摘要： Fiber laser keyhole welding as a popular metal joining process has been widely used in a variety of applications especially automotive, shipbuilding and aerospace industries. Although process parameters determination based on experiments is the frequently used in the practical welding, it is often a very costly and time consuming. Accurately predicting the weld bead without expensive trial experiments has great theoretical significance and engineering value for welding process parameters pre-selection. An innovative volume heat source model was proposed for weld bead geometry prediction through finite element analysis (FEA) in fiber laser keyhole welding. The hybrid heat source model consists of a double ellipsoid heat source and a 3D Gaussian heat distribution model. To validate the effectiveness of the proposed heat source model, the fiber laser keyhole welding of the stainless steel SUS301L-HT has been carried out in this paper. The main three parameters, laser power (LP), welding speed (WS) and focal position (FP) have been taken into consideration as the design variables. Both of the predicted values from the FEA and back propagation neural network (BPNN) are compared with the experimental results. The FEA predicted results achieve good agreement with experimental results of weld bead shape and dimension and are better than BPNN predicted results. The objective variation trend is also analyzed by two prediction methods. From the discussion, it is revealed that the proposed prediction method of weld bead is effective for fiber laser keyhole welding process and replacing the expensive experiments.