AIRCRAFT DAMAGE DETECTION FROM ACOUSTIC AND …



© 1997 Institute of Noise Control Engineering of the USA, Inc.

AIRCRAFT DAMAGE DETECTION FROM ACOUSTIC AND NOISE

IMPRESSED SIGNALS FOUND BY A COCKPIT VOICE RECORDER*

Authors: Ronald O. Stearman Stuart M. Rohre

Aerospace Eng. and Eng. Mechanics Dept. Applied Research Labs

The University of Texas at Austin The University of Texas at Austin

Austin, Texas 78712 P.O. Box 8029

Austin, Texas 78713

Glen H. Schulze

Data Acquisition Systems ron@aeroel.ae.utexas.edu

5965 W. Morraine Ave. schulze@arlut.utexas.edu

Littleton, CO 80123 rohre@arlut.utexas.edu

INTRODUCTION

Currently, research is being conducted to detect damage through structural acoustics, signal processing, and transducer designs. The present study illustrates that damage detection may be carried out with an existing system acting as a latent signal transducer. One example involved a reliability problem in a commuter aircraft engine mount design where undetected crack growth created a critical whirl flutter condition destroying the aircraft. This reliability problem prompted the need for an in-place damage detection system to identify critical engine mount conditions. Signal analysis of data acquired by a cockpit voice recorder prior to and during the catastrophic aircraft whirl flutter event provided insight into critical signals that indicated the failure onset. Although regularly scheduled inspections failed to detect the problem, cockpit voice recorder signals contained a dynamic signature of this damage feature intermittently throughout the duration of the tape. It is highly probable that this damage signature existed for a much longer period of time, but due to the endless loop configuration of the cockpit voice recorder the earlier data was erased. This study indicated that even in the case of an unused cockpit voice recorder track, careful signal processing can extract surprising details about detecting potential damage along with extracting aircraft breakup signatures.

BACKGROUND

The motivation for the present study was a 1992 Airline Pilots Association (ALPA) accident report concerning a popular nineteen passenger commuter aircraft [1]. The aircraft had been temporarily removed from its fleet operations for an evening training mission. The ALPA accident report concluded that an inflight right engine separation had occurred during this mission. The free right engine then traveled back striking the tail of the aircraft destroying most of the horizontal surfaces. This caused a complete loss of control of the aircraft resulting in an inflight breakup which was fatal to all occupants. Since this occurrence was prior to recent FAA legislation requiring flight data recorders on all commercial airliners, the cockpit voice recorder was the only on board flight record available to supply clues as to the cause of the accident [2]. This included not only the voice communication at the critical event but structural acoustic as well as other acoustical sounds and noise sources during the breakup phase. In essence, the lack of a flight data recorder prompted an extended study of the CVR tape to determine whether the CVR recorder acted as a latent signal transducer. That is, did the CVR record events other than voice signatures that would be critical to determining if an in flight breakup of the aircraft occurred? The voice signatures on the tape indicated that a final catastrophic event occurred without warning to the pilots. No significant voice stress characteristics were identified up through and including the final event.

RELIABILITY STUDY

Past history has taught us that propeller whirl flutter is a catastrophic event that will remove engines and destroy aircraft, but which does not generally occur unless the supporting engine and propeller structure has been significantly damaged. This consideration first prompted an engine mount reliability study to estimate the characteristic service life of the engine truss and to determine if a history of engine truss service difficulties did in fact exist [3].

An initial search through the Federal Aviation Agency’s (FAA’s) Service Difficulty Reports (SDR’s) indicated that at least six engine truss redesign cycles have occurred over the ten year life history of the aircraft. Such a large number of engine mount design changes suggest a significant engine mount service life problem. In view of this, the question was then pursued as to whether or not there was a reliability improvement as the redesign cycles progressed through the ten year period.

Once the service difficulty reports (SDR’s) had been reviewed, it was evident that the mount had been experiencing frequent cracking, particularly in the vicinity of welded joints near the engine attachment points. The actual truss life was noted to be far below the intended and predicted design life of 20,000 hours. Consistent with extreme value studies, only first-time cracking failures were considered in the reliability study. That is, a repaired truss reentered into service was not considered. Six truss types were identified in the preliminary investigation. Only four of the six were found to have statistical relevance to the study, as the remaining two truss types were only in limited use and the sample data was insufficient. The various truss types were indicative of redesigns that evolved over a ten year period and were developed with the goal of extending the characteristic life of the truss. Modifications included increases in truss tube diameters and tube wall thickness along with the addition of gusset plates to the elements and joints noted to crack most often.

