In view of the low precision of reverse software for 3D point cloud model hole repair, a repair strategy based on neural network algorithm on the basis of software repair was proposed. Taking the hole of the inner cavity of the bucket tooth of a excavator as an example, the mind evolution algorithm (MEA) was used to optimize the initial weight and threshold of the BP neural network. The improved neural network model MEA-BP was obtained, and the repair data were adjusted, and the results were compared with those of the PSO-BP neural network and the GA-BP neural network hole repair.The results show that the MEA-BP neural network model has higher efficiency and better repair effect, and can repair holes effectively and accurately under the condition that the prediction accuracy is improved.