Roll Speed and Roll Gap Control with Neural Network Compensation

Abstract

In this paper the detailed procedure of roll speed and roll gap control strategy development for a laboratory scale rolling mill is given. The core of the control strategy is the incorporation of feed-forward compensations based on neural network models for the roll force and roll torque, which are the major disturbances introduced during the rolling operation. An integrated computer simulation model is developed to investigate the performance of the proposed control strategies, and results show significant improvement over the traditional feedback control scheme. Based on the control strategies and the integrated simulation model, a major upgrading scheme is undertaken on an existing laboratory scale rolling mill. The new mill data acquisition and control systems, including the upgrading of the drive and gap motors, are currently under commissioning. After the mill upgrading system is fully commissioned, further work such as online adaptation of the neural network prediction model and the fine-adjustment of the feed-forward compensation need to be investigated for consistent control performance under changing rolling conditions.

Actuator Fault Detection, Isolation Method and State Estimator Design for Hot Rolling Mill Monitoring

“…The model parameters, such as the inertia, the stiffness of the shafts and the frictions of the individual components are supposed to be known and constant (Mahfouf et al, 2005).…”

Actuator/Sensors Fault Diagnosis for an Experimental Hot Rolling Mill - A Case Study

“…The model parameters, such as the inertia, the stiffness of the shafts and the frictions of the individual components are supposed to be known and constant (Mahfouf et al, 2005). According to the sensors available on the experimental hot rolling Hille mill, a roll speed discrete state space representation can be established as follows: faulty case, the roll speed discrete state space representation becomes as:…”

“…As illustrated in Fig. 2, based on the roll torque estimation performed by a multilayer neural network model (Mahfouf et al, 2005), a feedforward control loop is cancelling its effects upon the roll speed to achieve a fast speed response and to have better disturbance rejection ability coming from the load torque fluctuation during rolling. …”

Research on the Comprehensive Control Strategy of Over-welding Based on Decoupling- Design for Tandem Cold Mills

“…(1) Systematic function structure of dynamic over-welding strategy Figure 4 presents the systematic function structure of dynamic over-welding process, where, G is the implementation inner seam of roll gap; B is the implementation inner seam of bending force; G SET is the over-welding roll gap giver; B SET is over-welding bending force giver; TMi is the ith stand outlet tension detector [13,14].…”