Immune Genetic RBF-BP Neural Network Method For Cold Rolled Flatness Pattern Recognition

In recent years, with the surplus Production of steeling enterprises, the steelproducts with high added value are paid more and more attention. The higherrequirement will be put forward for the modern manufacture of cold-rolled strip.Therefore, achieving the transformation from quantity to quality for strip productionbecomes a problem needed to be solved imminently in modern steeling enterprises.As one of the important quality indicators of strip production, the shape controlbecome a significant development direction of modern rolling technology with highprecision.In this paper, the recognition of cold-rolled strip shape defect is chosen as aresearch subject by using artificial intelligence theory. By means of the analysis forthe cold rolled plate shape defect pattern, a shape defect recognition method usingRBF-BP neural network model based on immune genetic algorithm (IGA）is proposedin this paper. Namely, the combined RBF-BP neural network is optimized by adoptingimmune genetic algorithm, and the optimized network is as a recognition model forthe cold rolled strip shape defect, the memberships relative to six basic patterns ofcommon plate shape defects are identified. This method syncretizes the advantages ofRBF and BP neural network, and there are very fast approaching speed and highprecision of network recognition. Immune genetic algorithm introduces the antibodydiversity and concentration regulation mechanism of immune algorithm to geneticalgorithm, so that the optimization effect of genetic algorithm is improved. From thestudy and comparison for kinds of networks as BP, RBF-BP, RBF-BP optimized bygenetic algorithm, and RBF-BP optimized by immune genetic algorithm(IGA-RBF-BP), the results prove that IGA-RBF-BP neutral network has higherrecognition accuracy. And immune genetic algorithm can overcome the defects ofgenetic algorithm which the individual diversity decreases quickly, and thephenomenon of premature convergence is easy to arise, so that the plate shaperecognition effect is made better. And it is more suitable for real-time shape control.