Update: the tweak works but 8mH was needed. So it is considerably nonlinear.
I wanted to post this as it may help someone who has been experiencing an issue with the bank angles being too large/too small with the KI 256 and the KFC autopilot.
Short story: My KI-256 went bad and it was replaced with an overhauled unit. My KFC 200 A/P worked flawlessly prior to the KI-256 failure. After KI-256 was swapped out everything worked fine BUT I noticed the pressure gage (Bonanzas have pressure systems not vacuum) reading slightly high (top of green arc).
Here is why I made the fatal mistake and broke the cardinal rule “if it ain’t broke, don’t fix it” and replaced with my own rule, “if it ain’t broke it, fix it until it is!” I adjusted the regulator to bring the gage into the middle of green arc. BIG MISTAKE. The vaccum/pressure regulators are set/calibrated during manufacture and really should not be touched as they are VERY sensitive.
After “fixing” my gage reading problem my max bank angle with A/P on went to 30 degrees. Too much. Unfortunately at that time I did not link in my mind that my adjustment of regulator caused it so I tried to “fix” it by adjusting the roll gyro gain on the KC 295 computer. No joy. Even at the least gyro roll gain, the bank would still go to 30 degrees. Then, finally, I replaced the pressure regulator and noticed my max bank angle immediately lowered to about 16-18 degrees. That is why I realized the culprit was the defective pressure regulator. Finally, I adjusted the gyro roll gain back to specs as shown in the maintenance manual and voila the max bank angle is now 22 degrees and everything works perfect again.
The moral: DO NOT TOUCH the vacuum/pressure regulator. If you adjust it, you may permanently damage it (as was the case in my case).
An interesting point… however I reckon that those people whose KI256 is over-outputting in roll (and causing the shortfall in autopilot-controlled roll angle) are also having the KI256 over-outputting in pitch for the same reason, and producing pitch control loop instability, reported as light porpoising.
Avionics people try to solve this porpoising with
The AP does not fly with a different pitch if the KI256 is over- or under-outputting on the pitch signal. This is because the pitch value is not an external setpoint. Even in the most basic PIT mode it merely captures the current pitch attitude. And in VS mode there will be some internal lookup table (etc) for loading a given pitch value when you select e.g. +500fpm. The value loaded you don’t see, because if it is wrong, the AP will adjust it once it starts to get a baro-derived VS value from its internal barometer.
It kind of makes sense to me…
1970’s Navomatic 400B autopilots in Cessna SEPs (and up) do have a pressure switch in the pitot-static system to switch between a high and a low gain depending on IAS. Ideally it should also factor groundspeed (not TAS) and, in a way, they do, since they have a G-switch which will only be used to disconnect a/p if a certain g-threshold is exceeded. Clever bit of 1960’s analog engineering…
Back to the OP, what’s stopping you from carefully messing around with the gain pots under A/P supervision as I suggested above? At least you won’t be messing around with gains in the other channel…
Depends on the detail.
In classical control systems theory, yes.
In reality, the derivative term is problematic unless the system is really tight i.e. no dead zone, no backlash, no hysteresis, no lag. This almost never exists in the real world. And in GA there are extra challenges because you need tons of phase margin because it has to be stable (well, unless it is an STEC 
) all around the loading envelope and over all the speeds from Vs to Vne, with some margins. Most GA autopilots have no airspeed sensing and e.g. the proportional gain term varies with something like the square of the IAS. So any working solution is going to be highly sub-optimal. And a derivative term will wear out servo motor brushes a lot faster, although the mfg doesn’t care much about that because that’s good for business.
Peter wrote:
I am not sure avionics use the derivative term
Isn’t that important to get a quick response to a disturbance?
I agree with your general argument (I am not sure avionics use the derivative term, but they definitely use proportional and in most cases integral) but I think the case where the wrong roll angle is being achieved is already a case where the loop gain is wrong. So if one fixed the autopilot-commanded “supposedly Rate 1 or supposedly 23 degrees” roll angle, one ought to also be fixing the loop gain to its design value – on the roll axis at least.
The right way to bodge an inadequate roll angle would be to attenuate the output from the roll half of the LVDT.
The really right “certified” way would be to adjust it inside the autopilot, but this begs the question of why is it wrong to start with, once a KI256 was replaced with a freshly overhauled KI256, which was overhauled with the specific instruction to not touch the pitch/roll pickoff section because if this is messed with, one has to readjust the autopilot, which cannot be done in-flight and needs the $2000 bus extender (which I have – I bought it 50/50 with another KFC225 owner) and needs a messy setup on the ground, with a KI256 extender lead, a vacuum pump (which I have) and ideally an adjustable pitch/roll table (which very few avionics shops have and most of those who have it have “lost the plot” commercially).
My theory is that the overhauler did mess with the pitch/roll pickoffs and perhaps set them up as per the MM, but the MM test is at 400Hz or 440Hz while the KFC225 uses 510Hz (bizzarely and stupidly, but this is Honeywell!) and prob99 the KI256 has a higher gain at 510Hz… hence the problem. Of course nobody upstream will admit to anything.
It will matter since it will affect control gains.
A smaller gain will mean it will not be as aggressive in maintaining the commanded value (pitch or roll) and hence it will be slower to achieve and be less precise in maintaining it.
See, for example, in response to a gust , or a level off command or…
There will of course be some margin for error with little impact, but you will observe that behavior will change. This will be especially obvious when in approach mode.
One way to tweak the roll angle when on autopilot is to vary the excitation signal going to the KI256.
The KI256 input inductance is 11mH so if you add the appropriate inductance in series that will reduce the outputs and will increase the roll angle flown in proportion.
No idea if this will work because that will affect both pitch and roll at the same time, so one is hoping that the different pitch output won’t matter because it is controlled by the autopilot.
