The characteristic parameters can be extracted from both experimental and numerical specimens and the mapping relations can be established by BP artificial neural network. The output parameters can be then predicted based on the massive input data. Friction stir welding is taken as example. The input parameters include rotating speed

welding speed and the distance from weld centerline. Weld hardness is taken as output parameter. 3-10-1 3-layer topology model of6061-T6 aluminum alloy friction stir welding head hardness BP artificial neural network is established. The data from 13 conditions with different rotating and moving speeds are selected for training and testing in BP artificial neural network model. The comparison with experimental data shows the validity of the established BP artificial neural network model.The test results show that BP artificial neural network has good ability for prediction of joint hardness

which provides a new technical method for prediction of mechanical property of weld.