AVIATION Reports - 2005 - A05F0047

1.6 Meteorological Information

The reported weather at the time of departure from Varadero (0600) was as follows: winds variable at two knots, visibility 8000 m, few clouds (less than 2/8 sky coverage) at 1800 feet above ground level (agl), temperature 14ºC, dew point 12ºC, altimeter setting 1021 millibars. The weather at Varadero at the time of landing (0800) was reported as follows: winds variable at two knots, visibility 7000 m, few clouds at 1800 feet agl, temperature 12ºC, dew point 11ºC, altimeter setting 1020 millibars.

At the time of the occurrence, the flight crew was in night visual flight conditions, and no turbulence was reported.

1.7 Aids to Navigation

There were no reported problems with navigational aids.

1.8 Communications

1.8.1 Air Traffic Control

TSC961 levelled off at FL 350 at 0701. As a result of the in-flight problem, TSC961 climbed nearly 1000 feet, but there was no other traffic in the area; this altitude incursion did not result in a loss of separation. TSC961 was initially being guided for an approach in Fort Lauderdale, but the aircraft subsequently returned to Varadero.

The crew was in contact with the controllers of four separate ATC sectors between the time of the occurrence and landing at Varadero. An emergency was not declared.

1.8.2 Crew/Company Communications

At 0717, a phone patch was initiated with Air Transat dispatch in Montréal through New York aeronautical radio incorporated (ARINC) using their high frequency radio. The flight problem was discussed with dispatch and maintenance.

1.8.3 Communication Between the Flight Deck and the Flight Attendants

After hearing the abnormal loud noise, the FD contacted the flight deck via the interphone. The flight crew was unable to respond at the time because of the control situation. Shortly thereafter, as per the company's prescribed abnormal/emergency communication procedure, the captain called the FD and provided the TESTRA briefing:

• T - Type of problem: autopilot not responding, flight diverting to Fort
       Lauderdale
• E - Evacuation (land or ditch): no evacuation
• S - Signals (standard or alternate): standard signals
• T - Time available before landing: 10 minutes
• R - Relocation of passengers: not necessary
• A - Announcement to passengers done by (captain or FD): captain

The captain did not ask the FD for a briefing with respect to the cabin environment and none was provided. In abnormal and emergency situations, it is neither the flight crew procedure nor practice to ask the FD if he/she has information to provide. It is assumed that any information that may assist in decision making will automatically be provided. Air Transat's procedure for communicating in abnormal situations calls for flight crew to ask if there are any questions following the TESTRA briefing, which they did. None of the FAs that were in the area of the aft galley contacted the FD or the flight crew to provide information in reference to the abnormal events encountered because they assumed that the flight crew and the FD were aware of the severity of what was felt in the back.

In accordance with the applicable regulations and standards and as per the operator's approved training program, all crew members, pilots and FAs had received training with respect to crew communication. As well, they attended crew resource management training, which also addresses crew communications. Such training is provided during initial and annual training. During annual training, FAs and pilots also participate in joint crew communication training sessions during which communication skills and procedures are reviewed during simulated emergency situations.

For initial FA training, the prescribed communication training objective is to teach the importance of, and the procedures for, effective communication in normal, abnormal, and emergency situations. Emphasis is placed on

• the responsibility of FAs to provide complete and accurate information to the pilot-in-command to assist in decision making;
  
• the potential hazards to flight safety if communication is not effective; and
  
• the consequence of poor communication in aviation occurrences.

FAs are taught that they must communicate any on-board safety concerns they may have witnessed or that may have been communicated to them by passengers. When communicating safety concerns during normal or abnormal operations, FAs are to adhere to the line of authority when possible.

However, if FAs notice an emergency situation developing, including unusual noises, they must contact the flight crew immediately via the interphone, stating their position and the nature of the problem. Training stresses that FAs should never assume that the flight crew is aware of everything that is happening. When information is not communicated, its potential value to flight safety is lost.

There are procedures that set out the requirement for and the manner in which the FD must provide information to the FAs in abnormal and emergency situations. However, no such procedure or guideline was identified with respect to the FD collecting information from the FAs. As well, there is no requirement for the FD to provide flight crew with a structured briefing regarding the cabin environment in those situations.

1.9 Aerodrome Information

TSC961 used Runway 06 at Varadero/Juan Gualberto Gómez International Airport (MUVR), Cuba, for the initial arrival, departure, and the subsequent return that night. Runway 06 is 11 490 feet long and 148 feet wide, with an asphalt surface, and is served by an instrument landing system. Air Transat had maintenance personnel on site at the airport.

The Fort Lauderdale/Hollywood International Airport (KFLL), Florida, has a set of parallel runways and a crossing runway. TSC961 was being guided to Runway 27R, which is 9000 feet long and 150 feet wide, with an asphalt surface, and is served by an instrument landing system. Air Transat had maintenance personnel available at KFLL, but customs services were not available at night.

The Miami International Airport (KMIA), Miami, Florida, has four runways: 08/26L, 08/26R, 09/27 and 12/30. Runways 12, 08R, 09, 26L and 27 are equipped with an instrument landing system. Air Transat did not have maintenance personnel on site in Miami.

The Aircraft Rescue and Fire Fighting category of the three airports that could potentially have received TSC961 on the night of the incident exceeded the minimum response requirement for rescue and firefighting services for an aircraft the size of an Airbus A310.

1.10 Flight Recorders

1.10.1 Digital Flight Data Recorder

The aircraft was fitted with a Honeywell/Sundstrand model universal flight data recorder (UFDR), part number 980-4100-DXUN, serial number 10623. The recorder used an eight-track Mylar tape. The recording system consisted of a data frame of 64 words per second, recording over 300 parameters, with a minimum capacity of 25 hours. The digital flight data recorder (DFDR) was received in very good condition. The recorder was disassembled and the tape was removed from the crash-protected memory cartridge for playback on an eight-track reel-to-reel instrumentation recorder at slower speed. A total of 25.3 hours of data were recovered from the recorder.

1.10.2 Cockpit Voice Recorder

The cockpit voice recorder (CVR) on the aircraft was a Loral Fairchild model A100-A, part number 93-A100-80, serial number 60662, and was received in very good condition. The recorder contained four 30-minute audio tracks. Tracks 1 and 2 contained the radio channels of the captain and co-pilot, track 3 was the cockpit area microphone channel, and track 4 contained public address/interphone and radio communications. The quality of the recording was good.

The aircraft flew for 1 hour 17 minutes after the loss of the rudder. The CVR audio of the rudder-loss event was overwritten, resulting in the loss of information, including the noises heard in the cockpit during the rudder failure. The CVR recording started with the aircraft en route to Varadero, approximately 15 minutes before landing. The last 15 minutes were recorded on the ground in Varadero; the crew had not disabled the recorders. As a result of the TSB investigation into the Swissair Flight 111 accident in Nova Scotia, the Board, in 1999, made two recommendations that CVRs installed on aircraft be required to have a recording capacity of at least two hours (A99-01 and A99-02). As a result, aircraft manufactured after 31 December 2002 must retain information recorded during the last two hours of aircraft operation. Aircraft manufactured before this date, however, continue to require CVRs with a minimum of 30-minute recording capacity.

There was no company procedure describing how to disable the recorders after landing. Current requirements in Canada are set out in TC's Aeronautical Information Manual (AIM) under General Information, Section 3.0, Transportation Safety Board of Canada, Subsection 3.4.3, Protection of Occurrence Sites, Aircraft, Components and Documentation, which states in part

The AIM is consistent with the Transportation Safety Board Regulations, Section 9 (1), Preservation of Evidence Respecting Reportable Accidents and Incidents.

1.10.3 Direct Access Recorder

On the aircraft, there is a direct access recorder (DAR) with an optical disk device having a storage capacity of 128 megabytes. The data frame had a configuration of 128 words per second, recording approximately 127 parameters, identical to that of the DFDR. Both the DAR and DFDR recorded flight data from identical sources; however, the recorded samples were not identical due to differing sample times. Data acquisition for both DAR and DFDR is handled by the digital flight data acquisition monitoring unit. The unit, manufactured by SAGEM, combines both the digital flight data acquisition unit function for the DFDR and the data management unit function for the DAR, feeding data to both recorders. The DAR optical disk was not originally requested by the TSB. Arrangements were subsequently made to transfer all applicable DAR data to the TSB. A total of 977 hours (not continuous) of DAR data applicable to the incident aircraft, including the incident flight, were obtained from Air Transat.

The DAR data were scanned for possible airborne and ground events. From the DAR data available, there were no significant events recorded that indicated lateral acceleration excursions, severe turbulence, or rudder doublets.⁸ Similarly, there were no significant ground events recorded that might indicate an impact to the rudder.

1.10.4 Data Sampling Rates

The DFDR and DAR data were manually time-synchronized and the data showed good correlation, with the exception of the lateral acceleration data for approximately two seconds at the start of the rudder-loss event (see Appendix A). The differing data were the result of a highly dynamic event. Both the DFDR and DAR sampled lateral acceleration at a rate of 4 Hz. At this rate, it was not possible to identify any lateral acceleration frequencies above 2 Hz.⁹ The determination of the specific frequencies involved in the rudder-loss event was not possible due to these low sampling rates of the recorded lateral accelerations.

