Genetic Algorithm Coupled with the Neural Network for Fatigue Properties of Welding Joints Predicting

Abstract

The prediction of fatigue life of metal welded joints plays an important role at lower manufacturing costs and reduces accidents for engineering materials, the response of metal welded joints to fatigue properties has highly non-linear, so it is difficult to establish an accurate theoretical model using traditional method to predict its fatigue life. It is appropriate to consider modeling methods developed in other fields in order to provide adequate models for metal welded joints behavior on fatigue properties. Accordingly, a new system predict method, based on a hybrid genetic algorithm (GA) with the Back-propagation neural network (BPNN), for the simultaneous establishment of a predict model structure of fatigue life of metal welded joints and the related parameters is proposed. Based on the self-learning ability and approximation of non-linear mapping capability of the BPNN, by taking the advantages of the powerful ability of global optimization, implicit parallelism and high stability of the GA, the optimal parameters have been automatically determined, we establish a parameter adaptive optimization of GANN model to fit and predict the fatigue life of metal welded joints. GANN establishes the mapping relationship between the fatigue properties of metal welded joints and a variety of influencing factors, having greatly increased the computational efficiency for the fatigue properties of metal welded joints, also had a higher predict accuracy. The superiority of GANN had been tested by the prediction of the fatigue life of welded joints in different process parameters.