Planar Delaunay Mesh Smoothing Method Based on Angle and a Deep Q-Network

Mesh quality is critical to the accuracy and efficiency of finite element calculations, mesh smoothing an essential means reducing number poor elements improving quality. The deep Q-network-based optimization algorithm for planar Delaunay (unconstrained DQN) has attracted increasing attention due its advantages in autonomous optimization. However, unconstrained DQN model does not constrain movement area central node during training process, easily falls into a local optimum, resulting low generalization model. In this paper, updateable iterative inner polygon proposed as constraint limit node’s control element’s angle. Next, performance different neural networks when same dataset analyzed, appropriate network selected. After that, effectiveness method were analyzed. Finally, results compared with those obtained by existing methods. show that can improve minimum angle global shape elements, trained high generalization.