Under current regulations (Standard 625, Schedule 3, Aeroplane Digital Flight Data Recorder (DFDR) Specifications, of the Canadian Aviation Regulations [CARs], which are harmonized with Part 121, Appendix M, of the United States Federal Aviation Regulations), the sampling intervals for lateral and longitudinal acceleration are 4 Hz and vertical acceleration is 8 Hz. These rates meet the performance standards as recommended by the European Organisation for Civil Aviation Equipment (EUROCAE) minimum operational performance specifications (MOPS) for Crash Protected Airborne Recorder Systems (ED112).

1.10.5 Filtering of Recorded Data

The control surface position data recorded on the DFDR and DAR, including rudder position, were filtered by the system data analogue converter before recording. The filtered data are fed to the cockpit instrument displays, and the filtering process is designed to smooth out the data to remove unwanted spikes and prevent erratic indications. This same information is also recorded on the DFDR, and due to sampling and filtering, does not accurately represent the true control surface positions under dynamic conditions. Since the rudder loss on TSC961 was a dynamic event, critical information concerning the flight controls was potentially lost due to filtering.

1.10.6 Summary of Flight Recorder Data

At the time of the occurrence, the aircraft was in steady level flight at approximately 35 000 feet and 270 knots (Mach 0.795), with no significant control movements or turbulence. The aircraft had not exceeded any load or airspeed boundaries of its structural design envelope.

Approximately 50 seconds after levelling off at FL 350, a dynamic oscillation in lateral acceleration occurred, lasting for approximately two seconds. This was the first indication in the DFDR/DAR data of the rudder-loss event. At the start of the oscillations, the lateral acceleration changed from +0.006 g to -0.073 g, indicative of a lateral force applied to the aircraft. Within one second of the dynamic oscillations in lateral acceleration, the heading decreased by 2º and the aircraft began to roll left from wings level. At the same time, the autopilot commanded aileron and spoiler deflections (right-hand spoilers 5, 6 and 7 extended) for right roll. The recorded rudder position indicated movement to the right from 1.2º left of neutral (0º with the 1.2º bias removed) to approximately 0.3º left of neutral (0.9º right of neutral, with the 1.2º bias removed). A pitch increase from 2 to 3º nose-up occurred, with a corresponding increase in vertical acceleration to +1.28 g.

A yawing/rolling oscillatory mode, consistent with Dutch roll, commenced within two seconds of the rudder-loss event, as the dynamic oscillations in lateral acceleration decreased. At this time, a slight increase in altitude was followed by a decrease in pitch (from 3º to 2º nose-up), and a reduction in engine thrust (N₁ decreased from 90 to 77 per cent). A gradual reduction in speed followed. The roll attitude reached 6º left-wing-low and then reversed direction. Approximately six seconds into the event, the recorded rudder position reached 6.2º right of neutral (approximately 7.4º right, with the bias removed). At the speed of 270 KIAS, the recorded rudder deflection was beyond the YD authority of ±3.7º.

Approximately seven seconds into the event, the No. 2 autopilot was disengaged, followed immediately by disengagement of the auto thrust mode (manual throttle armed). The aircraft began to climb above FL 350 approximately 18 seconds into the event. With the autopilot disengaged, the oscillatory motion decreased in amplitude as the aircraft climbed through 35 200 feet, and as airspeed decreased through 256 KIAS. The speed decreased to a minimum of 248 KIAS. The altitude briefly peaked at 35 900 feet and the aircraft then began to descend.

Autopilot No.1 command mode was briefly engaged as the aircraft descended through 35 000 feet. With autopilot engagement, the yawing/rolling oscillatory motion increased in amplitude. After approximately 17 seconds, the autopilot was disengaged and the oscillations subsequently began to decrease in amplitude. As the aircraft descended through 27 900 feet and the speed decreased through 258 KIAS, the oscillatory motion ceased.

1.11 Wreckage and Impact Information

1.11.1 Miscellaneous Damage

Some ceiling panels inside the passenger cabin had partially popped out of position. The displacement was very slight and did not impede passenger movement. The interior of the fuselage compartment behind the aft pressure bulkhead was inspected; there were no indications that the loads and vibrations associated with the rudder separation had caused any structural damage.

The aircraft exterior was inspected, and there were no missing panels or structural components that may have come loose and struck the rudder. Apparent scrapes on the fuselage side, directed upward toward the tail, were determined to be poorly adhered, peeling paint and were not the result of foreign object damage (FOD). There was also blue-colour paint transfer visible on the left side of the tail cone, just aft of the rudder, probably the result of a piece of the rudder striking the tail cone during separation. There was a series of puncture holes in the fuselage skin on the upper right side near the base of the VTP. These punctures were the result of the impact, during rudder breakup, of the mechanical fasteners that attach the rudder leading edge fairing to the rudder.

1.11.2 Vertical Tail Plane Damage

1.11.2.1 General

Photo 2 shows the VTP and its rudder residuals being removed from the aircraft. The damage to the VTP trailing edge panels was generally limited to minor paint chipping. There was no damage suggesting that the rudder had been battered due to extreme travel from side to side.

1.11.2.2 Main Attachment Fittings

The VTP main attachment fittings were examined. On the fuselage side, these fittings are constructed of metal. Following the occurrence, they were subjected to visual and NDI, and no damage was found. The six CFRP main attachment fittings on the VTP side were subjected to ultrasonic NDI. Delamination damage was found in the two aft main attachment fittings.

When the VTP is loaded in lateral bending, the two rear main attachment fittings are the most severely loaded. A full-scale test of the VTP conducted during the initial certification involved fatigue testing for three lifetimes followed by static testing, where the specimen main attachment fittings failed at over 1.9 times the limit load. In addition, three further static load tests conducted during an earlier investigation resulted in attachment fitting failures at greater than 1.8 times the limit load. It is noted that design ultimate load corresponds to 1.5 times the limit load and that, in order to meet certification requirements, a structure must withstand design ultimate load for at least three seconds. These tests demonstrated that the design exceeds certification load requirements.

An analysis conducted in support of this occurrence investigation determined that, in order to cause the damage observed to the VTP main attachment fittings, the load experienced during the occurrence exceeded the design ultimate lateral fin bending load. However, it was not possible to quantify the precise load value attained.

A 3D finite element analysis was conducted and included details of the main attachment fittings delamination on the occurrence aircraft, as reported by the NDI. This analysis was validated by a test on a damaged rear attachment fitting. It indicated that, when ultimate loads were applied to the model, strain levels varied only slightly from those of the undamaged model and were well below the levels required to cause a fracture of the main attachment fittings. Therefore, the delamination had a minimal effect on the strength and stiffness of the main attachment fittings. Consequently, after the rudder-separation event, the aircraft was not in danger of losing the VTP during the flight either through loss of static strength or loss of stiffness.

1.11.2.3 Hinge Arms

Laboratory examination of hinge arm 1 found the right bolt under tolerance at the attachment of the hinge arm to the VTP. The fitting on the aft face of the VTP showed no visible signs of damage, such as cracks in the paint or sealant, but ultrasonic NDI found delamination around the mechanical fasteners. The forward ends of the hinge arms did not show any indication of upward travel as found at hinge arm 5. The rudder residuals were still attached. All three electrical grounding wires, two on VTP side and one on rudder side, were fastened with no indication of burning. There was no sign of any extreme side-to-side travel as with hinge arms 5 and 6. There were impact marks on the rudder leading edge fairing caused by the hinge arms. This damage was restricted to the centre region, and the damage lines up with the hinge arms when the rudder is not deflected.

At the hinge arm 2, 3, and 4 positions, the hinge arms are co-located with the hydraulic actuators. De-synchronization of the hydraulic actuators can result in force-fighting between them, which could lead to damage at their attachment points. Visual and NDI of the attachments did not find any sign of damage to the structure or to the mechanical fasteners. There was no indication of any structural damage that would degrade the stiffness of the actuator attachment. All the electrical grounding wires were fastened with no sign of burning. There was no indication of any extreme rudder travel as was the case at hinge arms 5 and 6. The z-strut located above hinge position 4 is designed to transfer vertical loads from the rudder into the VTP. The attachment fitting at the upper end of the z-strut showed no visual sign of damage, such as cracking of the paint or sealant, but there were paint chips on its top surface. This damage was probably caused by the upper end of the rudder separating and dropping vertically.

At the hinge arm 5 position, the metal hinge arms were still securely fastened to the fitting on the aft face of the VTP rear spar, and the fitting showed no visible signs of damage, such as cracks in the paint or in the sealant. The forward ends of the hinge arms showed damage consistent with the hinge arms having moved upwards. The rudder-side hinge fitting was still attached along with a short section of the rudder spar, roughly 23 cm high by 26 cm wide. Ultrasonic NDI found no delamination around the mechanical fasteners that attach the CFRP fitting to the VTP rear spar, but the shim layer used to adjust the thickness of the CFRP fitting was mostly disbonded.

All the electrical grounding wires were fastened with no sign of burning. The rudder-side bonding cable was badly frayed at the forward end, roughly at the position of the hinge bolt. The hinge arm was damaged by extreme side-to-side travel of the rudder hinge fitting, reaching nearly 90º deflection in each direction.

