摘要：AbstractCold rolling is a process that finishes the production of flat steel and must therefore guarantee high strip precision. However, the strip thickness produced in the roll gap cannot be measured directly which makes its observation in the roll gap challenging. In this paper, the model of both the mill frame as well as the cold rolled strip are optimized online using measured process data. A Recursive Least Squares parameter estimator is used to determine mill modulus and offset of the roll stand, while the rolling model of the steel strip is adapted using Gaussian Process Regression. The adapted models are then used in a model based controller which adjusts the roll gap accordingly. Experimental results show that the precision of the models is enhanced using online measurements. As a result the desired strip thickness is achieved despite initial model uncertainties.
关键词：KeywordsCold rollingModel adaptationGaussian Process RegressionRecursive Least Squares EstimationSelf-calibrationReal-time control
