Predicting the temperature distribution in friction stir welding thick 2219 aluminum alloy plate based on LSSVM

During friction stir welding (FSW) thick 2219 aluminum alloy plate, there exists large temperature gradient in direction of thickness and width of the workpieces welded. Temperature distribution affects mechanical properties of the welded joint. However, the method of predicting temperature distribution of thick plate is still in exploration stage. The authors propose a method to predict the temperature distribution in FSW thick 2219 aluminum alloy plate based on least square support vector machine algorithm. A least square linear system is used as the loss function to establish the regression model. The model takes the distances from points of the weldment to weld center and lower surface of the weldment as independent variables, and the peak temperature as the output. All-factor FSW experiment is conducted with welding speed from 75 to 125 mm/min and rotation speed from 300 to 450 r/min. The peak temperature of sampled points measured by thermocouple experiment and corresponding position information are collected as the dataset for the model. The radial basis function (RBF) is used for optimizing regression model. The optimal combination of core parameters of RBF is determined based on the grid search method. After training model, the peak temperature of any positions in the direction of thickness and width of the weldment is predicted. The maximum relative error between the predicted results and the experimental results is 2.95