Aircraft Accident Report

EgyptAir Flight 990
Boeing 767-366ER, SU-GAP
Nantucket, MA
October 31, 1999

NTSB/AAB-02/01
PDF


*Please note - the following items are not included here, but are available in the PDF and printed copies of the document:
Other items are available from the public docket, including 
Report of Investigation of Accident submitted by Egyptian Civil Aviation Authority


 
Accident Number.: 
DCA00MA006 
Operator/Flight Number:
EgyptAir flight 990 
Aircraft and Registration: 
Boeing 767-366ER, SU-GAP 
Location:  60 miles south of Nantucket, Massachusetts 
Date:
October 31, 1999 
Adopted On:
March 13, 2002

FACTUAL

On October 31, 1999, about 0152 eastern standard time (EST), EgyptAir flight 990, a Boeing 767-366ER (767), SU-GAP, crashed into the Atlantic Ocean about 60 miles south of Nantucket, Massachusetts. EgyptAir flight 990 was being operated under the provisions of Egyptian Civil Aviation Regulations (ECAR) Part 121 and U.S. 14 Code of Federal Regulations Part 129 as a scheduled, international flight from John F. Kennedy International Airport (JFK), New York, New York, to Cairo International Airport, Cairo, Egypt.1 The flight departed JFK about 0120, with 4 flight crewmembers, 10 flight attendants, and 203 passengers on board. All 217 people on board were killed, and the airplane was destroyed. Visual meteorological conditions prevailed for the flight, which operated on an instrument flight rules (IFR) flight plan.

HISTORY OF FLIGHT

On October 30, 1999, the accident airplane departed Los Angeles International Airport (LAX), Los Angeles, California, as EgyptAir flight 990, destined for Cairo, with a scheduled intermediate stop at JFK. EgyptAir flight 990 landed at JFK about 2348 eastern daylight time (EDT)2 and arrived at the gate about 0010 EDT on October 31, 1999.

Because of the 10-hour scheduled en route flight time from JFK to Cairo, ECAR Part 121, Subpart Q, required that the accident flight have two designated flight crews (each crew consisting of a captain and first officer). According to the EgyptAir flight dispatcher who accompanied the two accident flight crews from their hotel in New York City to the airport, they departed the hotel about 2330 EDT on October 30 and arrived at JFK about 40 minutes later, about the same time as the airplane, inbound from LAX, arrived at the terminal gate.

According to air traffic control (ATC) records, by 0101, the pilots of EgyptAir flight 990 had requested, received, and correctly read back an IFR clearance from ATC. ATC transcripts further indicated that between about 0112 and 0116, air traffic controllers issued a series of taxi instructions to EgyptAir flight 990. At 0117:56, the pilots advised the local controller that they were holding short of the departure runway (runway 22 right [22R]) and that they were ready for takeoff. The local controller instructed EgyptAir flight 990 to taxi into position and hold on runway 22R and, at 0119:22, cleared the accident flight for takeoff. The first officer acknowledged the takeoff clearance, and, about 0120, the airplane lifted off runway 22R.

Shortly after liftoff, the pilots of EgyptAir flight 990 contacted New York Terminal Radar Approach (and departure) Control (TRACON). New York TRACON issued a series of climb instructions and, at 0126:04, instructed the flight to climb to flight level (FL) 2303 and contact New York Air Route Traffic Control Center (ARTCC). According to ATC and cockpit voice recorder (CVR) records, at 0135:52, New York ARTCC instructed EgyptAir flight 990 to climb to FL 330 and proceed directly to DOVEY intersection.4

According to the CVR transcript,5 about 0140 (20 minutes after takeoff), as the airplane was climbing to its assigned altitude, the relief first officer suggested that he relieve the command first officer at the controls,6 stating, "I'm not going to sleep at all. I might come and sit for two hours, and then...," indicating that he wanted to fly his portion of the trip at that time. The command first officer stated, "But I...I slept. I slept," and the relief first officer stated, "You mean you're not going to get up? You will get up, go and get some rest and come back." The command first officer then stated, "You should have told me, you should have told me this, Captain [relief first officer's surname].7 You should have said, '[command first officer's first name]...I will work first.' Just leave me a message. Now I am going to sit beside you. I mean, now, I'll sit by you on the seat. I am not sleepy. Take your time sleeping and when you wake up, whenever you wake up, come back, Captain."

The relief first officer then stated, "I'll come either way...come work the last few hours, and that's all." The command first officer responded, "No...that's not the point, it's not like that, if you want to sit here, there's no problem." The relief first officer stated, "I'll come back to you, I mean, I will eat and come back, all right?" The command first officer responded, "Fine, look here, sir. Why don't you come so that...you want them to bring your dinner here, and I'll go to sleep [in the cabin]?" The relief first officer stated, "That's good." The command first officer then stated to the command captain, "With your permission, Captain?"

At 0140:56, the CVR recorded the sound of the cockpit door operating. About 1 second later, the command first officer stated in a soft voice, "Do you see how he does whatever he pleases?" At 0141:09, the command first officer stated, "No, he does whatever he pleases. Some days he doesn't work at all." At 0141:51, the CVR again recorded the sound of the cockpit door operating. Sounds recorded during the next minute by the CVR (including a whirring sound similar to an electric seat motor operating, a clicking sound similar to a seat belt operating, and some conversation) indicated that the command first officer vacated and the relief first officer moved into the first officer's seat.

Flight data recorder (FDR) and radar data indicated that the airplane leveled at its assigned altitude of FL 330 at 0144:27. At 0147:19, New York ARTCC instructed EgyptAir flight 990 to change radio frequencies for better communication coverage. The command captain of EgyptAir flight 990 acknowledged and reported on the new frequency at 0147:39.8

At 0147:55, the relief first officer stated, "Look, here's the new first officer's pen. Give it to him please. God spare you,"9 and, at 0147:58, someone responded, "yeah." At 0148:03, the command captain stated, "Excuse me, [nickname for relief first officer], while I take a quick trip to the toilet...before it gets crowded. While they are eating, and I'll be back to you." While the command captain was speaking, the relief first officer responded, "Go ahead please," and the CVR recorded the sound of an electric seat motor as the captain maneuvered to leave his seat and the cockpit. At 0148:18.55, the CVR recorded a sound similar to the cockpit door operating.

At 0148:30, about 11 seconds after the captain left the cockpit, the CVR recorded an unintelligible comment.10 Ten seconds later (about 0148:40), the relief first officer stated quietly, "I rely on God."11 There were no sounds or events recorded by the flight recorders that would indicate that an airplane anomaly or other unusual circumstance preceded the relief first officer's statement, "I rely on God."

At 0149:18, the CVR recorded the sound of an electric seat motor. FDR data indicated that, at 0149:45 (27 seconds later), the autopilot was disconnected.12 Aside from the very slight movement of both elevators (the left elevator moved from about a 0.7° to about a 0.5° nose-up deflection, and the right elevator moved from about a 0.35° nose-up to about a 0.3° nose-down deflection)13 and the airplane's corresponding slight nose-down pitch change, which were recorded within the first second after autopilot disconnect, and a very slow (0.5° per second) left roll rate, the airplane remained essentially in level flight about FL 330 for about 8 seconds after the autopilot was disconnected. At 0149:48, the relief first officer again stated quietly, "I rely on God." At 0149:53, the throttle levers were moved from their cruise power setting to idle, and, at 0149:54, the FDR recorded an abrupt nose-down elevator movement and a very slight movement of the inboard ailerons. Subsequently, the airplane began to rapidly pitch nose down and descend.

Between 0149:57 and 0150:05, the relief first officer quietly repeated, "I rely on God," seven additional times.14 During this time, as a result of the nose-down elevator movement, the airplane's load factor15 decreased from about 1 to about 0.2 G.16 Between 0150:04 and 0150:05 (about 10 to 11 seconds after the initial nose-down movement of the elevators), the FDR recorded additional, slightly larger inboard aileron movements, and the elevators started moving further in the nose-down direction. Immediately after the FDR recorded the increased nose-down elevator movement, the CVR recorded the sounds of the captain asking loudly (beginning at 0150:06), "What's happening? What's happening?," as he returned to the cockpit.

The airplane's load factor decreased further as a result of the increased nose-down elevator deflection, reaching negative G loads (about -0.2 G) between 0150:06 and 0150:07. During this time (and while the captain was still speaking [at 0150:07]), the relief first officer stated for the tenth time, "I rely on God." Additionally, the CVR transcript indicated that beginning at 0150:07, the CVR recorded the "sound of numerous thumps and clinks," which continued for about 15 seconds.

According to the CVR and FDR data, at 0150:08, as the airplane exceeded its maximum operating airspeed (0.86 Mach), a master warning alarm began to sound. (The warning continued until the FDR and CVR stopped recording at 0150:36.64 and 0150:38.47, respectively.)17 Also at 0150:08, the relief first officer stated quietly for the eleventh and final time, "I rely on God," and the captain repeated his question, "What's happening?" At 0150:15, the captain again asked, "What's happening, [relief first officer's first name]? What's happening?" At this time, as the airplane was descending through about 27,300 feet msl, the FDR recorded both elevator surfaces beginning to move in the nose-up direction. Shortly thereafter, the airplane's rate of descent began to decrease.18 At 0150:21, about 6 seconds after the airplane's rate of descent began to decrease, the left and right elevator surfaces began to move in opposite directions; the left surface continued to move in the nose-up direction, and the right surface reversed its motion and moved in the nose-down direction.

The FDR data indicated that the engine start lever switches for both engines moved from the run to the cutoff position between 0150:21 and 0150:23.19 Between 0150:24 and 0150:27, the throttle levers moved from their idle position to full throttle, the speedbrake handle moved to its fully deployed position, and the left elevator surface moved from a 3º nose-up to a 1º nose-up position, then back to a 3º nose-up position.20 During this time, the CVR recorded the captain asking, "What is this? What is this? Did you shut the engine(s)?" Also, at 0150:26.55, the captain stated, "Get away in the engines,"21 and, at 0150:28.85, the captain stated, "shut the engines." At 0150:29.66, the relief first officer stated, "It's shut."

Between 0150:31 and 0150:37, the captain repeatedly stated, "Pull with me." However, the FDR data indicated that the elevator surfaces remained in a split condition (with the left surface commanding nose up and the right surface commanding nose down) until the FDR and CVR stopped recording at 0150:36.64 and 0150:38.47, respectively. (The last transponder [secondary radar] return from the accident airplane was received at the radar site at Nantucket, Massachusetts, at 0150:34.)22

Information about the remainder of the flight came from the airplane's two debris fields and recorded primary radar data from long-range radar sites at Riverhead, New York, and North Truro, Massachusetts, and the short-range radar site at Nantucket. The height estimates based on primary radar data from the joint use FAA/U.S. Air Force (USAF) radar sites indicated that the airplane's descent stopped about 0150:38 and that the airplane subsequently climbed to about 25,000 feet msl and changed heading from 80º to 140º before it started a second descent, which continued until the airplane impacted the ocean.

Airplane wreckage was located in two debris fields, about 1,200 feet apart, centered at 40° 21' north latitude and 69° 46' west longitude. The accident occurred at night in dark lighting conditions.

PERSONNEL INFORMATION

The Safety Board reviewed the command and relief flight crew's flight and duty times and found no evidence that they were outside the limits established by applicable regulations. Because the command captain and the relief first officer were identified as being the only two crewmembers in the cockpit during the accident sequence, information on only these two crewmembers is included in this section.23 The cabin crew comprised 10 flight attendants. In addition, several nonduty EgyptAir flight crewmembers were on board the accident airplane.

Command Captain Information

The command captain, age 57, was hired by United Arab Airlines24 on July 13, 1963. He held an Egyptian airline transport pilot certificate with Boeing 707, 737-200, and 767-200 and -300 type ratings. The command captain's most recent medical certificate was issued on October 21, 1999, and he was found to be medically fit to fly with glasses in accordance with the standards specified in ECAR Part 67, "Medical Standards and Certification." According to his family, the command captain had suffered from chronic back problems but was addressing them and had no recent changes in his health.25

The command captain's most recent proficiency check was satisfactorily completed on March 9, 1999, and his most recent recurrent training was satisfactorily completed on August 14, 1999. According to EgyptAir records, at the time of the accident, the command captain had flown approximately 14,384 total flight hours, including 6,356 hours in the 767. The Safety Board's review of EgyptAir training records for the command captain indicated that he had accomplished all required checkrides and satisfactorily performed all required maneuvers.

The command captain arrived in New York the afternoon of October 28, 1999, after serving as a captain on EgyptAir flight 989 from Cairo to JFK. (Additional information about the command captain is contained in the public docket on this accident.)

Relief First Officer Information

The relief first officer, age 59, was hired by EgyptAir on September 8, 1987. He held an Egyptian commercial pilot certificate with 737-200 and 767-200 and -300 type ratings.26 The relief first officer's most recent medical certificate was issued on July 28, 1999, and he was found to be medically fit to fly with glasses in accordance with the standards specified in ECAR Part 67. According to a close friend, the relief first officer had no family history of major medical difficulties and did not complain of headaches, indigestion, or other medical problems before the accident.

The relief first officer's most recent proficiency check was satisfactorily completed on June 19, 1999, and his most recent recurrent training was satisfactorily completed on December 19, 1998. According to EgyptAir records, at the time of the accident, the relief first officer had flown approximately 12,538 total flight hours, including 5,191 hours in the 767. The Safety Board's review of EgyptAir training records for the relief first officer indicated that he had accomplished all required checkrides and performed all required maneuvers.

Before EgyptAir hired him, the relief first officer was a flight instructor, first for the Egyptian Air Force and later for a Government-operated civilian flight training institute in Egypt. The relief first officer became a Major in the Air Force before he transitioned to the flight training institute, where he eventually became the chief flight instructor.

The relief first officer arrived in New York City the afternoon of October 28, 1999, after serving as a first officer on EgyptAir flight 990 from LAX to JFK. (Additional information about the relief first officer is contained in the public docket on this accident.)

AIRPLANE INFORMATION

The accident airplane, SU-GAP, a 767-300 series airplane27 (model 767-366ER [extended range]), serial number (S/N) 24542, was manufactured by Boeing and delivered new to EgyptAir on September 26, 1989. According to EgyptAir records, it had 33,354 total hours of operation (7,594 flight cycles)28 at the time of the accident. It was configured to seat a maximum of 10 first-class, 22 business-class, and 185 economy-class passengers and to carry cargo.

The accident airplane was equipped with two P&W 4060 turbofan engines. Company maintenance records indicated that the No. 1 (left) engine, S/N 724126, was installed on the accident airplane on April 19, 1998, and had operated about 25,708 hours since new and that the No. 2 (right) engine, S/N 724127, was installed on the accident airplane on June 3, 1998, and had operated about 19,316 hours since new.

767 Longitudinal Control System Information

Because the accident sequence involved a sustained unusual motion about the airplane's pitch axis, the Safety Board examined the 767's longitudinal flight control system. According to the Boeing 767 Maintenance Manual, the 767's longitudinal flight control system includes two (left and right) sets of linked elevator surfaces (inboard and outboard), which are attached to the rear spar of the movable horizontal stabilizer by hinges. Each outboard elevator surface is driven by three power control actuators (PCA). Because the outboard and inboard surfaces are linked, the inboard elevator surfaces move when the outboard elevator surfaces are driven. Hydraulic power for elevator PCA movement is provided by the 767's three independent hydraulic systems--each hydraulic system powers one of each elevator surface's PCAs, which provides redundancy within the elevator control system. (Components in the elevator control system are shown in figures 1a and 1b.)

Two parallel sets (one operated from the captain's side, the other from the first officer's side) of flight control components move the elevator surfaces. Control column inputs made at the captain's position are linked directly to the actuators for the left elevator surface, whereas control column inputs made at the first officer's position are linked directly to the actuators for the right elevator surfaces. The two parallel sets of flight control components are linked together at the forward and aft override mechanisms/linkages and slave cable interconnects. Flight control commands from the captain's and first officer's control columns are transmitted through linkages and cables29 from the front of the airplane to the left and right aft quadrant assemblies, respectively. The aft quadrant assemblies then translate the inputs to the respective bellcrank assemblies and the input control rods for each of the three elevator PCAs for each outboard elevator surface.

After control cable movement is translated to input control rod movement, the control rods move control valves inside the PCAs, allowing high-pressure hydraulic fluid to flow to one side or the other of the actuators' pistons (depending on the direction of the input), resulting in elevator movements that correspond to the direction of the input. When the elevators reach the commanded position, feedback linkages move the control valves to a position in which the hydraulic fluid is blocked off, resulting in no further movement of the actuator piston or elevator.

Testing, evaluation, and analysis of the 767 elevator system showed that any movement of the control columns (whether pilot-induced or not) would have resulted in concurrent, identifiable movements of the elevators, which would have been recorded on the FDR.

An elevator feel-and-centering unit transmits hydraulic and mechanical feel forces to hold the elevator at the neutral (trimmed) position when no control column force is applied. It also provides feedback (or feel) force to the control column that increases as the control column is moved forward or aft. The feel forces provided are essentially equal at both pilot positions because of the connections between the left and right elevator systems.