In the parameter identification to characterize the behavior of the failing trusses, the data was analyzed with the versatile Weibull model. The resulting four analyses, one for each of the four trusses included in the study, indicated an alarming trend. Rather than increasing the characteristic life of the truss, the augmented strength in each redesign of the truss resulted in an overall decreasing trend in the characteristic life. A summary of those results are depicted in Figure 1. (Truss types are indicated in chronological order of development from left to right).

In addition, analysis of the shape function, a characterization of the type of failure rate, indicated that this important value was also decreasing and, as of the last design, indicated the possibly of failures in the infant mortality region of the "hazard rate curve”. Figure 2 illustrates the critical transition of the latest truss design toward an infant mortality mode of failure (shape parameter < 1).

[pic][pic]

Figure 1: Characteristic life as a function of Figure 2: Shape parameter as a function of

chronological truss development chronological truss development

WHIRL FLUTTER STUDY

Due to the sudden nature of the aircraft breakup and the growing unreliability of the aircraft engine mount designs, a catastrophic whirl flutter event seemed quite plausible. To further check on this possibility, two completely independent flutter analyses were carried out employing two different whirl flutter software codes. One study employed a recently developed MSC/NASTRAN 68 Whirl Flutter program resident in the Aeroelasticity option of this large commercially available finite element modeling code [4]. The second study was conducted by an outside flutter consultant employing his refinement of a whirl flutter software code developed by Vought Aeronautics [5].

The whirl-flutter results, presented in Figure 3, were found to be within reasonable agreement when the basic assumptions in propeller aerodynamics are considered for each code. That is, the NASTRAN code employs the simplifying assumption of a windmilling propeller aerodynamic model, which results in a conservative (lower speed) estimate of the whirl flutter speed. In contrast, the Vought code employs an actual thrusting propeller aerodynamic modeling and is thought to provide a closer estimate to the actual flutter speed. Whirl flutter frequency estimates were also included in Figure 3 as they were important in the CVR analysis. Since the maximum operating speed of the aircraft is 250 KTS, the whirl flutter speeds are seen to be critical to the aircraft only for the damaged truss cases. That is, when the truss cracking occurs in two or more tubes, the whirl-flutter speed lies well within the aircraft flight envelope. It is also interesting to note that, based upon the data from the last radar contacts, the estimated aircraft speed at the time of our projected breakup was approximately 185 to 195 KTS.

|Type |NASTRAN | |Brock | |

|Analysis |Velocity |Frequency |Velocity |Frequency |

|Undamaged Truss |621 KTS |0.034 Hz |710 KTS |3.96 Hz |

|1550 RPM | | | | |

| |1049 FPS | |1200 FPS | |

|Two Tubes Failed |167 KTS |0.710 Hz to |193 KTS |1.10 Hz |

|# 105 & # 107 | |0.880 Hz | | |

|1550 RPM |283 FPS | |326 FPS | |

|Three Tubes Failed |166 KTS |0.770 Hz |185 KTS |0.89 Hz |

|# 101, # 105, | | | | |

|& # 107 |280 FPS | |312 FPS | |

|1550 RPM | | | | |

Figure 3: Whirl flutter speeds and frequencies

The results of the reliability and whirl flutter analyses suggested that the CVR tape should first be checked for evidence of a catastrophic or explosive type acoustical signature which would be indicative of a whirl flutter breakup. The study would then continue to look for frequency signatures characteristic of the range of flutter frequencies shown in Figure 3, as well as the first blade and fourth blade passage frequencies of the propeller, which occur at approximately 26 Hz and 104 Hz, respectively, for the right propeller at the flight idle condition of 1550 RPM at the time of breakup.

