach to U.S. Army Rotorcraft CBM and the Resulting Tangible Benefits,” Proceedings of AHS International Specialists’ Meeting on Condition Based Maintenance, Huntsville, AL, Feb. 2009. [6] Damian Carr, “AH-64A/D Conditioned Based Mainte- [13] T. Kim, W. Cho, E. J. Powers, W. M. Grady, and A. Arapostathis, “ASD system condition monitoring using cross bicoherence,” 2007 IEEE electricship technologies symposium, pp. 378-383, May 2007. [14] B. Jang, C. Shin, E. J. Powers, and W. M. Gardy, “Machine fault detection using bicoherence spectra,” Proceeding of the 21st IEEE Instrumentation and Measurement Technology Conference, vol. 3, no. 1, pp. 16611666, May 2004. [15] D. Coats, Cho Kwangik, Yong-June Shin, N. Goodman, V. Blechertas, and A. Bayoumi, “Advanced TimeFrequency Mutual Information Measures for ConditionBased Maintenance of Helicopter Drivetrains,” IEEE Transactions on Instrumentation and Measurement, vol. 60, no. 8, pp. 2984-2994, Aug. 2011. [16] J. D. Gibbons, “Nonparametric Methods for Quantitative Analysis,” Columbus, Ohio: American Sciences Press, 1985. [17] R. B. Randall, “Fault Detection,” in Vibration-based Condition Monitoring: Industrial, Aerospace and Automotive Applications, Wiley, 2011. 7 B IOGRAPHY [ Abdel E. Bayoumi is Director of the USC Biomedical Engineering Program and Professor of Mechanical Engineering at the University of South CarolinaColumbia. He chaired the Department of Mechanical Engineering from 1998 through 2006. Before joining USC, he was a Professor and Director of the Manufacturing Program at North Carolina State University-Raleigh, North Carolina (1996-1998); Assistant, Associate, Professor, and Distinguished Boeing Manufacturing Professor at Washington State University (1983-1996); a Project Manager at Hewlett-Packard Company-Corvallis, Oregon (1993-1995 on professional leave from WSU); and a visiting scholar for a year at the American University in Cairo (AUC) - Egypt (1991-1992 - a sabbatical leave from WSU). During his tenure at the University of South Carolina, North Carolina State University, Washington State University, the American University in Cairo, and Hewlett-Packard Company, Dr. Bayoumi has been actively involved in developing strong research and educational programs. His current areas of interest can be grouped into three categories, (1) Study of Condition-Based Maintenance (CBM) of military aircraft in which diagnosis, prognosis and health monitoring systems are effectively utilized using informatics and sensing technologies, (2) MicroElectro Mechanical Systems (MEMS) and Mechatronics in which a MEMS device is designed, fabricated and used to sense and control mechanical or biological systems, and (3) Design and applications of efficient energy resources and system. Mohammed A. Hassan received the B.S. degree with honors in Electrical Engineering from Cairo University Fayoum Campus, Fayoum, Egypt, and the M.S. degree in Electronics and Communication Engineering from Cairo University, Cairo, Egypt, in 2001 and 2004, respectively. He is currently pursuing his PhD degree in the Electrical Engineering department at the University of South Carolina. His current research interests include advanced digital signal processing techniques: time-frequency analysis and higher order statistical signal processing, and its application in condition based maintenance. David Coats is pursuing his doctoral degree in Electrical Engineering at the University of South Carolina-Columbia. He received his B.S. degree with summa cum laude honors from the University of South Carolina, and he was the recipient of the NASA EPSCoR Undergraduate Scholarship, Magellan Scholar, and Valedictorian Scholars awards. His undergraduate research efforts culminated in receipt of the National Science Foundation Graduate Research Fellowship in 2010. His research interests include power electronics, condition based maintenance in aging aircraft and aerospace components, and digital signal processing in cable diagnostics. Kareem Gouda is a graduate research assistant at the Condition Based Maintenance (CBM) research Centre, University of South Carolina (USC). He is currently pursuing a doctoral degree in Mechanical Engineering. Kareem had his BS in Chemical Engineering. As an disciplinarian, Kareem’s research focus is on lubrication analysis and gearbox health monitoring for the improvement of the Condition Indicators (CIs) of the AH-64 drive train gearboxes. Yong-June Shin received his B.S. degree from Yonsei University, Seoul, Korea, in 1996 with early completion honors and the M.S. degree from The University of Michigan, Ann Arbor, in 1997. He received the Ph.D. degree from the Department of Electrical and Computer Engineering, The University of Texas at Austin, in 2004. Upon his graduation, he joined the Department of Electrical Engineering, The University of South Carolina in which he is now an Associate Professor. His area of research is power engineering/power electronics, with emphasis on power quality and harmonics. His research interests include advanced signal processing theory: time-frequency analysis, wavelets, and higher order statistical signal processing. He is a recipient of National Science Foundation CAREER award in year 2008 and GE Korean- American Education Commission scholarship. 8