A study on genetic algorithm to select architecture of a optimal neural network in the hot rolling process

In the face of global competition, the need for the continuously increasing productivity, flexibility and quality (dimensional accuracy, mechanical properties and surface properties) has imposed a major change on steel manufacturing industries. Conventional rolling force formulas, however, provide not more than reasonably exact approximations. The mathematical modelling of the hot rolling process has long been recognised to be a desirable approach to aid in rolling operating practice and the design of mill equipment to improve productivity and quality. However, many factors (strain hardening and strain rate hardening, friction condition, flattening of rolls, deflection of the rolling mill and temperature, as well as their interactions) make the theoretical analysis of the rolling process very complex and time-consuming. This paper proposes a new genetic algorithm (GA) to select the optimal architecture of the neural network and compared with that of engineer’s experience. It is shown that learning approach with the optimal structure of neural network could be applied to predict the rolling force for lot of change in hot rolling process and compared between the predicted rolling force by calculated from developed rolling force model and actual rolling force.