Accelerometer to get Position

[inabv]

My customer wanted us to use an accelerometer to get the curve of the velocity over the position.
We used a 3 axis sensor, with analog output. Here we used a 12 bit ADC.
Now all the effort has been done, but we cant get the exact position of the movement, which is a movement in just one axis. Customer may want to bring new axes in future as well. As we did integration over the acceleration and then over the velocity, we get very diverse results, but none of them matches by far with the real distance which was carried out.
Especially if you put two sensors on the same movement they will surely give very different results.

Now customer wants dearly a result for the position with max 10% error.
We are a little bit clueless what to do to get the exact position, we even envisage in using a different sensor with more precision. But still there would be error at integration, which cant be avoided.
Has anybody some experience with getting position out of acceleration?

I am really thankful to any suggestion.
Regards, Ina

 

[dford]

Ina:

You will save yourself grief if you take a look at some of the articles relative to inertial navigation errors available on the web.

Perhaps you should consider GPS solutions?

Good Luck,

Doug

 

[GregLocock]

If you want to measure displacement then use a displacement sensor. While it is possible to integrate some transducers, problems with zero drift and so on are legion.



Cheers

Greg Locock

SIGlease see FAQ731-376 for tips on how to make the best use of Eng-Tips.

 

[inabv]

GPS solutions are not possible, as we need to measure a distance of 45 cm with 10 % accuracy.

Displacement sensor i cant also use, as everthing should be tiny, work on battery and inside a box full of materials. So a chip solution is desirable.

 

[BobM3]

Make sure your accels are rated for DC output. Subtract out the offset from the accel data before integrating (and also the resultant velocity data before integrating it).

 

[GregLocock]

"As we did integration over the acceleration and then over the velocity, we get very diverse results, but none of them matches by far with the real distance which was carried out."

Yup. And that's the way it is. Sorry about that. Bob's suggestion may make things better, but equally it could make things worse (imagine what it would say about the displacement of a body in free fall).


Cheers

Greg Locock

SIGlease see FAQ731-376 for tips on how to make the best use of Eng-Tips.

 

[GregLocock]

Actually I have thought of a couple of possible solutions.

1)Do you have a large amount of data showing measured acceleration, accurate velocity and accurate displacement?

2) Is the system well defined? That is, are there constraints on how the thing moves?

3) Is it possible to sense the exact location at some point in the machine's cycle?



Cheers

Greg Locock

SIGlease see FAQ731-376 for tips on how to make the best use of Eng-Tips.

 

[inabv]

Thanks BobM3 and GregLocock.
To your questions:
1. Yes we are sampling with 100 Hz. We may try to sample with 1000 Hz. The sampling period is usually 5 seconds.
2. the current movement we are looking into is running in just one axis. It is on a rail.
3. for testing purposes we can sense the exact location. But when it is running we have no influence on the speed, stop movement, backward movement and thats what we need to sense.

 

[BobM3]

Another consideration is that acceleration is absolute (with a DC accelerometer) and displacement is normally relative. You would need to define a "zero" point for displacement.

 

[MikeHalloran]

Stripe the rail and use optical retroreflective sensors to get incremental position. You need two mounted in quadrature to get direction.



Mike Halloran
Pembroke Pines, FL, USA

 

[inabv]

Sorry but we have no access to the rail, we are just inside one box, which is running on this rail.

 

[MikeHalloran]

At this point, given the constraints that you have stated, I'd look for an analog artificer, and do the integrations in hardware.

Yes, linear circuits, which do not suffer from sampling error or quantization error.

When they were in vogue, 'tiny' meant something different than it does today. That could be a challenge.





Mike Halloran
Pembroke Pines, FL, USA

 

[patprimmer]

an the box have a transparent insert in the bottom, over the rail.

Regards

eng-tips, by professional engineers for professional engineers
Please see FAQ731-376 for tips on how to make the best use of Eng-Tips Fora.

 

[EricRICPDAS]

It matters very much how long the integration must run. If it is a few pulses of high acceleration over a long time of low acceleration, that is a problem. Your 12 bits must then capture the high acceleration, but may not do well at the low acceleration. Because the time is long, the digitization error of the low acceleration becomes huge. This is also true of the DC offset error. If you have moderate acceleratoins for most of the integration time, and this time is short, then this approach can work. From what you say, I doubt this is the case. You may need to invest in very accurate devices if you have to do this over a long time period.

Eric Ratliff

http://www.icpdas-usa.com