Manufacturer's drawings indicate that, at a rudder deflection of 45º, the rudder leading edge fairing cut-out strikes the hinge arm. At 60º, the rudder-side hinge fitting strikes the hinge arm. At 84º, the rudder side panel strikes the VTP trailing edge panel. There was no damage to the hinge arm where the leading edge fairing would have struck as the rudder passed through 45º of travel, and no damage to the VTP trailing edge where the rudder would have struck while passing through 84º of travel. The absence of such damage indicated that the damage from the extreme side-to-side travel occurred after detachment of the rudder and would have started as the rudder passed through 60º of travel, progressing until the rudder reached about 90º of travel.

At the hinge arm 6 position, the metal hinge arms were still securely fastened to the fitting on the aft face of the VTP. The fitting showed no visible signs of damage. The forward ends of the hinge arms did not show any sign of upwards travel. The rudder-side hinge fitting was still attached along with a short section of the rudder spar, roughly 15 cm high by 22 cm wide. Ultrasonic NDI found delamination around the mechanical fasteners that attach the CFRP fitting to the VTP rear spar. All the electrical grounding wires were fastened, with no sign of burning. The bolt that attaches the left side of the hinge arm to the VTP was nearly seized.

The hinge arm had been damaged by extreme side-to-side travel of the rudder, where the rudder-side hinge fitting had struck the hinge arm, reaching nearly 90º deflection in each direction. The damage was less severe than that at hinge arm 5. Manufacturer's drawings indicated that, at a rudder deflection of 43º, the rudder leading edge fairing cut-out strikes the hinge arm. At 70º, the rudder-side hinge fitting strikes the hinge arm. At 84º, the rudder side panel strikes the VTP trailing edge panel.

There was no damage to the hinge arm where the leading edge fairing would have struck as the rudder passed through 43º of travel, and no damage to the VTP trailing edge where the rudder would have struck while passing through 84º of travel. The absence of such damage indicated that the damage from the extreme side-to-side travel occurred after detachment of the rudder; the damage would have started as the rudder passed through 70º of travel and would have progressed until the rudder had reached about 90º of travel.

At the hinge arm 7 position, the VTP-side CFRP attachment fitting had fractured and separated.

1.11.3 Rudder Damage

1.11.3.1 General

As will be discussed later in this report, subsequent analysis found that the rudder loss during flight was progressive, and by the time the aircraft landed, most of the rudder had separated from the aircraft. Photo 3 shows the empennage after landing in Varadero. The separated pieces fell into the ocean and none were recovered. Rib 0 remained attached, as did the length of rudder spar up to hinge point 4. A small piece of rudder side panel from each side remained attached to the spar in the region between hinge points 2 to 4, and also in the corner where the spar meets rib 0. The leading edge fairings between hinge points 2 and 4, and below hinge point 1, were still attached. The leading edge fairing between hinge points 1 and 2 had separated and some pieces were found jammed between the rudder and the VTP. At hinge points 5 and 6, small pieces of the rudder spar remained attached to the hinge arm. At hinge point 7, the VTP-side hinge bracket had fractured and separated, so none of the rudder remained.

1.11.3.2 Detailed Description of Rudder Damage

The front face of the rudder spar was cleaner at the hinge positions, consistent with what would be expected if the hinge areas had been cleaned for inspection. The aft face of the rudder spar was generally clean along its entire length, becoming slightly dirtier toward the bottom. There were drip stains oriented downwards originating at the lightening holes, consistent with normal in-service staining caused by dripping hydraulic fluid, corrosion inhibitor, or other fluids. There were dark stains observed on the interior of the side panels where the reinforcing bolts pass through the GFRP blocks at the hinge points. These stains originated at the bolts and progressed in a downward direction. The foils, which normally cover the lightening holes on the rudder spar, were missing, and the pattern of dirt around each lightening hole suggested that the foils had been removed for a considerable period before the occurrence. The top surface of rib 0 was visibly dirty. There were no stains on the interior bottom of the rudder to suggest that fluid had been pooling in the bottom of the rudder. The fluid drain holes and drain paths at the bottom of the rudder were not plugged. There were no stains in the honeycomb cells to suggest the presence of trapped fluids; however, very little of the honeycomb remained for examination.

Cross-section examination of the rudder exterior skin revealed 10 layers of paint composed of primer, anti-static, filler, and topcoat. There was an accumulation of three paint re-sprays. It was calculated that the mass of the extra two re-sprays was approximately 19.3 kg. The total mass of a rudder with the nominal paint system is approximately 190 kg.

The small amount of surviving rudder was examined for indications of contact with maintenance equipment, FOD, or damage by misuse. The only finding was a circular grinding mark on the exterior of the right rudder side panel. Cross-section examination determined that the grinding mark only extended down into the first few layers of paint, with no damage to the CFRP or discolouration due to heating.

On each side of the rudder, there are three LPPs running chordwise. On the occurrence rudder, recent maintenance had involved the replacement of the lower right LPP in May 2004. A short length of this LPP remained and its fractured end was bent forward. A section of the side panel at this LPP was taken for subsequent laboratory analysis.

The right side panel's outer face sheet exhibited many small multidirectional surface marks. There were similar marks on another aircraft (MSN 600), whose rudder was inspected and found undamaged. A section cut through these marks on the occurrence rudder revealed that they were cracks that had originated in the paint, caused by excessive paint thickness. It was further found that, when a paint crack was parallel to the direction of the CFRP fibres, the crack could extend down into the CFRP resin matrix. These cracks were limited to the matrix and did not damage the fibres.

The CFRP face sheets had separated from the honeycomb core and the separation had a different appearance depending on whether it was an interior or exterior face sheet. The interior face sheets had generally separated from the honeycomb very cleanly near the bond line. However, the exterior face sheets had separated from the honeycomb in a very jagged manner with separations occurring at different depths in the honeycomb.

Microscopic examination of the inner skin separation found that they were mostly cohesive failures within the bond line through the meniscus.¹⁰ Since the honeycomb had been so badly damaged during the occurrence, it was not possible to distinguish damage to honeycomb cells that may have been caused by freezing of trapped water. At the actuator locations, the interior CFRP face sheets had separated into four plies. There were no significant gaps in the coverage of the splice bond adhesive at the edges of the honeycomb sheets. In regions where a separation occurred near a honeycomb splice bond, the separation tended to occur in the weaker density honeycomb, and not in the bond line. Exposed regions of honeycomb that were no longer supported by the CFRP had tended to split into many small chordwise "fingers," each about 25 to 50 mm wide.

The rudder spar had fractured just above the hydraulic actuator attachments. Examination of fractured fibres indicated that the spar separated in an up and aft direction. The metal strip along the z-section at the front edge of each side panel had also fractured at this location, and examination revealed that it was a ductile overload failure.

Photo 1 shows that there was no significant amount of rudder side panel still attached between hinge positions 1 and 2. In this region, more honeycomb remained on the right side, but more inner skin remained on the left side. The joint between the side panels and the spar, which uses blind mechanical fasteners, had not failed and the fasteners were intact. Examination of the fractures at the joint between the front spar and the side panels revealed that the side panels or part of the side panels separated toward the outboard. Along the length of rudder spar between hinge points 1 and 2, the z-sections had fractured and separated along with the side panel on both sides. Since the leading edge fairing attaches to the z-sections, this explains why the leading edge fairing was missing in this region. The metal actuator attachment fittings at the hinge point 2, 3, and 4 positions did not show indications of damage, deformation, or looseness.

The joint between the side panels and rib 0, which uses blind mechanical fasteners, had not failed and the fasteners were intact. There was a wipe mark across the top of rib 0, consistent with a fractured section of the left side panel moving towards the right and downwards. At the left side panel separation, more of the z-section had remained than on the right side. The fastener holes had been torn out towards the bottom, suggesting that the left side panel or part of the side panel separated from rib 0 in a downwards or outwards direction. The left side panel also had compression damage, suggesting the inboard skin moved downwards during separation. A failure in the z-section remains suggests that the outboard skin moved outboard during separation.

At the right side panel separation, a length of the z-section had fractured and separated. Marks were observed on the remaining CFRP edge and their spacing corresponded to the spacing of the missing mechanical fasteners. These marks suggest that the right side panel or part of the side panel separated in an upwards direction. Examination of the right side panel showed that it separated from rib 0 in a tension flexion failure. The metal strips along the z-section had failed by overstress, a combination of tension and bending to the outside. There was a skin buckle on each side panel consistent with rib 0 moving upwards. There was a crack at the tip of rib 0 whose orientation was consistent with rib 0 twisting to the right.

Examination of other rudders as part of the fleet inspections following the occurrence found side panel damage at the hoisting points and the trailing edge fasteners. Since none of these areas of the occurrence rudder were recovered, it was not possible to examine them. Furthermore, since the entire upper end of the rudder was not recovered, the area around the 1997 lightning strike damage could not be examined.

Only a short section of the rudder spar at hinge point 5, roughly 23 cm high by 26 cm wide, remained attached. The lower surface of the spar section included the edge of a lightening hole. The rear reinforcement plate was still securely fastened to the spar and all its fasteners were still present and appeared undamaged. The separation between the honeycomb and the CFRP skin had generally occurred in the honeycomb, at varying depths and not along the honeycomb/CFRP bond line. On the front surface of the spar, the rudder-side hinge bracket had fractured. The fractured surface appeared typical of a tensile/bending overload failure with no indication of fatigue. Metallurgical analysis determined that the fittings were made of the correct aluminum alloy and temper.

