At work and at home, I do a lot of design and fabrication work. It’s straightforward enough to do fabrication work once you have a plan. But any time design work is involved up front, the question comes up: How much do you rely on theory, and how much should be driven by experimentation?
The classic case of this is Edison and Tesla. For a time, Tesla worked in Edison’s lab. Two brilliant engineers with two completely different approaches to their work. Edison was a diehard experimental engineer, and would happily test the most off-the-wall corner cases, just to get the null result that proved they didn’t work. Tesla was more of a practical theorist, and insisted on doing the back of the envelope calculation before considering an idea for testing. Needless to say, the two drove each other up the wall. No event spells this out better than Tesla’s opinion of Edison insisting he test bamboo slivers as a possible filament for the light bulb. I’m pretty sure Tesla pulled out half of his hair before that set of experiments was done.
But there’s a danger in relying too heavily on theory, too. It all boils down to assumptions and initial conditions. The phrase “theory clearly states that…” has been used to shut down a lot of good ideas that probably would have worked. Why? Because the theory made assumptions that weren’t true in the real-world case. And if the theorist who’s busy shutting down the idea isn’t completely aware of the assumptions their theory makes and the initial conditions of the real-world case, they really aren’t in a position to speak.
Nowhere does this come into play quite as obviously as when discussing the stability of a KAP rig. Kite aerial photography is a pretty simple idea: hang a camera from a kite line and use it to take pictures. But when you start trying to stabilize the camera, life gets complicated. And because most textbooks aren’t written with kites in mind, the available body of theoretical knowledge often doesn’t apply the way people assume it does. The forces acting on a load suspended from a kite line are not the same as the forces acting on a load suspended from a fixed point (aka pendulum), or the forces acting on an airplane, or the forces acting on a free-floating body. They’re close. They’re all quite close. But they’re all different enough that the theories just don’t apply. Here are two cases in point from my own mistakes:
A couple of years ago I made a fully gimballed, balanced suspension for a KAP rig. I based the design off the gimbal and sled design of the Steadicam. Hey, the Steadicam has been used in Hollywood since The Shining, and was directly responsible for some of the most innovative camera shots in the last three decades. They have to know what they’re doing, right? (Theory) And a KAP rig is just like someone holding a Steadicam gimbal, right? (Assumptions)
Wrong. A person holding a Steadicam gimbal has six completely independent degrees of freedom in their hand: translation in X, Y, and Z, and rotation about X, Y, and Z. A kite line can translate freely in two axes X’ and Y’, and is semi-rigid in a third axis, Z’, oriented along the kite line. Translational forces are relatively slow acting in X’ and Y’, but very rapid with very high impulse in Z’. A KAP rig is free to rotate about Z’, but not about X’ or Y’. Typically it can rotate around Y, though. Since the camera is in the XYZ frame rather than the X’Y’Z’ frame, there’s a coordinate system rotation that needs to be taken into account. But since the XYZ and X’Y’Z’ frames aren’t orthogonal to each other, this means at least two of the axes will couple. How this worked out with my fully gimballed KAP suspension is that any sort of rotation about Z’ led to a very rapid very high impulse rotational force around Y. It rendered the rig less stable rather than more stable. It’s not that the theory behind the Steadicam was wrong. It just didn’t apply to KAP. The real-world conditions didn’t match up with the assumptions.
My second example is a lot shorter: Some years before this test, there was a discussion about using a MEMS gyro to provide feedback to a servo to stabilize a KAP rig. I was one of the ones that said it wouldn’t work, and that a 50Hz servo loop was simply too slow to take out an appreciable amount of motion. Control theory clearly stated that it simply couldn’t keep up – that the idea wasn’t even worth pursuing. (I think that was how I worded it at the time.) Thank goodness someone ignored me and decided to try the experiment anyway. Not only did it work, it worked well. It worked so well and was so feasible, he made some custom boards to replace the PCB in the servo that included the MEMS gyro. Voila! A self-contained gyro-stabilized servo. And so the GS-1 servo was born.
In this case it wasn’t the theory that didn’t apply. It was the windbag (me) who insisted that the theory said something that it didn’t. The lesson here is to beware the expert who thinks they’ve understood the problem well enough to do their back of the envelope calculation. If their understanding is flawed, so will the theoretical results.
All of which goes to explain why I lean toward experimentation rather than raw theory. It’s still possible to make a bad experiment, but it’s darn near trivial to mess up a theory and fool yourself into thinking something is true that isn’t. Given the choice, I’d rather pull out the bamboo slivers and pass some current through them, just to make sure the answer really is no.