CVR TAPE MEASUREMENT AND ANALYSIS

CVR as a Latent Transducer. Four track 30 minute tape loop Cockpit Voice Recorders (CVRs) are now mandatory on civilian airliners worldwide. Their purpose is to provide a survivable analog tape recording of the voice communications among the aircraft flight crew and radio communications between the aircraft and the ground Air Traffic Control (ATC). One of the four tracks is dedicated to the secondary but highly important task of recording the flight deck cabin area sounds via a centrally located cabin area microphone (CAM). It is not unusual in commuter aircraft to find one of the four tape tracks unused or “silent” (underdriven) when four microphones are not used. These silent tracks have been found, to be rich in non-speech sounds, which were fortuitously recorded onto the CVR tape.

The measurements and analyses of non-speech signals from all four tracks of the CVR are usually complicated by the wide range of signal amplitudes encountered and the sometimes short time periods of these signals. For this accident investigation case a Hewlett-Packard 3566A/67A Signal Analyzer was capable of extracting meaningful time series and FFT spectral content from recorded events as short as 330 milliseconds. The 80 dB dynamic range of the 3566A/67A Analog-to-Digital converter was fully used in this investigation in order to produce digital data files with high SNR content.

Non-speech signals from the subject accident CVR pilot’s voice tracks and the silent track were highly informative in providing key evidence in establishing the accident cause. The auditory listening properties of the human ear became highly instrumental in discovering that an operating CVR in a disintegrating aircraft was functioning as an accelerometer with memory, i.e., a “latent transducer”. The otherwise constant 500 Hz aircraft landing gear warning horn recorded on the pilot’s audio track was heard to be warbling or unsteady in it’s frequency during the last seconds of this flight. A violent vibration associated with the first propeller blade passage frequency would be expected as the engine and rotating propeller system tore loose from the aircraft resulting in a significant mass unbalance condition for the rotating propeller. It appears that the CVR capstan drive mechanism, designed to transport the magnetic tape over the record head at a constant speed, was found to be “sympathetic” with the vibrations it was receiving from the disintegrating airframe. Specifically, airframe vibrations modified the normally constant tape speed by superimposing these vibration frequencies onto the 1d ips tape speed. A second contributor to the warbling could also be the modulation of the landing gear warning horn vibrating diaphragm as it was violently shaken by the disintegrating airframe at the propeller blade passage frequency.

When this CVR tape was replayed at constant speed, the recorded vibration components were transformed into frequency domain modulations of any and all carrier tones, which were being delivered to the tape heads. There were at least four such carrier tones present on different tracks of this aircraft’s CVR including a landing gear warning horn. This horn signal consisted of 500 Hz tone bursts 330 ms in duration and spaced 700 ms apart. It was this highly audible horn signal which was heard to be varying in pitch (frequency) as the airframe was disintegrating. This discovery was surprising in that the tone bursts were only 330 milliseconds long and contained only 8 cycles of low percentage deviation frequency modulation (FM) at a 24 Hz rate. Nevertheless, this modulation was first detected by the human ear and later confirmed by electronic instrumentation.

Special copy tapes were made of these EOT (end-of-tape) FM modulated warning horn tones, where the airframe was disintegrating, and presented to listening panels. Not all listeners could clearly testify they had heard the FM warble at 24 Hz. It was therefore necessary to use FFT spectral analysis data plots in order to clearly show that 24 Hz FM modulation did, in fact, develop at the end of the CVR tape. Figures 4A, 4B, 5A, and 5B, depicting horn harmonics with and without FM carrier erosion and with and without FM sidebands, show the described FM modulation. These figures also illustrate spectral broadening of the higher frequency harmonics of the 500 Hz horn tone burst. The 24 Hz sidebands are also present on the third and fourth harmonics of this tone. FM modulation indices and deviation ratios of the higher harmonics of the landing gear warning horn signal resulted in clear erosion of the higher frequency carrier signals, as well as the generation of the 24 Hz FM sidebands.

Electroacoustics and CVR Recordings. The characteristic audible whine of this jet engine turbine is usually found in a frequency band occupying a range between 3000 Hz ( idle power) and 6000 Hz ( full power). These frequencies are well within the recording bandwidth of today’s CVR systems even though the specified nominal bandwidth range is 150 to 5000 Hz. The ability to recover these frequencies from a CVR accident tape, in conjunction with the pilot’s speech, provides a powerful insight into the pilot’s actions coupled with engine performance and applied throttle settings.

