On Fri, Oct 11, 2013 at 12:47:07PM +1300, Michael Fincham wrote: >On Thu, 10 Oct 2013 16:26:52 -0700 (GMT-07:00), David Weinshenker wrote: >> Yes, please put it up for download >> somewhere > >I've put it up online here: > ><http://finch.am/u/nasa-saturn-manual-control-pdf> > >It'll probably hang around for a while at that URL if anyone wants to >grab a copy. I just got around to having a look at it. A few things stand out. For one, this wasn't manual control as in "something that would work if all the computers fail". The pilot wasn't given eight levers, one for each control signal (pitch and yaw for the four gimbaled F-1 engines), and told "have at it... you can control this thing, sure you can, I mean you have ten fingers, and there are only eight signals, so you have two fingers to spare". Instead his control input was sent to the control computer for the launch vehicle, which translated it into engine movements. If any computer was cut out of the equation, it was the control computer for the spacecraft, which was also involved in normal flight... but it seems like that computer was mostly just relaying data from the gyros in the spacecraft (although that part isn't described well in the paper, and others may wish to correct me as to the true way the two computers interacted). In any case, technically, using the joystick didn't cut either computer out of the loop; instead the pilot's signals were added to the computer's -- but it seems like the joystick had enough control authority to thoroughly override the computer's choice. That is, as long as the computer was working and obeying the joystick; "computer failure" does not appear on the list of failure scenarios they considered. Besides the joystick, the pilot was also given six switches to turn off parts of the automatic control loop. Those were in case various sensors failed. But they considered those sensor failures to be low probability, and the ability to override them not a big contributor to the overall benefit of the system. Skimming through the procedures for sensing those faults and flicking those switches (Appendix B), they read like things that, these days, could and should be done in software. They found that it was important to give the pilot a "load relief system", meaning lift sensors: he had a display showing the output of accelerometers mounted near the center of mass of the vehicle, so that what they sensed (at least in two dimensions) was aerodynamic lift. The idea was to fly so as to minimize that lift -- which, interestingly, was to be done even before any failure had occurred, so as to give "a greater margin of safety in the event of a system failure". I don't know whether the astronauts actually ended up doing that. From simulating one particular failure (engine gimbal actuator hard over, the failure mode they figured was the most probable) in "95% wind", they gave the automatic system an "effectivity" of 0.488, the piloted system with lift sensors an "effectivity" of 0.322, and the piloted system with no lift sensors an "effectivity" of 0.045. In each case that number is the probability of the launcher being broken up by wind and other forces, so a lower "effectivity" is better (making it a poor choice of word -- but at least they weren't being modern and politically correct, and using "piloted" as a euphemism for "manned": here "piloted" actually means piloted). But in some of the other failure scenarios the pilot didn't help: for the "loss of thrust in one engine" scenario (another thing they thought there was a big chance of, and rightly so), the differences in success rates were marginal, and vehicle loss was highly probable. These days, introducing extra lift sensors and only giving access to them to the human would be cheating: the normal thing to do would be to let the computer code use them too, for cross-checking and/or for flying in a degraded mode. But back in the days when every byte was precious and computers were programmed in assembler (if not in machine code), it was a normal sort of thing to do. (For the simulations they did for the paper, they didn't even use a digital computer; instead they used "a 400-amplifier analog computer complex with extensive function generation capability".) But given that this wasn't computers versus wires-and-cables-and- hydraulics but rather computers in automatic mode versus computers in joystick mode, it's also permissible to wonder whether the joystick mode was what it should have been. Were the control parameters altered appropriately for the engine-out scenario, for instance? The computer knew that the engine was out, and could have altered them -- but did it alter them, and if so did it do it well? Likely not, since if it did it well, why would a human be needed in the first place? Or maybe no possible control action would work in those cases -- they were pushing the system rather hard, considering cases near max-Q and with high winds and high wind shear, but they don't address the question of whether the system was controllable in those failure cases, or whether no possible set of commands would work. (It's the kind of question you could throw a lot of computer power at, these days, but they didn't have a lot of computer power.) -- Norman Yarvin http://yarchive.net/blog