Research On Defects Recognition Of Flatness Shape Based On Genetic Algorithms

Steel is the material basis for developing national economy. Being one of the mainsteel products, strip has been widely used in automotive, construction, householdappliances and military industry. Plate shape is one of the quality indexes of steel. Theplates and strips have been applied to various national economy departments widely. Oneof their important quality indicators is the plate shape. Plate shape defect recognition isthe basis and key point of flatness control.By using of the analysis for the cold rolled plate shape defect pattern, a recognitionmethod based on GA and RBF-BP neural network model has been proposed. A flatnessrecognition RBF-BP neural network model with only6-input and3-output has beenestablished based on Legendre orthodoxy polynomial. At the same time, combined withthe characteristics of the genetic algorithm to optimize the weights and threshold of theinput layer and hidden layer and output layer of the RBF-BP model, formed a kind ofmethod with the combination of genetic algorithm and neural network, that is, theGA-RBF-BP model. It can identify the degrees of membership for six kinds of commonbasic pattern of shape defects. Combining BP and RBF neural network’s advantages withgenetic algorithm’s characteristics, the GA-RBF-BP model recognition method has theadvantages of approaching speed and high recognition accuracy. Based on the GA, thenchoose the error function between the actual output and the expected output as the fitnessfunction of the GA, and use the GA to optimize the weights and the threshold, and finallyreturn to the combination of neural network, after many simulation, it make the errorreach the expected value. A simulation study about the model was conducted andcompared with the simulation study of BP and RBF-BP neural network. The results showthat the GA-RBF-BP recognition method has the better effect than using BP neuralnetwork or RBF-BP network in a single way. And it also more suitable for real-timeflatness control. Finally, a set of measured data are simulated, and the simulation resultsdemonstrate the feasibility of the method.