Neural network for identification of roll eccentricity in rolling mills

The roll eccentricity in a rolling mill may define the limit of achievable thickness tolerances and thus is subject of interest for the automation equipment in hot rolling mills as well as in cold rolling mills. Today's demand on thickness tolerances less than 0,8% require efficient methods for roll eccentricity identification and compensation. This paper should present a solution for identifying roll eccentricity by using a neural network with a comparison to other methods in order to show the advantages and disadvantages for further use in a roll eccentricity compensation. The solution is verified on measured data sets of a cold rolling mill.