Only a short section of the rudder spar at hinge point 6, roughly 15 cm high by 22 cm wide, remained attached. The lower surface of the spar section included the edge of a lightening hole. The rear reinforcement plate was still securely fastened to the spar, and all its fasteners were still present and appeared undamaged. The separation between the honeycomb and the CFRP skin had generally occurred in the honeycomb, at varying depths, and not along the honeycomb/CFRP bond line. On the front surface of the spar, the rudder-side hinge bracket had fractured. The fractured surface appeared typical of a tensile/bending overload failure with no indication of fatigue. Metallurgical analysis determined that the fittings were made of the correct aluminum alloy and temper.

1.11.4 Chemical Attack and Contamination

The rudder residuals were examined to study the possibility that they had been contaminated and degraded by exposure to chemicals. The manufacturer provides a list of approved consumables, as well as procedures to follow for the approval of materials not on that list. No indication was found that unapproved consumables were being used by the operator. There is no in-service experience to suggest that there was a systemic problem with chemical attack by approved consumables. During the material qualification process at certification, extensive testing was conducted to understand the interaction between the materials and possible contaminants, including hydraulic fluid. However, the bond between the honeycomb and the CFRP face sheets was not included in these tests because it was in the interior structure and considered to be sealed from such exposure.

A water and hydraulic fluid mixture may react under certain concentrations to form phosphoric acid, which can attack epoxy resin creating irreversible damage to the core/face sheet interface. Microscopic examination of the rudder of aircraft MSN 361, which was known to be contaminated by hydraulic fluid, revealed that hydraulic fluid had attacked the matrix of the GFRP layer adjacent to the honeycomb, weakening the bond, but not leading to a disbond.

Based on service experience with the aircraft MSN 361 and MSN 545 rudders, the access path for hydraulic fluid into the sandwich structure is around the blind fasteners at the front and bottom edges of the side panels. Three methods were used to search for the presence of hydraulic fluid contamination: energy dispersion X-ray spectroscopy (EDX), X-ray photoelectron spectroscopy (XPS), and infrared (IR) spectroscopy. The rudder of aircraft MSN 361, which was known to be contaminated by hydraulic fluid, was used to calibrate these three analysis methods. EDX testing of regions that were visibly stained by hydraulic fluid found roughly 2 per cent phosphorus content, and XPS testing found roughly 0.8 per cent phosphoric acid-ester content.

The top surface of rib 0 of the occurrence aircraft was visibly dirty, and an area at the front near the spar was analyzed. EDX results indicated 0.4 per cent¹¹ phosphorus, considerably lower than the 2 per cent associated with the visibly contaminated region of the rudder of aircraft MSN 361. An area of inner skin (non-honeycomb side) from the left side panel front bottom corner was analyzed. EDX results indicated less than 0.1 per cent phosphorus, considerably lower than values from the aircraft MSN 361 rudder. An area of the inner skin (honeycomb side) from the left side panel front bottom corner was analyzed. EDX results indicated 0.3 per cent phosphorus, and XPS results indicated 0.18 per cent phosphoric acid-ester, both considerably lower than values from the rudder of aircraft MSN 361. An area of inner skin (honeycomb side) from the right side panel front bottom corner was analyzed. EDX results showed less than 0.1 per cent phosphorus and XPS results indicated 0.07 per cent phosphoric acid-ester, both considerably lower than the values from the aircraft MSN 361 rudder.

Since the suspected hydraulic fluid ingress path was around the blind fasteners, specimens were taken around blind fasteners on both side panels at the front spar and at rib 0. Measurements were taken inside the sandwich structure at the inner face of the skin. EDX results for phosphorus on interior surfaces were all below the 0.1 per cent detection limit. EDX results on the external surfaces at rivet positions showed readings as high as 3.0 per cent. In addition, cross-section microscopic examination of the bond area did not reveal visual indication of chemical attack. Therefore, these results show the presence of hydraulic fluid contamination on exterior surfaces, but no indication of seepage into the structure.

1.11.5 System Inspection and Testing

The inspection of the rudder system on the occurrence aircraft showed that the rudder control, in cruising flight at 270 knots, would not have exceeded 7º of travel per side; the RTL control would have prevented it (RTL systems do not only limit the pedal inputs, but also limit the sum of inputs from trim, pedals or APYA, and YD). The autopilot was active at the time of the occurrence. The YD was also active (YD is active in manual flight also). The YD was restricted to moving the rudder to no further than 3.7º either side to compensate for the natural Dutch roll tendency of the aircraft.

The last rudder servo-control synchronization check, performed by Air Transat maintenance on 01 March 2005, revealed that no anomalies and no adjustments were required. In Varadero, the synchronization check showed no movement between the neutral position of the three servo-controls; there was no force-fighting between servo-controls. Therefore, the synchronization between the servo-controls was within the Airbus aircraft maintenance manual (AMM) parameters before and after the event.

The inspection and investigation of the aircraft flight control system and related subsystem components was performed by the investigation team in Varadero after the occurrence and revealed no anomalies. The rudder control system was checked and tested for proper operation in Varadero with no anomalies found. The rudder servo-controls and actuating spring rods were then removed, inspected, and laboratory tested, and no anomalies that would have affected the normal operation of the rudder system were found. In addition, all safety features that are part of the servo-controls and spring rods to ensure safe operations in case of servo-control malfunction were operational.

The free play was measured from hinge point 1 to hinge point 6; one free-play measurement was out of tolerance on the hinge line bearing at hinge point 2. In addition, 3 out of 10 VTP-side hinge arm bearings were partly seized but could still be rotated. Airbus specifies that the hinge arm bearing free play has no impact on the structural integrity of the rudder. Free play at the hinge arm bearings would result in detectable rudder vibration that will trigger a specific troubleshooting inspection process. No in-flight rudder vibrations had been reported. Operators are provided with some troubleshooting guidelines that list the most probable causes when vibrations are felt. However, the main cause of rudder vibration is play at the servo-control bearings, rather than at the hinge arms. The Airbus troubleshooting philosophy is such that, if there are no findings of free play at the servo-control bearings, or if they are replaced and the situation is not improved, the operator will contact Airbus for investigation. Airbus in-service experience has confirmed the relevancy of this approach.

1.11.6 High-Intensity Radiated Fields Investigation

The possibility that high-intensity radiated fields (HIRF) interference could affect the normal in-flight operation of the rudder system was investigated. Any oscillation from the YD system at a frequency of 20 Hz, representing the difference of frequency between the radar and rudder synchronization frequencies, will be attenuated by the YD actuator and the three servo-controls that would be acting as filters. A review of the theoretical rudder deflection when the YD system is subjected to HIRF, assuming susceptibility at 20 Hz, led to the conclusion that the maximum rudder deflection would be less than 0.1º. Therefore, the investigation determined that the effect of HIRF would have a negligible impact on the rudder surface control.

1.11.7 Examination of Pre-Occurrence Photos

Photographs of the aircraft taken before the occurrence showed curious visual features on the rudder. Photo 4 shows an example of one of these features. It was taken 11 days before the occurrence and shows light-coloured vertical lines on the left side of the rudder below the hydraulic actuators. There were also earlier photographs that showed arc-shaped lines on the left side panel just aft of the hydraulic actuators, and white spots on the trailing edge. These features were not present on the most recent photographs.

There was insufficient resolution in the photographs to conduct a photogrammetric analysis that would determine whether these vertical lines represented an out-of-plane deformation such as a disbond bubble. Since the vertical line features were observed on photographs taken on different days under different lighting conditions by different photographers, they were actual physical features on the rudder, and not simply reflections or dirt on the camera lens.

The vertical line features first appeared in photographs starting in early 2003, and the aircraft was subject to all its regular inspections in the intervening time. Examination of other aircraft found that staining of the rudder near the hydraulic actuators was not unusual. Subsequent testing found that hydraulic fluid could dissolve the Air Transat tail decal material and analysis of a vertical streak on sister aircraft MSN 600 found the streak to be composed of a mixture of hydraulic fluid and dissolved decal material.

1.12 Tests and Research

1.12.1 General

A series of laboratory examinations and tests were conducted on the residuals of the occurrence rudder, on other rudders, and on test specimens. In addition, analyses of flight dynamics, flutter, and radar data were conducted. This work was performed by the manufacturer at its facilities, with the participation of the involved national investigation authorities and specialist advisors from France, Germany, United States, and Canada. Test progress was monitored and validated by the investigation team.

A number of the tests and analyses helped to eliminate avenues of inquiry and allowed the investigation to concentrate on others that proved germane. For the sake of clarity, the report focuses on this second group of activities.

A small number of test programs developed as a consequence of this occurrence are being continued under the auspices of other entities. In each case, the objectives of the program have shifted away from this investigation to broader issues.

1.12.2 Examination of Two Sister Rudders

The rudders from aircraft MSN 592 and MSN 614 were removed from service and inspected at the manufacturer's facilities. These two rudders were selected because their side panels were produced by Soko as part of the same production batch as the occurrence rudder. Both of these rudders were subjected to visual inspection and elasticity laminate checker (ELCH) testing under TSB supervision. No damage was found.