[pic][pic]

Figure 4A: Horn harmonics w/o FM carrier erosion Figure 4B: Horn harmonics w/ FM carrier erosion

[pic][pic]

Figure 5A: Horn harmonics w/o FM sidebands Figure 5B: Horn harmonics w/ FM sidebands

The most direct means of coupling of the engine sounds onto the CVR tape is via the CAM or the pilot’s mikes. These signal paths are usually dominated by air flow noise past the cabin area or pilot’s speech activity. As such it is normally possible to capture only small segments of the uncontaminated engine acoustics, particularly in high pilot-voice stress conditions. Experience has shown that CVR tape tracks, not driven by a mike, can also capture these engine sounds, possibly better than a track fitted with a microphone. By a transduction process best described as “electroacoustics," the characteristic jet engine turbine whine sounds can be recorded onto CVR silent tracks as shown in Figure 6, Jet turbine signals during landing/taxiing [6]. This figure, derived from the HP 3566A/67A Signal Analyzer System in the Spectrograph Color Mode, clearly shows the landing, two engine braking and single engine 180 degree runway turn-around. Turbine generated frequencies during these engine conditions ranging from idle (3000 Hz) to near full power (6000 Hz) were all found on the silent track of the CVR. The assertion of a “sneak path” mechanism whereby the jet engine sounds were recorded onto a silent track can be met with significant skepticism. A more common example of sneak path signal dynamics can be demonstrated with a portable radio placed toward the inside rear of a large station wagon with the turn signal indicator activated. The metallic sound of the indicator timing switch located under the dashboard can be electronically heard over the speaker of the portable radio set to almost any carrier frequency in the AM band. Other examples of electroacoustic transduction methods are triboelectric, magneto-electric and piezoelectric effects [7]. Triboelectric currents are generated by charges created between a vibrating conductor and its insulator due to friction. Free electrons rub off the conductor and create a charge imbalance that causes the current flow. The most likely explanation of finding signals recorded on the silent CVR tape track is the triboelectric effect acting on the unterminated signal lines leading to the CVR input amplifiers. When these jet turbine sounds are recovered from the silent track, simultaneously with the pilot’s voice corroborating engine throttle activity, little doubt remains as to signal source authenticity.

[pic]

Figure 6: Jet turbine signals during landing/taxiing

Close inspection of the amplified (30 dB) time series from the CVR silent track also revealed a very periodic set of transient components occurring at a frequency of 0.86 Hz. Furthermore, this frequency correlated with an independent structural dynamic and flutter analysis of the engine mount damage, which is evident from Figure 3. This transient frequency was found throughout the 32 minutes of this CVR tape indicating the condition that generated this phenomenon was an ongoing long term condition rather than a rapid onset event. These transient components are thought to be typical of impact signatures that would occur from a broken tube end impacting on the tube joint where the fracture occurred. These transients became more frequent about 15 seconds before the airframe disintegrated and then diminished to nearly zero at the end of the flight where the CVR suddenly stopped. Figure 7, End-of-tape time series shows the presence of these 0.86 Hz transients which were demonstrated by independent structural and flutter analyses to be quite close to the frequency experienced by a damaged engine mount. (See Figure 3. and appendix A)

Non-Speech Sounds at the End of CVR Tape. It is highly desirable to analyze and explain all of the non-speech sounds at the end of a CVR accident tape. Although this is difficult to accomplish it should not be deemed impossible. With today’s DSP (Digital Signal Processing) including new signal classification techniques such as “wavelets” and “voice recognition” this task becomes much easier [8]. The seemingly simple categorization of the non-speech signals by amplitude level discrimination is a powerful first step. The CAM (Cabin Area Mike) CVR track is usually the track richest in captured non-speech sounds. It is important that the signal levels from the CAM track be recovered using a calibrated magnetic test tape as a means to precisely determine their recorded magnetic flux level. Also, when the non-speech sounds on the CAM track are compared to the level of the outside airflow noise, usually the dominant component of this CVR track, a further determination can be made. The newest technique of voice recognition via DSP appears to offer a significant advantage in the analysis and classification of non-speech CVR signals, an application for which this methodology was not envisioned. These newer DSP voice recognition programs have progressed to a stage where they should be applied to CVR non-speech analysis if for no other reason than to discover their potential.