The captain's and first officer's control columns have authority to command full travel of the elevators under most flight conditions and normally work together as one system. However, the two sides of the system can be commanded independently because of override mechanisms at the control columns and aft quadrant. Therefore, if one side of the system becomes immobilized, control column inputs on the operational side can cause full travel of the nonfailed elevator. In addition, in many cases, control column inputs on the operational side can also result in nearly full travel of the elevator on the failed side through the override mechanisms. The elevator PCAs are installed with compressible links located between each bellcrank assembly and PCA input control rod to provide a means of isolating a jammed PCA, thus allowing the pilots to retain control of that elevator surface through its two remaining (unjammed) PCAs.

767 Elevator Blowdown Information

During ground operations, the 767 elevator PCAs can drive the elevators through a range of motion from 28.5° in the nose-up direction to 20.5° in the nose-down direction. However, in-flight elevator deflections can be limited by the aerodynamic forces acting on the elevator. The maximum position to which the elevator can move is that which balances the aerodynamic forces that are acting on the elevator surfaces against the force produced by the elevator PCAs and is referred to as its "blowdown" position. Thus, as the airplane's airspeed increases (increasing the aerodynamic forces acting against the elevator PCAs), the elevators' range of motion is increasingly limited.

The maximum output force produced by the elevator PCAs is generated by the hydraulic system pressure acting on the PCAs' piston area; if all three elevator PCAs are working properly, the total output force for each elevator surface is the sum of the forces produced by all three of that elevator's PCAs. When a dual elevator PCA failure occurs,30 the forces produced by the two failed PCAs would overpower the opposing force produced by the one nonfailed PCA. The resultant initial force on the elevator surface in the failed direction would be equivalent to a single functioning PCA operating at 100 percent of its maximum force. The failed PCAs would resist the backdriving force31 with a force equivalent to about 130 percent of a single functioning PCA. The high internal pressures required for activation of the PCAs' pressure relief valves allow the PCAs' pistons to resist the aerodynamic backdriving movement with more force than normal operating pressures would allow. Therefore, if a dual PCA failure occurred in flight, the elevator would initially move to a position consistent with a single functioning PCA operating at 100 percent of its maximum force, balanced against the aerodynamic forces affecting the elevator surface. As the airspeed increases, the failed elevator surface would remain at this initial position until the backdriving forces exceeded those of a single PCA operating at 130 percent of its normal capability, at which point the deflection of the failed elevator surface would decrease.32 (Figure 2 is a comparison of the elevator positions recorded by the accident airplane's FDR with failed and nonfailed elevator positions following a dual PCA failure.)

767 Autoflight Systems Information

The 767 autoflight systems include the autopilot/flight director, yaw damper, automatic stabilizer trim, Mach trim, maintenance monitoring, instrument landing system deviation monitor, and thrust management systems. The thrust management system includes autothrottle control.

767 Autopilot Information

The 767 autopilot/flight director system consists (in part) of three separate autopilot systems that can be used singly or in combination to provide automatic control of the ailerons, elevator, stabilizer, and rudder control systems when operating in selected flight modes. Any one of the three autopilot systems can control the airplane in the normal climb, cruise, descent, and approach modes.

The 767 autopilot system controls the airplane's movement about the pitch axis by using the elevators for dynamic control of the airplane's pitch and the horizontal stabilizer to trim out steady-state elevator deflections. When the autopilot is engaged and the airplane is in a steady-state flight condition, the autopilot is designed to keep the elevators near their neutral (or faired) position, using the elevators primarily for short-term dynamic adjustments (such as those necessitated by atmospheric disturbances). The elevators are also used for small trim adjustments, such as those necessitated by fuel consumption during flight. As these small elevator adjustments accumulate over time, the elevator deflections move further from their neutral (or faired) position. When the elevators' deflections reach a threshold value, the autopilot "retrims" the horizontal stabilizer and the elevator returns to a neutral (or faired) position. According to Boeing, when the autopilot system is disconnected, the force applied by the autopilot actuator to the elevator control system is removed, and, if the horizontal stabilizer has not been adjusted recently, small elevator movements result. Boeing representatives indicated that the following circumstances could result in elevator movements at the time of autopilot disconnect:

  • Differences between the neutral position recognized by the autopilot and the actual neutral position of the elevator feel-and-centering unit would result in the autopilot actuator holding a force that would be released when the system is disconnected.
  • The autopilot may have moved the elevators since it last trimmed the stabilizer, placing the elevators at a position other than their neutral position at the time of disconnect. When the autopilot is disconnected, the elevators would return to the neutral position commanded by the feel-and-centering unit. (During steady-state flight conditions, this situation occurs because of the effect of fuel consumption on the airplane's center of gravity.) According to Boeing, "this type of elevator motion upon autopilot disconnect is inherent in the operation of the autopilot system."
  • Pilot forces on the control column at the time of manual autopilot disconnect can affect the movement of the elevator. (The autopilot can be disconnected manually by double-clicking the control yoke-mounted autopilot disconnect switch.)
  • Mechanical aspects of the elevator control system (including friction, the effects of compliance in the system,33 variations among individual autopilot actuator units, and variations in the centering detent force) can cause elevator movement at the time of autopilot disconnect.
Boeing's 767 Maintenance Manual indicates that if the autopilot disconnects because of a system failure, the following cockpit warnings and annunciations would occur:
  • the red autopilot disconnect warning light illuminates,
  • the red master warning light illuminates,
  • the engine indication and crew alerting system computer displays an autopilot disconnect message, and
  • a siren alert sounds.
Although these autopilot disconnect warnings and annunciations are also generated when the autopilot is disconnected by pressing the autopilot manual disconnect switch on the control wheel, pressing the manual disconnect switch a second time within 0.5 second resets, and thus cancels, the system's disconnect warnings and annunciations before they are displayed to the flight crew. The 767 autopilot warnings and annunciations system contains multiple redundancies. For example, two warning signals are generated for each of the warning functions listed above: one warning signal uses software logic that is powered by normal power (which would be inhibited by a loss of normal power or a computer failure), and the other uses hardware logic that is powered by 28-volt alternating current standby power.

767 Autothrottle Information

The 767's thrust management system provides autothrottle control based on selected modes, existing conditions, and engine limitations. The autothrottle can be operated independently of or with the autopilot system. The autothrottle servomotor generator is connected to the throttle levers through a clutch pack assembly, which, when overridden,34 allows the pilots to make manual thrust inputs when the autothrottle is engaged. Movement of the throttle levers aft of the autothrottle commanded position for a given flight condition would require a manual force of about 9 lbs at the throttle levers to override the autothrottle servomotor clutch.

When the autothrottle function is engaged, it controls throttle lever movement. The maximum autothrottle commanded throttle lever movement rate for a normally functioning autothrottle system is 10.5° per second. Manual throttle lever inputs can exceed this rate; for example, the accident airplane's FDR recorded throttle lever movement at a rate of 25° per second at the beginning of the accident sequence. The minimum throttle lever position that the autothrottle can command varies as a function of the airplane's speed and the autothrottle mode selected. For the accident airplane's flight conditions and the selected autothrottle mode at the beginning of the accident sequence, this position would have been 40° to 50°. The FDR recorded a throttle lever position of about 33° at the beginning of the accident sequence.

Reported Autopilot Anomalies in the Accident Airplane

During interviews conducted at the request of the Egyptian Government on February 21, 2001, an EgyptAir captain who had flown the accident airplane from Newark International Airport (EWR), Newark, New Jersey,35 to LAX on October 30, 1999, reported that he had experienced difficulties with the autopilot during a portion of that flight.36 The captain told investigators that the autopilot was "hunting" for the glideslope at 8,000 to 10,000 feet msl during the approach to LAX and that, because he was uncomfortable with the autopilot's performance, he disconnected it. The captain reported that his three subsequent attempts to reengage the autopilot to intercept the glideslope in flight were unsuccessful; therefore, he continued the approach and landed the airplane manually. This captain told investigators that the autopilot operated normally when he engaged it on the ground after landing at LAX. Examination of the accident airplane's maintenance logbooks revealed no autopilot-related maintenance writeups, and no subsequent autopilot anomalies were verbally reported.

Examination of the FDR data for the October 30th flight to LAX revealed that at the time the captain reported he disconnected the autopilot because it was "hunting" for the glideslope during the approach to LAX, the autopilot was operating in its LOC (localizer approach) mode, which does not have glideslope intercept capability. The FDR data indicated that, later in the approach to LAX, when the captain tried to reengage the autopilot using the APP (approach) mode, which has both localizer and glideslope intercept capability, the airplane had descended far enough below the glideslope that the autopilot system could not capture the glideslope signal.37

The Safety Board's review of the FDR data revealed that nine autopilot disconnects were recorded on the accident airplane's 25-hour-long FDR tape: one just before landing at Cairo the day before the accident; one just before its next landing at EWR; four during the approach to LAX (during which the reported autopilot difficulties occurred); one on the ground at LAX; one just before landing at JFK the night of the accident flight; and one immediately preceding the accident sequence. No elevator movement was recorded after the autopilot disconnect that occurred on the ground at LAX. The elevator movements recorded following the other eight autopilot disconnects were primarily in the trailing-edge-down (TED) direction and were less than 0.88° in magnitude. According to Boeing, the elevator movements recorded by the accident airplane's FDR were consistent with the movements that would be expected as a result of the normal operation of the autopilot on a properly rigged 767.

Accident Airplane Maintenance Information

During its investigation of the EgyptAir flight 990 accident, the Safety Board reviewed EgyptAir's maintenance program and maintenance recordkeeping procedures and conducted a detailed examination of the accident airplane's maintenance records. The Board's review revealed that the accident airplane had been maintained in accordance with EgyptAir's continuous airworthiness maintenance inspection program for its 767 fleet. Additionally, the accident airplane's maintenance records indicated that all applicable airworthiness directives (AD) had been complied with; no related discrepancies were noted. Further, the Board's review of the accident airplane's technical log sheets from July 29 to October 30, 1999, revealed no pertinent unresolved discrepancies.

FAA Service Difficulty Reports (SDR)38 and accident and incident data from all operators flying 767s between 1990 and 2000 were also reviewed by investigators. Although some elevator-related SDRs were noted,39 there were no documented maintenance trends or anomalies that were relevant to the circumstances of this accident.

767 Bellcrank Anomalies

On March 8, 2000, Boeing personnel reported to the Safety Board that Boeing had been informed of an air carrier incident involving a 767 in which failed bellcrank shear rivets were found in the left inboard and left center elevator PCA bellcrank assemblies.40 The bellcrank shear rivets are designed to shear if an elevator PCA jam occurs, the compressible links between the bellcrank assemblies and the PCA input arms are bottomed out,41 and a force of about 50 lbs is applied to the control column. Research and testing indicated that sheared rivets in a bellcrank assembly could result in an elevator PCA disconnect. Such a failure is discussed briefly later in this section and in detail in the section titled, "Potential Causes for Elevator Movements During the Accident Sequence."

Boeing and the FAA conducted additional tests and research to further investigate why the rivets failed and what the possible repercussions of such a failure would be, including metallurgical examination of high-time bellcranks, material properties testing on old and new bellcranks, review of bellcrank failure rate data obtained from 767 operators, and examination of maintenance procedures to determine whether changes in procedures and/or intervals were warranted. The Safety Board monitored the FAA's and Boeing's tests and research into the bellcrank shear rivet failures.

The research conducted by Boeing and the FAA revealed that single bellcrank shear rivet failures had occurred on other 767s, some of which might not have been detected during the single hydraulic system maintenance check that is to be conducted by 767 operators every 400 flight hours.42 On August 17, 2000, Boeing issued Service Bulletin 767-27A0166, which described methods by which failed bellcrank shear rivets that might not be detected during the single hydraulic system maintenance check could be identified. Subsequently, the FAA issued AD 00-17-05, effective September 11, 2000, which required all 767 operators to perform a one-time functional check of one shear rivet in all six elevator PCA bellcrank assemblies within 30 days, reworking or replacing the bellcrank assembly if needed. AD 00-17-05 indicates the following:

[F]ailure of two [of the three] bellcrank assemblies on one side can result in that single elevator surface [but not both surfaces] moving to a hardover position independent of pilot command resulting in a significant pitch upset recoverable by the crew. Failure of [all] three bellcrank assemblies on one side can cause an elevator hardover that may result in loss of controllability of the airplane...the FAA has received no factual information that indicates that this incident is related to [the EgyptAir flight 990] accident....The cause of that accident is still under investigation.
Because the FAA received reports that the one-time functional check required by AD 00-17-05 revealed failed shear rivets on several 767-300 airplanes, on March 5, 2001, the FAA issued AD 01-04-09, effective March 20, 2001, which required all 767 operators to perform repetitive functional testing of the elevator control system to determine whether the elevator PCAs are properly rigged and accomplish followup actions (including depth penetration inspection of the shear rivets),43 as necessary. AD 01-04-09 required operators to perform the repetitive testing of the elevator control system at least every 400 flight hours, beginning within 90 days of the AD's effective date. Although the cause of the bellcrank shear rivet failures has not yet been determined, Boeing and the FAA are continuing to study the issue.

One of the mechanical failure conditions evaluated by the Safety Board during the EgyptAir flight 990 investigation involved disconnection of the input linkages to two of the three PCAs on one elevator surface. This failure condition could be caused by the failure of any of the components that comprise the elevator PCAs' input linkage systems, including the bellcranks. As further discussed in the section titled, "Potential Causes for Elevator Movements During the Accident Sequence," the Board's tests and simulations indicated that the nonfailed elevator and the airplane are controllable from either control column with a dual PCA disconnect on one elevator surface. Those tests showed that neither a dual disconnection nor a triple disconnection (such as would result from a triple bellcrank failure) on one elevator surface would produce elevator deflections that matched the FDR data from the accident sequence.

METEOROLOGICAL INFORMATION

The Safety Board's review of data from the National Climatic Data Center National Radar Mosaic (from about 0100 through 0230 on October 31, 1999) and other meteorological data revealed no record of significant meteorological conditions in the area at the time of the accident. No pilot reports indicating any significant meteorological event were transmitted in the accident area between about 2300 EDT on October 30 and 0700 on October 31, 1999.

FLIGHT RECORDERS

The FDR and CVR were recovered from the Atlantic Ocean by U.S. Navy remote-operated vehicles on November 9 and November 14, 1999, respectively. Upon recovery, they were immediately packed in water to prevent/delay the onset of corrosion and shipped to the Safety Board's laboratory in Washington, D.C., for readout.

Cockpit Voice Recorder

The CVR installed on the accident airplane was a Fairchild model A-100, S/N 3193. Although the CVR unit exhibited external and internal structural damage and the recording medium (magnetic tape) was wet, the tape was otherwise in good condition. The CVR recording consisted of four channels of audio information, the following three of which recorded usable audio information: the cockpit area microphone (CAM) and the hot microphones at the captain's and first officer's positions.44 The quality of the audio information recorded by the CAM was good, whereas the quality of the audio information recorded by the hot microphone at the first officer's position was excellent until 0141:11, after which time it was poor.45 The audio information recorded by the hot microphone at the captain's position was difficult or impossible to decipher throughout most of the recording.46

The CVR recording started at 0119:13, as the flight was cleared for takeoff from runway 22R at JFK. As previously discussed, the cessation of the CVR recording at 0150:38.47 (shortly after the FDR recorded the airplane's loss of engine power) was consistent with the loss of electrical power to the recorder that occurred after the engines were shut off.

Two transcripts were prepared of the entire 31-minute 30-second recording, one in Arabic/English words and phrases exactly as spoken on the accident flight and the other with Arabic words translated to English. As stated previously, throughout the CVR transcript, the Cockpit Voice Recorder Group provided as direct a translation as possible, without attempting to interpret the words or the intent of the speaker. According to participants in the Cockpit Voice Recorder Group (which included several Arabic/English speakers), occasionally the direct translation of Arabic words into English resulted in awkward or seemingly inappropriate phrases.

Cockpit Voice Recorder Sound Spectrum and Speech Studies

The Safety Board conducted CVR speech and sound spectrum studies to document any unknown sounds and to verify and expand on the information contained in the CVR transcript.47 The results of these studies are discussed in the following sections.

Audio Information Recorded by First Officer's Hot Microphone

The Safety Board's study of the CVR information recorded by the hot microphone at the first officer's position during the accident flight revealed that, at 0141:03, the CVR recorded a decrease in the audio level of the first officer's hot microphone system, and, at 0141:11, the CVR recorded a rustling sound through the first officer's hot microphone system. According to a member of the Speech Examination Study Group, this rustling sound resembled the sound of the headset being stowed as the command first officer prepared to leave the first officer's position. Until this time, the hot microphone at the first officer's position had recorded the first officer's utterances clearly, as well as some additional cockpit noises and conversations; however, subsequently, this microphone (which is a part of the first officer's headset assembly) provided muffled recordings of some, but not all, of the cockpit conversations. Command and relief first officer statements after 0141:11 were recorded more clearly by the CAM.

The study concluded that the recording quality of the first officer's hot microphone was excellent while the command first officer wore the headset/microphone and poor when the headset/microphone was believed to be stowed.48 After 0150:24, the first officer's hot microphone stopped recording cockpit conversation and started recording a sudden increase in background noise. The speech evaluation study indicated that this most likely occurred because a pilot inadvertently activated the air-to-ground/interphone button on the back of the control wheel and thereby altered the amplitude of the recording to the amplitude level set at the individual pilot position.

