Kalman Filters & accelerometers - linear vs extended, what do I need?

In summary, the author is trying to integrate a single accelerometer reading in order to gain insight into the resulting velocity and position curves with respect to time. He is also implementing a LVDT in an attempt to directly measure displacement with time. He is suspecting that he will require an extended Kalman filter (EKF) in order to model the nonlinear accelerometer data and LVDT displacement curve. He is looking for feedback on whether this is feasible and if he is on the right track.
  • #1
TTM
18
0
Hello,
I have a problem for work where I am attempting to integrate (a single) accelerometer readout in order to gain insight into the resulting velocity and position curves with respect to time. From experience I know this is a tricky task due to drift and error being amplified when integrated.

Just some background: displacement of our subject will most likely occur over a short time span - millisecond scale due to an impact force during the test. A http://en.wikipedia.org/wiki/LVDT" will also be implemented during the test in an attempt to directly measure displacement with time. However, I believe this data will be erroneous and spotty at best due to the high sample rates involved (possibly tens of khz) Worst case, the LVDT will give us a very accurate total displacement value with which we can check our accelerometer against.

Based on my reading and intermediate understanding of the linear Kalman filter, I am suspecting that I will require an extended Kalman filter (EKF) in order to model the nonlinear accelerometer data and LVDT displacement curve.

Since both of these tools can eventually measure the same thing (displacement) it is my thinking that their results can be combined and filtered to eventually wind up with a displacement curve w.r.t time that agrees with the total displacement measured by the LVDT. Please see this flow chart:

[Accelerometer data] + [EKF] -> [Double integration] --> [Displacement curve 1]

[LVDT] + [EKF] --> [Displacement curve 2]

EKF {[Displacement curve 1] + [Displacement curve 2]} --> [Most trustworthy displacement curve]

I want to figure out from those with more experience if this is feasible and if I am on the right track..? Thank you for reading and I appreciate any tips/criticism or suggested reading you may offer.
 
Last edited by a moderator:
Engineering news on Phys.org
  • #2

FAQ: Kalman Filters & accelerometers - linear vs extended, what do I need?

1. What is a Kalman filter and how does it work?

A Kalman filter is a mathematical algorithm used for filtering and predicting data based on noisy measurements. It combines past measurements with a prediction of the current state to produce a more accurate estimate. It essentially uses a series of equations to update and refine the state estimate over time.

2. What is the difference between a linear and extended Kalman filter?

A linear Kalman filter assumes that the system being modeled follows linear dynamics, meaning that the relationship between the current state and the previous state is linear. An extended Kalman filter, on the other hand, can handle non-linear systems by using a linear approximation of the system dynamics. This allows for a more accurate estimate in non-linear systems.

3. When is a linear Kalman filter appropriate to use?

A linear Kalman filter is appropriate to use when the system being modeled follows linear dynamics and the measurements are Gaussian (normally distributed). This is often the case in engineering and physics applications.

4. When is an extended Kalman filter appropriate to use?

An extended Kalman filter is appropriate to use when the system being modeled is non-linear and the measurements are Gaussian. This is often the case in applications involving sensors, such as accelerometers, which measure non-linear motion.

5. What equipment do I need to implement a Kalman filter with accelerometers?

To implement a Kalman filter with accelerometers, you will need the accelerometers themselves, a microcontroller or other computing device, and software for implementing the filter algorithm. You may also need additional sensors, such as gyroscopes, to improve the accuracy of the filter. Additionally, knowledge of programming and mathematics is necessary for implementing and fine-tuning the filter.

Similar threads

Replies
1
Views
1K
Replies
24
Views
10K
Replies
5
Views
2K
Replies
1
Views
2K
Replies
1
Views
8K
Replies
1
Views
2K
Replies
9
Views
9K
Replies
4
Views
5K
Back
Top