DFC90 & DFC100: Using the Existing Servos

This post was written by Mark Krebs, our Vice President of Engineering for Guidance and Controls.

How do the servos work?

One of the first questions many people have asked was how well the autopilot performs without digital smart servos.  Plenty of smart folks have been confounded by the idea of a modern digital autopilot being coupled to the the existing Cirrus servo system: The question was even the topic of a COPA forum thread. "How could that work?" they asked.  The answer is "very well indeed."

If that seems a surprising answer, well at first it was for me too; I was daunted by the initial challenge to utilize the existing servos.  We had a few things going for us though: good understanding of the airplane's flying qualities, our high performance AHRS, and not least a modern digital flight control system.  Over time, we overcame the challenges and became very comfortable with the Cirrus & STEC servos.

With three planes in continuous service now, we have accumulated hundreds of flight hours on multiple systems, including one with no STEC pitch servo at all, and they are all flying great.  We are very confident the Cirrus trim servos and linkages provide a perfectly viable foundation on which to build a flight control system. I will provide the details here, along with correcting some misconceptions.

First, regarding digital servos.  In all cases the servo boils down to a motor which is an analog device that converts current and voltage into torque and speed. Computer control makes it a "digital smart" servo, and as for computers, we got one.  The point is, the tricky bit is about the "smarts" not about where they're located. With a fully capable actuator & harness already in place, it makes sense to keep the computation centralized. If there is any value in further discussion of servos, I'm happy to engage, just send me your questions.

Next, some basics on the flight control system. As you know, the stock Cirrus comes with pitch and roll trim servos, driven by DC motors through planetary reduction geartrains and closely coupled to the flying surfaces by spring cartridges. These motors are reliable and extremely strong: no way could a human ever overpower one.  Enter the need for the spring cartridge. Whenever you move the yoke, you are overpowering that spring (but not budging the servo).  That's why discussions of "wearing out" the spring cartridge are misplaced: it receives the greatest workout when the plane is hand flown!  There are linkage issues we'll discuss in a moment, but basically, springs don't wear out, and we haven't seen any sign of it. It's a pretty good system. In fact, when I met with Cirrus aerodynamic designers to discuss the controls some time ago, they proudly touted the fact that you could fly the plane with just the trim hat.  That might be a tricky proposition since the trim system only gives you one speed (full speed) but it is true, and is a unique feature of the Cirrus aircraft design.

The dynamics of the spring cartridge can be characterized informally by one word: "detent."  What that detent is, is two springs preloaded to hold the yoke in the middle, against a center stop so that for the first several pounds of force applied, nothing happens. The cartridge won't budge.  That's something you understand implicitly almost as a "muscle memory," and so when you feel the detent, you quickly and forcefully power through it. Once you've overcome the preload in either direction, the spring responds linearly thereafter and control surface deflection is proportional to the force applied.  All this detail is relevant because as a human you have three special advantages over the autopilot. First, your arm is both a wonderful force sensor and a fast actuator. Second, you are attached to the elevator almost entirely without the slop (or "deadband," in the lingo) through which the trim servo must act.  Lastly, your hand is a force sensor: you can feel the detent and know to push through it, but the autopilot servo can not.  It doesn't know what it's pushing against, or how hard. Also, autopilot servo has some backlash of its own, which is the result of geometry and tension in the capstan bridle cable: that acts like another spring in the system, stiffer than the spring cartridge but softer than the preload.

So now, to summarize the autopilot's problem (again, from the viewpoint of the servo) it must push back through bridle backlash against unknown aerodynamic force from the elevator, which is itself attached through a sloppy linkage to the trim servo that introduces the detent characteristics just described. The aerodynamic force is called "hinge moment" and it's highly variable.  Dependent on indicated airspeed and how much weight you want to put on the tailplane as a result of g-loading and cg location, the hinge moment can change dramatically during the flight and that's why there's a trim system: so you don't have to fight it all the time.  In summary, the elevator servo is facing a backlash, a variable hinge moment, a deazdone, a preload and another spring. Finally, since we are driving a DC motor with a sticky, high ratio geartrain to push, it's going to take several Volts worth of determination to move at all! 

That summarizes the challenges.  What do we do about them? Two things...  

Handling the last problem is the easiest, since we have a "smart" servo drive circuit. If we want to move at one volt speed, we command four volts, because it takes three volts just to "unstick" the motor.  This is a very big part of the solution because it means we get proportionate servo response when we ask for it.  The details are not much more complicated than that and you can easily imagine what a big improvement it makes: after all, control systems are all about getting what you ask for. In fact this is about all you need in the roll axis, where the spring cartridge is never deflected. There's been a surprising amount of confusion over the roll channel, with STEC introducing another whole servo and Garmin eschewing the trim system altogether. I don't know why. The Cirrus roll trim servo system is, in a word, excellent.

The pitch problem is a lot more interesting and the solution is both fun and elegant.  First, consider that with all those different pieces of hardware, and all their individual dynamics, it is not clear who is flying the airplane!  Is it the autopilot, the trim servo, or you?  Just to illustrate with an example, suppose the aerodynamic hinge moment is large enough to overcome the spring cartridge detent.  Then, if the automatic pitch trim is working, it will have been adjusting the trim servo to unload the autopilot servo and consequently the spring in that cartridge is deflected to "hold" the elevator force. We are in the linear range. "So what?" you may ask, but consider, that means the autopilot servo may be hanging "slack" in its bridle. It will have to turn a bit just to take a strain, before it can move the elevator at all. Until that happens, trim servo is actually flying the airplane!  When this happens to you in your STEC you experience it as wandering uncontrolled pitch because the trim servo does not care about your glideslope deviation.

We experienced exactly this in flight test at high speed, during strong pull-ups, and sometimes on the glide slope, so, while intermittent, it can really matter.  Our solution harks back to the rudder / elevator mixers on a Beech Bonanza: we simply command both servos to move the elevator.  That way it doesn't matter which one has the solid linkage to the elevator because they're both doing the right thing all the time.  When there's a trim signal (and this is sensed by the autopilot servo just the way it is in your STEC right now) then we command the servos in opposite directions, to soak up the  load in that spring cartridge.  That's it. In the long tradition of ruddervators, elevons, and spoilerons, we have a "trimmervator."  It's a little bit funny, but legitimate. From the Bonanza to the B2, space shuttle, pegasus rocket and any number of other missiles, control mixing is a commonplace solution to cross coupled surface effectivity, which is what we've got here.

With these two simple ideas, we are able to master the complex Cirrus pitch flight control system, and get very good response out of the airplane.  The system performs almost as well with just a trim servo although it suffers in speed and in situations where the hinge moment is large it just can't fly quite as precisely.  We do recommend the pitch servo upgrade, but you will find that, even without it, your airplane will fly much much better with our autopilot.

This hardly begins to scratch the surface of the autopilot innovations. An important characteristic to think about is that "digital autopilot" is just a buzzword anyone can use.  The devil's in the equations and not all equations are created equal.  Probably, we won't share all the details of how Avidyne's advanced the control systems in the DFC, but a general discussion of some of those ideas will be forthcoming in future posts.  The point, I suppose, is that I'm very proud of the uncompromising work we've done, and confident you'll see it expressed in robust, high performance capability of the autopilot. Some of the COPA posts are starting to show that already.

Happy flying!

Mark