Speech Sample Information

All speech samples analyzed in the speech study were captured through the CAM located in the overhead panel. Investigators identified recorded speech samples for six EgyptAir crewmembers that were in the cockpit at various times during the accident flight, including the command captain, the relief first officer, the command first officer, the EgyptAir 767 chief pilot, and two nonduty first officers on board the airplane.

All utterances made after the captain departed the cockpit (at 0148:18) were analyzed to the extent possible; however, in part because of occasional loud background noise in the cockpit after that time, only 15 of the 23 utterances recorded by the CAM after such time were strong enough (relative to background noise) to be analyzed by computer for fundamental frequency (pitch) and formant dispersion49 information.

Fundamental Frequency and Speech Duration Information

Research50 has shown that fundamental frequency and speech duration vary characteristically among speakers and often convey information about the speaker's psychological stress. The Safety Board has used the following guidelines51 with regard to fundamental frequency for evaluating the degree of psychological stress experienced by a speaker:
  • An increase in fundamental frequency of about 30 percent (compared with that individual's speech in a relaxed condition) would be characteristic of a stage 1 level of stress, which could result in the speaker's focused attention and improved performance.
  • An increase in fundamental frequency of between 50 to 150 percent would be characteristic of a stage 2 level of stress, which could result in the speaker's performance becoming hasty and abbreviated; however, the speaker's performance would not display gross mistakes.
  • An increase in fundamental frequency of between 100 to 200 percent would be characteristic of a stage 3 level of stress, or panic, which could result in the speaker's inability to think or function logically or productively.
On the basis of these guidelines, the CVR speech study concluded that the relief first officer exhibited no more than a 25 percent increase in fundamental frequency, compared to what he exhibited during routine flight, when he made any of his "I rely on God" statements and when he stated, "it's shut," during the emergency sequence. However, the speech study concluded that the command captain exhibited an increase in fundamental frequency of 29 percent when he stated, "what's happening?," shortly after he returned to the cockpit and of between 47 and 65 percent when he stated, "get away in the engines," "shut the engines," "pull," and "pull with me," during the emergency sequence, compared to what he exhibited during routine flight.

Previous research has also shown that speech duration often becomes shorter (that is, speaking rate becomes faster) when psychological stress increases. Speech duration measurements were performed on the phrase "I rely on God" (repeated by the relief first officer 11 times between 0148:39 and 0150:08). The CVR transcript indicated that the first utterance of the phrase "I rely on God" was spoken faintly, about 1 minute 6 seconds before the autopilot was disconnected, and had a duration of 1.02 seconds. The second utterance of this phrase, which occurred about 5 seconds before the throttle levers were moved to idle and while the airplane was still in level flight, had a duration of 0.81 second. The remaining nine utterances of this phrase, which began about 8 seconds later (as the airplane began its abrupt nose-down pitch and steep descent), varied in duration from 0.73 to 0.87 seconds, with pauses of 0.51 and 0.70 seconds between successive utterances. According to the speech study, the relief first officer's rate of speech did not increase significantly when saying, "I rely on God," during the pitchdown and descent.

The Safety Board also examined the length of time between the relief first officer's "I rely on God" statements for evidence of psychological stress. About 67 seconds passed between the first and second utterances of the phrase, 8.1 seconds passed between the second and third, and 0.51 to 0.70 seconds passed between subsequent utterances of the phrase. According to the speech study, after the second utterance, the data suggested a "rhythmic repetition of the phrase rather than an accelerating trend, as might be expected with increased psychological stress."

The speech study concluded that, although the relief first officer's speech displayed some evidence of increased psychological stress between the first and second utterance of "I rely on God" (when the airplane was still in level flight at cruise altitude), there was no evidence of increased psychological stress in the relief first officer's speech after he uttered the phrase the second time. As previously discussed, after the second utterance of the phrase, the airplane departed level flight to a steep nose-down pitch attitude and experienced an increased nose-down pitch attitude and rate of descent and a decrease in its load factor (to negative Gs) while the relief first officer repeated, "I rely on God," the last nine times.

Unintelligible Comment

The Safety Board's audio examination and sound spectrum analysis of the unintelligible comment that was recorded by the CAM at 0148:30 showed that it appeared to have characteristics consistent with human speech. It consisted of three syllables, with the accent on the second syllable, and was probably spoken very softly (as shown by very poor speech signal definition). The speech examination study indicated that the comment was preceded by 19.2 seconds without speech and followed by 9.2 seconds without speech, suggesting that it was an isolated statement rather than part of a conversation. Unfortunately, the speech segment was not long or clear enough to determine what was said and who said it. However, two speech characteristics of the unintelligible comment--fundamental frequency and formant dispersion--displayed values that, of the six pilots' speech that had been recorded earlier on the CVR tape, most closely resembled the speech values displayed by the relief first officer.

As previously discussed, and as noted as follows in the CVR transcript:

The five Arabic speaking members of the [CVR] group concur that they do not recognize this as an Arabic word, words, or phrase. The entire group agrees that three syllables are heard and the accent is on the second syllable. Four Arabic speaking group members believe that they heard words similar to 'control it.' One English speaking member believes that he heard a word similar to 'hydraulic.' The five other members believe that the word(s) were unintelligible.
Because the content of the comment (the word[s] and the language in which it was spoken) could not be positively identified, the members of the Cockpit Voice Recorder Group agreed to characterize the comment as "unintelligible."

Flight Data Recorder

The FDR installed on the accident airplane was a Sundstrand Data Corporation (now named Honeywell Aerospace Electronic Systems) Universal Flight Data Recorder, S/N unknown. Although the FDR unit exhibited external and internal structural damage and the recording medium (magnetic tape) was wet, the tape was otherwise in good condition. After waveform recovery techniques were used to correct areas of weak FDR signals, a complete set of accident flight data, from takeoff through the last recorded FDR parameter (which was recorded at 0150:36.64),52 was prepared.

Flight performance parameters recorded by the FDR included the following: pressure altitude; airspeed (computed); engine rpm; pitch; roll; heading; angle of attack; normal (vertical), longitudinal, and lateral acceleration (load factors); left and right elevator positions; left and right inboard and outboard aileron positions; left and right trailing edge flap positions; rudder position; and horizontal stabilizer position. In addition, the FDR recorded speedbrake handle position, throttle resolver angle, autopilot engagement/disengagement, engine low oil pressure, and engine fuel cut signals. The FDR was not required to and did not record control wheel, control column, or spoiler positions nor did it record control wheel and column forces.53 Excerpts from the FDR data plots and CVR transcript are shown in figures 3a through 3h.

WRECKAGE INFORMATION

About 70 percent of the airplane was recovered during the initial recovery operations, which began on the morning of October 31 and ended on December 22, 1999. Subsequent recovery efforts conducted between March 29 and April 3, 2000, resulted in the recovery of the left engine and additional pieces of airplane wreckage.

Sonar mapping of the wreckage site depicted two distinct underwater debris fields, which were identified by recovery personnel and investigators as the western and eastern debris fields. These debris fields were about 366 meters (1,200 feet) apart from center point to center point. The western debris field, which was estimated to be 62 meters by 66 meters and was centered about 40° 20' 57" north latitude, 69° 45' 40" west longitude, contained mainly parts associated with the left engine and various other small pieces of wreckage (including portions of two wing panels, fuselage skin, horizontal stabilizer skin, and the majority of the nose landing gear assembly). The eastern debris field, which was estimated to be 83 meters by 73 meters and was centered about 40° 20' 51" north latitude, 69° 45' 24" west longitude, contained the bulk of the airplane's fuselage, wings, empennage (including the outboard tips of the right and left elevators and all recovered elevator PCAs), right engine, main landing gear, and flight recorders. Many pieces of floating wreckage (including pieces of the right and left elevator surfaces)54 were recovered from the water's surface in or near the eastern debris field shortly after the accident; specific recovery locations for some of these pieces were not noted. The small size of most of the recovered pieces of wreckage was consistent with the airplane impacting the water at a high speed. The locations of the two main wreckage debris fields were consistent with the accident airplane's flightpath, as indicated by the primary radar data.55

The Safety Board leased a commercial vessel to recover the wreckage that had settled on the ocean floor. Pieces of wreckage were recovered from a depth of about 230 feet using a clamshell scoop and a crane, loaded (using a front loader) into containers on the recovery vessel, and moved to shore. Upon reaching shore, the containers of wreckage were lifted off the recovery vessel and rinsed thoroughly twice. The containers were then moved into the hangar at Quonset Point, Rhode Island, where they were tipped onto their sides. The wreckage was then moved out of the containers onto the floor using rakes and shovels. Once on the hangar floor, the wreckage was spread evenly by a front loader to assist the drying process. During this process, FBI and Safety Board investigators examined the recovered wreckage for evidence of fire or explosion damage. The FBI placed identification tags on some of the debris; accident investigators then documented all of the debris.

Four of the elevators' six PCAs (the center and outboard right elevator PCAs and two elevator PCAs whose positions could not be determined)56 were recovered. Postaccident examination revealed that all four of the recovered PCAs exhibited impact-related damage. One of the four also exhibited the following two unusual characteristics on its internal mechanisms: (1) the pin that attaches the spring guide to the valve slide was sheared, and (2) a portion of the bias spring (about one full coil) was improperly positioned over the head portion of the spring guide. It could not be determined whether these conditions existed before impact or whether they were impact related. The Safety Board's measurements of these components indicated that the inside diameter of the servo valve cap into which the bias spring and spring guide fit was 0.872 inch and that the outside diameter of the spring guide at its widest point was 0.749 inch, leaving a clearance of 0.123 inch between the spring guide and the servo valve cap. Measurements indicated that the bias spring wire had a diameter of 0.031 inch. Impact marks and damage were observed on other components in this PCA; however, there was no evidence of scraping, abrasion, or other marks on the improperly positioned bias spring or adjacent surfaces that would indicate that these metal parts had jammed in the PCA.57

Five of the elevators' six bellcranks (all three right elevator and two of the left elevator bellcrank assemblies) were recovered. Postaccident examination of the recovered bellcrank assemblies revealed that all of the shear rivets in the recovered bellcrank assemblies were sheared,58 with the sheared surfaces appearing consistent with shear overstress. However, the rivets in some of the bellcrank assemblies sheared in a direction opposite to others; shear rivets in the two recovered bellcrank assemblies from the left elevator surface and in the inboard bellcrank assembly from the right elevator surface were sheared as if the bellcrank arms were moving to a higher relative angle, whereas the shear rivets in the middle and outboard bellcrank assemblies from the right elevator surface were sheared as if the bellcrank arms were moving to a lower relative angle. Most of the recovered elevator control linkages were found broken or otherwise damaged.

Examination of the fracture surfaces on the recovered pieces of wreckage revealed that the fractures were consistent with failures generated by a high-speed impact. None of the fracture surfaces examined exhibited any sign of preexisting fatigue or corrosion. No evidence of foreign object impact damage or pre- or postimpact explosion or fire damage was observed.59

Examination of the left engine (which was recovered relatively intact) revealed evidence of little, if any, rotation at the time of impact. The right engine was severely broken up, and only about 80 percent of it was recovered. Examination of the recovered portions of the right engine showed evidence of little, if any, rotation at the time of impact. The observed deformations on the right engine were consistent with a steep impact angle, whereas observed deformations on the left engine were consistent with an inverted, slightly aft-end-down impact angle. Although the recovery location of and damage to the left engine were consistent with it separating from the airplane before impact, no evidence of any preimpact catastrophic damage or fire was observed on either engine.60

TESTS AND RESEARCH

Review of Radar Data

Five radar sites detected primary and/or secondary returns from EgyptAir flight 990. These sites are located at North Truro, Massachusetts; Riverhead, New York; Gibbsboro, New Jersey; Oceana, Virginia; and Nantucket, Massachusetts. The Safety Board's examination of the available radar data revealed that four of the five radar sites recorded no sequence of primary or secondary radar returns that intersected EgyptAir flight 990's position at any time nor did they reveal any radar returns consistent with a projectile or other object traveling toward the accident airplane. Although the Riverhead radar site recorded numerous radar returns near the flightpath of EgyptAir flight 990 within 5 minutes of the accident, none of the radar sites with areas of coverage that overlapped this area of Riverhead's coverage recorded similar radar returns. Consultation with the USAF Radar Evaluation Squadron revealed that the primary returns in question from the Riverhead radar site were caused by radio frequency interference from the Bucks Harbor, Maine, long-range radar site.

No secondary radar returns were received from EgyptAir flight 990 after 0150:36 (about the time the CVR and FDR stopped recording); however, after this time, several radar sites recorded primary radar returns that continued along the accident airplane's extended flightpath from its last recorded radar position. As previously discussed, these primary radar data (with extrapolated FDR data and simulation results) indicated that after the airplane's FDR and CVR stopped recording, the airplane descended to an altitude of about 16,000 feet msl, then climbed to about 25,000 feet msl and changed heading from 80º to 140º before it began its second descent, which continued until it impacted the ocean.61

Accident Sequence Study

The Safety Board used the FDR, radar, winds aloft, and 767 performance data to determine the accident airplane's motions and performance during the accident sequence. These data (and associated calculations) indicated the following:
  • Aside from the very slight movement of both elevators (the left elevator moved from a 0.7° to about a 0.5° nose-up deflection, and the right elevator moved from a 0.35° nose-up deflection to about a 0.3° nose-down deflection)62 and the airplane's corresponding slight nose-down pitch change, which were recorded within the first second after autopilot disconnect at 0149:45, and a very slow (0.5º per second) left roll rate, the airplane remained essentially in level flight about FL 330 for about 8 seconds after the autopilot was disconnected.
  • At 0149:53, the left and right throttles were retarded to the aft idle stop (equivalent to a throttle lever angle of about 33º) at a rate of about 25º per second.63 About 1 second after the start of the throttle movement, the FDR recorded slight motion in the inboard ailerons, the left elevator surface moved to about a 3.4º TED position, and the right elevator surface moved to about a 3.8º TED position.64
  • At 0149:54, the airplane began to pitch nose down, reaching a pitch attitude of about 40º nose down at 0150:15. During the dive, the wings remained within about 10º of level and the heading remained about 80º, increasing to about 85º between 0150:20 and 0150:33.
  • Between 0150:05 and 0150:06, the FDR recorded additional movements in the inboard ailerons, and the left and right elevators moved an additional 1.5º TED to about 5.5º TED. Before this time, the load factor had been about 0.2 G; after this time, the load factor decreased to about -0.1 G. Between 0150:06 and 0150:10, the FDR began to record "Low Engine Oil Pressure" signals for both engines; the FDR recorded these signals until after the load factor increased to above 0 G between 0150:17 and 0150:21.65
  • At 0150:08, as the airplane passed through about 30,800 feet msl, the airplane exceeded its maximum operating airspeed (0.86 Mach), and the Master Warning alarm sounded. The maximum rate of descent recorded during the dive was about 39,000 fpm at 0150:19, as the airplane descended through about 24,600 feet msl. At 0150:23, the airspeed reached its peak calculated value of 0.99 Mach, as the airplane descended through about 22,200 feet msl.
  • At 0150:15 and about 27,300 feet msl, the left and right elevator surfaces started to move slowly (about 0.6º per second) in the trailing-edge-up (TEU) direction, back toward their neutral position. The pitch angle, angle of attack, and load factor also started to increase at this point, so that when the FDR recorded the last data for the accident flight at 0150:36.64, the pitch angle had increased to about 8º nose down, and the airplane was experiencing about 2.4 Gs.
  • Between 0150:18 and 0150:27, the FDR recorded TEU movements of the left and right outboard ailerons and the left inboard aileron.66
  • At 0150:21, the left and right elevator surfaces started to split (that is, to move asymmetrically). The right elevator surface started to move TED, whereas the left elevator surface moved TEU. This split between the left and right elevator surface positions continued to the end of the FDR data, varying in magnitude but averaging about 4º difference between the surfaces (see figure 2).
  • Between 0150:21 and 0150:23, the engine start lever switches for both engines moved from the run to the cutoff position.
  • Between 0150:24 and 0150:25, both throttle handles moved full forward.
  • Between 0150:25 and 0150:26, the speedbrake handle moved to its fully deployed position. Coincident with this activity, between 0150:24 and 0150:27, the left elevator surface moved briefly in the TED direction (from 3º TEU to 1º TEU) before it returned to 3º TEU.
  • Almost immediately after the speedbrakes were deployed at 0150:26, the left elevator surface deflection increased further, reaching its maximum deflection of more than 3.8º nose up about 0150:30.67 After 0150:30, the left elevator's nose-up deflection gradually reduced, until the data for that parameter ended at 0150:36 with a left elevator deflection of about 2.3º nose up.
  • Between 0150:21 and 0150:24, the right elevator surface's nose-down deflection increased gradually, then increased rapidly until just after 0150:25, when the nose-down deflection briefly reduced from about 2.35º to about 1.9º nose down. Between 0150:21 and 0150:23, the engine start levers moved to the cutoff position. At 0150:26, the right elevator's nose-down deflection began to increase again, reaching its maximum nose-down deflection of about 3.2º at 0150:29. Subsequently, the right elevator's deflection moved generally toward a nose-up position, with occasional movements in a nose-down direction; when the FDR data ended, the right elevator deflection was 0.2º nose down.
  • Between 0150:27 and 0150:32, the FDR recorded a split condition in the outboard ailerons (the left outboard aileron maintained its approximate presplit deflection, while the right outboard aileron began to move in a TED direction).68 The outboard ailerons had been moving in a TEU direction since 0150:18.
  • During the elevator split, the larger movements of the left and right elevators individually corresponded with changes in the load factor (see figure 4). For example, between 0150:30 and 0150:36, the recorded movements of the right elevator (lower graph) are reflected in the load factor profile (upper graph).
  • No secondary radar returns were received from the accident airplane after the last data were recorded by the FDR at 0150:36.64.
  • Performance calculations based on primary radar returns indicated that the airplane's rapid descent stopped at an altitude of about 16,000 feet msl. The primary radar returns indicated that the airplane then began to climb, reaching about 25,000 feet msl about 0151:15. During this climb, the airplane's heading changed from about 80º to about 140º.
  • After 0151:15, the data indicated that the airplane began a second rapid descent that continued until it impacted the ocean.
Seven primary radar returns from the airplane were recorded during the second dive; the altitude estimates from these returns are subject to potentially large errors, which introduces significant uncertainty into the performance calculations during the second dive. However, the data indicate that the airplane impacted the ocean about 0152:30, with an average descent rate during the second dive of about 20,000 fpm.
 
