Host: The Society of Instrument and Control Engineers, Control Division
Pages 49
A modeling error has a time-variant factor like the influence of roll wear and a factor by rolling conditions like the size characteristic. If both factors are not distinguished when learning the modeling errors and adaptation of a setup control for the following rolling piece are executed, the appropriate compensation is not obtained, then accuracy gets wores. We propose the synchronous learning algorithm for both factors with learning gain scheduling to improve learning efficiency. Finally, the application simulation using hot-rolled width data shows that standard deviation for modeling error can be reduced 20% compared with the conventional method.