1.12.3 Elasticity Laminate Checker Test

1.12.3.1 Elasticity Laminate Checker Test Description

The ELCH test is an NDI method developed by Airbus to detect disbonded face sheets on honeycomb-core panels. The machine applies a vacuum to a small area of the outer surface of a panel and measures deflection. If there is a disbond of either the inner or outer face sheet, a greater-than-normal deflection will be measured. The advantage of this test method is that it can be used to find disbonds on the inner face sheets, which are not easily accessible. Airbus Report TN-EV37-579/90 (18 December 1990) describes the qualification results for the ELCH inspection procedure.

1.12.3.2 ELCH 1-Complete Grid Examination of 24 Rudders of Similar Design

The exteriors of the rudder side panels were marked with a 75 mm grid, and an ELCH reading was taken at each grid point. There were approximately 2000 test points per side on each rudder. This would permit finding defects down to a size of 120 mm in diameter. A total of 24 rudders were tested, which included the flight cycle and flight hour fleet leaders. No disbonds were found.

1.12.3.3 ELCH 2-Focused Examination of Rudders

Thirteen rudders close in serial number to rudder serial number HF1090 (aircraft MSN 361) were selected for further ELCH examination. The test area was a 50 mm grid along the complete z-section, around the hoist points, and along the trailing edge fasteners. No large disbonds were found.

1.12.4 Examination of Other Rudders

During the course of the investigation, some rudders of other operators that had been damaged were inspected in greater detail.

The rudder from aircraft MSN 251 was accidentally damaged at its lower end by contact with maintenance equipment during normal scheduled maintenance activities. A repair was carried out, which included the replacement of the lower LPP using heat-assisted cure. A tap test was conducted following the repair, and no defects were found. An ELCH test was then conducted as a precaution. It found that the heat applied during the LPP replacement had not caused a disbond of the inner skin in the area of the LPP replacement. However, the area aft of the trailing edge fasteners around the repair to the contact damage failed a tap test, and subsequent laboratory examination found that the inner skin was almost completely disbonded within the repair area.

The rudder from aircraft MSN 361 was damaged at the trailing edge by unintentional contact with a maintenance dock during maintenance. Subsequent inspection found a disbond on the left inner-face sheet near the front bottom corner, approximately 830 mm long by 350 mm high. Stress analysis determined that the disbond was not caused by the impact with the maintenance dock. This internal disbond had not been detected by the AOT-1 (all operators telex - see Section 1.14.1) external tap test. Further examination revealed that hydraulic fluid had seeped into the left side panel around the blind fasteners at the front spar as shown in Photo 5, and that this fluid had weakened the bond between the honeycomb and the face sheet. In addition to the hydraulic fluid contamination, this rudder exhibited signs of water ingress around some trailing edge fasteners and at the leading edge immediately aft of the z-section.

Approximately one litre of water was drained from the area around the right lower aft hoisting point of the rudder from aircraft MSN 378 during inspection before re-paint. Subsequent inspection revealed that excessive grinding of the old paint had resulted in exposed cells in the upper GFRP area. The manufacturer then subjected the entire area of both side panels to X-ray examination and found additional water ingress at some of the trailing edge screws. Endoscopic examination around the hoist points revealed that honeycomb cell walls had been damaged or torn in some affected areas.

During the AOT-1 inspection of the rudder from aircraft MSN 530, two indications of possible irregularities were found on the exterior face of the right side panel. These were 80 cm² and 670 cm² in size. There was no visual indication of impact damage at these locations. Destructive laboratory examination determined that it was a core crush, not a disbond. It was further determined that the core crush had occurred during cure at original manufacture. A review of manufacturing records did not indicate any concession related to this damage. There was no indication that this manufacturing deviation had grown in service. Stress analysis determined that this deviation had a negligible effect on the rudder structural strength. This rudder was pre-modification 8408 and had lower density honeycomb, which was more susceptible to core crush. The occurrence aircraft was post-modification 8408, with higher density honeycomb.

Trapped fluid was found in the lower nose area of the rudder of aircraft MSN 701 during the AOT-2 inspection (see Photo 6). The fluid was not formally identified, but was reported to be mostly water with some dirt, possibly containing hydraulic fluid and sealant chips. The fluid level was reported to be up to the lower inspection hole. The rudder box aft of the spar was dry. The two drain openings in the nose section were found to be clogged. The tap test did not show any sign of disbond. An X-ray and thermography inspection of side panels adjacent to the fluid did not reveal any fluidity entrapped inside the sandwich.

1.12.5 Fluid Contamination Program

As a result of the findings on aircraft MSN 361, a separate investigation was launched by the National Transportation Safety Board (NTSB) with support from Airbus to address the issue of hydraulic fluid contamination and its effect on structural strength. This program is ongoing, but preliminary conclusions include the following:

• Hydraulic fluid contamination does not have an immediate effect on mechanical strength; the effect takes time to develop.
  
• The effect of a hydraulic fluid/water mixture is more severe than that of hydraulic fluid alone.
  
• The effect of hydraulic fluid/water or hydraulic fluid is not reversible whereas the effect of water alone is reversible.

1.12.6 Impact Damage Tests

A series of impact tests was conducted to determine whether it was possible to cause a disbond between the honeycomb and the CFRP face sheets. Drop tower tests were conducted with spherical impactors of nose radii of 12.5 to 100 mm. The test panels had honeycomb cores 40 mm thick with 32 kg/m³ density, and face sheets with one CFRP ply and one GFRP ply. Advanced drop tower tests were conducted with a spherical impactor of 2000 mm nose radius and a cylindrical impactor of 100 mm radius. These tests resulted in crushed core or face sheet perforation depending on the energy level, but no disbonds.

Chapters 12-21-11 (Cleaning), 12-31-11 (Anti Icing), and 12-31-12 (Ice & Snow Removal) of the Airbus AMM warn that the impact pressure from fluid jets must not be more than 0.35 bar. High-pressure jets can potentially damage honeycomb structure. A survey of maintenance and de-icing facilities found that nozzle pressures in the range of 3.4 to 4.1 bars were common, with maximum up to 10.2 bars. However, the impact force that results from a spray jet depends on many factors, including the width of the spray and the distance of the nozzle from the aircraft. Airbus reports that it has no knowledge of damage occurring with pressures as high as 1 bar at the impact point. No fluid impact damage tests were conducted in the course of this investigation. However, it is considered that the misuse of a high-pressure spray would most likely result in damages similar to those caused by blunt impact, and would include core crush but no disbonds. It was reported that the occurrence aircraft was last de-iced on 19 February 2005, approximately two weeks before the occurrence.

1.12.7 Laboratory Tests of Rudder Residuals

1.12.7.1 General

Only a small amount of the occurrence rudder was recovered. The remaining honeycomb was examined and its density was consistent with production drawings.

Light microscopy examination of specimens from the occurrence rudder showed the characteristic kidney-shaped cross-section of Toray T300 carbon fibres, consistent with the design specification.

Specimens were taken from the CFRP face sheets of both side panels, rib 0, the front spar, and the hinge point 7 fin-side bracket. Only one deviation from drawings was found; an additional reinforcement layer on the front spar around one of the access holes was in the wrong orientation. A review of manufacturing records did not indicate any concession related to this deviation. The rupture of the front spar did not pass through this deviated lay-up. Stress analysis determined that this deviation had a negligible effect on strength and stiffness. Apart from this exception, all the lay-ups corresponded to manufacturer's drawings.

1.12.7.2 Bond Between Honeycomb and Face Sheets

The quality of the bond between the honeycomb and the face sheets is normally evaluated using a climbing drum peel test. Insufficient undamaged residuals of the occurrence rudder remained to conduct drum peel tests. The sole alternative approach to evaluating this bond was by examining the shape of the meniscus at the interface between the honeycomb and the face sheets. A number of cross-sectional specimens were taken to examine the meniscus. Meniscus formation was generally found to be similar to the baseline comparison panels.

1.12.7.3 Bond at z-section

The front and bottom edge of each rudder side panel is trimmed with a z-section. Specimens of the inner skin bond taken near the z-section along the front edge of the left side panel of the occurrence rudder had a meniscus whose appearance was consistent with insufficient bonding pressure within a width of 20 mm. The investigation revealed that this condition could be caused by insufficient caul plate pressure during cure, either resulting from mispositioning of the z-section or accumulation of tolerances of the components. In Figure 6, the upper sketch shows correct orientation during cure. The lower sketch shows how dimensional tolerances can result in low bonding pressure next to the z-section.

1.12.7.4 Inter-laminar Bond Between Face Sheet Layers

The quality of the inter-laminar bond between the individual face sheet layers was evaluated by examining sections under microscope. No unusual features were observed that suggested a poor laminate quality.

1.12.7.5 Splice Bond

A compound was used to bond around the perimeters of the honeycomb blocks. Small gaps in the bonding were observed at some locations, but these had a minimal effect on the strength of the joint. IR spectroscopy tests confirmed that the correct splice bonding material had been used.

1.12.7.6 Type of Resins

Resins in the residuals were identified using a combination of IR spectroscopy tests and visual microscopic examination. Results found that the resins in the occurrence rudder were consistent with Hexcel F550 for the CFRP and EHG250 or EP112 for the GFRP. It was not possible to distinguish between EHG250 and EP112 since they are identical in formulation and manufacturing process. It was confirmed that approved resin types had been used.

1.12.7.7 Cure of Resins

Differential scanning calorimetry tests were conducted to determine the adequacy of curing. All cures exceeded 98 per cent. For these resin systems, any value greater than 95 per cent is acceptable.

