Using accelerometer measurements?

Question

I need to know how to make real-world use of accelerometer measurements. We have a Summit Instruments 3-axis unit with a 1K rolloff frequency (hardware filtering cuts that back to about 100 Hz).

Just calculating gravity has proved to be more difficult than I had thought, as even with a very long low pass filter, I am still left with wiggles that can't possibly be from a change in gravity.

The integration error getting to velocity is huge, orders of magnitude larger than my known average velocity. Determining displacement remains just a dream.

I have had some success in caging my errors since I do have a known average velocity and can detect points in the data where all motion has stopped, providing me with some known zero velocity points.

Can anyone steer me in the direction of application-level references (theory books are pretty much useless)?

I keep running into Kalman filters. What is special about them and why would they be applied to the accelerometer problem?

Grounding should be good. But I'm not sure what you mean about the matching.

Mostly I am facing my own ignorance about how to properly use the information - what sequence of operations I need to do - and am looking for a useable reference.

I should also mention that the time interval involved would typically be on the order of an hour or so, sampling every 4 milliseconds.

BTW, I just found out about Runge-Kutta integration, which I have _not_ been doing.

Thanks for your efforts.

My real problem here is not that I suspect problems with the instrumentation but rather that I simply don't know what I'm doing.

I need help in finding references that will tell me how to set up the problem properly. For example, all of the references that I have found regarding the use of a Kalman filter always assume that the gravity vector has already been removed, yet I don't know how to properly remove the gravity so that I can integrate the accelerations.

When I low pass the data I am left with wiggles that almost certainly aren't due to gravity, yet if I make the filter longer (it is already effectively 50 ft long) I feel that I may go the other direction and lose actual changes in the gravity vector due to changing instrument orientation.

Answer 1

Maybe you have a scaling problem.

1) Is the transducer matched to the instrument? I.e., does the instrument have an input that matches the output of the transducer, e.g., similar milivolts range?

2) Check the span (min-max range) of the tranducer.

a) your variable, gravity, should fall within the span

b) your variable should be at least 50% of the span. The smaller the signal viz-a-viz the range/span, the higher the relative error, and the wiggle.

c) the narrower the span, the more sensitive your reading.

FILTERING WILL IMPROVE ACCURACY, BUT MAKE SURE YOU HAVE A SOLID READING FIRST.

Answer 2

Yes all accelerometers will have a tough time measuring speed with any accuracy. This is because of the relationship between velocity (speed) and acceleration. Acceleration is the derivative of speed. Or to go backwards.. Velocity is the integral of acceleration. if you're confused by the math terms don't worry I'll try to explain it simply. Acceleration is a measure of how much the velocity changes over time. Think of holding a phone or a wii remote with an accelerometer. At first you are holding it still. Then you move the controller rapidly up. The accelerometer essentially measures the force you put into the controller. It sees that there is no change in velocity. Then it sees that the velocity is changing in the upward direction. Unfortunately it cannot see that if you are moving it a constant speed how face it it going. It only sees the difference between the starting velocity and the final velocity. Picture yourself using a wii remote on a plane. The wii remote is only going to see the changes in velocity of your hand. It will have no idea that you are flying 300 miles per hour above the earth. That is why velocity measurements, even if you track them accurately, will always be less accurate than say, a pitot tube or even gps. In math terms the accelerometer is only evaluating the integral of speed. But it has no idea of the initial conditions. This theoretically can be compensated for by using calibration techniques with the a known mass and numerical integration techniques, however there are several places where error will be introduced and the calibration is not impossible, but quite hard. It has been done with some success on very simple systems with know mass and velocity profiles but with complicated systems like in the aerospace industry it is much more accurate to use a pitot tube. Hope that helps

Answer 3

It sounds like you might have a grounding / shielding problem with your cables connecting the gauges to the amplifiers.

Another possiblity (although probably not by your description) is that the amplifiers are not matched to the gauges.

Have you looked on Summit's website? They may have application field engineers that can do a phone consultation to troubleshoot your problem.

.