Machine Direction Adaptive Control on a Paper Machine

Abstract

Control of industrial sheet and film processes involves separate controllers and actuators for minimizing both temporal variations along the machine direction (MD) and spatial variations along the cross direction (CD). Model-based control methods such as model predictive control (MPC) have gained widespread implementation for controlling both the MD and CD processes. One limitation of model-based methods is that changes in the true process pose significant identification challenges for operators which are often resolved with costly open loop identification experiments. The predominant industrial measurement technology acquires a signal of mixed MD and CD variations that requires separation. This work compares various model-free approaches for MD-CD separation as a prerequisite to effective MD control. To address the challenges of model-based control, this paper introduces an adaptive control method for the MD process. Closed loop identification experiments are conducted and compared to benchmarks on an industrial paper machine simulator.

Multiobjective Tuning Technique for MPC in Grinding Circuits

“…The authors have observed a dramatic performance change between the initial guess and final solution. In a similar vein, [19] proposed an adaptive MPC strategy for paper machines under uncertainty and model degradation. The algorithm relies on exciting the process inputs to compute and minimize the variance of a Fisher information matrix defined by the process model.…”

“…The algorithm relies on exciting the process inputs to compute and minimize the variance of a Fisher information matrix defined by the process model. By recalculating the process model instead of re-tuning the controller algorithm when deviation from the nominal operation point is detected, [19] achieved adequate tracking performance under model uncertainty, but since the decision variables of the optimization problem are dependent on the prediction horizon and the number of inputs and outputs of the model, and moreover since a exhaustive search method was proposed, the method might not be feasible for real time applications in large systems. In [9], a dynamic model has been developed for a grinding system and its steady-state form was used to calculate robust H ∞ controllers that maximize throughput while maintaining the SAG mill power draw as close as possible to its upper bound.…”