I haven't popped on here for a while but someone pointed me to it... I did that KFC225 "bench investigation" some years ago, after I got fed up with the smell of burning servos getting into the cockpit. Some random comments:
The red disconnect switch is not easy to replace because (as stated above) it carries the clutch current of all three servos and with these being heavily inductive loads there is a lot of arcing. One would need to add some arc suppression arrangement (more than just a clamping diode because that will just clamp the arc to 28V). A switch which won't burn out for years will have to be a big one - too big to fit into the available space - and using a relay is adding a point of failure which you do not want.
The KS27xC servos are mechanically pretty good. They will outlast the life of the airframe in most cases. What wears out is the motor and to a lesser extent its integral gearbox and especially the motor brushes. The motors are "cheap and nasty" motors made by Globe in the US and Globe refuse to enter into correspondence, or sell brushes, because ... HBK have contractually prevented them from doing so
Why do HBK use Globe? Probably because Globe have put -55C as the bottom end temp on their datasheet so the motor ticks the boxes.
There is an issue in the pitch servo where the torque sensing strain gauges (they use a cheap standard industrial part, worth about $5) are subject to too much strain (flexing) due to the cantilever being not stiff enough, so the glue breaks, or the strain gauge breaks. This is trivial to fix: a stiffer cantilever and a bit more gain on the torque circuit to get the 3V output back.
There is also a problem with the Faulhaber tachometers but they are available, for about $60
Cheap and dirty, have as many as you want
But a tachometer is a horrible solution!
So there is no problem getting parts for the servos to fix them, or make new ones. Globe still make those motors. However, for the 21st century, a new (and plug-compatible) design should use a brushless system, which would most likely be a stepping motor (rather than a 3 phase brushless) because then you don't need a tachometer.
The servos burn out not due to something inside the servo but due to the KC225 computer occassionally getting itself into a mess and sending in a fast waveform which the servo cannot follow and (because the current limit circuit limits at a current high enough to melt the armature winding in some tens of seconds) this kills the motor, which goes short circuit and this melts the MOSFETs which have no heatsink; they are just soldered to a PCB.
One can prevent the burnouts in an obvious way (reduce the current limit by about 5x, just one resistor, plus some trivia) and I told some HBK reps about this years ago, including a meeting (which they requested) at Aero Friedrichshafen (EDNY) but nothing came of it. The latest SB for the servos just shorts out the current limit circuit which is the most dumb thing you can do.
There is an almost incidental other issue: the KI256 signal decoding is badly done in the KC225. Also panel vibration of the KI256 is transmitted to the autopilot as pitch/roll and this all helps to wear out the motors a lot faster than it should.
A proper fix is a fix for the KC225 software but since "everybody" left HBK c. 2003, I doubt there is anyone who can find the source code, let alone understand it, fix it, and recompile it. This is why other HBK autopilots which use the same servos don't burn them out. They don't have the dodgy software.
To get any trust back, HBK need to recruit some real software and hardware engineers, not the old guys whose careers have been in paper pushing and who are counting the days to their old age pension. And they need to put an engineer in charge, not somebody who just uses one of the several corporate bull***t generators which you can find online.
HBK should redesign the servos so that existing owners have a plug-compatible service path. They sell of the order of 1k servos a year (mostly to replace burnt out ones; the burnt PCB is declared BER) which is a nice business. Only then should they bring out a new computer.
The KI256 is a hassle but Castleberry make an electrically driven equivalent. It is TSOd, but not STCd for anything. They used to make the KI256 for a while, many years ago. A good company. Anyway, an AHRS based KI256 replacement is not hard to do, although you would probably need to bring in GPS data to get it certified as a primary pitch/roll source.
That said, the KFC225 is a great autopilot in performance terms.