 

Potential Causes for Elevator Movements During the Accident Sequence

Investigators used Boeing's six-degree-of-freedom, full-flight engineering simulator (which incorporated, to the maximum extent possible, the flight characteristics of the 767) to evaluate whether the accident airplane's recorded pitch motions were consistent with the elevator position movements recorded on the FDR.69 The results showed that the elevator movements required to make the simulator duplicate the pitch motions and flightpath recorded on the FDR were consistent with the elevator movements recorded by the FDR throughout the recorded data, even during the time that the data indicated a split between the left and right elevator surfaces.70 The investigation attempted to determine if any mechanical failures could have caused these elevator movements.

The Systems Group reviewed numerous potential failure scenarios to evaluate whether any of them might have been capable of causing the elevator surface movements recorded on the FDR during the accident sequence, including failures associated with the elevator system's flight control cables,71 failures associated with elevator surface PCAs,72 and other system-related failures.73 On the basis of the results of failure modes and effects analyses, the Safety Board ruled out all but four of these potential failure scenarios because they failed to reflect the accident flight's elevator movements in obvious and significant ways.74 For example, it was determined that neither an autopilot malfunction nor EMI75 would have caused any elevator movements during the accident sequence.76 Although some of the other scenarios could have caused some elevator movements, the nature and degree of those movements differed so greatly from the elevator movements recorded during the accident flight that they did not warrant further consideration.

However, the failure modes and effects analyses showed that the following four elevator failure scenarios (each of which involves two failures) warranted further study because they could potentially cause nose-down elevator movements or a split elevator condition that might resemble some portions of the data recorded on the accident flight's FDR:

  1. Disconnection of the input linkages to two of the three PCAs on the right elevator surface. This failure scenario could be caused by the failure of any of the components that comprise the actuator input linkage system, including the bellcranks.
  2. A jam of the input linkages or servo valves in two of the three PCAs on the right elevator surface. In order for this failure scenario to occur, the internal slides of the affected servo valve would first have to be moved (by manual or autopilot input) to an offset position and then jam. Although such jams could theoretically occur in either direction, all tests and simulations involving jammed elevator PCAs were intentionally configured to produce nose-down (rather than nose-up) elevator input.
  3. A jam of the input linkage or servo valve in one PCA and the disconnection of the input linkage to another PCA on the right elevator surface.
  4. A jam in the elevator flight control cable connecting the right-side control column to the right aft quadrant assembly combined with a break in the same cable. (Four variants of this scenario were studied. For additional information, see the Systems Group Chairman's Factual Report and its addendum regarding the cable break/jam and PCA jam with high breakout force [compressible link] ground testing.)
For further evaluation, the Systems Group conducted ground tests on an instrumented 767 to record the elevator system's response to each of these failure scenarios. During the ground tests, the test airplane's systems were configured to simulate the accident airplane's altitude and airspeed. The Systems Group also studied the effect that each failure scenario would have on the elevator control system and calculated the effect on the elevators that each scenario would have had at the specific conditions of the accident flight at the time of the initial pitchdown. The results of the tests, studies, and calculations were as follows:
  1. Disconnection of the input linkages to two of the three PCAs on the right elevator surface.

    During the ground tests, the failed elevator surface was driven to its full nose-down position and would not respond to nose-up flight control inputs from either control column. A study of the elevator control system indicated that if this scenario occurred in flight, it would result in an initial deflection of the failed surface to a position consistent with a single functioning elevator PCA operating at 100 percent of its maximum force (as limited by aerodynamic blowdown forces); the failed elevator surface would resist being backdriven with a force equivalent to about 130 percent of a single functioning PCA.77 Calculations showed that at 280 knots (the accident airplane's airspeed when the initial descent began), this position would initially have been about 6º nose-down elevator deflection, and the degree of deflection would be reduced as the airplane's speed increased above 290 knots. See figure 2 for additional elevator blowdown position information.78

    During the ground tests, the nonfailed elevator surface remained in its prefailure position unless it received inputs from either control column. A study of the elevator control system's force balance and calculations of the effect of this failure under the conditions of the accident flight indicated that the nonfailed surface would remain in its prefailure position.

    During the ground tests, either control column could be used to control the nonfailed elevator surface and to command the full travel of that surface available at the existing flight condition. The Safety Board's study of the elevator control system indicated that under the accident flight conditions, inputs from either control column would have resulted in corresponding movement of the nonfailed elevator surface.
     

  2. A jam of the input linkages or servo valves in two of the three PCAs on the right elevator surface.
  3. During the ground tests, the failed elevator surface was driven to its full nose-down position and would not respond to nose-up flight control inputs from either control column. A study of the elevator control system indicated that if this scenario occurred in flight, it would result in an initial deflection of the failed surface to a position consistent with a single functioning elevator operating at 100 percent of its maximum force (as limited by aerodynamic blowdown forces); the failed elevator surface would resist being backdriven with a force equivalent to about 130 percent of a single functioning PCA. As discussed in connection with the previous failure scenario, calculations showed that, under the conditions of the accident flight, this position would initially have been about 6º nose-down elevator deflection, and the degree of deflection would be reduced as the airplane's speed increased above 290 knots.

    During the ground tests, the nonfailed elevator surface moved to about 4º nose-down deflection in the same direction as the failed surface. A study of the elevator control system's force balance and calculations of the effect of this failure under the conditions of the accident flight indicated that the nonfailed surface would move to a position corresponding to 30 lbs of feel force.79 Calculations showed that under the conditions of the accident flight when the initial descent began, the degree of deflection for the nonfailed surface would be the same as during the ground tests (about 4º).

    During the ground tests, either control column could be used to control the nonfailed elevator surface and to command that surface in either the nose-up or the nose-down direction.80 The Safety Board's study of the elevator control system indicated that under the accident flight conditions, inputs from either control column would have resulted in corresponding movement of the nonfailed elevator surface.
     

  4. A jam of the input linkage or servo valve in one PCA and the disconnection of the input linkage to another PCA on the right elevator surface.
  5. During the ground tests, the failed elevator surface was driven to its full nose-down position and would not respond to nose-up flight control inputs from either control column. A study of the elevator control system indicated that if this scenario occurred in flight, it would result in an initial deflection of the failed surface to a position consistent with a single functioning elevator operating at 100 percent of its maximum force (as limited by aerodynamic blowdown forces); the failed elevator surface would resist being backdriven with a force equivalent to about 130 percent of a single functioning PCA. As discussed in connection with the previous failure scenarios, calculations showed that, under the conditions of the accident flight, this position would initially have been about 6º nose-down elevator deflection, and the degree of deflection would be reduced as the airplane's speed increased above 290 knots.

    During the ground tests, the nonfailed elevator surface moved to about 2.1º nose-down deflection in the same direction as the failed surface. A study of the elevator control system's force balance and calculations of the effect of this failure under the conditions of the accident flight indicated that the nonfailed surface would move to a position corresponding to 15 lbs of feel force. Calculations showed that under the conditions of the accident flight when the initial descent began, the degree of deflection for the nonfailed surface would be the same as during the ground tests (about 2.1º).

    During the ground tests, either control column could be used to control the nonfailed elevator surface and to command that surface in either the nose-up or the nose-down direction. The Safety Board's study of the elevator control system indicated that under the accident flight conditions, inputs from either control column would have resulted in corresponding movement of the nonfailed elevator surface.
     

  6. A jam in the elevator control cable connecting the right-side control column to the right aft quadrant assembly combined with a break in the same cable.

    During the ground tests, the left elevator surface moved to nose-down deflections of 1.2º to 3.9º and the right elevator surface moved to nose-down deflections of 1.4º to 5.0º, depending on the scenario tested. Analysis of the elevator system indicated that if such failures occurred in flight, the resultant elevator surface positions would not have varied as a result of changes in the aerodynamic forces acting on the elevator in the same manner as the previous three failures because all three PCAs would still be functioning properly.

    During the ground tests for all break/jam combinations, either control column could be used to control both elevator surfaces. Testing showed that a pull from either control column of 25 lbs would result in sufficient movement of both elevators in a nose-up direction to be evident on the FDR. A pull from either control column of 50 to 100 lbs would result in sufficient movement of both elevators in a nose-up direction to either reverse or significantly slow the airplane's nose-down dive.

Pilots from Boeing, EgyptAir, the FAA, and the Safety Board evaluated the controllability of the airplane following the first three of these failure scenarios in Boeing's fixed-base engineering simulator. The simulations assumed that the right elevator was affected by the failure scenario being evaluated and duplicated the airplane's response to the occurrence of that scenario. As previously mentioned, the simulator reflected the flight characteristics of the 767 to the maximum extent possible. Although all flight conditions (for example, airspeed, altitude, roll attitude, load factor, and pitch attitude) were calculated correctly in the simulator, the fixed-base simulator could not duplicate the physical sensations that would have resulted from these flight conditions. For example, the load factors that would be produced under actual flight conditions were not produced in Boeing's fixed-base simulator nor were the actual attitude changes that would have occurred in flight.81 In addition, the simulator did not model the override mechanisms between the two control columns.

Simulations of the three failure scenarios showed that it was possible to regain control of the airplane using either control column and return it to straight and level flight using normal piloting techniques and that the airplane could be trimmed to hands-off level flight after each of the three failure scenarios occurred.82 Further, for all three failure scenarios, full recovery was possible even when no efforts were made to recover the airplane until 20 seconds after the failure occurred. Although additional force beyond that required for recovery from a dive of this magnitude without a failure was necessary in all tested scenarios, the nonfailed surface responded immediately to nose-up inputs and recovery could be accomplished by a single pilot using either the left or right control column. Although the recovery was easier and the required control column force was reduced when stabilizer trim was used, it was not necessary to use stabilizer trim to recover from any of the three failure scenarios. Further, the simulations also demonstrated that the engines could have been restarted throughout most (if not all) of the recovery from the dive and/or the subsequent climb and that the airplane could have been returned to straight and level flight after the recorders stopped recording. The elevator deflections resulting from the fourth scenario were less extreme, and would therefore be easier to recover from, than those resulting from the first three failure scenarios.
 

Additional Information

Submissions

The Safety Board received submissions from EgyptAir,83 Boeing, and P&W.84 (Note: The fourth failure scenario [a cable jam with a break of the same cable] was studied after these submissions were received; therefore, these submissions do not reference this scenario.)

EgyptAir's April 28, 2000, Presentation

During EgyptAir's April 28, 2000, presentation to the Safety Board, its representatives stated, in part, the following:
  • The suicide scenario is not consistent with data and facts of [the EgyptAir flight 990] accident.
  • [Left and right] elevator deflection as a result of right elevator dual PCA jam is consistent with the FDR data where Boeing data is valid.
  • In the area beyond the airplane normal design envelope where the data is not valid, all the flight control behavior is uncertain, control surfaces are subject to flutter.
  • The right elevator middle and outboard bellcrank rivets shear direction is consistent with a jammed PCA reacting against pilot input to move the elevator up.
  • Analysis revealed that there are a lot of [radar] returns forming continuous paths crossing the flightpath of [EgyptAir flight 990], which may reflect deliberate [evasive] action by one of the pilots.

EgyptAir's August 11, 2000, Submission

In its August 11, 2000, submission, EgyptAir asserted that ground tests and simulations conducted during the investigation were flawed because simulations were conducted using Boeing's published 767 data and "did not reflect the actual operation of the airplane"; steady-state values were used to calculate the control column forces in various dynamic flight conditions, thus invalidating conclusions; and extrapolation of data for calculated airplane speeds in excess of those for which test data existed (Mach 0.91) "cannot produce accurate results."

EgyptAir also argued in its submission that the relief first officer did not deliberately cause the accident. According to the submission, "the deliberate act theory was based, in large part, on the initial inaccurate translation of an expression repeated several times by the [relief] first officer...[which] has been eliminated not only by credible evidence and analysis but also by accurate translation of the CVR." EgyptAir further stated that (1) the relief first officer had no motive to kill himself or others aboard EgyptAir flight 990; (2) the relief first officer did not use his seniority to insist that he be allowed to fly the airplane; (3) the relief first officer may not have been alone in the cockpit at the onset of the dive; and (4) the captain returned to the cockpit almost immediately after the dive started, and there was no indication of a struggle or disagreement between the two flight crewmembers. EgyptAir further stated that "the cockpit conversations showed an effort at teamwork rather than a crew working at cross purposes." Further, EgyptAir indicated that several of the relief first officer's actions were not consistent with a deliberate attempt to crash the airplane. For example, it stated that the cockpit door was not closed and locked; the throttle levers were moved to idle, whereas engine power would have accelerated the descent; and more radical flight control inputs were available (more nose-down elevator deflection or aileron and rudder with elevator deflections) but not used.

EgyptAir also contended in its submission that at least three flight crewmembers were in the cockpit during the descent, as evidenced "by the fact that if either the captain or [relief] first officer had let go of their control columns to shut the engines or to deploy the speedbrake...the aircraft would have pitched down at the same time." The EgyptAir submission further stated that the split elevators "do not support the conclusion that there was a struggle in the cockpit" because (1) the CVR provides no indication of a struggle, argument, or refusal to follow a command; (2) the FDR recorded control surface positions but did not record the control column position or forces--"accordingly, one cannot conclude from examining only the FDR data that pilot input to his control column caused the elevators to be in a given position"; and (3) "at the same moment the elevators split, both outboard ailerons moved upward...when this unusual aileron movement occurred during the dive, the aircraft's speed was approaching Mach 1.0, and no published performance data is available to predict what will occur to ailerons at these high speeds. It is likely, however, that aerodynamic shocks or flutter were occurring at the control surfaces."

In its submission, EgyptAir summarized its position as follows:

  • Accordingly, from an impartial review of the factual evidence gathered during the investigation, it is clear that the [relief] first officer did not intentionally dive the aircraft into the ocean.
  • At this point in the investigation considering the factual evidence gathered, it is clear that the first officer did not commit suicide. Further investigation of the elevator control system's design in conjunction with the other factual information available is necessary before a conclusion can be reached regarding the true cause of this accident. Specifically, further engineering analysis, including wind tunnel tests, is necessary to examine the dual actuator malfunction in the speed ranges for which current data is not available. In addition, further investigation of radar data is also necessary to completely rule out the possibility of conflicting traffic. Until this work is accomplished, the cause of this accident cannot be truly established.
  • An analysis of the facts and of the elevator control system's design indicates that malfunctions in two PCAs on the right elevator may have precipitated the airplane's dive. This dual PCA malfunction may have consisted of a latent or nearly latent failure in one PCA that may have existed for a period of time followed by a jam of a second PCA shortly before the dive.
  • The facts do not support the initial, and widely reported, theory that the [relief] first officer deliberately dove the plane toward the ocean.
  • Without further information concerning the data from military and FAA radar, one cannot rule out the possibility that the [relief] first officer may have been attempting to avoid or maneuver the aircraft out of a perceived dangerous situation at the time the dive occurred.

Boeing's Submission

Boeing's October 31, 2000, submission indicated that none of the mechanical failure modes examined during the investigation were consistent with the FDR data because (1) "the FDR elevator positions did not displace to the positions [predicted by the failure mode and effects analysis] during the initial pitchover" and (2) "the elevator motions after the initial pitchover indicate that both surfaces were functioning normally."

Boeing also considered several operational scenarios, including collision avoidance, rapid descents, response to engine oil pressure lights, and loss of thrust on both engines. Boeing's submission stated that, "EgyptAir 990 crew actions were determined to be inconsistent with the performance of standard Boeing recommended operating procedures and training for the 767 airplane."

In its submission, Boeing summarized its position as follows:

  • Flight control surface movements recorded on the [FDR] are capable of generating the airplane flight path recorded by the [FDR] and radar.
  • Based on the examination of the recovered wreckage, Boeing did not find any evidence of a failure condition within the airplane flight control system that could have caused or contributed to the initial pitchover, or prevented recovery from the dive.
  • Boeing participated in examining all potential failure conditions developed during the investigation and could not find a failure condition that: (1) matched the data recorded by the [FDR] or (2) resulted in a condition that was not recoverable by the pilot.
  • Therefore, Boeing does not believe that the loss of EgyptAir 990 was the result of a mechanical failure of the aircraft or aircraft systems.

