Perhaps an intelligent machine? Starting with a fix table and then adding correction data, after measuring actual current they had with fix table data?


Hmm, such parabola fits can be dangerous: If two calibration points are too close to each other and they deviate only slightly from ideal curve, parabola methode tries to compensate this by a heavy square term. Then, all the three points will go through the parabola somehow, but curve looks totally wrong.

This effect is the more relevant the higher terms are used. I had an example, where fifth order polynom approximation was used. But customer didn't follow the instructions of manual and set certain calibration points too close to each other. As his calibration was rather 'noisy', he ended up with a totally bizarre calibration curve, full of spiky mountains and deep ravines.

Antoher trap is rounding errors of calculation of coefficients. There are certain situations, where higher precision is needed than expected, otherwise the coefficients can look very strange. So, calculation of coefficients can be troublesome.

In many applications the good old piecewise point to point linear approximation methode gives best results.

In your situation I would first try to improve the table entries. Check, whether the deviations are from mains distortion or voltage dips. If everything is checked out, then your table will very probably compensate all machine typical effects. Then you should try the whole stuff at a total different location, maybe a factory of friends, where you can check your machine again. Then, there should not develop further deviations.

Kai