The Precise Prediction of Rolling Forces in Heavy Plate Rolling Based on Inverse Modeling Techniques

Abstract

The industry scale production of heavy steel plates is performed in plate mills using reversing mill stands and roll schedules with 30 or more passes. Ideal pass scheduling is dependent on a precise prediction of the roll force in each pass. To obtain these forces process models based on the slab theory together with semi-empirical material models are most frequently used. The material parameters necessary to calibrate the models for a specific steel grade are conventionally generated on a lab scale via time-consuming and expensive compression tests. This paper will introduce a concept that allows obtaining the material model parameters directly from the rolling process on an industrial scale by an inverse modeling technique. In this context, the parameters of the material model are adjusted according to the current offset between measured and calculated roll forces. The adjustment is enabled by a non-linear optimization that aims for a minimum deviation between measured and calculated roll forces for each pass. Several results gathered in an optimization run with industrial data of about 2650 produced plates are presented. The concurrence between measurement and prediction using the optimal parameters is in detail shown for three different schedules. In summary, inverse modeling seems to allow utilizing data of past rolling processes to determine material model parameters with high accuracy.

Advanced automatic pass schedule design for hot rolling by coupling reinforcement learning with a fast rolling model

“…For the coupling with RL, an already existing fast rolling model is used, which was developed and validated at IBF by Lohmar et al [11]. It consists of several modules allowing the prediction of deformation, temperature and austenite grain size evolution as well as rolling forces and torques.…”

“…Beynon and Sellars [10] present a rolling model called SLIMMER, which is able to describe the microstructure evolution and predict the rolling force and torque during hot rolling. Inspired by their work, Lohmar et al [11] extended a similar model to include height resolution within the workpiece and considered the influence of shear during deformation. In addition to these classical approaches, data-driven methods have also been increasingly used in recent years to model the hot rolling process, for example by Shen et al [12] to predict rolling forces.…”

Coupling of an analytical rolling model and reinforcement learning to design pass schedules: towards properties controlled hot rolling

“…To achieve this, three steps regarding the FRM and the reinforcement learning algorithm are necessary. As mentioned before, an existing rolling model (Lohmar et al, 2014b) is used for the coupling with a reinforcement learning algorithm. The model is able to calculate inter alia the roll force and the austenite grain size, but it is not capable of predicting final properties after the hot rolling process.…”

“…Seuren et al (2014), showed that such a model can be extended to incorporate shear strain determined in FE simulations and in turn improved the microstructure prediction over roll stock thickness. Furthermore, Lohmar et al (2014a) combined inverse modeling and fast rolling models to determine semiempirical material model parameters based on industrial rolling data.…”

Modeling steel rolling processes by fluid-like differential equations

“…In [22] quantitative results for the accuracy of a fast model used in the steel industry is given. For roughly 40.000 single roll passes the relative deviations between the model and the measured roll force were computed.…”