Velocity through Accelerometer without Drift

[educosta]

I’m making a project about Control and Optimization of the Electromagnetic Suspension Operation of a MAGLEV Vehicle.
In this, I need to measure inertial acceleration and make integration to get velocity of the Vehicle , but I’m having a PROBLEM WITH DRIFT.
I’m using the ADXL 105EB accelerometer to measure acceleration of the Vehicle in heave axis.
In the start, the Microcomputer with 12 bit A/D converter reads values of acceleration with 5000 Hz Sample rate by one minute and calculates the average (reference 0g).
The velocity is the integration of: reference 0g – acceleration in an instant. But after some seconds appear a drift in the velocity.

I’d like to know if would have some solution to measure this velocity without drift.


Thank you,

 

[GregLocock]

Integration of signals is very risky if you are worried about drift. This is because any DC error will accumulate over time.

I suggest you use a velocity measuring device, or, measure displacements and differentiate.

Cheers

Greg Locock

 

[MikeyP]

Do you start the measurement from a state of rest and finish it at a state of rest? If you do, then you can use frequency domain integration. Take the WHOLE acceleration signal from beginning to end (including any time required for vibrations to die away), then perform a Fourier transform to transform the signal to the frequency domain. Divide each spectral line by i*omega. Then perform the inverse Fourier transform to return the signal to the time domain. If there are still problems with drift, simply set the first few spectral lines to zero before performing the inverse transform. You will lose information about the very lowest frequencies in the signal but in most circumstances, this information is not of much use anyway.

M

 

[EnglishMuffin]

Is the "drift" linear ? Usually when you integrate an acceleration signal over time there is a slight dc offset which gets summed and produces an apparant superimposed linear change in velocity over time. When I have done this in the past I have simply written software to subtract out the linear drift. The toughest part is figuring out how to read the file data. But I was fortunate in my case since it was obvious what the drift was - I was actually finding the displacement of an automobile valve by mounting an accelerometer on the valve itself. When you integrate a second time to get displacement there is usually an additional error - which this time is due to A to D round-off - converters are only 12 bit usually. This question seems to come up a lot on this site and you can also find discussions of it on the internet itself.

 

[tomirvine]

I recommend using a Bruel & Kjaer 2635 signal conditioner with a charge mode accelerometer.

This signal conditioner integrates the accelerometer signal. The output is a velocity signal.

The conditioner also performs a highpass filter of the signal prior to integration. This removes the drift.

All of these operations are performed on the analog signal. The output is thus an analog velocity signal.

Tom Irvine

http://www.vibrationdata.com