Neural Network-Based Prediction Welding Residual Stress Of Aluminum Welded Joints

By the end of 2014, China’s high-speed railway has more than 16,000 km, high-speed trains and high-speed EMU have made tremendous contributions for passengers and economic development of our country. Aluminum has high toughness, low density, high strength, good corrosion resistance, etc, the manufacturing quality of car body about 30%lower than that of the steel material, is the best material for lightweight car body design.High-speed train body in the welding process, due to uneven heat input welding, the weld joints will produce welding residual stress. Related literatures show that have a potential impact on the stability of the structure of welding residual stress, the static strength, fatigue,corrosion resistance, etc., are important factors affecting the quality of welding speed train body, so by computer-aided engineering and artificial neural network algorithm to predict welding residual stress values, in order to take reasonable technical measures to reduce the harm of great significance.In this paper, Based on the 6005- T6 aluminum alloy as the research object, using SYSWELD software process of aluminum alloy thin plate welding temperature field, stress field numerical simulation analysis, the simulation analysis of the influence of process parameters on the welding stress field distribution; Process parameters are established and the stress field parameters of BP neural network forecasting model and RBF neural network prediction model;Finally, based on neural networks for high-speed EMU body sidewall maximum longitudinal welding residual stress analysis were forecast. The main contents are as follows:Based on the numerical simulation experiments, using BP neural network and RBF Neural networks to establish a welding parameter(welding current,welding voltage,welding speed) and the welding process(maximum residual stress, maximum longitudinal residual stress, maximum Vonmises stress) forecast model. This model can be the maximum transverse residual stresses, maximum longitudinal residual stress, maximum von Mises stress to predict specific guidance on selecting reasonable processing parameters.Build a prediction model of neural network on the basis of the established high speed EMU car 8 side wall of seam welding process parameter(welding current, welding voltage,welding speed, preheating temperature) and the forecasting models of maximum longitudinal residual stress. The prediction results show that the model prediction error within the permitted range for high-speed EMU identified the body sidewall welding manufacturing process parameters to provide guidance.