[pic]

Figure 7: End-of-tape time series

Figure 8 depicts a sudden loud sound at the end of the subject CVR accident tape. This 25 millisecond long event can be seen to be much louder, about 15 dB, than the cabin outside air flow noise. Also, when compared to the amplitude of the erase head power shut-off transient pulse we find this event to be the largest amplitude signal recorded on the entire 32 minute long CVR tape. Although the short 25 ms length does not provide adequate audio listening time there is certainly enough signal time and amplitude to perform wavelet and voice recognition analysis. Particularly revealing would be the analytical ability to say that non-speech sounds at the end of a CVR tape have the same mathematical transforms found on an earlier CVR accident tape of the same type of aircraft involved in a similar accident.

[pic]

Figure 8: End-of-tape erasure gap w/ loud event

Figure 8: End-of-tape erasure gap w/ loud event

Explosion Acoustics and the CVR. One of the most recent acoustical concerns dealing with CVR tape studies relates to the question of “explosion acoustics”; the study of the acoustical pressure waves resulting from an explosion aboard an aircraft and recorded as an acoustical event on the CVR. This question was first voiced by the British AAIB (Air Accidents Investigation Branch) after the disintegration of Air India Flight 182 [9] over the North Atlantic in 1985 and again with the disintegration of Pan Am 103 [10] over Lockerbie in 1989. The British CVR experts, after announcing their analysis could not prove that evidence of an explosive device was found on the Pan Am 103 CVR tape, called for a CVR/FDR (Flight Data Recorder) study of explosive acoustics. The primary goal of this study was to determine if a bomb explosion could be differentiated from a cabin decompression event. The British postulated that a differentiation might be proven by use of the initial pressure wavefront directionality and gradient. Possibly a bomb would produce a sudden positive increase in air pressure whereas a cabin decompression would produce a sudden decrease in air pressure. Would the CVR microphones, record electronics, and magnetic heads along with the CVR playback equipment sufficiently preserve the pressure wavefront polarities of these two different events and permit ascertaining the initializing cause? What modifications to the existing CVRs and FDR's would be necessary to provide this capability?

As of this date and to our knowledge no polarity sense studies of explosive acoustics recorded by CVR's or FDR's have been published. Such studies appear to be long overdue and could have possibly helped to more rapidly and more efficiently explain the recent disintegration of TWA 800 over Long Island in July 1996. An explosive blast, recorded and reproduced from a high quality audio tape system is shown in Figure 9.

[pic]

Figure 9: Explosive blast from audio recorder

SUMMARY

To determine the reason for an engine loss and the resulting in-flight breakup of a turboprop commuter airliner, an engine mount reliability study and propeller whirl flutter analyses were first carried out. The results of these studies indicated that for engine mount damage conditions similar to those found in service, a whirl flutter event was possible for the flight speeds experienced by this aircraft. An analysis, involving the CVR tape, provided the final evidence. The speech and acoustical sounds at the EOT indicated an explosive high level short duration signal with no prior warning to the pilots. No voice stress signatures were present. Just seconds before the EOT, a significant FM modulation was detected both audibly and with the aid of a spectrum analyzer. The modulation was at the propeller fundamental rotation frequency and was due to the dynamic mass unbalance generated by a rotating propeller as it tore loose from its mounting system. An intermittent amplitude modulation of the noise signatures was found on the silent track by electroacoustic transduction. This occurred over the entire thirty minutes of the tape recording. The frequency of the modulation correlated with the predicted damaged engine mount frequency just prior to and at the onset of the explosive whirl flutter event. The CVR acted as a latent transducer to not only confirm the whirl flutter event, but also to warn of the existing engine mount damage at least thirty minutes prior to the catastrophic whirl flutter event. The potential also exists for the CVR to act as a latent sensor to discriminate between explosive and catastrophic decompression events in the aircraft.

REFERENCES

1. “Airline Pilots Association Report, Beechcraft 1900C N811BE, 1992; see also "Petition for Reconsideration of Probable Cause," Air Line Pilots Association, International, N811BE-ALPA 97008, June 25, 1997.