EgyptAir's January 12, 2001, Response to Boeing's Submission

In its January 12, 2001, response to Boeing's submission, EgyptAir stated that Boeing did not "account for or...comment on" the FAA's ADs regarding bellcrank shear rivet failures in its submission. However, EgyptAir's response to Boeing's submission indicated that it could not determine whether the bellcrank shear rivet failures were involved in the accident, "[the FDR data were] remarkably consistent with test data of a jam of two right elevator servos in the trailing edge down position." EgyptAir further stated that "the differences between the test data and the FDR can be adequately explained as either performance variances within normal limit or limitations of the test facilities and protocols."

Additionally, EgyptAir's response to Boeing's submission indicated that there was evidence of a mechanical malfunction of the elevator system; specifically, EgyptAir cited the reported autopilot difficulties during an approach to LAX the day before the accident and the downward elevator deflections recorded by the FDR at autopilot disconnect. EgyptAir stated the following:

[This evidence] shows that an anomaly existed in the flight 990 elevator system even before the aircraft left New York for Cairo on October 31, 1999--a latent defect that could not be detected by the crew. In light of these facts, it is plausible to believe that--just as [the captain of the flight into LAX] had done a day earlier--the [relief] first officer on flight 990 disconnected the autopilot after observing some unusual movement of the column.
The EgyptAir response to Boeing's submission also stated that examination of the recovered wreckage "indicated damage to the elevator system prior to impact." Specifically, EgyptAir asserted that the damage to the right outboard PCA (sheared pin and improperly positioned bias spring) and right elevator bellcrank shear rivets (failed in opposite directions) likely occurred before the airplane impacted the ocean.

Further, in its response to Boeing's submission, EgyptAir reiterated its position that its analysis of the elevator's motions85 indicated that the elevator split observed in the FDR data was not the result of a struggle between the captain and the relief first officer. The response stated that the split "may [have been] the result of the loss of the right elevator." To support this statement, EgyptAir asserted that (1) there was no indication on the CVR transcript that a struggle occurred in the cockpit; (2) the uncommanded elevator positions might have resulted from unique aerodynamic phenomena as the airplane's speed increased; (3) when an FDR records elevator position, it is actually recording the sensor position, which does not, by itself, indicate the position of the elevator or the control column; and (4) the pitch and roll motions recorded during the last 15 seconds of FDR operation were "much closer to the expected aircraft performance if the right elevator is missing."

EgyptAir's response to Boeing's submission also stated the following:

EgyptAir has determined that the FDR flight profile after the split is consistent with the expected aircraft performance only if the right elevator has departed the airplane....this conclusion is based upon the absence of expected rolling moment that would have been induced by a differential deflection of the elevators...shown on the FDR.
In addition, EgyptAir's response stated the following:
  • Boeing's engineering simulator did not provide an accurate model of real aircraft performance.
  • Boeing often ignored the more reliable ground test results.
  • Boeing's selective use of test data resulted in inconsistent conclusions.
  • Boeing's conclusions regarding crew actions are erroneous.
In its response to Boeing's submission, EgyptAir summarized its position as follows:
  • Boeing's submission to the NTSB [National Transportation Safety Board] dated October 31, 2000, contains many inaccuracies, omissions, and the selective use of evidence.
  • Boeing's own ground test and simulator data does not support its conclusion that FDR data is inconsistent with a dual jam scenario.
  • A dual PCA control valve failure on the right elevator is consistent with the EgyptAir flight 990 FDR data.
  • There is physical evidence consistent with a malfunction in the elevator control system which might be a plausible cause for the accident.

P&W's Submission

In an October 6, 2000, letter, P&W stated that it believed that "the facts, as gathered to date, sufficiently represent Pratt & Whitney's perspective on this crash. Therefore, Pratt & Whitney will not be providing a further submission to be considered during the development of the final report."

ANALYSIS

General

The command captain and relief first officer were properly certificated and qualified and had received the training and off-duty time prescribed by applicable regulations and company requirements. (For more detailed information regarding the background and recent activities of all EgyptAir flight 990 crewmembers, see the Operational Factors Group Chairman's Factual Report and the Human Performance Group Chairman's Factual Report and their addendums.)

The accident airplane was properly certificated and was equipped, maintained, and dispatched in accordance with applicable regulations and industry practices.

The Safety Board's review of air traffic control (ATC) information revealed no evidence of any ATC problems or issues related to the accident. Further, examination of the recovered airplane wreckage and cockpit voice recorder (CVR), flight data recorder (FDR), ATC, weather, and radar data revealed no evidence that an encounter with other air traffic or any other airborne object was involved in the accident or that weather was a factor in the accident.

Examination of the wreckage revealed no evidence of preexisting fatigue, corrosion, or mechanical damage that could have contributed to the airplane's initial pitchover.86 (The condition of the recovered elevator power control actuators (PCA) and bellcrank shear rivets is discussed in the next section titled, "Mechanical Failure/Anomaly Scenarios.") No evidence of explosion or fire damage or foreign object impact damage was found.

Additionally, the Safety Board's examination of the accident airplane's maintenance records revealed no evidence of any mechanical problems that could have played a role in the accident sequence. Although during interviews conducted at the request of the Egyptian Government more than 1 year after the accident an EgyptAir 767 captain reported that he had experienced autopilot difficulties in the accident airplane during the approach to Los Angeles International Airport (LAX), Los Angeles, California, the day before the accident, these difficulties were likely the result of improper autopilot approach mode selection. Additionally, as previously noted, neither the captain nor the first officer of the flight to LAX reported any autopilot anomalies in the airplane's maintenance logbooks, and the first officer of that flight did not mention any autopilot difficulties during interviews conducted 3 days after the accident. Although the captain reported several minor anomalies during these interviews (including an autopilot anomaly), he told investigators that the airplane was "almost perfect." No autopilot difficulties were reported by the flight crew that flew the airplane from LAX to John F. Kennedy International Airport (JFK), New York, New York, immediately before the accident flight nor did they report any autopilot anomalies in the airplane's maintenance logbooks. Further, the Board's examination of the FDR data before and after all recorded autopilot disconnects in the 25 hours of data recorded by the FDR (including the accident flight) revealed no evidence of abnormal autopilot or elevator surface behavior.

The Safety Board's review of ATC, FDR, CVR, and radar information indicated that the airplane's movements during the accident flight were routine until about 0149:54 (9 seconds after the autopilot disconnect occurred), when an abrupt sustained nose-down elevator motion occurred. A review of the FDR data indicated that the accident airplane's pitch motion before and during the accident sequence was consistent with the elevators' recorded movements. Boeing's full-flight engineering simulator was used to evaluate the consistency of the elevator positions with the pitch motions recorded on the FDR. During these evaluations, the elevator movements required to make the simulator duplicate the pitch motions recorded by the accident airplane's FDR and the flightpath developed from the available data closely matched the elevator movements recorded by the FDR. Further, the recorded load factors were consistent with the recorded movements of both elevator surfaces throughout the recorded data, even during the time that the data indicated a split between the left and right elevator surfaces (see figure 4).87

The results of the Safety Board's examination of CVR, FDR, radar, airplane maintenance history, wreckage, trajectory study, and debris field information were not consistent with any portion of the airplane (including any part of the longitudinal flight controls) separating throughout the initial dive and subsequent climb to about 25,000 feet mean sea level (msl). It is apparent that the left engine and some small pieces of wreckage separated from the airplane at some point before water impact because they were located in the western debris field about 1,200 feet from the eastern debris field. Although no radar or FDR data indicated exactly when (at what altitude) the separation occurred, on the basis of aerodynamic evidence and the proximity of the two debris fields, it is apparent that the airplane remained intact until sometime during its final descent. Further, it is apparent that while the recorders were operating, both elevator surfaces were intact, attached to the airplane, and placed in the positions recorded by the FDR data and that the elevator movements were driving the airplane pitch motion, and all associated recorded parameters changed accordingly.

Mechanical Failure/Anomaly Scenarios

The Safety Board evaluated possible mechanical failure and pilot action scenarios in an attempt to determine whether they were consistent with the elevator movements made during the accident sequence. As previously discussed in the section titled, "Potential Causes for Elevator Movements During the Accident Sequence," the investigation ruled out all but four possible anomalies and failure scenarios as potential factors in the accident because they diverged too far from what was reflected on the accident flight's FDR to warrant further consideration.88 Analysis showed that the effects of four failure scenarios (each of which involves dual failures) bore some resemblance to some portions of the accident flight's FDR data. Specifically, initially it appeared that each of these failure scenarios could potentially cause nose-down elevator movements or a split elevator condition that might resemble those recorded on the accident flight's FDR. Those four failure scenarios were (1) disconnection of the input linkages to two of the three PCAs on the right elevator surface,89 (2) a jam of the input linkages or servo valves in two of the three PCAs on the right elevator surface,90 (3) a jam of the input linkage or servo valve in one PCA and the disconnection of the input linkage to another PCA on the right elevator surface,91 and (4) a jam in the elevator flight control cable connecting the right-side control column to the right aft quadrant assembly combined with a break in the same cable.92 Therefore, the wreckage from the accident airplane was examined for possible evidence of PCA anomalies, and the predicted elevator movements resulting from these failure scenarios were evaluated and compared with the data from the accident flight.

As previously mentioned, one of the recovered PCAs was found with a pin (that attached the spring guide to the servo valve slide) sheared and one coil of the bias spring improperly positioned over the head portion of the spring guide. However, there were no marks on any of the surfaces or any deformation of the spring coil to suggest that the spring coil had become jammed between the servo cap and the spring guide, as would be expected if such a jam had occurred.93 Moreover, investigators measured the clearances between these components and determined that those clearances were large enough that even if a coil of the bias spring had become misplaced between the spring guide and the servo valve cap, no jam would have resulted.94 Further, the FDR data preceding the accident sequence do not show any evidence of a single jammed PCA.95

Most of the recovered elevator control linkages were broken; however, this type of damage is typical following a high-speed water impact and underwater wreckage recovery operations. The shear rivets in the recovered elevator bellcrank assemblies were sheared in different directions; however, the Safety Board considers it likely that the rivets sheared as a result of impact or recovery-related forces. Nonetheless, on the basis of the examination of the structure alone, the absence or presence of a jammed or disconnected input linkage or a jam in the servo valve in one of the accident airplane's elevator PCAs could not be established.

However, ground tests, studies, and calculations showed that each of the first three failure scenarios would have resulted in airplane and flight control movements that were inconsistent with the accident airplane's elevator movements. Specifically, each of those three failure scenarios would have caused the failed elevator surface to move to, and remain at, a position consistent with a single functioning PCA operating at 100 percent of its maximum force. The failed elevator surface would resist being backdriven with a force equivalent to about 130 percent of a single functioning PCA and would not have responded to nose-up flight control inputs. If one of these scenarios occurred at the accident airplane's indicated airspeed at the time of the initial dive (280 knots), the failed elevator surface would have initially moved from its prefailure position (close to neutral) to about a 6º nose-down position.96 However, the initial elevator movement (for both elevator surfaces) on the accident airplane recorded during the accident sequence was to a nose-down position of only about 3.6°.97

The Safety Board also compared the recorded elevator movements following the initial upset to elevator movements resulting from the first three failure scenarios. As the airplane's speed increased after the initial upset, the maximum deflection value associated with the three failure scenarios would have decreased in response to the increased aerodynamic forces on that surface. However, subsequent movements of both elevator surfaces on the accident airplane deviated repeatedly, for sustained periods of time in both the nose-up and nose-down directions,98 from the maximum deflection values that the failure scenarios would have produced, at times exceeding the maximum deflection values by several degrees. As shown in figure 2, the elevator movement profile from the accident flight differs significantly throughout the accident sequence from the elevator movement profile that would have resulted from any of these three failure scenarios,99 indicating that neither elevator surface on the accident airplane was limited by a mechanical failure but, rather, that both surfaces were responding normally to flight control inputs. Therefore, the first three failure scenarios are inconsistent with the elevator movements recorded after the initial upset.

Similarly, the elevator movements that would have followed any variant of the fourth failure scenario are also inconsistent with the accident airplane's recorded elevator movements after the initial upset. The Safety Board notes that, in one of the four variations of this scenario (a jam in the aft portion of the elevator control cable combined with a cable break forward of the jam), the initial elevator positions match those on the accident airplane. However, this similarity between the failure scenario and the accident airplane's elevator movements lasts only a few seconds. For the remainder of that variation of the scenario (and for the entire duration of the other three variations of this failure scenario), the elevator positions are inconsistent with those of the accident airplane.

In addition, if one of the first three failure scenarios had occurred, the nonfailed surface would have responded immediately to any nose-up flight control inputs from either control column and would have resulted in an increase in the magnitude of the difference between the two elevator surface positions (because the failed surface would remain at its failure-induced position). If the fourth failure scenario had occurred, both elevator surfaces would have responded immediately to nose-up inputs from either control column. However, the FDR data from the accident flight showed that there was no significant nose-up elevator movement or difference between the two elevator surface positions for the first 28 seconds of the accident sequence--until the captain returned to the cockpit.100 If a failure had actually occurred, this would indicate that no attempts were made to recover the airplane for the first 28 seconds after the initial pitchdown. Further, after the captain returned to the cockpit, both elevator surfaces began moving together in the nose-up direction, indicating that neither surface was limited by a mechanical failure but, rather, that both surfaces were responding normally to flight control inputs. Similarly, later in the accident sequence, when the elevator split occurred, the right elevator deflected well beyond the maximum position possible for a failed elevator surface in any of the first three failure scenarios. However, the elevator movements during the split are well within the limits of a pilot-commanded movement.

After reviewing all of the inconsistencies between the effects of the four potential failure scenarios evaluated in depth, the actual behavior of the airplane, and the controllability of the airplane in the event of such failures, the Safety Board determined that none of these failure scenarios occurred during the accident sequence.101 Therefore, these four failure scenarios can be ruled out along with all of the other potential failure scenarios considered during this investigation.

The Safety Board also conducted simulations in which pilots from Boeing, EgyptAir, the Federal Aviation Administration (FAA), and the Board evaluated the controllability of the airplane following an initial upset that might have been caused by any of these failure scenarios. During these simulations, the pilots were consistently able to regain control of the airplane and return it to straight and level flight using normal piloting techniques, and the airplane could be trimmed to hands-off level flight. In fact, the 767's redundant actuation system is designed to allow pilots to overcome dual failures such as these.

Even though increased control forces were necessary, recovery could be accomplished by a single pilot using either the left or right control column.102 Further, the simulations also demonstrated that the airplane could climb to about 25,000 feet msl with the engines shut down, even with the speedbrakes extended. The simulation also documented that the engines could have been promptly restarted and (assuming there were no opposing pilot inputs) that the airplane could have been recovered during the climb after the recorders stopped recording. Although the Safety Board recognizes that the simulator did not duplicate the accident airplane's actual flight conditions in every way,103 such limitations are not uncommon in simulations, and the Board takes those limitations into account when evaluating simulator results. In this case, the Board determined that the differences were not significant and did not affect the validity of the results of the simulations.

Immediately after the airplane's initial nose-down dive, the relief first officer would have felt an immediate uncomfortable sensation as the airplane's load factor decreased to near 0 Gs. He should also have noted sudden changes in the airplane's pitch attitude, pitch rate, airspeed, and altitude. In response to these obvious cues, the relief first officer did not attempt to counter the dive by commanding nose-up elevator, a largely intuitive pilot response to initiate a recovery.

Nor did the relief first officer exhibit any audible expression of anxiety or surprise or call for help during the airplane's initial dive or at any time during the remainder of the recorded portions of the accident sequence. Further, the relief first officer did not respond to the captain's repeated question, "What's happening?" after the captain returned to the cockpit. Rather, he continued his calm repetitions of the phrase "I rely on God" (which began about 74 seconds before the airplane's dive began) for 2 to 3 seconds, and then became silent, despite the captain's repeated requests for information. The absence of any reaction from the relief first officer (such as anxiety or surprise, a nose-up elevator input to regain control of the airplane, or a request for assistance) to the airplane's sudden departure from cruise flight to a steep descent is not consistent with his encountering an unexpected mechanical problem. Whereas the captain's audible alarm and the content of his statements in reaction to the situation upon returning to the cockpit were consistent with the reaction of a pilot who has encountered an unexpected flight condition, the passive behavior of the relief first officer was not.

The primary radar data indicated that the airplane climbed for about 40 seconds after the FDR stopped recording before it rapidly descended again and impacted the ocean. Therefore, the relief first officer and captain had about 83 and 69 seconds, respectively, from the time the airplane began its initial nose-down pitch until it began its second (final) descent, in which to regain control of the airplane; return it to level flight and restart the engines; or at least establish the airplane in a gradual, controlled glide while attempting an engine restart. (If control of the airplane had been regained during this time, the flight crew would have had several minutes in which to restart the engines.) However, a successful recovery--although possible--was not accomplished.

In summary, the investigation did not reveal any evidence of a failure condition within the airplane's elevator system that would have caused or contributed to the airplane's initial pitchover or prevented the flight crew's successful recovery from the airplane's rapid descent. Further, the relief first officer's reaction was inconsistent with his having encountered an unexpected airplane anomaly. Therefore, the investigation determined that neither the nose-down elevator movements nor the failure to recover from those movements could be explained by a mechanical failure.

