钣金件的自动特征识别技术能从参数化模型中提取加工特征信息，为后续的展开与工艺规划提供基础支撑。然而，现有钣金件特征识别方法难以区分拓扑结构相同的折弯特征与焊接特征，同时钣金件中包含的大量面又限制了特征识别效率的提升。本文提出了一种基于图简化的钣金件折弯特征快速识别与展开优化方法。该方法重新定义了折弯特征，通过构建折弯特征树来表征不同折弯特征之间的关联，实现对拓扑结构相同的折弯特征与焊接特征的区分。为了提高折弯特征识别的效率，在折弯特征树构建前，引入基于内环与厚度面的图简化算法，以简化扩展属性邻接图。对于包含焊接特征的钣金件，基于折弯特征树优化其展开方法并获取其展开模型。测试结果表明，本文算法能够区分钣金件中拓扑结构一致的折弯特征与焊接特征。对于扩展属性邻接图中含环状结构的钣金件，本文算法能将其扩展属性邻接图转化为树结构的折弯特征树，进而实现钣金件的展开。典型案例显示折弯特征识别效率相比于图简化前可提升89%。

The automatic feature recognition technology for sheet metal parts can extract machining feature information from parametric models, providing a fundamental basis for subsequent unfolding and process planning. However, current feature recognition methods for sheet metal parts struggle to distinguish between bending features and welding features that share the same topology, and the large number of faces in sheet metal parts also limits the improvement of feature recognition efficiency. This paper proposes a graph-simplification-based rapid recognition for bending features and unfolding optimization method in sheet metal parts. The proposed method redefines the bending feature and constructs a bending feature tree to represent the relationships among different bending features, thereby enabling the distinction between bending and welding features with identical topological structures. To improve the efficiency of bending feature recognition, a graph simplification algorithm based on inner loops and thickness faces is introduced before the construction of the bending feature tree to simplify the extended attribute adjacency graph. For sheet metal parts containing welding features, the unfolding process is optimized based on the bending feature tree to obtain the unfolded model. Experimental results demonstrate that the proposed algorithm can effectively distinguish bending features from welding features with identical topology. For sheet metal parts whose extended attribute adjacency graphs contain loop structures, the proposed method can transform them into tree-structured bending feature trees, thereby enabling effective unfolding. And typical case show that the efficiency of bending feature recognition improves by up to 89% compared with that before graph simplification.