2. Sendzimir, V., “Black Box”, Invention and Technology, Fall 1996, pp. 26-32.

3. Buschow, Monte, “A Study of the Reliability of In-Service Engine Mounts,” M.S. Thesis, The University of Texas at Austin, Spring 1997.

4. MSC/Nastran - Handbook for Aeroelastic Analysis Vol. 3, Version 68 Mac Neal-Schwendler, 1995, USA.

5. Brock, B., “Personal Communication Concerning the Modified XC 142 Whirl Flutter Software Code” 1966.

6. Frederick V. Hunt, Electroacoustics ( ASA Trade Paper, ISBN: 0-88318-401-X, 1982).

7. Keithley Company, Low Level Measurements, 4th Edition, no date.

8. “Wavelet analysis,” A. Bruce, & D. Donoho, Hong-Ye Gao, IEEE SPECTRUM, Oct., 1996.

9. Indian Government Report of The Court Investigation, “Accident to Air India Kanishka on 23rd June 1985."

10. UK Air Accidents Investigation Branch. “Report No. 2/90 on Boeing 747-121, N739PA at Lockerbie, Dumsfriesshire, Scotland on 21 December 1988."

*Appreciation is expressed to Arthur Wolk of Wolk and Genter for bringing this problem to our attention and for providing additional insight on the accident. Our thanks are also due to John Murphy, who was helpful in providing us with data along with a professional pilot's interpretation of the event, and to Monte Buschow who was quite beneficial in the preparation of this manuscript.

For a lay language version of this manuscript, see in on the World Wide Web at

Appendix A

In our cockpit voice recorder (CVR) tape studies four structural acoustic signatures were found on the CVR silent track. Silent track refers to the track identified as without a microphone but still attached to the CVR input amplifiers. These four signatures included:

1.) Aircraft breakup just prior to record head shut-off transient.

2.) The speed signatures of both engines including tracking of propeller rpm changes.

3.) Noise of tires upon impacting runway.

4. Structural acoustic signatures of fractured engine truss tubes impacting the surrounding structure.

Spectrographic analysis, magnetic development of the original CVR tape, and time series signatures of the silent track have all been utilized to demonstrate the presence of these four signatures. As one would expect, the last of these signatures (4.) would be the most debated. Frequency variations from the theoretical aeroelastic analyses of three independent studies are inconclusive, ranging from estimates of 0 Hz for a divergence instability to 0.86 Hz for the most current work to approximately 1.1 Hz for the earlier 1900D aircraft studies. The new 1900C studies with modified propeller inertias also yield a range of frequencies, from 0 Hz for a three-tube damage to 0.64 Hz for a two-tube damage. However, as I mentioned before, none of the analyses have involved tuned data to a ground vibration test as is the conventional procedure. Current test results shown in Enclosure 14 carried out on a small E.A.A. Star-Lite aircraft under controlled laboratory conditions suggest that item 4.) is certainly plausible. This study was carried out in our Aeroelasticity Laboratory working with Stuart Rohre of our Applied Research Laboratories. The study consisted of impacting the Star-Lite’s tubular engine mount with an instrumented hammer in a welded joint area. Meanwhile, two twisted pairs of Teflon coated copper wires picked up the impact signals, as illustrated by Figures (a) and (b) of Enclosure 14. One twisted pair of wires, about two feet in length, was wrapped around an engine mount tube located away from the impacted joint. The other wire pair, about nine feet long, was taped along the outside of the fuselage from the firewall to the tail of the aircraft. Figure (a) shows good time series phase correlation, in the presence of significant background noise, between the impacts and the response of the aircraft, measure by the fuselage wire. Figure (b) illustrates a similar correlation between engine truss impacts and wire response, as measured on an engine mount steel truss tube away from the impact location. The last figure of Enclosure 14, Figure (c), shows the EMI field from a strobe light flashing about 28 inches away from the wire. Note the distinction between a mechanical impact pickup and an EMI pickup wave form. Possibly more work is needed here, such as a wavelet analysis, to further classify some five different wave form spikes observed. This one signature out of the four earlier mentioned structural acoustic signatures, that is of engine truss tubes impacting the surrounding structure, may still be open to question. However, the above controlled laboratory tests indicate, however, that such signatures are plausible.

[pic]

[pic]

................
................

In order to avoid copyright disputes, this page is only a partial summary.

Google Online Preview   Download