Optimization of Metal Inert-Gas Welding Process for 5052 Aluminum Alloy by Artificial Neural Network

5052 aluminum alloy MIG welding is carried out with filler wire ER5356. The welding for thirty six samples is carried out with three process parameters, which are current (70–100 A), arc voltage (14–20 V) and welding speed (7–9 mm/s). Artificial neural network (ANN) is used to reversely establish the multi-output welding process optimization model to directly obtain the welding process parameters under certain mechanical properties. By comparing the model performance under different input layer activation functions, it is concluded that the model performance is optimal when the input layer activation function is tanh (Hyperbolic Tangent) function. The number of middle layer neurons varies from 1 to 25 at the step of 2 by grid search method. It is found that when the number of middle layer neurons is 17, the model performance is optimal. Finally, by observing the model training process curve, when the epoch of model is 750, the model has the best performance. The test set error of the model is 4.37%, and the optimal welding process parameters are obtained. Welding current is 81.1 A, arc voltage is 16.2 V, and welding speed is 9.4 mm/s.