I found this article interesting and I am exposing my naivety here, but on my AP (the Century IIB), I have the POH and I am aware of the checks, which I do for anything more than a local flight where I wont use the AP. However I don't test the override part, which the manual says should require a force of 15lbs, but when I have tried it, I have exerted a 'lot' of force and it hasn't disengaged. This either means I am not actually applying 15lbs or more (maybe because I cant equate how much I am applying), or there is something wrong with the system. Perhaps foolishly I haven't applied extra force in case I brake it, and the plane ends up in the workshop with a large bill.
But I always assumed if there is a problem with it, then I can just switch the system off via the control switch? I expect the answer is "no" but is that safe enough? Thinking aloud, bviously an electrical failure would prevent me from turning it off this way, but if one of the servo's got stuck, I wonder if this old system has a disengage override when power is not applied to it. The POH doesn't say.
In my opinion those POH mandated autopilot pre-flight tests are like the "read and accept license before installing" dialogs in modern software. It is just unrealistic to take the time and do it all. The instructions for my new DFC90 would take longer than the rest of the pre-flight. It's typical US lawyer stuff to protect the company.
One should always have the attitude that the autopilot was designed by people that want to kill you. Never trust it.
The way the S-TEC servos are built and installed, I am not concerned that there could be a failure mode where they overpower me or block the flight controls.
Also, never be afraid of overpowering an autopilot. It is designed to withstand that.
Also, never be afraid of overpowering an autopilot. It is designed to withstand that.
Yes, certainly. But be aware that there are variants which won't re-engage after having been overpowered. So if you have something like a "TCS" or "CWS" button on your yoke it's always better to decouple the servos (temporarily) using that button rather than by brute force. Mind you, I have come across quite a few pilots who have been flying their aeroplane for years yet had no idea what that button was for...
Peter: Any autopilot with a pitch servo should have the pitch overpower test.
Every aircraft without "power controls" (e.g. hydraulics) and an autopilot has to have a pitch servo somewhere. From SEP over MEP, turboprop and Bizjet to regional airliner. All those servos have the same, or similar, disengagement forces. Yet this overpower test really seems to be specific to some types only. I have flown all the stuff on my little list above and have not come across that test until you started this thread yesterday!
PiperArcher - you will find the exciting 411 page IM for the Century 2B here.
Have a look on e.g. page 9 of the PDF
4. Roll Servo - The Roll Servo receives power from the console amplifier and applies force to the aircraft controls. A solenoid controls engagement to the controls and a force limiting clutch is provided to limit the maximum force the servo can apply. The setting of the clutch is determined during F.A.A. certification and its maximum setting as contained in the servo data should never be exceeded.
On a very quick look the 2B is a roll control only autopilot. The 3B shows a pitch servo etc.
It is from 1973 - just a year after the last Apollo moon landing.
Every aircraft without "power controls" (e.g. hydraulics) and an autopilot has to have a pitch servo somewhere. From SEP over MEP, turboprop and Bizjet to regional airliner. All those servos have the same, or similar, disengagement forces. Yet this overpower test really seems to be specific to some types only.
I am speaking of light GA stuff. I have no idea how it is done with hydraulic-assistance systems, but I would assume the servos are at the "pilot end" of the link otherwise they would need a lot more power.
I have flown all the stuff on my little list above and have not come across that test until you started this thread yesterday!
My guess is that - as with so much in aircraft systems certification - somebody took the view that if you enforce a periodic removal of the servos and checking the clutch torque (which is always a nice bit of money for the service industry) then the pilots do not need to be concerned with it.
In GA maintenance, the tendency is to skip anything remotely non-trivial, tick every box in the manufacturer's service schedule, and sign a release to service which says it was done IAW the manufacturer's service schedule, but in reality all the shop did was to have a nose around and squirt some lube at the obvious things. Most owners are unaware of this - until something seizes up totally.
But be aware that there are variants which won't re-engage after having been overpowered
I think all airliner autopilots will disengage if you e.g. lean against the yoke. They won't re-engage. Pre 9/11 I asked the question to a DC10 crew mid-Atlantic and they got quite worried and sent me back to my seat pronto 
But they did tell me it would disengage the AP.
I am not suggesting anybody does this test in flight! The procedure is to simply press the AP button which, on most 3-servo systems, will engage the autopilot in wings-level and pitch-hold mode. Usually the yoke will start slowly moving on both axes, and when it hits the pitch stop the pitch trim servo starts running, etc. But you do the overpower test long before it gets anywhere near any of the stops. The engine has to be running on the King ones, usually, because the AP won't engage if the vacuum AI is more than about 20 degrees off being straight-up.
As a non-engineer, I seem to be regularly amazed at what passes for engineering in the world of certified GA.
"One should always have the attitude that the autopilot was designed by people that want to kill you. Never trust it."
I can see the sense in approving an autopilot that 'lets go' at some inconvenient moment - after all, the pilot should be able to fly the plane manually. But to approve an autopilot that could fail in some kind of 'locked on' mode that could kill you in doing so just seems plain lunacy. It brings a whole new meaning to the idea of 'certified'.