Pilot Action Scenario

Simulations showed that certain combinations of pilot inputs could result in elevator motions consistent with those recorded by the accident airplane's FDR and a flightpath consistent with the FDR and radar data for the accident airplane.104 Therefore, the Safety Board evaluated the actions of the pilots as recorded on the CVR, in the context of all of the evidence gathered in this investigation, to determine whether pilot action provided a possible explanation for the accident scenario.

Events Before and During the Initial Descent, While the Relief First Officer Was Alone in the Cockpit

About 20 minutes after takeoff (about 0140), the relief first officer suggested that he relieve the command first officer. A transfer of control this early in the flight was contrary to the EgyptAir practice typically agreed-upon by flight crews of waiting until 3 or 4 hours into the flight before relieving the command crewmembers. The command first officer initially reacted with surprise and resistance to the relief first officer's suggestion that he assume first officer duties at that time, indicating that the relief first officer's suggestion was unexpected. However, after some discussion, the command first officer agreed to the change, and sounds recorded by the CVR indicated that, about 0142, the command first officer vacated and the relief first officer moved into the first officer's seat.

About 0147, the relief first officer asked an unidentified crewmember to return a pen to another first officer, who was in the cabin. The unidentified crewmember agreed and left the cockpit. At 0148:03, the command captain excused himself from the cockpit, saying that he wanted to "take a quick trip to the toilet...before it gets crowded." While the command captain was excusing himself, the CVR recorded the sound of an electric seat motor, presumably the captain's, as he maneuvered to leave his seat and the cockpit.105 At 0148:18.55, the CVR recorded a sound similar to the cockpit door operating.

The Safety Board considered whether another flight crewmember might have been in the cockpit with the relief first officer during this time period. However, careful laboratory examination of the CVR recording indicated that the CVR did not record any speech or human sounds other than those attributed to the captain and relief first officer from 0148:30 until the end of the recording at 0150:38.47.106 The Board determined that the possibility that another person, especially a pilot, was present during the airplane's sudden transition from cruise flight to steep descent and did not audibly express surprise at the abrupt change in the flight situation (as the captain did when he returned to the cockpit) or offer help/suggestions on how to deal with the emergency situation was extremely unlikely. Therefore, the evidence indicates that the relief first officer was alone in the cockpit from about 0148:19, when the command captain left the cockpit, to 0150:06, when he returned to the cockpit.

Ten seconds after the unintelligible comment was made (at 0148:40), the relief first officer stated quietly, "I rely on God." At 0149:18, the CVR recorded a "whirring sound similar to [the] electric seat motor operating." Because the relief first officer's seat was likely moved into an aft position because the command first officer had vacated the seat, and in light of the autopilot disconnect and subsequent flight control movements, the whirring sound is consistent with the relief first officer moving his seat forward into a position from which he could manually fly the airplane.107 Thus, all manual flight control inputs made after 0148:19, until the command captain's return to the cockpit at 0150:06, must have been made by the relief first officer.

The absence of an autopilot disconnect warning tone on the CVR recording when the autopilot disconnected at 0149:45 is consistent with the autopilot being manually disconnected by rapidly double-clicking on the control yoke-mounted autopilot disconnect switch. Because the relief first officer was alone in the cockpit, the evidence indicates that he manually disconnected the autopilot. The Safety Board's examination revealed no evidence in the CVR, FDR, ATC, or radar data of any system malfunction, conflicting air traffic, or other event that might have prompted the relief first officer to disconnect the autopilot; therefore, there was no logical operational reason for the relief first officer to disconnect the autopilot while in cruise flight over the ocean. Further, as previously stated, the Board's testing and evaluation of the 767 elevator system showed that none of the failure modes examined during this investigation would have resulted in control column movements without concurrent identifiable movements of the elevators, which would have been observed in the FDR data. The FDR did not record any unusual or alarming elevator movements before the autopilot was disconnected; therefore, it is unlikely that the relief first officer was prompted to disconnect the autopilot because he sensed unusual control column movements.

Aside from some very slight elevator movements and a very gradual left roll, the airplane remained in level flight at flight level 330 for about 8 seconds after the autopilot was disconnected. As previously discussed in the section titled, "767 Autopilot Information," such slight movements are normal and expected when the autopilot is disengaged and the pilot takes manual control of the airplane. There was no indication of an upset or loss of control at this time.

At 0149:48, the relief first officer again quietly stated, "I rely on God." At 0149:53, the throttle levers were retarded (moved from their cruise power setting to idle). This throttle lever movement occurred at a rate that was more than twice that which the autothrottle can command. Further, the throttle levers moved 10º to 15º beyond the minimum position that the autothrottle would have been able to command at the existing flight conditions to the throttle levers' full aft idle stop, about 33º.108 Movement of the throttles aft of the autothrottle commanded position requires a manually applied force of about 9 pounds on the throttle levers to override the autothrottle servomotor clutch. Thus, it is apparent that the throttle lever movements at 0149:53 were caused by the relief first officer's manual inputs and were not the result of autothrottle commands.109

At 0149:54, the FDR recorded a very slight movement of the inboard ailerons and both elevator surfaces beginning to rapidly pitch nose down (to about 3.6° nose-down deflection). The nose-down elevator movement began after the throttle levers started to move to idle; therefore, the relief first officer did not move the throttle levers to idle in response to the nose-down elevator movement. As previously noted, the relief first officer did not audibly express surprise or seem anxious or disturbed by the airplane's sudden and extreme nose-down movement or the reduction in load factor to near 0 G, nor did he call for help during the accident sequence. Again, there was no evidence in the CVR, FDR, ATC, or radar data of any system malfunction, conflicting air traffic, or other event that would have prompted the relief first officer to adjust the throttle levers at all, let alone take an action as drastic as moving the throttle levers to the idle position while in cruise flight at night over the ocean or to then command a sustained nose-down elevator movement.

About 11 seconds after the initial nose-down movement of the elevators, the FDR recorded additional (larger) movements of the inboard ailerons and the elevators started to move further in the nose-down direction, decreasing the airplane's load factor to negative G loads. The relief first officer would have been gripping the control wheel with his hand(s) when he applied these significant nose-down elevator control column inputs. It is unlikely that he could make such significant control column inputs without (intentionally or unintentionally) also affecting the control wheel's lateral position and thus providing some input to the ailerons. Therefore, these inboard aileron movements, and those that occurred at 0149:54 (both of which were coincident with changes in the relief first officer's inputs to the control column), are consistent with evidence indicating that the relief first officer was providing manual inputs to the flight controls during the accident sequence.

Events After the Command Captain Returned to the Cockpit

Immediately after this increase in nose-down elevator movement, at 0150:06, the CVR recorded the command captain exclaiming, "What's happening? What's happening?," as he returned to the cockpit.110 At 0150:08, the captain repeated his question. While the captain was still speaking and moving toward his seat in the forward portion of the cockpit (at 0150:07 and again at 0150:08), the relief first officer quietly repeated, "I rely on God."111 However, the relief first officer did not answer the captain's question. The Safety Board considers it unlikely that the captain--who was likely focusing on getting into his seat, troubleshooting the upset, and attempting to regain control of the airplane--would have suspected at this point that the relief first officer's actions were directly contributing to the airplane's dive.112 Rather, the captain likely would have assumed that the relief first officer was also attempting to regain control of the airplane and would work cooperatively with him.

As previously discussed, the relief first officer's passive behavior in response to the airplane's nose-down movements and the captain's questions is not consistent with what would be expected from a pilot who was dealing with an unexpected or undesired airplane problem. To the contrary, the timing of the increased nose-down elevator movement and the corresponding decrease in load factor was consistent with the relief first officer having increased the forward control column pressure when the captain returned to the cockpit.

At 0150:15, as the airplane continued to descend rapidly in a 40° nose-down attitude, the captain again asked, "What's happening, [relief first officer's first name]? What's happening?" Again, the relief first officer did not respond to the captain's question. Although the relief first officer remained unresponsive to the captain's queries, there is no specific evidence to indicate that the captain suspected at this point that the relief first officer's actions were causing the airplane's dive.

At the same time, as the airplane was descending through about 27,300 feet msl, both elevator surfaces began moving to reduced nose-down deflections. Shortly thereafter, the airplane's rate of descent began to decrease. Because there was no evidence that the relief first officer had attempted to regain control of the airplane before this, the Safety Board considers it likely that these movements were the result of nose-up flight control inputs made by the captain after he returned to the cockpit.113 Six seconds later (at 0150:21), both elevator surfaces passed through their neutral positions into nose-up deflections. However, less than 1 second later, the right surface reversed its motion and moved back in the nose-down direction, and the left surface continued to move in the nose-up direction.

According to Boeing's tests and research, with the elevator PCAs operating normally, the accident airplane's elevators would have only been minimally affected by the aerodynamic forces that would have resulted from the small sideslip angle, roll rates, and the Mach numbers that existed during the accident sequence. Therefore, it follows that the elevator split recorded by the FDR was the result of flight control inputs to each elevator surface and not the result of aerodynamic forces on those surfaces.114 (In contrast, Boeing indicated that an outboard aileron split recorded between 0150:27 and 0150:32 could be explained by the aerodynamic effects of the small sideslip angles and roll rates calculated to have been present at that time.)115

Testing confirmed that the left and right elevator surfaces could be moved in different directions by differential column movements from the relief first officer and captain in the cockpit. As intended by the elevator control system design, the elevators would split, each surface following the movements of the control column on its side (the left elevator moving in response to the left column movement, and the right elevator moving in response to the right column movement). The opposing control column inputs likely existed during the 7 to 8 seconds before the elevator split (when both elevators were moving in a trailing-edge-up direction); however, the elevator split would not occur until the difference between the two control column forces was great enough to engage the override mechanism. Tests conducted in a 767 simulator and airplane (on the ground) demonstrated that pilots with heights and weights similar to those of the command captain and relief first officer could apply enough force on the control column to produce and maintain the split elevator condition recorded by the FDR.

The captain's actions just after the elevator split began were consistent with an attempt to recover the airplane and the relief first officer's were not. In rapid sequence, just after the elevator split began, the engine start lever switches were moved to the cutoff position, the throttle levers were advanced to full throttle, and the speedbrakes were deployed.116 After the throttle levers were advanced (but the engines did not respond), the captain reacted with surprise, asking the relief first officer, "What is this? What is this? Did you shut the engine(s)?"117 The timing and direction of the left elevator motions during this time suggest that the captain, who had likely been using both hands to pull aft on the left control column, released his right hand to advance the throttles and deploy the speedbrakes, resulting in a decrease in his total aft pressure on the control column, which was reflected in the decrease in the left elevator's nose-up deflection that was recorded by the FDR at this time. Subsequently, when the captain likely had returned his right hand to the control column, the FDR recorded a corresponding increase in the left elevator's nose-up deflection. As previously stated, tests and simulations demonstrated that a pilot seated in the captain's position could easily have advanced the throttles, moved his hand a little to the left, and deployed the speedbrakes in the 3 to 4 seconds it took for these events to occur.

Concurrent with the brief downward motion of the left elevator that was recorded when the throttles were advanced and the speedbrakes deployed, a brief downward motion of the right elevator was recorded. This movement of the right elevator suggests that when the captain's aft pressure on the left control column decreased, the relief first officer's sustained forward pressure on the right control column caused that column to move forward briefly. Although it would have been physically possible for the relief first officer to have advanced the throttles and deployed the speedbrakes, the evidence does not support the notion that the relief first officer performed these actions. Rather, the evidence indicates that the relief first officer moved the engine start lever switches to the cutoff position (a counterproductive action, in terms of recovery), whereas the captain deployed the speedbrakes in an attempt to arrest the airplane's descent.

Additionally, the surprised reaction from the captain when the engines did not respond to the throttle movement ("What is this? What is this? Did you shut the engine(s)?") suggested that it was he (not the relief first officer) who advanced the throttle levers. This response clearly indicated that the captain was unaware that the engine start lever switches had been moved to the cutoff position, that such an action was at odds with his intentions, and that it was, therefore, not part of a mutual, cooperative troubleshooting exercise between the captain and relief first officer.

At 0150:26.55, the captain stated, "Get away in the engines," and at 0150:28.85, he stated, "shut the engines."118 At 0150:29.66, the relief first officer responded for the first (and only) time after the captain returned to the cockpit, stating, "It's shut." Between 0150:31 and 0150:37, the captain repeatedly asked the relief first officer to "pull with me" on the control column. However, the FDR data indicated that the elevator surfaces remained in a split condition (with the left surface commanding nose up and the right surface commanding nose down) until the last data were recorded by the FDR at 0150:36.64.

As with the earlier portion of the accident sequence (before the captain's return to the cockpit), the relief first officer's responses during this portion of the accident sequence did not indicate that he was surprised or disturbed by the events. Similarly, his rate of speech and fundamental frequency when he repeated, "I rely on God," and stated, "It's shut," did not indicate any significant increase in his level of psychological stress. In contrast, the captain's fundamental frequency was about 65 percent higher when he repeatedly asked the relief first officer to "pull with me" during the elevator split period than it was during routine flight, reflecting an increased level of psychological stress.

As previously discussed, simulations showed that even if a failure condition had affected the elevator system, it would have been possible to regain control of the airplane at any time during the recorded portion of the accident sequence and to have restarted the engines and recovered the airplane during the climb after the recorders stopped. However, those simulations assumed that there were no opposing pilot inputs. The captain's failure to recover the airplane can be explained, in part, by the relief first officer's opposing flight control inputs. It is possible that efforts to recover the airplane after the airplane lost electrical power were also complicated by the loss of electronic cockpit displays.

In summary, the evidence establishes that the nose-down elevator movements were not the result of a failure in the elevator control system or any other airplane system but were the result of the relief first officer's manipulation of the airplane controls. The evidence further indicates that the subsequent climb and elevator split were not the result of a mechanical failure but were the result of pilot inputs, including opposing pilot inputs where the relief first officer was commanding nose-down and the captain was commanding nose-up movement. The Safety Board considered possible reasons for the relief first officer's actions; however, the Board did not reach a conclusion regarding the intent of or motivation for his actions.

Summary

  1. The accident airplane's nose-down movements did not result from a failure in the elevator control system or any other airplane failure.

    There was no evidence of any failure condition within the elevator system of the accident airplane that would have caused or contributed to the initial pitchover or prevented a successful recovery.

    No mechanical failure scenario resulted in airplane movements that matched the flight data recorder data from the accident airplane.

    Even assuming that one of the four examined failure scenarios that the investigation evaluated in depth had occurred, the accident airplane would still have been recoverable because of the capabilities of the Boeing 767's redundant elevator system.
     

  2. The accident airplane's movements during the initial part of the accident sequence were the result of the relief first officer's manipulation of the controls.

    At the relief first officer's suggestion, a transfer of control at the first officer's position occurred earlier than normal during the accident flight.

    The relief first officer was alone in the cockpit when he manually disconnected the autopilot and moved the throttle levers from cruise to idle; there was no evidence of any airplane system malfunction, conflicting air traffic, or other event that would have prompted these actions.

    The nature and degree of the subsequent nose-down elevator movements were not consistent with those that might have resulted from a mechanical failure but could be explained by pilot input.

    There was no apparent reason for the relief first officer's nose-down elevator inputs.

    The relief first officer's calm repetition of the phrase "I rely on God," beginning about 74 seconds before the airplane's dive began and continuing until just after the captain returned to the cockpit (about 14 seconds into the dive), without any call for help or other audible reaction of surprise or alarm from the relief first officer after the sudden dive is not consistent with the reaction that would be expected from a pilot who is encountering an unexpected or uncommanded flight condition.

    The absence of any attempt by the relief first officer to recover from the accident airplane's sudden dive is also inconsistent with his having encountered an unexpected or uncommanded flight condition.

    The relief first officer's failure to respond to the command captain's questions ("What's happening? What's happening?") upon the captain's return to the cockpit is also inconsistent with the reaction that would be expected from a pilot who is encountering an uncommanded or undesired flight condition.
     

  3. The accident airplane's movements after the command captain returned to the cockpit were the result of both pilots' inputs, including opposing elevator inputs where the relief first officer continued to command nose-down and the captain commanded nose-up elevator movements.

    Nose-up elevator movements began only after the captain returned to the cockpit.

    Testing showed that recovery of the airplane was possible but not accomplished.

    Seconds after the nose-up elevator movements began, the elevator surfaces began moving in different directions, with the captain's control column commanding nose-up movement and the relief first officer's control column commanding nose-down movement.

    After the elevator split began, the relief first officer shut down the engines.

    The captain repeatedly asked the relief first officer to "pull with me," but the relief first officer continued to command nose-down elevator movement.

    The captain's actions were consistent with an attempt to recover the accident airplane and the relief first officer's were not.

Probable Cause

The National Transportation Safety Board determines that the probable cause of the EgyptAir flight 990 accident is the airplane's departure from normal cruise flight and subsequent impact with the Atlantic Ocean as a result of the relief first officer's flight control inputs. The reason for the relief first officer's actions was not determined.

