These questions relate to the Altimatic V FD, but I think they are fairly generic for that generation of autopilot.
1) I was adjusting the angle of bank in a fully developed turn (ie a commanded 90° turn to ensure that it wasn’t being tentative about a small turn.) I found that if I turned the heading bug 90° left, and once it was in a left turn, rapidly moved the bug right by say 120° (ie crossing back through the current heading the roll gyro disengaged with a jolt and then would not re-engage for about five minutes, though altitude was held. Is that normal?
2) Does anyone know for reasonable certainty what all 11 pots do? I would love an explanation for how they interact, because they are not obvious when you start fiddling:
Pitch Gradient
Pitch Trim
Roll Trim
Roll Gradient
Roll Gain
Pitch Gain
Roll FB
Pitch FB
Intercept
G/A Pitch up
Pitch Comm Max
3) I have two controllers, which are not quite the same, in that one has a GS button which I would like to have. But the trouble is that the preferable controller is a 12v model, whereas my aircraft is 24v. Does anyone know whether there is a simple adjustment or component replacement which will make it 24v?
Timothy,
1. seems like a fault.
2. You need to have an extension lead so you can have the controller on your lap in flight while adjusting. Put a drop of Corrosion-X on each of the potmeters and let it soak. I have a lead for my Bendix FCS810, I don’t know if it will fit.
The Pitch/Roll Trims are used in level flight to set up straight and level flight.
The Pitch/Roll Gain are used to compensate for over/undercorrection.
‘Somewhere’ I have the manual pages for my Bendix, I will have a look when I return from the PBN full day course later tonight.
3. Do you like smoke in the cockpit?
I have already discovered that Trim is to do with centering (ie being in the middle of the heading bug and reducing climb/descent to zero approaching the Alt hold centre position).
I am also aware that Intercept is the angle of interception of the HSI and has to be very much reduced from factory settings to work with the Aspen.
G/A Pitch up seems self evident, and I assume that Pitch Comm max is a preset to prevent over pitching.
But I don’t understand, and really want to understand, the relationship between Gradient, Gain and FB.
I am sure @wigglyamp knows more but here is my input:
Pitch Gradient – no idea
Pitch Trim – no idea
Roll Trim – could be a null adjustment (sets wings-level in straight flight)
Roll Gradient – no idea
Roll Gain – probably the control loop gain; determines stability in roll and needs to be tweaked for worst-case loading and speed envelope case to not oscillate
Pitch Gain – as above – the prime case of porpoising is when this is too high
Roll FB – no idea
Pitch FB – no idea
Intercept – no idea
G/A Pitch up – self explanatory?
Pitch Comm Max – no idea
Presumably, this is a PID controller, so Gain is the proportional term, Gradient is the derivative term and FD (by elimination) is the integral term.
The relationship between them is somewhat unintuitive, and you may find that you get very weird behaviour if you move them around. In particular, messing with the proportional term will eventually lead to the system becoming unstable, and messing with the derivative term can make it “wobbly” whenever there’s a small disturbance.
I am not sure many (or any) GA autopilots are PID.
They certainly have the P (proportional gain).
I doubt any have D (derivative term).
They all have I (integral term) for NAV tracking, so any error is gradually eliminated by feeding an integrated part of it into the heading. But probably none do this on the HDG mode, where any error just stays. The KFC225 seems to have an integral term on altitude hold.
The whole business is done mostly badly or very badly, which is why so many STEC installations porpoise in some loading conditions, and is one reason we have so few retrofit installations (STEC got in under the wire, and the more recent ones would have to do it properly).
(Sorry Timothy, going off on a tangent here)
@Peter you may well be right. I suspect this is also why the newer ‘experimental’ autopilots are so successful: they are proper controllers with properly thought out parameters which work pretty well across most light GA aircraft, and have sensible built in safeties to prevent them going completely haywire. It doesn’t take much to implement a good controller, and the last bit of optimisation can easily be done on each aircraft individually – most aircraft have broadly similar control forces and pitch/roll rates. These “legacy” autopilots must either be simple analogue controllers which are much much harder to tweak and a pain because you need to implement so many different loops, or they are operating on incomplete data.
A half-way sensible autopilot implementation would look something like this:
Take in attitude (in roll and pitch), rate of change (in roll, pitch and yaw), altitude, altitude rate of change, and instantaneous vertical G. Apply limits based on attitude (don’t put the aircraft into anything too crazy and don’t move too quickly around any axis) and G (don’t pull or push too hard). Filter the inputs attitude with a simple Kalman filter (making sure it deals with any reasonable acceleration), and feed them into the controller. Make sure the control outputs are reasonable and put some simple runtime limits in place: like not pulling any harder when G is approaching the positive limit.
Then you just need different control loops depending on the mode, and some modes would involve scheduling a target – eg turns you would schedule a roll angle based on speed and heading error remaining, and the PID loop would run on the error vs the scheduled roll angle (GPSS is supposed to do exactly this, with a roll angle calculated in the GPS navigator itself).
For trim, you would need a second loop with a very low gain, adjusting commanded trim position (not trim rate, but perhaps with some limits on max rate) based on control force. Probably also a high-pass filter to stop it constantly adjusting when it’s near enough.
I think the problem is that an aircraft is nonlinear, and on top of that the small signal response is not the same as large signal response. And say a case where it is loaded at the front end of the envelope and at Vs the stability will be a lot worse than when loaded at the aft end and at Vne. And in all cases the loop gain will have an element which is proportional to airspeed^2 so without IAS input you have to make it stable at Vs and front-loaded (unless you are STEC in which case you just sell the STC 
) so you have to make the gain about 1/4 of the ideal value (1/4 because most GA planes have a roughly 2:1 IAS range).
On a homebuilt you can just sell the system and let people fiddle with it, but in the certified market it has to work well enough.
It is interesting that Timothy has all these tweaks. It looks like a general purpose box. But then how was the installer supposed to set it up? Obviously not while flying, because that would take many flights and a lot of work (with bags of potatoes in the plane). I suppose one could have a bench rig which exercises the box… I don’t think any of the King autopilots have anything like this. I don’t know how they did the parameters, except that the KFC225 (which along with one of the others – KAP or KFC – is implemented in software) has a “certification diskette” (a 3.5" disk with the control loop values, which is very tightly controlled and only the specially authorised dealers can get it) which is loaded into a laptop which uploads it to the AP.
When you order an S-TEC autopilot, you specify the airframe. The factory then adjusts the pots based on data they have.
