Eccentricity filter for hot rolling mills
Industrial process control using neural networks

When using thickness gauges to control the thickness of hot strip, reducing the influence of disturbances such as roll eccentricity and chock on the rolling force signal is a key factor in the quality of the control system.

A large number of filter-based systems are available to perform this task, but while they compensate for these shortcomings, they also have the disadvantage of distorting the useful signal, introducing a phase shift in particular, and thus reducing the stability of the control loop. Since disturbances are also non-linear, compromises have to be made if linear filters are chosen.

The non-linear structure of the artificial neural network filter offers new degrees of freedom for optimization.

To achieve optimum control quality, an eccentricity filter must satisfy the following requirements:...