 


1.Under the provisions of Annex 13 to the Convention on International Civil Aviation, the investigation of an airplane crash occurring in international waters falls under the jurisdiction of the airplane's country of registry (in this case, Egypt). At the request of the Egyptian Government, the National Transportation Safety Board assumed full responsibility for the investigation. Parties to the investigation included the Federal Aviation Administration (FAA), Boeing Aircraft Company, and Pratt & Whitney (P&W) Aircraft Engines. The Egyptian Civil Aviation Authority (ECAA) designated an accredited representative to the investigation on behalf of the Egyptian Government. EgyptAir provided a technical advisor to the ECAA and the investigation. The Federal Bureau of Investigation (FBI) also assisted in the investigation. The National Oceanic and Atmospheric Administration, U.S. Navy, and U.S. Coast Guard assisted in the search and recovery operations.

2.At 0200 EDT on October 31, 1999, local time in the eastern United States changed from 0200 EDT to 0100 EST. Unless otherwise indicated, all times in this document are EST, based on a 24-hour clock.

3.FL 230 is 23,000 feet mean sea level (msl), based on an altimeter setting of 29.92 inches of mercury.

4.This clearance resulted in EgyptAir flight 990 passing through a type of special-use airspace referred to as a "warning area." New York ARTCC and U.S. Navy records indicated that the warning area was not in use by the U.S. Navy at the time of the accident. For additional information, see the Air Traffic Control Group Chairman's Factual Report and its attachments.

5.A complete, English-language transcript of the CVR is attached to this report.

6.When two flight crews are used, EgyptAir designates one crew as the command flight crew and the other as the relief flight crew. Although EgyptAir has no written or formal procedures for command/relief flight crew transitions, postaccident interviews with EgyptAir flight crewmembers indicated that the command and relief flight crews typically agreed upon transfer-of-control procedures for a flight before departure. The interviews indicated that the most common procedure involved the command flight crew flying the airplane for the first 3 or 4 hours of the flight, then the relief flight crew assuming control until about 1 to 2 hours before landing. The command flight crew would then resume control of the airplane and complete the flight.

7.Postaccident interviews with several EgyptAir pilots indicated that the relief first officer was often addressed as "captain" as a title of respect because he had instructed many of the EgyptAir pilots at the Egyptian flight training institute before he was hired by EgyptAir.

8.This was the last transmission to ATC from the accident airplane. Although some irregularities in ATC handling were noted during the investigation, they were not relevant to the accident. For additional information, see the Air Traffic Control Group Chairman's Factual Report and its attachments and addenda.

9.The context of this statement indicates that the relief first officer was talking to the command first officer and that the "new first officer" to whom the relief first officer was referring was a pilot who had been in the cockpit earlier in the flight and who was seated in the cabin at the time of this statement. (According to the Cockpit Voice Recorder Group Chairman's Factual Report, an Arabic-speaking member of the Cockpit Voice Recorder Group identified the voices of six flight crewmembers and one flight attendant recorded in the cockpit at various times during the accident flight.)

10.According to the CVR transcript, "the five Arabic speaking members of the [CVR] group concur that they do not recognize this as an Arabic word, words, or phrase. The entire group agrees that three syllables are heard and the accent is on the second syllable. Four Arabic speaking group members believe that they heard words similar to 'control it.' One English speaking member believes that he heard a word similar to 'hydraulic.' The five other members believe that the word(s) were unintelligible." For additional information regarding the computer analysis of this comment, see the section titled, "Cockpit Voice Recorder."

11.This phrase (recorded on the CVR in Arabic as "Tawakkalt Ala Allah") was originally interpreted to mean "I place my fate in the hands of God." The interpretation of this Arabic statement was later amended to "I rely on God." According to an EgyptAir and ECAA presentation to Safety Board staff on April 28, 2000, this phrase "is very often used by the Egyptian layman in day to day activities to ask God's assistance for the task at hand."

12.No autopilot disconnect warning tone was heard on the CVR recording. According to the system design, an autopilot disconnect warning is generated unless the autopilot is disconnected manually, either by clicking the control yoke-mounted autopilot disconnect switch twice within 0.5 second or by moving the autopilot switch on the instrument panel.

13.Throughout the FDR data for the accident airplane (including data recorded during uneventful portions of the accident flight and during previous flights and ground operations), small (less than 1°) differences between the left and right elevator surface positions were observed. The left and right elevator surface movements were consistent (that is, moved in the same direction about the same time) where these offsets were observed. According to Boeing, there are several factors that could result in differences between the left and right elevator surfaces, including rigging of the elevator control system, tolerances within the system's temperature compensation rods, routing differences between the left and right elevator control cables, friction distribution within the system, the accuracy of the sensors used to measure elevator position, and differences in FDR sampling times for the left and right elevator parameters.

14.Although earlier statements made by the relief first officer were recorded by the hot microphone at the first officer's position, the "I rely on God" statements were not, which was consistent with these statements being spoken relatively quietly. For additional information, see the section titled, "Audio Information Recorded by First Officer's Hot Microphone."

15.An airplane's normal load factor is approximately perpendicular to the airplane's wings. Although the terms "vertical load factor," "vertical acceleration," and "normal load factor" are often used interchangeably, for the purposes of this document, the term "load factor" is used.

16.A G is a unit of measurement of force on a body undergoing acceleration as a multiple of its weight. The normal load factor for an airplane in straight and level flight is about 1 G. As the load factor decreases from 1 G, objects would become increasingly weightless, and at 0 G, those objects would float. At load factors less than 0 G (negative G), loose objects would float toward the ceiling, and, at -1 G, those objects would accelerate toward the ceiling.

17.The cessation of the FDR and CVR recordings was consistent with the loss of electrical power to the recorders that resulted from the engines being shut off. Although the FDR recorded different parameters at different sampling rates and at slightly different times, the last subframe of recorded data was recorded at 0150:36.64.

18.According to calculations based on FDR data, the airplane's maximum rate of descent was about 39,000 feet per minute (fpm); this rate was recorded at 0150:19.

19.The engine start lever switches control the flow of fuel to the engines and are located on the center console between the pilot positions. When these levers are moved to the cutoff position, fuel flow to the engines is stopped, and the engines stop operating within about 5 or 6 seconds. They are spring-loaded, lever-lock design switches that must be pulled up to release from one detent before they can be moved to the other position, where they will engage in another detent.

20.The Safety Board's simulator tests demonstrated that an EgyptAir pilot similar in size to the command captain was able to occupy the captain's seat without physical interference; brace himself against the center console or floor structure; readily apply back pressure on the control column; and reach the throttles, speedbrakes, and other controls on the central console with the seat in its aft position. (The Board recognizes that the simulations could not duplicate the near 0 G loads recorded by the FDR during the accident sequence; however, such near 0 G loads were present only momentarily after the recovery started and should not have substantially affected the fore-and-aft forces either pilot could generate once normally seated and effectively braced.)

21.According to participants in the Cockpit Voice Recorder Group (which included several Arabic/English speakers), occasionally the direct translation of Arabic words into English resulted in awkward or seemingly inappropriate phrases. Throughout the CVR transcript, the Cockpit Voice Recorder Group provided as direct a translation as possible; however, it did not attempt to interpret or analyze the words or the intent of the speaker.

22.Surveillance radars fall into two categories: primary (also known as "search") and secondary (also known as "beacon"). Secondary radar broadcasts an interrogation signal to which equipment on board an airplane automatically responds by transmitting information to the ground-based site for processing and display. Secondary radar returns contain an identification code and altitude data. Primary radar broadcasts radio waves and detects the reflections of the waves off objects (including airplanes). Primary radar reflections do not contain any unique identification information. (For additional information, see the Aircraft Performance Group Chairman's Aircraft Performance Study.)

23.For more detailed information regarding the background and recent activities of all EgyptAir flight 990 crewmembers, see the Operational Factors Group Chairman's Factual Report and its addendum and the Human Performance Group Chairman's Factual Report and its addendum.

24.In 1971, United Arab Airlines was renamed EgyptAir.

25.There was no mention of treatment for chronic back problems in the captain's records at EgyptAir.

26.The relief first officer did not upgrade to captain even though he was eligible to do so in the early 1990s. Colleagues stated that he did not upgrade because he preferred the benefits of seniority in the first officer position. According to EgyptAir, the relief first officer became ineligible to upgrade after his 55th birthday in February 1995.

27.The 767-300 is a low-wing, twin-engine, transport-category airplane.

28.A flight cycle is one complete takeoff and landing sequence.

29.The cables for the captain's (left-side) system are routed below the floor boards, and the cables for the first officer's (right-side) system are routed above the cabin ceiling.

30.The Safety Board thoroughly examined the dual elevator PCA failure scenarios during its investigation of this accident. For more information, see the section titled, "Potential Causes for Elevator Movements During the Accident Sequence."

31.The term "backdriving" refers to the effect of aerodynamic forces that act on the elevator surface and move the surface in the direction opposite to that being commanded (by the two failed PCAs, in the case of a dual elevator PCA failure). This backdriving force increases as an airplane's airspeed increases.

32.For additional information, see the section of this report titled, "Potential Causes for Elevator Movements During the Accident Sequence."

33.Compliance in the elevator system can occur as a result of cable stretch; yield, give, or elastic deformation in linkages (that does not damage the linkages but allows additional motion); and variations in tolerance buildups throughout the system.

34.A manual force of about 9 pounds (lbs) is required to override the clutch.

35.The October 30, 1999, EgyptAir flight from Cairo was scheduled to land at JFK but diverted to EWR because of weather.

36.During interviews conducted 3 days after the accident, the captain that had flown the airplane from EWR to LAX on October 30, 1999, described several noncritical anomalies (a deactivated thrust reverser, an intermittent air conditioning pack "inoperative" light, a full aft lavatory holding tank, and the autopilot anomaly previously mentioned) but stated that the airplane was "almost perfect." The first officer of the flight to LAX did not describe the autopilot anomaly.

37.According to Boeing, the autopilot is designed to capture the glideslope signal when the proper autopilot mode is selected if the airplane is within 80 feet of the glideslope.

38.Through its SDR program, the FAA collects information about mechanical failures from reports submitted by aircraft operators or maintenance facilities, as required by regulations.

39.The SDRs included two reports of anomalous elevator behavior on the same United Airlines 767, the first incident occurred on September 12, 1994, and the second on June 20, 1996. Both incidents involved "stiff" or "frozen" elevator flight controls, and, in both cases, the pilots regained control of the elevator by applying higher-than-normal pressure on the control column. Postincident examination of the elevator system components revealed no discrepancies.

The Safety Board is also aware of the following two similar, more recent incidents:

  1. On March 27, 2001, an American Airlines 767 experienced elevator control difficulties during an approach to land. The pilots landed safely using horizontal stabilizer trim for pitch control and reported that as they taxied to the gate, they "broke [the elevator] free" by applying a higher-than-normal force on the control column. Postincident examination revealed no discrepancies in the elevator's mechanical flight control rigging, PCAs, pushrods, bellcranks, or shear rivets; however, during postincident examination, investigators observed water dripping directly on elevator system components in the empennage.
  2. On April 23, 2001, the pilots of another 767 experienced elevator control binding during the approach to land. The pilots applied additional force to the control column, and the elevator binding released. Postincident examination revealed no evidence of mechanical anomalies; however, investigators observed an accumulation of water and ice in the empennage around the elevator system components.
Additional tests indicated that water could freeze on the elevator components and create the effects described by these flight crews and observed in the FDR data of the two recent incidents. (FDR data were not available for the two earlier incidents.) The Safety Board compared the FDR data from the two recent incidents with that from EgyptAir flight 990 and found no similarities. Boeing and the FAA are evaluating possible corrective actions related to preventing or limiting water from entering the 767 empennage, freezing at altitude, and impinging on elevator system components.

40.This anomalous condition was discovered when a drooping elevator surface was observed during a preflight inspection; there were no reports of in-flight anomalies before this discovery. The air carrier's maintenance personnel found sheared rivets in the bellcranks, which they repaired. The system was functionally checked after the repair, and the airplane was returned to service. The air carrier reported the anomalous condition and repair to Boeing and has reported no further anomalies. FDR data were not available.

41.For the purposes of this report, the compressible links are described as "bottomed out" when they have been deflected to the full extent of their travel in either direction.

42.The single hydraulic system maintenance check tests the operation of each PCA individually by powering each of the airplane's three hydraulic systems, one at a time. An inoperative elevator PCA will not operate the elevator when powered by its hydraulic system. A PCA with a failed bellcrank shear rivet will not operate the elevator properly.

43.Indications of an improperly rigged PCA can occur as a result of yielded or failed shear rivets in a bellcrank assembly.

44.The fourth channel of audio information recorded by the CVR is usually recorded through audio equipment at a cockpit jumpseat position. The FAA does not require a fourth channel to be installed/used on airplanes equipped with CVRs.

45.The Safety Board uses the following categories to classify the levels of CVR recording quality: excellent, good, fair, poor, and unusable.

  • An excellent recording is one that is very clear and easily transcribed.
  • A good recording is one in which most of the crew conversations can be accurately and easily understood. The transcript that is developed may indicate unintelligible several words or phrases. Any loss in the transcript can be attributed to minor technical deficiencies or momentary dropouts in the recording system or to a large number of simultaneous cockpit/radio transmissions that obscure each other.
  • A poor recording is one in which a transcription is nearly impossible because a large portion of the recording is unintelligible.
The quality of audio information recorded by the hot microphone at the first officer's position is discussed further later in this report.

46.The captain apparently did not use the hot microphone system; however, depending on the nature and volume of the captain's communications, the sounds were recorded by the CAM.

47.For additional sound spectrum and speech study information, see the Cockpit Voice Recorder Group Chairman's Factual Report/Sound Spectrum Study and the Speech Examination Study Factual Report.

48.The study stated that the exact stowage location of the headset was unknown; however, according to an EgyptAir representative, it would normally be stowed in the storage console, which located at the first officer's right side, or in his flight bag, which is located just aft of the storage console.

49.Formants, which determine many aspects of perceived speech, are frequencies at which the vocal tract above the larynx (acting as a filter because of its normal modes of vibration) will allow maximum energy to pass from the sound produced by the vocal cords. Formant dispersion refers to the relative spacing between successive formants.

50.For additional information, see Williams, C. E. and Stevens, K. N. 1981. "Vocal Correlates of Emotional States." Speech Evaluation in Psychiatry . Grune & Stratton. New York, New York; Ruiz, R.; Legros, C.; and Guell, A. 1990. "Voice Analysis to Predict the Psychological or Physical State of a Speaker." Aviation, Space, and Environmental Medicine. Vol. 61. p. 266-71; Johannes, B.; Salnitski, V. P.; Gunga, H.; and Kirsch, K. 2000. "Voice Stress Monitoring in Space--Possibilities and Limits." Aviation, Space, and Environmental Medicine. Vol. 71. p. A58-65; Brenner, M.; Doherty, E. T.; and Shipp, T. 1994. "Speech Measures Indicating Workload Demand." Aviation, Space, and Environmental Medicine . Vol. 65. p. 21-6; Brenner, M.; Mayer, D.; and Cash, J. 1996. "Speech Analysis in Russia." Methods and Metrics of Voice Communications . Ed. B. G. Kanki and O. V. Prinzo. Department of Transportation, Federal Aviation Administration, and Office of Aviation Medicine. DOT/FAA/AM-96/10. Washington, DC; and National Transportation Safety Board. 1999. Uncontrolled Descent and Collision with Terrain, USAir Flight 427, Boeing 737-300, N513AU, near Aliquippa, Pennsylvania, September 8, 1994. Aircraft Accident Report. NTSB/AAR-99/01. Washington, DC.

51.For additional information, see NTSB/AAR-99/01.

52.As previously discussed, the last FDR parameter was recorded at 0150:36.64, and the cessation of the FDR data was consistent with the loss of electrical power that resulted from the engines being shut off.

53.Although control column position was not recorded by the FDR, the Safety Board's testing and evaluation of the 767 elevator system showed that any movement occurring at the control columns would have resulted in concurrent, identifiable movements of the elevators, which would have been recorded on the FDR. For additional information, see Flight Data Recorder Group Chairman's Factual Report and its attachments. Also, see the section of this report titled, "Tests and Research," for a discussion of the airplane's performance during the emergency/accident sequence, as determined by the Safety Board's evaluation of the available FDR, radar, weather, and airplane performance data.

54.In total, about 37 percent of the total elevator surface area was recovered.

55.For additional information, see the section titled, "Review of Radar Data."

56.The right elevator center PCA was identified as such because of its location in recovered horizontal stabilizer wreckage. The right elevator outboard PCA was identified as such by EgyptAir personnel, who matched the PCA's S/N to their maintenance documents for the accident airplane. The condition of the other two recovered PCAs precluded identification of their location on the airplane.

57.The Safety Board recognizes that a jam between two surfaces can occur without leaving any physical evidence. However, as discussed in the Board's report on the September 8, 1994, accident involving USAir flight 427, tests conducted in connection with that accident investigation showed that physical evidence of a jam was always observed after tests involving hardened steel chips jammed and/or sheared in a PCA.

58.As previously discussed, the shear rivets are designed to fail when they are subjected to about 50 lbs of force or more at the control column, the PCA is jammed, and the compressible links are bottomed out. In addition, shear rivets may fail as a result of impact or recovery-related forces.

59.For additional information, see Systems Group Chairman's Factual Report and its appendixes and addendum, Materials Laboratory Factual Report, and Structures Group Chairman's Factual Report and its appendixes and addendum.

60.For additional information, see Powerplants Group Chairman's Factual Report.