1.12.7.8 Lightning Protection Plate Replacement

The region of the recently repaired lower right LPP was examined. IR spectroscopy found that the adhesive used to fasten the new LPP was Hysol EA934 NA, as specified in the SRM. A cross-section specimen of the CFRP layer beneath the LPP was examined under microscope, and the only unusual feature was matrix cracking in the CFRP. To provide a baseline for comparison, an LPP was peeled from a comparison rudder side panel. A tap test of that comparison specimen found that the peeling had not resulted in a disbond, and a microscopic examination of the CFRP cross-section found matrix cracking similar to that in the occurrence rudder.

1.12.7.9 z-Strut Paint Chips

The area around the paint chips on the z-strut was examined by EDX. This analysis found no traces of titanium or steel residues in the paint chipped areas that would have been caused by contact with mechanical fasteners from the upper part of the rudder separating in a downward direction.

1.12.7.10 Explosion Damage

The rudder residuals and the VTP were visually examined by explosives specialists from the German police. There was no indication of damage or residue radiating outwards from an origin point, as would be the case if there had been an explosion.

1.12.8 Double Cantilever Beam Tests

The American Society for Testing and Materials (ASTM) Specification D5528 DCB test was adapted to determine the fracture toughness of the interface between the core and the face sheet. The standard test coupon is 50 mm wide, 220 mm long, and has an initial crack length of 35 mm. For this particular investigation, the test coupon was modified from the standard to include 0.8 mm and 1.6 mm-thick aluminum doublers over the face sheets to achieve peel angles that were more representative of disbond growth. Static and fatigue tests of double cantilever beam (DCB) specimens are ongoing.

1.12.9 Disbond Growth

1.12.9.1 History of Earlier Design Aramid Fibre-Reinforced Plastic Rudders

In the earlier design of the rudder, GFRP was used as the bridging layer between honeycomb and CFRP only in the reinforced region around the hydraulic actuators. Elsewhere, AFRP was used. The AFRP system resulted in poor bond strength of the interior skin. Those rudders experienced large in-service disbonds of the inner skins. Early AFRP rudders were of the design type that had two-part side panels - upper and lower. In the rudder of aircraft MSN 237, growth of the disbond was stopped by this joint and did not progress beyond it. Disbonds were discovered in service during routine inspection and had not led to rudder separation or adverse aircraft performance. AFRP rudders are no longer in service.

1.12.9.2 Temperature Effects

Tests were conducted to investigate the effect of temperature on the out-of-plane strength in terms of climbing drum peel strength and fracture toughness (G1c) measured by a modified DCB test. Climbing drum peel tests were conducted at room temperature and at -55ºC. At room temperature, the failures tended to be in the honeycomb core, whereas at -55ºC, they tended to be in the interface bond line between the honeycomb and the face sheet. It was found that the peel strength at cold temperature was significantly reduced to about 45 to 67 per cent of its value at room temperature. The steep angle involved in this method of test is not representative of the shallow angle at which a disbond in the rudder would propagate; therefore, DCB tests were also conducted.

DCB tests were conducted at room temperature and at -55ºC. In all the tests, the failure tended to be in the honeycomb core, regardless of temperature. However, the cold temperature tests had fracture toughness values about 20 per cent lower. In addition, disbond growth at room temperature was steady and continuous, whereas at cold temperature, it was unstable.

1.12.9.3 Vacuum Cycling Tests

Test panels with various natural and artificial damage were placed in a vacuum chamber and cycled to a differential pressure of -0.7 bar to simulate ground-to-air cycles. The tests were accelerated, with each cycle lasting 90 seconds from ground to altitude and back to ground. In real time, a flight cycle lasts 540 minutes. One aircraft lifetime is 48 000 cycles. To provide conservative results, the test panels were fully sealed at the edges to impede breathing.

The following results were obtained:

• A specimen from the damaged left side panel of the rudder of aircraft MSN 361, which included the disbonded area (approximately 2256 cm²), was subjected to vacuum cycling at room temperature. The area of the damage almost doubled instantly at a pressure of 0.44 bar (absolute). The rapid propagation event was reported to be explosively loud and violent, resulting in some damage to the interior of the test chamber. During the test, a surface crack developed in the CFRP face sheet at the panel edge (Photo 7). This fracture halted any further damage growth by removing the pressure differential.

 

• A specimen from the right side panel of aircraft MSN 361 rudder was also tested. Since it was undamaged, an artificial disbond (approximately 338 cm²) was introduced. It was subjected to vacuum cycling with a change of pressure (DELTA P) of -0.7 bar at room temperature. The specimen completed the test program without failure.
• Eight test panels were constructed, which included 24 and 32 kg/m³ core densities and 30 and 40 mm thicknesses. Damage was introduced by sweeping out a disbond damage with a knife. The artificial damages had diameters of 100 to 250 mm. When exposed to cyclic vacuum loading, four of these panels demonstrated slow steady growth of the disbond and eventually failed in rapid propagation.
• A test specimen from the blunt impact test program, which had experienced crushed core but no disbond, was cycled for 10 000 flight cycles and showed no propagation of the damage.

1.12.10 Computer Simulation of Disbond Growth

Computer simulations were conducted using the LS-DYNA software to study the growth of the MSN 361 rudder disbond. The following conclusions were drawn:

• The MSN 361 rudder side panel with the disbond damage as found at the time of detection was sufficiently large to achieve unstable growth at reduced pressure altitude. Since it did not propagate in flight, it is most probable that it experienced leakage during the period before detection.
• A disbond of a size corresponding to the initial contaminated region of the MSN 361 rudder did not grow under reduced pressure altitude.
• When a disbond size corresponding to the initial contaminated region of the MSN 361 rudder was combined with a disbond at the z-sections at the lower front corner of the side panel, unstable disbond growth occurred at reduced pressure altitude.

1.12.11 Effects of Temperature and Moisture

Liquid water trapped in the honeycomb will expand as it freezes, and the repeated freeze-thaw cycle can gradually damage the cell walls. In other rudders where water ingress into the honeycomb has been documented, this infiltration has tended to occur around the hoisting points or near the trailing edge fasteners. These areas are the most likely entry paths for water since these design features involve a break in the continuity of the CFRP face sheets. On the occurrence aircraft, these areas of the rudder were not recovered, and it was not possible to determine whether the occurrence rudder had experienced liquid water ingress into the honeycomb.

The epoxy matrix in the face sheets and the Nomex® core material can absorb moisture from humid air or condensed water. Moisture absorption degrades the mechanical properties, especially of face sheets that are matrix or matrix-interface dominated, particularly at elevated temperatures. At certification, static strength of the composite structure was substantiated assuming a worst case of maximum service moisture content and by conducting tests at elevated temperature.

The temperature of an aircraft can reach extreme levels while parked on the ground due to the ambient temperature, solar radiation, and the colour of the aircraft. The rudder of the occurrence aircraft was painted dark blue, and its previous paint scheme had a black stripe across the rudder. United States Department of Transport paper DOT/FAA/AR-04/30 describes tests involving CFRP laminate panels of different colours, whose temperature was measured during exposure to sunlight. The highest measured value was 82ºC at an ambient temperature of 33ºC. No thermal analysis was conducted for the A310 rudder, but a thermal analysis for the VTP fin box of another model of transport aircraft found maximum temperatures of 91ºC in still air on the ground, and 76ºC when moving.

High temperature can have an adverse effect on the mechanical properties of composite materials. If the material's glass transition temperature is exceeded while under load, the structure can experience plastic deformation. The onset glass transition temperature for F550/EHG250 resin is 102ºC dry or 75ºC wet. Given the occurrence rudder's age, its onset glass transition temperature had most probably reached its saturated or equilibrium state, and was closer to the 75ºC value.

1.12.12 Age-Related Structures and Materials Degradation

The possibility was studied that there might be some unknown phenomenon at work that could cause a reduction in structural stiffness with age. Such a reduction in stiffness could result in a reduced flutter speed and lead to flutter. In 2004, Airbus conducted GVT in support of its MRTT program. The testing was conducted on an aged A310 aircraft (MSN 523) that had accumulated over 28 000 flight hours. This test aircraft had the same design of VTP and rudder as the occurrence aircraft. GVT results found that fin bending and rudder rotation frequencies of the MRTT test aircraft were consistent with those obtained during the original A310-300 certification. No indication was found to suggest that stiffness had reduced with age.

Airbus Report TN-ESWCG-1181/02 documents an earlier investigation of material properties changes with age. The glass transition temperature (Tg) was measured for specimens of CFRP materials Hexcel F913 and F550, and GFRP material EHG250. These results were compared with those from specimens analyzed 14 years earlier during material qualification, and there was no indication of age-related Tg degradation.

Climbing drum peel tests were conducted with specimens from the two unused baseline reference panels, which were roughly the same age as the occurrence rudder. Tests results were in the range expected, and there was no indication of age-related degradation in peel strength.

