Cold Rolling Force Prediction Base On BP Neural Network

Rolling process is complicated and non-linear forming process. In fact, the modeling of rolling force is a non-linear mapping problem from many factors affecting rolling force to rolling force.Traditional theoretical formulas of rolling force,or returning model, are simplified and deduced under many assumptions. So it's difficult to define the properties of rolling process and compute the rolling force accurately.Because of its unique ability to process information, the technology of neural networks is very appropriation for the computing of rolling force .Here, a mathematical model of rolling force is founded using the technology of rolling force. Some developments are made in the modeling:1. Analyzing the properties of rolling process of strip,selecting the main factors affecting the rolling force.2. Presenting a new mathematical model of aluminum rolling process according to BP neural networks.After the studying and testing of this system using VAI production data, the fact is that the maximum error is below 6 percent. It's better than the maximum error of traditional rolling force formula, which is more than 40 percent. The accuracy of the model is fit for engineering application.The BP neural network, which adopted Levenberg-Marquardt optimized algorithm has been used to predict rolling force on cold rolling mill. In this algorithm, the parameter μ can be adaptively adjusted, the network convergence speed is higher. The precision of rolling force prediction has been improved. It has provided a new precise and high efficiency way for rolling force prediction on cold rolling continuous mills.