Selbstoptimierende Banddickenregelung zum hochpräzisen Kaltwalzen mit piezoelektrischen Aktoren

  • Self-optimizing automatic gauge control in high precision cold rolling with piezoelectric actuators

Wehr, Matthias; Abel, Dirk (Thesis advisor); Hirt, Gerhard (Thesis advisor)

Aachen : RWTH Aachen University (2021)
Dissertation / PhD Thesis

Dissertation, Rheinisch-Westfälische Technische Hochschule Aachen, 2021


The cold rolling process is of great economic and technological importance in the process chain for processing semi-finished products. In order to achieve high productivity, it is operated industrially at high strip speeds of up to several thousand meters per minute and uses automatic gauge control (AGC) systems. AGC systems require precise process models and fast actuators to compensate disturbances and uncertainties. Here current systems reach their limits. Piezoelectric actuators (PSA) meet the requirements for speed and positioning accuracy. In this thesis they are tested for roll gap adjustment in a rolling mill for narrow strips on an industrial scale. In addition, the models of the roll stand and the forming process are adapted online in order to improve the roll force prediction already during the process run time and to save time during machine setup.For machine control, a cascaded structure consisting of control loops for the PSA and the roll gap is developed. For the positioning of the PSA a sliding mode controller is used, which robustly compensates unmodeled nonlinearities. At the same time it takes into account the stiffness of the roll stand and strip as well as the occurring loads. Due to their small travel, PSA cannot realize the stroke for forming themselves. Therefore, an electromechanical spindle drive (EST) is used to support the PSA. A model predictive control (MPC) coordinates the two actuator types for roll gap adjustment. By measuring the strip thickness before the roll stand, a predicitve roll gap adjustment can be performed. Roll eccentricities are detected by a roll gap observer and compensated by the MPC.The higher-level control loop realizes the strip thickness control by self-optimization of the models. The required measurement values are synchronized, selected and processed online. The parameters of the roll stand characteristic curve are determined using the recursive least squares method which allows precise adjustment of the rolls even without calibration. To optimize the rolling force model, a non-parametric model is used on the one hand, which is determined by the Gaussian process regression (GPR). This allows a model to be easily determined in stationary operation, while it guarantees reliable extrapolation for unknown operating points. On the other hand, a parameter estimation of the yield stress and friction for the rolling model developed by Bland & Ford is performed using the Nelder & Mead algorithm.The PSA are positioned in high dynamic operation with a mean absolute error of 0.44 µm. They are suitable for use in the rolling mill at low forces and with support from the EST. The required force is determined by the model adaptations and the roll gap is adjusted accordingly. The control system achieves a strip thickness of 950.9 µm +3.4 µm -3.2 µm for a reference value of 950 µm.