Peter Nachtwey
Member
About Kalman Filters
There are Kalman filters. This is a predictor corrector method and the Kalman gains ( corrector gains ) are computed each iteration. This is time consuming and complicated. This is too much for a PLC. In addition, a proper Kalman filter requires knowing the process noise and the measurement noise. What the Kalman filter does is weight a process prediction with the current measurement based on the noise. The sad truth is that even on the engineering forums the engineers generally don't know the process noise and the measurement noise so the 'fudge' it. Well what is the point of that?
There are static Kalman filters. Static Kalman filters assume the process noise and measurement noise don't change so the Kalman gains only need to be calculated once. It is possible to implement a static Kalman filter on a PLC BUT, what is the process noise and measurement noise.
Finally, there are alpha-beta-gamma filters. These are dumbed down Kalman filters. They aren't optimal bit neither are Kalman filters if you don't know the process and measurement noise. We implement alpha-beta-gamma filters on our newest products. The customer only needs to select a bandwidth that works for him. There is no need to know the process or measurement noise. Like the Kalman and static Kalman filter there is a predictor and corrector phase so one gets 95% of the benefit of a Kalman filter with 5% of the effort. I think that is a good bargain. There is no problem implementing an alpha-beta-gamma filter in a PLC.
http://deltamotion.com/peter/Maxima/ABG.html
There are Kalman filters. This is a predictor corrector method and the Kalman gains ( corrector gains ) are computed each iteration. This is time consuming and complicated. This is too much for a PLC. In addition, a proper Kalman filter requires knowing the process noise and the measurement noise. What the Kalman filter does is weight a process prediction with the current measurement based on the noise. The sad truth is that even on the engineering forums the engineers generally don't know the process noise and the measurement noise so the 'fudge' it. Well what is the point of that?
There are static Kalman filters. Static Kalman filters assume the process noise and measurement noise don't change so the Kalman gains only need to be calculated once. It is possible to implement a static Kalman filter on a PLC BUT, what is the process noise and measurement noise.
Finally, there are alpha-beta-gamma filters. These are dumbed down Kalman filters. They aren't optimal bit neither are Kalman filters if you don't know the process and measurement noise. We implement alpha-beta-gamma filters on our newest products. The customer only needs to select a bandwidth that works for him. There is no need to know the process or measurement noise. Like the Kalman and static Kalman filter there is a predictor and corrector phase so one gets 95% of the benefit of a Kalman filter with 5% of the effort. I think that is a good bargain. There is no problem implementing an alpha-beta-gamma filter in a PLC.
http://deltamotion.com/peter/Maxima/ABG.html