Abstract
A major objective in the design of rolling mills is the achievement of required final temperature and grain size in the rolled bar. Therefore, the problem is posed as an inverse problem : Determine the input conditions and the pass sequence for the desired finish temperature and grain size. This paper presents a hybrid inverse method that uses FEM, DOE and ANN techniques for developing an inverse agent. The FEM and DOE techniques are used to calculate the phenomenological and microstructural process data which is needed for training the forward and inverse neural networks. These networks are then used for process parameter design. The results of this agent are verified using FEM simulation. It is seen that the inverse agent predicts the thermomechanical conditions well for grain sizes and finishing temperatures within the training envelope. For other values, a few predictor-corrector iterations are needed to converge to a good final solution.
