6Dof IMU design, how? / General Science and Electronics / Forums

Forums

4hv.org :: Forums :: General Science and Electronics

« Previous topic | Next topic »

6Dof IMU design, how?

1 2 next

Move Thread

LAN_403

Electroholic

Thu Jan 14 2010, 06:27PM

Registered Member #191 Joined: Fri Feb 17 2006, 02:01AM
Location: Esbjerg Denmark
Posts: 720

So here is my problem, I need to track free motion accurately for a short duration of time.
Engineering requirements
Gyros, +/-2000deg/s
Accel, +/-35g
duration : 5 seconds
100 sample per second or better
noise + integration drift after 5 seconds should be <3mm, and <1degree.

Is this even possible?
And how should I start about the design?
Most info I found are meant for designing long term guidance/navi systems, and I'm not sure if it applies here. Help?

Dr. Slack

Thu Jan 14 2010, 07:15PM

Registered Member #72 Joined: Thu Feb 09 2006, 08:29AM
Location: UK St. Albans
Posts: 1659

35g = 350m/s2, and after 5 seconds at that acceleration you'd be over 4km away doing over 600mph. So I guess the acceleration isn't going to be constant at that peak

It would help if you constrained the total volume needed to be covered, and the peak velocities expected. Maximum weight of the sensor?

Two things spring to mind. The first is to hack a Wii controller and just build something. It may not give you the precision, but it may be a fast way to get started and help develop the software or the specification. If the volume to be covered is small enough, then some variant of motion capture (as used in all big budget films these days) with perhaps web cams? Maybe not fast enough for 100/s, but maybe a start?

Mattski

Thu Jan 14 2010, 07:26PM

Registered Member #1792 Joined: Fri Oct 31 2008, 08:12PM
Location: University of California
Posts: 527

The requirements are a little challenging. For example, for the gyrometer you are asking for 1 degree out of a maximum 10,000. For accelerometer (1/2*9.8*35(m/s^2)*(5s)^2) = 4.3km is the maximum distance travelled (though presumably you can limit it much more than that), you are asking for 3mm out of 122m. Accuracy is always more difficilt to be had from your sensors if you want such high accelerations and turn-rates.

Guidance and nav systems are exactly what you want to be reading about, that's where inertial sensors are mainly used. In fact, what you have described is a guidance system without feedback to steering.

A huge part of this is going to be properly processing the data once you have it in digital, and figuring out exactly what specs you need in your system to achieve your error specs. You may need to input something like Kalman filtering.

What's your application? That always makes it easier to make suggestions.

Edit: beat to the punch slightly.

Electroholic

Thu Jan 14 2010, 08:09PM

Registered Member #191 Joined: Fri Feb 17 2006, 02:01AM
Location: Esbjerg Denmark
Posts: 720

Well, the specs are for capturing golf swings, and yes those are peak values, during a drive, at the handle. I'm using them to spec the. sensors. I understand the requirements are quite high, you can tell i dont know much about mems/what to expect from mems.

From what I've read, navigation IMU uses kalman filters and redundant geomagnetic sensors to calibrate for long term accuracy. I might be wrong, but I thought kalman filters are used for filtering out jerkyness in the readings, and isn't that kinda counter productive if I need to capture something really fast with high peak values?

I read about random noise and zero drift within the sensors and integration drift, but I'm not too sure how to translate them into uncertainties and drift of displacement and orientation.

tesla500

Thu Jan 14 2010, 08:16PM

Registered Member #347 Joined: Sat Mar 25 2006, 08:26AM
Location: Vancouver, Canada
Posts: 106

Perhaps a camera based motion capture system would be better for this? There are no integral errors to worry about there.

Carbon_Rod

Thu Jan 14 2010, 09:26PM

Registered Member #65 Joined: Thu Feb 09 2006, 06:43AM
Location:
Posts: 1155

Generally,
* 3-axis compass chips max out at around 100 samples/s
* 3-axis digital accelerometer chips max out at around 50 samples/s (but 200 samples/s are available)
* Most gyro-chips under $1000 max out at around 50 samples/s

At 35g... you will likely have bigger problems than just tying to find a sensor for that range.

As others have stated, you will have to be more specific about what you are trying to do.

Electroholic

Fri Jan 15 2010, 04:20AM

Registered Member #191 Joined: Fri Feb 17 2006, 02:01AM
Location: Esbjerg Denmark
Posts: 720

2000deg/s and 35g range is needed because of the nature of the motion, which is well defined and modeled. At the moment, I am trying to find commercially available mems sensors that will do the job, or a feasibility study.
I cant seem to find maximum sampling rate in the datasheets, how is it determined? And say I go with 50 samples per second, what type of sensors am I looking at?

Carbon_Rod

Fri Jan 15 2010, 07:01AM

Registered Member #65 Joined: Thu Feb 09 2006, 06:43AM
Location:
Posts: 1155

IIRC this company can provide special order units over 5000 degree/s

Expect the error rate +-50% larger than what their documents state.
Sampling rates depend on the technology used in the sensor.

Note there are better systems for corrected attitude in moving aircraft/spacecraft if you budget over $16000.

Cheers,

Mattski

Fri Jan 15 2010, 07:36AM

Registered Member #1792 Joined: Fri Oct 31 2008, 08:12PM
Location: University of California
Posts: 527

What Kalman filtering does, as I understand, is allow you to improve the inaccurate measurement data based on what you know about the system going in. For example, suppose you know the club's velocity at one time. The next sample time you might expect the magnitude to be close in value, but the direction to have rotated within a certain cone of probability. Since very fast gyros and high-g accelerometers usually have fewer bits or volts per degree or g, I think it might let you process it to a higher resolution.

Are you interested in doing something like having a golf pro hit some balls with the club, then having a complete record of the swing for analyzing the design of the club? Because I think that the IMU will have an effect on the club swing, it will be very challenging to make a design to your specs that will be both small, conformal, and light on the club's head. Based on that alone, it would be worth considering the use of cameras instead. Put some markers, or even better some IR LED's on the club and record it with several cameras. I'm sure it'll take a lot of processing though to get the appropriate data out.

I'm also not sure how well it would capture contact with the ball, do you know how long the club typically spends in contact with the ball? Very rapid events need really high sample rates to properly capture, as you can see from the Nyquist-Shannon sampling theorem. As a rough guess you would probably want at least 5 samples during the club/ball contact time.

You should start here to see what kind of sensors are available, and of course do the parametric search on digikey/mouser/etc.

There are also lots of specialized companies is this field who serve the aviation/nautical/military markets who have the best stuff, which is the most expensive.

Electroholic

Tue Jan 19 2010, 06:17PM

Registered Member #191 Joined: Fri Feb 17 2006, 02:01AM
Location: Esbjerg Denmark
Posts: 720

This device is meant to accurately capture the swing path of a driving swing, hence the high peak rate.
Basically a glorified wiimote.

Yes, I get the Nyquist theorem.
Questions
I have the spectral content of the linear acceleration and angular velocity, and that would determine the bandwidth, correct?

How is sampling rate defined? I still cant find it anywhere. For example, in this datasheet, how would I figure out the maximum sampling rate? I mean, the thing is analog, right? with proper buffering, is 50samples per second realistic?, how about 100?

1 2 next

Moderator(s): Chris Russell, Noelle, Alex, Tesladownunder, Dave Marshall, Dave Billington, Bjørn, Steve Conner, Wolfram, Kizmo, Mads Barnkob