1.12.13 Flight Dynamics Analysis

1.12.13.1 Background

To study the aircraft response during the occurrence, the Airbus A310 flight model was configured to simulate the aircraft's behaviour following a loss of rudder surface area. It was first necessary to determine the movement of the flight controls (ailerons, rudder, elevator, and spoilers). This was done using the merged DFDR and DAR data. It was also necessary to derive parameters not available directly from the DFDR, such as sideslip, and to modify existing parameters to take into account data latencies. The mass properties of the aircraft (that is, weight, inertia, CG) at the time of the incident were derived using the load and trim sheet for TSC961. Modifying the aircraft configuration to simulate rudder loss required modification of lateral-directional aerodynamic coefficients to take into account the loss of surface area. This modification was done by applying an adjustable ratio to the appropriate aerodynamic coefficients, allowing investigation of differing amounts of rudder loss.

1.12.13.2 Initial Simulations

Seven initial simulations were performed and the outputs were compared to the existing DFDR and DAR data to determine their applicability to the occurrence scenario. Of these seven simulations, four were performed with different amounts of rudder loss and with different aircraft mass properties. Three simulations were performed to evaluate additional aircraft behaviour related to the incident. The timeframe of interest was from 0701:57 (approximately 4 seconds before autopilot 2 was disconnected) to 0702:14 (approximately 13 seconds after autopilot 2 was disconnected). The actual event was not simulated, since there were significant structural dynamics issues with which the flight dynamics simulator could not deal.

The results of the simulations were presented graphically showing different lateral-directional flight parameters as determined by the simulation compared with flight data derived from the DFDR and DAR. In all cases, the properties of interest for comparison with the flight data were the magnitude and the frequency. Matching these two properties of interest for all aircraft parameters would indicate a simulation configuration similar to the aircraft at the time of the occurrence.

The only simulation that reasonably matched the frequency and amplitude of the DFDR and DAR data was one in which there was a reduction of the useable rudder by 84 per cent and a reduction in the yaw moment of inertia by 10 per cent. The moment of inertia calculations for the aircraft at the time of the incident were only accurate to 10 per cent, which means that the mass properties for this simulation were within the limits of the calculations for the aircraft.

1.12.13.3 Simulations to Determine the Lateral Force at Tail During the Event

The initial event was characterized by significant lateral acceleration excursions, suggesting that a lateral force was applied to the right-hand side of the aircraft during the event. The Airbus flight dynamics simulator was unable to investigate structural dynamic aspects of the initial event but could provide an indication of the magnitude of the force required to initiate the Dutch roll motion.

The only way the simulation could introduce this force into the aircraft was through rudder movement. The rudder input was constrained by the limitation of the rudder control jacks of 60º per second. The simulated lateral acceleration resulting from the rudder input was then compared to the DFDR and DAR data. The rudder motion required to obtain the lateral acceleration was significant when compared to the DFDR and DAR data. A movement of this magnitude and duration would have been recorded on the DFDR and DAR; therefore, the rudder was not the source of the force. The force generated by this rudder motion had a peak value of 108 000 newtons (N) applied at 0701:54. Although the source of the force is not known, this does give an indication of the magnitude of the lateral force involved in the initial event.

1.12.13.4 Simulations to Determine the Longitudinal/Vertical Force at Tail During the Event

In addition to the lateral acceleration, increases in pitch angle, angle of attack, and vertical acceleration were observed in the DFDR and DAR data during the initial event, indicating a nose-up motion, whereas the stabilizer position was commanding a nose-down motion. The Airbus flight dynamics simulator was used to investigate the magnitude of longitudinal and vertical forces required to produce this motion.

The forces were simulated through changes to the pitching moment coefficient, lift coefficient, and drag coefficient. The combination of these changes that best matched the DFDR and DAR data indicated that a downward vertical force of approximately 36 000 N at 28 m aft of the CG and a rearward horizontal force of approximately 35 000 N at approximately 9.5 m above the CG were applied to the aircraft approximately one second after the application of the lateral force described above or at 0701:55.

1.12.13.5 Additional Simulations

Additional simulations were conducted to better understand the results of the initial simulations. These additional simulations covered the time period from 0701:50 to 0702:15 and included the time of the excessive rudder deflection. Four simulations were performed with different rudder breakup scenarios. The simulation that best matched the DFDR and DAR data was 76 per cent rudder loss at 0701:28, increasing to 80 per cent rudder loss at 0701:32.5 to 0701:33, and increasing again to 84 per cent rudder loss at 0701:34.5 to 0701:35. An important result of these simulations was that the rudder movement recorded on the DFDR and DAR was made by a part of the rudder that was not aerodynamically effective.

Combining the results from these simulations with the investigation of the lateral, longitudinal, and vertical forces suggests the following scenario:

• At 0701:54, a lateral force of approximately 108 000 N was applied to the aircraft.
• One second later, at 0701:55, a horizontal force of 35 000 N was applied 9.5 m above the CG, and a vertical force of 36 000 N was applied 28 m behind the CG.
• The application of the horizontal and vertical force coincided with an initial rudder loss of 76 per cent.
• Over a period of approximately seven seconds after the initial rudder loss, the rudder continued to break up, with 80 per cent rudder loss occurring between 0701:59.5 and 0702:00, and 84 per cent rudder loss occurring at 0702:02.
• The results also suggest that the excessive rudder deflection recorded on the DFDR and DAR at 0702:00 was made by a part of the rudder that was not aerodynamically effective.

1.12.13.6 Yaw Damper Modelling

After the initial event, the DFDR and DAR data indicated that the rudder was moving in a sinusoidal pattern, which frequency analysis showed to be similar to the frequency of the lateral acceleration. This would indicate that the movement of the rudder was connected with the aircraft movement. One theory proposed for the rudder movement was the operation of the YD. To confirm this theory, the YD system was modelled after information on YD mechanization provided by Airbus.

When the YD model output was compared with DFDR and DAR rudder movement data, corrected to remove a rudder motion sensor bias and a time lag introduced by the system data acquisition concentrator, it showed that the rudder movement closely matched the predicted YD output. The only significant deviation occurred approximately five seconds after the initial event, where the DFDR and DAR data showed a large rudder motion that exceeded the mechanical limits of the rudder. This motion was attributed to the loss of rigidity in the rudder due to the breakup.

1.12.14 Effect of Disbond Bubble on Static Aerodynamic Loads

A rudder side panel disbond or in-plane core fracture, under the influence of aerodynamic loads, could result in the affected area bulging outwards with a bubble-like appearance. An analysis was conducted to determine the effect of such a bubble on static aerodynamic loads. The analysis was conducted using computational fluid dynamics based on geometrical information from the finite element analysis model. The analysis was repeated for rudder deflection angles of 1º and 6º, under the flight conditions that existed at the time of the occurrence. At 6º deflection, the disbond bubble caused a 1 per cent increase in rudder force and a 2.4 per cent increase in total VTP force. At 0º rudder deflection, which was the situation at the time of the occurrence, the disbond bubble had a negligible effect on rudder and VTP loads. Therefore, the presence of a disbond bubble would have a comparatively small effect on static aerodynamic loads.

1.12.15 Flutter Analysis

1.12.15.1 Analysis of a Dynamic Event

The lateral load traces obtained from the DFDR and DAR were coincident except for a two-second period at the beginning of the occurrence. This suggested the possibility of a dynamic event. Since the sampling rate was only 4 Hz, it was not possible to determine the amplitude and frequency of the dynamic signal. Therefore, the signal was analyzed using a simplistic manual experimental curve-fitting approach. Using this method of analysis, it was not possible to find a unique solution.

However, it was possible to fit the DFDR and DAR data by assuming both a high- and low-frequency component, a loss in damping, and by assuming that the first part of the signal was divergent and the second part was stable. Since there was a dynamic event, and flutter is a dynamic phenomenon, a flutter analysis was conducted to determine whether it was possible to find a scenario involving flutter before or after the rudder rupture, consistent with the available data.

1.12.15.2 Method of Analysis

A theoretical flutter analysis was conducted to study the effects of various failure scenarios on the aircraft's flutter characteristics. The analysis was done using a complete aircraft model, flutter computations with 70 modes, and 1 per cent structural damping. This same method of flutter analysis was used for the original certification, and at the time, was substantiated both by GVT and flight testing. The doublet lattice analysis method that was used does have limitations that may affect the accuracy of predictions in cases such as a disbonded skin panel on one side panel only. A preliminary investigation of advanced techniques was conducted, but the state of development of these approaches was not sufficiently advanced to be useful in the analysis.

1.12.15.3 Baseline Analysis

The baseline case represented an aircraft with no structural damage. The flutter analysis found that the basic coupling mechanism involved a 6.76 Hz fin bending mode and a 13.18 Hz rudder rotation mode. There were sufficient damping margins at both the occurrence speed of 270 KCAS and the design dive speed of 406.5 KCAS.

