Inertial measurement hardware: MEMS vs non-MEMS | Navigation | Forum

I have been reading/thinking about the hardware side of the inertial navigation system.  In some sense we won't need to make this decision for a while, because the meat of the work will be in the filtering algorithms, which are indifferent to the hardware, but it's never too soon to start thinking about things.  There isn't much of a point to this post, in that it doesn't make or ask for any design decisions, I'm just trying to keep people up to date on my reading/thinking.

Accelerometers and gyroscopes these days come in two broad categories: MEMS (Micro Engineered Mechanical Systems) accelerometers/gyros and what I'll call "macro" accelerometers/gyros, which is everything else.  The MEMS devices are relatively recent and are basically just silicon chips.  You can fit 3 accelerometers and 3 gyros into a unit smaller than a postage stamp!  These are the sorts of devices used in Wiimotes, in laptops (to park the hard drive if you drop it), in cars to activate airbags, etc.  Compared to the "macro" devices (which use "big" things like springs, bearings, lasers, mirrors, etc), MEMS devices are cheaper, smaller, lighter, more rugged, consume less power and take less time to start working after first powering up.  Their only downside is that they are currently quite a lot less accurate than the macro devices, which are the current de facto standard in aircraft navigation systems.

I'm not sure if we should necessarily count MEMS devices out entirely though.  For one thing, MEMS devices are getting more accurate all the time.  By the time we are actually ready to build something, it's possible the accuracy gap will have narrowed some.  For another thing, the difference in cost, size and weight is so great that you could easily average the signals from 10 or 20 separate MEMS devices before you started to approach a non-MEMS solution in any if these cost factors, which would go someway to improving the accuracy (not enough to get us close to modern non-MEMS solutions, but possibly close to 1960s non-MEMS, which we know to be good enough).  Thirdly, inertial navigation will not be our only means of navigation – we will be combining the inertial estimates with estimates from radio measurements, visual sightings of stars/Earth/moon, etc.  Until we have a better feel for the accuracy that these other methods will allow, we can't be too sure how important inertial accuracy is in the overall picture.  Finally, my understanding is that an important part of being able to filter noise from inertial measurements well, using a Kalman filter or the like, is being able to accurately model the dynamics of the vehicle.  In our case, this should be super easy.  We won't have to account for wind or turbulence like aircraft, and there should be very little unexpected acceleration.  Once we are in space, except for service engine burns and RCS use (which our guidance system can forewarn the navigation system about), the only acceleration will be due to gravity, and that will be almost perfectly constant when we are inbetween the Earth and moon's spheres of influence.

Basically, the situation is: a non-MEMS solution will definitely be good enough but it will be (relative to MEMS) heavy, expensive and power hungry.  A MEMS solution *might* be good enough, but might not, but all the advantages of a MEMS approach are great enough that I am reluctant to count them out too soon.

I welcome feedback on this from people who know more about this than me!

galonar said:

Just as an intro, I work as an Attitude Control and Navigation Systems Engineer in aerospace, and I'm going to poke a lot of holes in your arguments here.

Glad to have someone around who knows a little more what they are talking about. As long as you do your poking in a constructive manner, it's very welcome. Indeed, essential. Before I say anything else, since you seem relatively knowledgable about navigation issues as a whole, I want to draw your attention to the NGW suggested reading list on our Wiki – if you can contribute links to this it will help tremendously in getting the group up to speed.

Now, as for specifics…

For starters, your claim that algorithms are indifferent to the hardware is a gross oversimplification.  The purely mathematical concept behind the algorithm may be indifferent, but once you get into the implementation of it, the hardware becomes integral.

In what way? I know Kalman filtering requires estimates of variance for all the measurements, which is of course hardware dependent. I suspect you are talking about more than just this, though, because that's just a constant parameter.

MEMS gyros and accelerometers:  Yes, they are getting smaller and more accurate, but there is more to the equation than accuracy.  The environment in space is nothing like that on earth.  Reliability is one of the biggest concerns for space hardware.  Can the unit execute its task in the harsh environment (large temperature swings, radiation, etc) accurately and repeatably? Radiation plays havoc with electronics that aren't properly shielded, and even when they are, there's always a chance for problems to occur.  Macro devices, as you call them, are the de facto standard for a good reason: a metric known as TRL.  Technology Readiness Level assesses how mature a technology is for its environment.  In short, technology won't fly until it has reached TRL 8.  As far as i know, MEMS technology is not close to that.

Although I'm not sure it has been made explicit yet, I think the implicit idea all along has been that the intertial navigation stuff will be contained in the CM, which will be pressurised, temeprature controlled and radiation shielded by virtue of having our astronaut in it. It's possible that radiation shielding requirements for electronics can exceed those for humans and if so that is definitely something we will have to look into. The advantage of the MEMS approach is that we can easily afford to have of the order of 10 of the things running at once, so instantaneous errors or even complete failures of individual devices are not necessarily disasterous.

Star visuals using a Star Tracker can provide fantastic accuracy in determining your attitude.  Unfortunately, they require an equally impressing algorithm set commonly known as SIAD, Stellar Inertial Attitude Determination.

Do you know much about this? Most of our navigation related discussion so far has been heavily focused on inertial stuff, I would be very happy to see discussion of other options and the star visuals sounds like as good a place to start as any.

The space environment:  your assumption that we won't have anything to worry about once we get into space is flat out incorrect.  Solar pressure, gravity gradients, micrometeroid impacts, and more can all cause disturbances to our trajectory.

I am not convinced those things will be huge factors. Re solar pressure: Wikipedia says "Radiation pressure is about 10^-5 Pa at Earth's distance from the Sun". If the cross-sectional area of our CSM normal to the direction of this pressure is of an order of magnitude 10 m, then with that pressure we are talking about a force of 10^-4 N. Assuming the mass of our CSM is on the order of magnitude 1000kg (and really 10000kg is probably more likely) this gives an acceleration of 10^-7 ms^-2, which is negligible and probably significantly less than the random noise in most acclerometers. Re micrometeroid impact: Wikipedia says these are typically less than 1g and their velocity relative to spacecraft is on the order of 1km/s. A worst case impact with a 1g meteoroid going at 10km/s imparting all of its momentum to the CSM would give an impulse of 10sN. Assuming a CSM mass on the order of 1000kg (again probably an underestimate), this is a delta-v of 0.01 m/s. Admittedly that is not negligible, but presumably the impact rate for these is roughly isotropic (I'm happy to be corrected on this) so the effect is a random walk in velocity space with 0.01 m/s steps that cancel each other out quickly. Re gravity gradients: it's a bit more difficult to get a back of envelope estimate for this but I still think that for the bulk of the trip (when we are far from the Earth and moon) the gravity will be close enough to constant for our concerns. Even if it is not, this effect is at least deterministic – we can estimate gravity gradients from our position estimates. It may not be the case that we have literally nothing to worry about, but we have an awful lot less than, say, a jet fighter.

More to come as I think of it.

Looking forward to it!