Development of a neural network model for a cold rolling process

Abstract This paper describes the development of a neural network model for the flat rolling process. This neural network was based on the backpropagation paradigm. A nonlinear mathematical model based on the slab method was developed to guide and supervise the learning procedures. A near-optimal neural network structure was determined by using a development process that emphasized second-order derivative information. The application of this process yielded improvements in the learning errors, prediction errors, and training times. A robust and accurate model was obtained as a result of this process.