University of Bielefeld - Faculty of technology | |
---|---|
Networks and distributed Systems
Research group of Prof. Peter B. Ladkin, Ph.D. |
|
Back to Abstracts of References and Incidents | Back to Root |
This page was copied from: http://catless.ncl.ac.uk/Risks/3.79.html |
ACM Committee on Computers and Public Policy, Peter G. Neumann, moderator
Excerpted from 'Tumbledown Jumbo', an article in the Oct 86' issue of FLYING magazine, concerning the China Airlines 006 incident of Feb 86. ellipses and contractions in [square brackets] are mine. ... At one point the autothrottle brought the engines back to about zero thrust. ...as the throttles came forward again, the number-four engine did not respond. The flight engineer ... told the captain that the engine had flamed out. Maximum restart altitude is 30,000 feet [the plane started at 41,000]. The captain told the first officer to request a lower altitude. He then told the engineer to attempt a relight, even though the plane ... was still at 41,000. The restart attempt was unsuccessful. The captain ... released the speed and altitude hold on the autopilot. The autopilot was now programmed to maintain pitch attitude and ground track. The airplane continued to lose speed gradually ... and the captain eventually disengaged the autopilot completely and pushed the nose down. At the same moment, the airplane yawed and rolled to the right. The captain's attitude indicator appeared to tumble [as did two backups]. The airplane had now entered the clouds. At the same time ... the other three engines quit. [paragraph omitted, describing speed varying between Mach .92 and 80 knots, as crew attempts recovery under up to 5G accelerations.] After ... more than two minutes, the 747 emerged from the clouds at 11,000 feet and the captain was able to level it by outside reference. Coincidentally, he felt that the attitude indicators 'came back in' at this point. [engines 1,2, & 3 restart themselves, and 4 responds to a checklist restart]. Initially the captain decided to continue ... [but it was noticed that] the landing gear was down and one hydraulic system had lost all its fluid. ... the captain decided to land at San Francicso. The plane operated normally during descent, approach and landing. [Later analysis showed that engine four had NOT flamed out, but just stuck at low thrust due to a worn part. The others were also responding to the throttles very slowly, a common problem at 41,000 feet. The NTSB inquiry concluded that...] the captain had become so preoccupied with the dwindling airspeed that he failed to note that the autopilot, which relied on ailerons only, not the rudder, to maintain heading, was using the maximum left control- wheel deflection available to it to overcome the thrust asymmetry due to the hung outboard engine. When the right wing nevertheless began to drop, ... the captain didn't notice the bank on the attitude indicator ... . When he did notice it, he refused to believe what he saw. At this point, ... the upset had begun and the captain and first officer were both spatially disorientated. [...] Once the erroneous diagnosis of a flameout had been announced, ... the captain placed excessive reliance on the autopilot.... When he finally disengaged it, and put himself 'back into the feedback loop' it was at a critical moment, and he could not adjust quickly enough to the unexpected combination of control feel and instrument indications to prevent the upset. END OF QUOTATIONS. The rest of the article is devoted to RISKS-style analysis of use of automatic systems. To give a more down-to-earth (pun intended) analogy, suppose your car was equipped with an AI 'drivers assistant', which handled all normal highway driving. Suppose further, at night, with you drowsy and at 60 mph, the right front wheel blows out. The AI blasts the horn to alert you, and applies substantial left torque to the steering wheel to keep it straight. You realize your in trouble, grab the wheel, and turn off the AI. The wheel immediatally jumps out of your hands to the right (you didn't know how much torque the AI was applying), and the car swerves off the road... The use of automated systems to handle routine operations of critical systems, with dangerous situations suddenly dumped in the hands of human operators, presents a new Risk... that they may not fully understand the ramifications of the problem during the critical transition time. A co-worker of mine who has worked in both the Navy and civilian nuclear programs tells me that Navy reactor systems are designed to keep humans in the loop. The only thing the automated systems can do without a person is 'scram' or shut down the reactor. Changes in power level, opening and shutting valves, pulling control rods, operating pumps, etc, must be performed by the people constantly tending the reactor. Thus, the system cant very easily spring surprises on the operators.
Some of you may have missed a recent set of rather serious breaches of the integrity of the air-traffic control system. It is another important instance of a masquerading spoof attack typified by the Captain Midnight case (although via voice rather than digital signals). [Again note the October 86 issue of Mother Jones noting similar vulnerabilities and the ease of performing attacks.] Washington Post, 8 October 1986 MIAMI -- A radio operator with a ``bizarre sense of humor'' is posing as an air traffic controller and transmitting potentially dangerous flight instructions to airliners, and pilots have been warned about it, an Federal Aviation Administration spokesman said. Two fake transmissions have occurred in the last week, and one caused a premature descent, said Jack Barker of the FAA's southern region in Atlanta. ``There have been no dangerous incidents, but the potential for danger is there. It's more an annoyance than a safety problem,'' Barker said from an FAA meeting in Washington. Barker said the operator uses two frequencies that air traffic controllers use to tell pilots how to approach Miami International Airport. The transmissions began Sept. 25, and the last was Friday [3 Oct], he said.
The accident report involving a British Airways 737 at Manchester Airport was released recently. The aircraft suffered an engine compressor failure on take-off. The aircraft instruments indicated something else (I'm a little hazy about exactly what, I think it was a tire burst), and standard operating procedure was to turn clear of the runway, basically I believe to clear the runway for other traffic. This the pilots did, bringing the wind, which had been dead ahead to blow from the now burning engine and wing, onto the fuselage. Multiple lives were lost, etc. It would appear from this that had the pilots performed an abort and maintained the runway, all that would be required for safety reasons, the deaths could have been reduced or avoided. However the operating procedure, for operational (not safety) reasons mandated otherwise and worsened an otherwise pretty terrible situation. UUCP : {ihnp4|mtuxo}!naples!mjw Matthew Waugh ATTMAIL: attmail!mjw AT&T IS, Lincroft, N.J. Telephone : (201) 576-3362
Danny Cohen's point about accuracy is well taken. The incident I was trying to refer to was the crash of Eastern 212, a DC-9, in Charlotte, N.C. I apologise to Risks readers for not confirming this before posting. Danny and I have exchanged letters on the issue of *deliberate override*. Danny considers the action of turning off the LAAS to be both non-deliberate and not an override. I still consider it both deliberate and an override. It seems to hinge on whether habitual actions can be described as deliberate, and on whether not following prescribed procedure upon receipt of a warning can be considered an override. Peter Ladkin
This page was copied from: | http://catless.ncl.ac.uk/Risks/3.79.html |
COPY! | |
COPY! |
by Michael Blume |