Research On Shape Flatness Pattern Recognition For Cold Rolling Strip Mill

This thesis mainly researches on shape pattern recognition. It analyzes and compares with every method on recognizing flatness pattern, and puts forward the new methods.In this paper, we put forward using RBF network to flatness pattern recognition. RBF network is a kind of part approach to neural network. It is more capable of approach and classification and is higher speed of learning than BP network. Simulation results of RBF network method based on fuzzy distance indicate that the method is higher speed convergence and better recognition precision than FBP network.Though BP network indirect recognition method based on fuzzy distance is able to settle network structure fixedness as the plate width changes, which BP network is unable to settle, network must be differently trained according to different plate width to recognize shape caused by different plate width, so that the complexity of writingprogram is increased. According to the above problem, the paper puts forward improving on BP network direct recognition method. It can recognize shapes of every different plate width by only once learning and it simples program. The paper mainly uses GA-BP network optimizing arithmetic to make up slow convergence and local minimal in the BP network. Simulation results indicate that the method is feasible.Finally, the method of using support vector machine to recognize flatness is quested for. And some proposals are tabled by analyzing.