1.12.15.4 Failure Scenarios

Flutter analyses were conducted to study the effect of the following structural failure scenarios:

• Disbond - The scenarios involved varying degrees of disbond between the honeycomb core and the inner face sheet, as shown in Figure 7. Single-sided disbonds were studied on the left panel, the right panel, and both panels. Simulating a disbond between the core and the face sheet was accomplished in the analysis by reducing the stiffness of the face sheet to 0 or 20 per cent of its original value in the affected area. The analysis found that there was an increase in flutter tendency with greater disbonds, and that double-sided disbonds were more critical than single-sided ones. There was no significant difference between left-side and right-side disbonds. No flutter was observed when stiffness was reduced to 20 per cent. When it was reduced to 0 per cent, the double-sided 2.9 m² disbond scenario and the single-sided 5.6 m² disbond scenario resulted in violent flutter below 150 KCAS. A more detailed examination of the single-sided 5.6 m² disbond scenario, within the range between 0 and 20 per cent stiffness, found that the flutter speed decreased as the stiffness was reduced, and that this decrease was steady and continuous down to about 5 per cent stiffness where flutter started. Below 5 per cent stiffness, the frequency and damping decreased rapidly leading to flutter conditions.
• Free Rudder - This scenario simulated a chordwise fracture of the rudder just above the hydraulic actuators, resulting in a free upper rudder. This model demonstrated flutter of several modes below 270 KCAS, the speed at the time of the occurrence.
• Trailing Edge Screws Removed - This scenario simulated the removal of all the trailing edge screws. The analysis found that the flutter behaviour of this model was nearly unchanged relative to the baseline scenario. The reason for this was that the trailing edge connection was still intact so the torsion cell was still closed.
• Trailing Edge Connection Removed - This scenario simulated the removal, to varying degrees, of the trailing edge connection. The analysis found that the tendency to flutter increased with the increasing size of the damage, and that flutter within the flight envelope was possible with the larger damage areas.
• Fractured Front Spar - This scenario assumed that the rudder front spar was fractured above the hydraulic actuators. The analysis found that there was only a slight deterioration in the rudder-rotation/fin-bending flutter coupling relative to the baseline scenario.
• Loss of Rudder Below Hinge 2 - This scenario assumed that the section of rudder below hinge point 2 was lost. The analysis found that the rudder rotation frequency increased significantly due to the missing rudder mass, but there was no coupling.
• Upper-End Disbond - This scenario assumed a large one-sided disbond at the upper end of the rudder. Twenty percent stiffness was assumed in the disbonded area. The analysis found that there was only a small change in damping relative to the baseline scenario.
• Hinge Failures - Five scenarios were evaluated involving the following failures: failure of hinge 1; failure of hinge 7; failure of hinges 1 and 7; failure of hinges 6 and 7; and failure of hinges 5, 6, and 7. The analysis found that scenarios for failures of hinge 7, and failures of hinges 1 and 7 did lead to flutter, but this occurred beyond 400 KCAS, outside the design envelope. The remainder of the scenarios did not show any flutter-critical couplings.
• Hinge Stiffness - Three scenarios were evaluated involving the following failures: 75 per cent of nominal stiffness, all 7 hinges; 50 per cent of nominal stiffness, all 7 hinges; and 25 per cent of nominal stiffness, all 7 hinges. The analysis results showed that none of these scenarios presented any significant deterioration in coupling of the fin bending and rudder rotation modes compared to the baseline scenario.
• Hydraulic Actuator Stiffness - Three scenarios were evaluated involving the following failures: 75 per cent of nominal stiffness, all 3 hydraulic actuators; 50 per cent of nominal stiffness, all 3 hydraulic actuators; and 25 per cent of nominal stiffness, all 3 hydraulic actuators. The analysis results showed that decreasing the hydraulic actuator stiffness increased the tendency to flutter. At 25 per cent stiffness, a flutter speed of 268 KCAS was calculated.
• Single Hydraulic Actuator Disconnection - Three scenarios were evaluated involving the following failures: disconnection of actuator at hinge position 2; disconnection of actuator at hinge position 3; and disconnection of actuator at hinge position 4. The analysis found no flutter with the disconnection of one actuator. Further analysis found that, when simulating a double hydraulic failure, the reduction of rudder rotation mode frequency was only 0.33 Hz and did not lead to flutter.
• Actuator Stiffness versus Actuator Disconnection - A comparative analysis of the two previous scenarios found that disconnecting one hydraulic actuator has the same effect as all three actuators having 66 per cent of their nominal stiffness.
• VTP Attachment Failure - A flutter analysis was conducted to determine the effect of failed VTP main attachment fittings on flutter. The analysis found that the disconnection of one or both of the VTP rear attachments did not provide a flutter-critical coupling.
• Extra Rudder Mass - This design of rudder is not mass-balanced. Any condition that adds mass to the rudder and moves its CG aft, further away from the hinge line, has an adverse effect on the flutter margin. Excessive paint layers or fluid ingress are examples of conditions that could cause such an effect.
  • Extra Mass at Trailing Edge - The analysis simulated extra rudder mass by increasing the mass of the trailing edge screws. This analysis found that 142 kg distributed at the trailing edge was necessary to produce flutter at the occurrence speed of 270 knots. The analysis also found that, with 71 kg of extra mass, there was adequate damping at the occurrence conditions.
  • Extra Paint - The analysis simulated extra mass distributed evenly over the surface of the rudder. This analysis found that, at the occurrence speed of 270 knots, the addition of mass did not significantly reduce the damping. At high speeds, the addition of the initial 19.3 kg of extra paint significantly reduced the damping.
  • Pooled Fluid in Rudder - A flutter analysis was conducted to determine the effect of pooled fluid in the rudder leading edge as was found in the rudder of aircraft MSN 701. It was found that, since this extra mass is concentrated so close to the hinge line, it had a negligible effect on flutter.

1.12.15.5 Summary of Flutter Analyses

The analyses found that flutter could occur within the certification envelope for three damage scenarios:

• free rudder or free rudder section;
• significant decrease in hydraulic actuator stiffness; and
• significant reduction in rudder torsion stiffness by extensive disbond or failure of trailing edge connection.

In all these scenarios, the necessary damage was significant and exceeded certification requirements.

1.12.16 Time-Domain Flutter Analysis

1.12.16.1 General

The flutter analysis results described above were presented graphically by frequency and damping curves versus speed. In order to correlate this flutter analysis with the results of the flight dynamics analysis and the recorder data, which were presented as curves versus time, the following time-domain flutter analysis was conducted to observe the flutter amplitude versus time.

1.12.16.2 Method of Analysis

A time-domain flutter analysis was conducted for two of the failure scenarios: single-sided 5.6 m² disbond with less than 5 per cent stiffness of the original panel and free upper rudder. The aim was to produce time histories and correlate the resulting loads. The analysis was conducted in time steps of 0.001 second. The structure was excited with a lateral force of 1 kilonewton (kN) at the VTP tip acting as a step pulse starting at time step 100 with a 0.1-second duration.

The following variables were examined:

• rudder deflection (top, middle, bottom);
• loads at all seven hinges;
• forces at all three actuators;
• loads at VTP attachments (front, middle, rear); and
• lateral accelerations (cockpit, CG/DFDR, VTP attachment, VTP tip, rudder top and bottom)

1.12.16.3 Results

Upon excitation, the amplitudes of all the responses started to grow. Figure 8 shows a typical load response at the left rear VTP attachment. At design ultimate load, the maximum force acting at the aft VTP attachment is approximately 700 kN, and based on the attachment fittings damage, it is known that loads during the occurrence did exceed ultimate strength. Therefore, the time step where the load peaked above 700 kN was selected as a reference point for each scenario. All the remaining monitored variables were measured within an envelope around the reference time step, since all the maxima did not occur at the same time. Their peak value within this envelope was recorded. Both investigated scenarios showed no contradictions with respect to restraints like VTP attachment rupture load and maximum rudder rotation. The free upper rudder scenario did not exhibit any significant hinge loads in the fore/aft direction, and the VTP attachments would have failed before the hinges. Since this contradicts the observed damage, this scenario is excluded as a realistic damage scenario.

The single-sided 5.6 m² disbond with less than 5 per cent stiffness of the original panel scenario did exhibit significant hinge loads in the fore/aft direction. The time-domain flutter analysis determined that hinge 5 would be the first to fail and that it would fail before the VTP main attachment fittings were damaged. A second time-domain flutter analysis was conducted to study this scenario with a disconnected hinge 5. That analysis found that, after the failure of hinge 5, the loads of hinge 6 exceed allowable loads while VTP rear attachments reached a level above ultimate but below rupture.

1.12.17 Summary of Flutter and Time-Domain Flutter Analyses

Flutter and time-domain flutter analyses were conducted for a number of failure scenarios. One scenario, the single-sided 5.6 m² disbond with less than 5 per cent stiffness of the original panel, provided a credible response and showed good correlation with the observed damage, the recorder data, and the flight dynamics analysis.

1.12.18 National Transportation Safety Board Radar Data Analysis Study

The NTSB conducted a trajectory analysis of the Florida Keys (KEY), Melbourne (MEL), Tamiami (TMA), and Cudjoe Key (CUD) air route surveillance radar data to identify the numerous tracks in the vicinity of the track of TSC961 at the approximate time the rudder separated from the aircraft. The aim was to distinguish larger pieces of debris from smaller shards and examine relative timing of events. The caveat was that radar has many variables that cannot always be reverse-engineered. No discernable track of an initial separation or a sustained track that would resemble a large piece of aeroplane part was evident. The tracks of the returns resembled numerous small pieces floating with the prevailing winds.

1.12.19 Other Aircraft in Vicinity

ATC records indicated that there were no other aircraft with an instrument flight rules discrete transponder code at a similar altitude within 60 nautical miles (nm) of the occurrence aircraft at the time of the event. The nearest aircraft was at FL 320 feet approximately 70 nm to the east. All other aircraft were considerably lower and further away.

1.12.20 Space Objects

Military authorities reported that, at the time and place of the occurrence, records did not indicate any man-made objects re-entering the atmosphere. Records were unavailable for natural objects re-entering the atmosphere.

Previous | Next