How Does a Kalman Filter Impact PI Controller Performance?

[flo24601]

Hello,

I am using a Kalman filter with a PI controller.
The goal is to control a mirror to stabilize the position of a laser spot: I am sending a disturbance on the mirror and the PI sends command to the same mirror so the spot stays fix.
With a camera I am recording the image of the laser and I compute the centroid to get its position.
With the Kalman filter I also get an estimation of this position.

My problem is the following:

• Without the KF the PI controller works fine: after an overshoot the error goes to 0 (observing the measurements)
• With the KF I thought I would decrease the overshoot but the measurements are the same (slightly worse even) but the overshoot of the estimations is lower

My questions are:

• Which values should I consider (being the real position): the measurements or the estimations ?
• Why doesn't the KF have an impact on the measurements ?

I can of course clarify things if needed

Thanks

 

[donpacino]

Which values should I consider (being the real position): the measurements or the estimations ?

That entirely depends on the accuracy of the measurements and the accuracy of the model you're using for the kalman filter.

Why doesn't the KF have an impact on the measurements ?

perhaps you don't have an understanding of what the kalamn filter is supposed to do. A kalman filter is a tool to use noisy, bad, or partial data in conjunction with a system model to estimate a state. What is the kalman filter estimating?? If the kalman filter is estimating position, and you already have an accurate measurement of that position, the kalman filter will not help you at all. If you have a poor measurement, and the kalman filter improves that measurement, then you might get an improved step response. it all depends.

My favorite example of what a kalman filter can do is an avionics GPS system.
in aircraft there are generally two things used to estimate position. the gps unit and the inertial measurement unit. the gps and IMU are used in conjunction to get an accurate measurement. One is good at course measurement, but poor at fine measurement, and the other is the opposite. A kalman filter is used to merge the two poor and good data sources.

to decrease overshoot change the tuning on your PI controller