Also I have to say that I've come across NO instructor (or examiner) who has either suggested this overpower test (or pulled me up when I didn't do it). Do the instructors on this forum teach it?
As an electronics designer, what amazed me most is that the King autopilots, and most other avionics we fly with, do not contain a watchdog.
This means that if they crash they just freeze up.
One I like is the MAX705 - example pricing here. I pay about £1.50, 1000+.
However I do not believe all the designers are that stupid. They are not putting one in for a reason, and it probably is to do with certification-permitted failure modes. I would bet that the authorities permit a GPS navigator freezing solid, or an autopilot going haywire and melting its servos, but they don't allow it to silently reset itself and come back up where it was, say 20 secs later. Also such recovery would need a continuous saving of the context, etc. That said, if you pull the plug on a KLN94 it will come back up where it was before, with the current flight plan etc active in location 0 as usual.
But the calibre of electronics engineers in avionics has usually been fairly low. It is probably to do with the very slow pace of innovation in the business. So good people don't stick around.
Even Garmin, who churn out stuff at seemingly breakneck speed, are actually doing very little when you look at the actual hardware they produce. Everything is no more than an incremental improvement on previous stuff, and they run old cash cows (GNS boxes, etc) for 10+ years. For a design engineer working inside, not a lot is happening, though the software department is fairly busy, and the documentation department is extremely busy.
But to approve an autopilot that could fail in some kind of 'locked on' mode that could kill you in doing so just seems plain lunacy. It brings a whole new meaning to the idea of 'certified'.
It is quite common for autopilots to do very bad things and endanger people's lives. The most common malfunction is a "trim runaway". Try turning your trim wheel in your C172/PA28 etc. all the way back during level flight and then hold the altitude. Now imagine this in a much larger aircraft. The Cirrus SR20/SR22 do not have manual trim so in case of a trim runaway, there is nothing you can do other than applying elbow grease.
I was told by an autopilot maintenance company that they almost got killed once when test flying a Metroliner. They experienced a trim runaway and two persons in the cockpit barely had enough force to prevent the airplane from crashing.
If my TR182 is completely out of trim, it can be really difficult. In the beginning I often had the habit of overpowering the autopilot servo when I was in stress (e.g. avoid TCU). The result of that is that the autopilot senses the force and tries to eliminate it by turning the trim wheel. Depending on how long I overpower it, it will turn the wheel all the way forward/backward. I had some very unpleasant moments due to this. Now I am more experienced and my new DFC90 will finally bring CWS (control wheel steering) -- a button to momentarily disengage the autopilot and allow for a manual manoeuver.
Try turning your trim wheel in your C172/PA28 etc. all the way back during level flight and then hold the altitude
The pitch trim is supposed to warn after (from memory) 9 seconds of continuous running with TRIM IN MOTION and after something like 15 seconds it is supposed to disconnect the autopilot. Obviously the latter action will hand the aircraft back to you with the pitch massively out of trim... But it is only in the most unusual type of flying (probably, trying to reach stall while on autopilot) that you should even get the first message.
Now I am more experienced and my new DFC90 will finally bring CWS (control wheel steering) -- a button to momentarily disengage the autopilot and allow for a manual manoeuver.
Gosh they are only 30 years late 
I am amazed the Cirrus has no manual trim wheel (never noticed the absence in when I flew in them) because you don't exactly get a lot of leverage with the sidesticks.
SFAR 23.11 Electric trim tabs. The airplane must meet the requirements of FAR 23.677 and in addition it must be shown that the airplane is safely controllable and that a pilot can perform all the maneuvers and operations necessary to effect a safe landing following any probable electric trim tab runaway which might be reasonably expected in service allowing for appropriate time delay after pilot recognition of the runaway. This demonstration must be conducted at the critical airplane weights and center of gravity positions.
23.677 Trim systems. (a) Proper precautions must be taken to prevent inadvertent, improper, or abrupt trim tab operation. There must be means near the trim control to indicate to the pilot the direction of trim control movement relative to airplane motion. In addition, there must be means to indicate to the pilot the position of the trim device with respect to both the range of adjustment and, in the case of lateral and directional trim, the neutral position. This means must be visible to the pilot and must be located and designed to prevent confusion. The pitch trim indicator must be clearly marked with a position or range within which it has been demonstrated that take-off is safe for all center of gravity positions and each flap position approved for takeoff.
(b) Trimming devices must be designed so that, when any one connecting or transmitting element in the primary flight control system fails, adequate control for safe flight and landing is available with - (1) For single engine airplanes, the longitudinal trimming devices; or (2) For multiengine airplanes, the longitudinal and directional trimming devices.
(c) Tab controls must be irreversible unless the tab is properly balanced and has no unsafe flutter characteristics. Irreversible tab systems must have adequate rigidity and reliability in the portion of the system from the tab to the attachment of the irreversible unit to the airplane structure.
(d) It must be demonstrated that the airplane is safely controllable and that the pilot can perform all maneuvers and operations necessary to effect a safe landing following any probable powered trim system runaway that reasonably might be expected in service, allowing for appropriate time delay after pilot recognition of the trim system runaway. The demonstration must be conducted at critical airplane weights and center of gravity positions.
