Abstract: In order to realize the precise welding of T-joint between aircraft hyperbolic panel and the stringers, a method of solving inverse kinematics equations for cooperative welding robot with multiple manipulators based on neural network was presented. On the basis of building an effective DH model of cooperative welding robot with multiple manipulators, sample data was obtained considering the movement ranges of the robot joints. Based on BP and RBF neural networks, 18 joint subspaces are mapped to the workspaces of three manipulators. The high-dimensional and non-linear inverse kinematics problem was transformed into multi-input and multi-output prediction model. The results show that prediction model of solving cooperative welding robot kinematics equations has high accuracy, among which the BP prediction model is faster and the RBF prediction model is better.