61.For additional information, see the Aircraft Performance Group Chairman's Aircraft Performance Study.

62.Throughout the FDR data for the accident airplane (including data recorded during uneventful portions of the accident flight and during previous flights and ground operations), small (generally less than 1°) differences between the left and right elevator surface positions were observed. The elevator surface movements were consistent (that is, moved in the same direction about the same time) where these offsets were observed.

63.As previously mentioned, at the accident airplane's flight conditions at the beginning of the accident sequence, the minimum autothrottle commanded throttle lever position would have been between 40º and 50º; this value would have decreased as the airplane's airspeed increased. The maximum autothrottle commanded throttle lever movement rate for a normally functioning autothrottle system is 10.5º per second.

64.Elevator movement in the TED direction would result in a decrease in the airplane's lift and load factor and an increase in the airplane's nose-down attitude.

65.The Safety Board's examination of Boeing's 767 certification flight test data revealed that the low oil pressure warnings for P&W 4060 engines would occur when the engine's oil pressure drops below 70 lbs per square inch, as occurred when the accident airplane was operating at low (near 0) load factors. (For additional information, see Powerplants Group Chairman's Factual Report and appendixes 1 through 8.)

66.According to Boeing, these movements were consistent with the effects of blowdown on those surfaces as documented during flight tests. However, the outboard aileron split recorded by the FDR after about 0150:27, which is discussed later in this section, was not consistent with the flight test data.

67.As previously discussed, Safety Board simulations demonstrated that a pilot in the left seat could have moved his right hand from the control wheel to the throttle, advanced the throttles, moved his hand a little to the left, and deployed the speedbrakes in the 3 to 4 seconds it took for these events to occur.

68.Wind tunnel tests and computational fluid dynamics analyses show that a small sideslip angle and/or roll rate could produce large changes in the aerodynamic forces acting on the outboard ailerons at speeds approaching Mach 1.0, but these forces would not likely be strong enough to cause the split elevator condition recorded by the accident airplane's FDR. For additional information, see Aircraft Performance - Addendum #1.1, Addendum to Group Chairman's Aircraft Performance Study, including appendixes B and C (correspondence from Boeing, dated April 12 and 16, 2001).

69.The simulator data are based on wind tunnel tests and updated with available flight test data. The maximum Mach number for which the simulator is programmed (Mach 0.91) corresponds to the airplane's never-exceed airspeed. The maximum speed calculated for the accident airplane during the accident sequence was Mach 0.99 at 0150:23. To evaluate the performance of the airplane at Mach numbers greater than 0.91, the simulator's database was adjusted to reflect extrapolations, based on 777 wind tunnel tests. (The 767 and 777 have aerodynamically similar horizontal stabilizers and elevators.)

70.For additional information, see Systems Group Chairman's Factual Report and its appendixes and addendums, Flight Data Recorder Group Chairman's Factual Report and its attachments, Cockpit Voice Recorder Group Chairman's Factual Report and Sound Spectrum Study, and the Aircraft Performance Group Chairman's Aircraft Performance Study and its attachments and addendum.

71.The cable-related failures considered included a single failed elevator body cable; a failed slave cable; a failed component or other object falling on elevator cables; a cable tension regulator failure; an aft pressure bulkhead failure, resulting in cable displacement; and a cable break combined with a jam in the same cable.

72.The elevator PCA-related failure scenarios considered included an input rod/cable jammed at an offset position (position jam), an input arm for a single PCA jammed at an offset position to command a specific control surface rate of movement (rate jam), failure of the bellcrank assemblies on all three of the PCAs on a single elevator surface, jam of the input linkage or servo valve of one PCA with a high breakout force compressible link (a high breakout force compressible link would allow more force to be transmitted to the input linkages of the nonfailed side before compressing and negating the jammed PCA), disconnection of the input linkages to two of the three PCAs on a single elevator surface, a jam of the input linkage or servo valve on one PCA and the disconnection of the input linkage to another PCA on a single elevator surface, and a jam of the input linkages or servo valves in two of the three PCAs on a single elevator surface.

73.The other system-related failures considered included erroneous stick nudger activation; air in the hydraulic system and elevated return pressure; hydraulic system failure to one surface; elevator position transducer disconnect, resulting in erroneous indications on the FDR of an elevator surface offset or split; a single linkage disconnect downstream of feel unit; a failure of the elevator feel unit's attachment to aircraft structure; electromagnetic interference (EMI); and an autopilot malfunction such as a servo jam, resulting in a hardover autopilot output.

74.For additional information, see Systems Group Chairman's Factual Report and its appendixes.

75.EMI is electromagnetic radiation that is emitted and/or received by an electronic device and adversely affects the performance of that device or other devices.

76.An autopilot malfunction was ruled out as a potential cause of the elevator movements because the autopilot was disconnected before the beginning of the nose-down elevator movements. Even assuming that the autopilot was engaged during the accident sequence, the elevator movements recorded by the FDR exceeded the maximum inputs that could be commanded by the autopilot. EMI was ruled out because elevator surface movements are not electrically actuated (and, therefore, are not susceptible to the effects of EMI) except through the autopilot. (When the autopilot is not engaged, elevator surface movements on the 767 are mechanically signaled and hydraulically actuated.)

77.For a detailed explanation of why a failed surface would deflect to this position, see the Aircraft Performance Group Chairman's Factual Report and its addendum 1.1.

78.Elevator hinge moment data provided by Boeing were used to estimate the 767 elevator blowdown positions during the first three failure scenarios. Boeing extrapolated available elevator data based on Boeing 777 wind tunnel data, which were available for Mach numbers 0.91, 0.94, and 0.96. (As previously stated, the 767 and 777 have aerodynamically similar horizontal stabilizers and elevators.)

79.Feel force is the amount of force generated by the aircraft's feel-and-centering unit. In normal operation, the feel force is a function of control column deflection and aircraft flight condition. If a jam of the input linkages or servo valves in two of the three PCAs on a single elevator surface occurred, the airplane's feel-and-centering unit would provide a force to oppose the forces needed to deflect the compressible links on the input side of the failed elevator PCAs.

80.To avoid shearing the test airplane's bellcrank rivets during the ground tests evaluating the two failure scenarios that involved jammed PCA linkages and/or servo valves, full travel of the elevator surface was not commanded. However, a study of the elevator control system indicated that full travel of the nonfailed surface could have been achieved under these two failure scenarios, if commanded.

81.The simulator's cockpit displays did replicate the visual cues (cockpit instrument displays and out-the-window presentation) that would have been present during actual flight.

82.For additional information, see Systems Group Chairman's Factual Report and its addendum regarding the ground and simulation testing.

83.EgyptAir provided the Safety Board with a presentation and several documents and letters that documented its position, including the following: a presentation, dated April 28, 2000; a formal submission, dated August 11, 2000; and a document, dated January 12, 2001, titled, "Response of EgyptAir to October 31, 2000, Submission of The Boeing Company Regarding the EgyptAir Flight 990 Investigation." These documents and letters are available in the public docket for this accident.

84.The Egyptian Government also provided the Safety Board with additional documents, including the following: (1) the Egyptian Government's comments regarding the Board's draft report of this accident, (2) the Egyptian Government's own report regarding this accident, and (3) the Egyptian Government's addendum to its report regarding this accident. Although these documents are not submissions, and therefore are not discussed in this section, they are available for review. The Egyptian Government's comments regarding the Board's draft report are attached to this report, and the Egyptian Government's accident report and its addendum are available in the public docket for this accident.

85.For its analysis of the elevator motions, EgyptAir used the methods of Roskam (Airplane Design, Part VI, Roskam Aviation and Engineering Corporation).

86.For additional information, see the Systems Group Chairman's Factual Report and its addendum and the Materials Laboratory Factual Report.

87.The engineering simulator was modified to model the left and right elevator surfaces independently, and, using split elevator movements similar to those recorded on the FDR, the simulator was able to duplicate the FDR-recorded pitch history.

88.For additional information, see Systems Group Chairman's Factual Report and its appendixes.

89.This would result in the failed surface moving to a nose-down deflection of about 6º and the nonfailed surface remaining at its prefailure position.

90.This would result in the failed surface moving to a nose-down deflection of about 6º and the nonfailed surface moving to a nose-down deflection of about 4º.

91.This would result in the failed surface moving to a nose-down deflection of about 6º and the nonfailed surface moving to a nose-down deflection of about 2.1º.

92.This would result in the left elevator moving to a nose-down deflection of between 1.2º and 3.9º and the right elevator moving to a nose-down deflection of between 1.4º and 5.0º (depending on which variation of this scenario is being tested).

93.As previously mentioned, the Safety Board recognizes that a jam between two surfaces can occur without leaving any physical evidence. However, as discussed in the Board's report on the accident involving USAir flight 427, physical evidence of a jam would be expected if a hardened steel component (such as the bias spring in this case) were to become jammed between two surfaces because such evidence was always observed after tests involving hardened steel chips jammed/sheared in a PCA.

94.Even if two coils of the spring had somehow become displaced into the space between the spring guide and servo valve cap wall, there would still have been sufficient clearance to avoid a jam.

95.FDR and ground test data indicated that a single PCA failure would have resulted in much higher offsets between the two elevator surfaces than were recorded during the accident flight and on the ground before takeoff. During the accident flight, the slight offset that was recorded by the FDR was only 47 percent of the offset that would be expected if a latent single PCA failure had occurred. On the ground before takeoff, the recorded offset was only 27 percent of what would be expected if a latent single PCA failure had occurred.

96.As previously discussed, failure scenarios resulting in nose-up motions of the elevators were also possible but were not considered relevant to this accident investigation.

97.As discussed in the section titled, "Accident Sequence Study," the initial deflection for the left elevator surface was about 3.4º, and the initial deflection for the right elevator surface was about 3.8º. As shown in the graphical representations of the recorded elevator positions, a position of about 6º can be easily distinguished in the data from a position of either 3.4º or 3.8º.

98.The failure scenarios would not preclude additional commanded nose-down movement of the failed elevator surface. However, commanding additional nose-down movements would be inconsistent with an attempt to recover the airplane.

99.The Safety Board recognizes that there was some uncertainty in the aerodynamic hinge moment data used to calculate the elevator movement profiles for the failure scenarios depicted in figure 2, especially at Mach speeds greater than 0.91. However, for the initial elevator movement on the accident flight to match the elevator deflections in response to the failure scenarios, the aerodynamic hinge moment data used to calculate the failure scenario profiles would have to have been about 79 percent greater than assumed. The Board considers this amount of error to be extremely unlikely, particularly because the initial elevator movement occurred below Mach 0.91, where hinge moment data were validated by certification flight tests.

100.As previously mentioned, throughout the FDR data for the accident airplane (including data recorded during uneventful portions of the accident flight and during previous flights and ground operations), small (less than 1°) differences between the left and right elevator surface positions were observed. Even where these offsets were observed, the elevator surfaces always moved in the same direction about the same time. However, beginning at 0150:21, the elevator surfaces moved in opposite directions and remained there until the FDR ceased recording.

101.Further, although the first three failure scenarios evaluated in depth involved simultaneous dual PCA failures at the start of the accident sequence, as previously discussed, it is also clear from the FDR data that no latent jam of a single PCA occurred before the accident sequence.

102.As a former chief flight instructor with 5,191 hours in the 767, the relief first officer should have been readily able to regain control of the airplane.

103.For more information about the limitations of Boeing's simulator, see the section titled, "Potential Causes for Elevator Movements During the Accident Sequence."

104.For additional information, see Systems Group Chairman's Factual Report and its appendixes and addendums, Flight Data Recorder Group Chairman's Factual Report and its attachments, Cockpit Voice Recorder Group Chairman's Factual Report and Sound Spectrum Study, and Aircraft Performance Group Chairman's Aircraft Performance Study and its attachments and addendum.

105.This electric seat motor was recorded by the cockpit area microphone (CAM) but not by the hot microphone at the first officer's position, which (as previously discussed) was likely stowed at the first officer's side of the airplane. Because of its position on the right side of the airplane and its directionally sensitive nature, it is likely that all seat motions recorded by the hot microphone at the first officer's position after 0141 represented motions of the right seat.

106.As previously discussed, about 0148:30, the CVR recorded an unintelligible comment that could not positively be attributed to any previously identified crewmember. Two speech characteristics of the unintelligible comment (fundamental frequency and formant dispersion) more closely resembled values displayed by the relief first officer than by the other voices evaluated.

107.This electric seat motor sound was recorded by both the CAM and the hot microphone at the first officer's position, further confirming that this sound represented a motion of the relief first officer's seat.

108.This throttle lever position was consistent with manually input throttle lever positions recorded by the FDR earlier in the accident flight.

109.The Safety Board notes that several of its incident and accident investigations (including EgyptAir flight 990) might have benefited from a visual record of cockpit images/events. On April 11, 2000, the Board issued Safety Recommendations A-00-30 and -31. Safety Recommendation A-00-30 asked the FAA to require that all aircraft operated under 14 Code of Federal Regulations (CFR) Part 121, 125, or 135 and currently required to be equipped with a CVR and FDR be retrofitted with a crash-protected cockpit image recording system by January 1, 2005. Safety Recommendation A-0-31 asked the FAA to require that all aircraft manufactured after January 1, 2003; operated under 14 CFR Part 121, 125, or 135; and required to be equipped with a CVR and FDR be equipped with two crash-protected cockpit image recording systems. The Board specified that the cockpit image recording system should have a 2-hour recording duration and be "capable of recording, in color, a view of the entire cockpit including each control position and each action...taken by people in the cockpit." Safety Recommendations A-00-30 and -31 are currently classified "Open--Unacceptable Response."

110.Although the CAM recorded all of the captain's remarks, the "What's happening? What's happening?" comments at 0150:06 were of a poorer recording quality and less audible than similar remarks made at 0150:08 and 0150:15. The evidence from both microphones was consistent with the captain speaking from outside the cockpit or the rear portion of the cockpit when he made the earlier statement and from the forward portion of the cockpit when he made the later statements, suggesting that the captain was moving forward as he made these statements. Further, the content and tone of the captain's statements were consistent with his trying to understand an unexpected situation upon his return to the cockpit.

When the captain asked, "What's happening? What's happening?" at 0150:06, his words were not recorded by the hot microphone at the first officer's position; however, the hot microphone recorded the captain's subsequent remarks until it stopped recording cockpit conversation at 0150:25. (None of the relief first officer's comments during the accident sequence were recorded by the hot microphone. For additional information, see the section titled, "Audio Information Recorded by First Officer's Hot Microphone.")

111.The Safety Board considers it likely that the captain never heard any of the relief first officer's "I rely on God" statements. None of these statements were recorded by the hot microphone at the first officer's position, suggesting that they were spoken very quietly. (By contrast, the hot microphone at the first officer's position did record the captain's statements of "What's happening?" as he moved to his seat at the forward portion of the cockpit [at 0150:08] and again after he was seated in his seat [at 0150:15], despite the fact that the captain was farther from that hot microphone.)

112.The visual difference between pushing forward on the control column and pulling aft on the control column to create elevator movements of the magnitude recorded on the FDR would not have been readily apparent to the captain in the darkened cockpit during the unexplained crisis, especially when he was trying to understand the many abnormal events and sensations that were occurring during the dive.

113.During the Safety Board's tests and simulations, a pilot similar in height and weight to the EgyptAir flight 990 command captain was physically able to move from the aft cockpit into the captain's seat, to brace himself against the control console or floor structure, and to apply enough back pressure on the control column to match the physical pulling forces computed to have been required to generate the split elevator condition recorded by the FDR. However, the pilot stated that it was physically difficult or uncomfortable for him to manipulate the control column while kneeling on the floor or standing behind the captain's seat and suggested that, given his build and his need to manipulate the controls, the captain of EgyptAir flight 990 would almost certainly have attempted to enter his seat immediately upon his return to the cockpit.

As previously discussed, the simulator did not duplicate the accident airplane's actual flight conditions in every way; for example, the simulator did not duplicate the negative G loads recorded by the FDR. However, once the captain was normally seated and effectively braced, these forces should not have substantially affected the maximum fore-and-aft forces he could generate. Further, the G loads on the accident airplane did not remain negative for long; FDR data show that the G loads increased to greater than 1/2 G within 2 to 3 seconds of the start of the recovery.

114.For additional information, see Boeing's April 16, 2001, letter in the public docket for this accident.

115.For additional information, see Boeing's April 12, 2001, letter in the public docket for this accident.

116.Tests and simulations demonstrated that the magnitude of the elevator split would vary, but a split could be maintained even when the pilot in the left seat temporarily removed his right hand from the control yoke to advance the throttles and deploy the speedbrakes and the pilot in the right seat temporarily removed his left hand from the control yoke to move the engine start lever switches to the cutoff position.

117.The Safety Board notes that the captain's statement "What is this? What is this? Did you shut the engine(s)?" might reflect the beginning of a suspicion that the relief first officer's actions were not appropriate for recovery.

118.This sentence, "Get away in the engines," is an example of a phrase where direct translation of the Arabic words into English with no attempt to interpret or analyze the words resulted in an awkward or seemingly inappropriate phrase. In this case, it is possible that the captain, surprised to realize that the engines had been shut off, was trying to tell the relief first officer to leave the engines alone. However, research indicates that poor word choice, improper grammar, and the use of incomplete phrases can be symptomatic of high levels of psychological stress in a speaker.