形状特征与领域知识驱动的多特征多领域铸件模型检索算法

Multi-Feature and Multi-Domain Casting Model Retrieval Algorithm Driven by Shape Feature and Domain Knowledge

针对铸造典型应用领域中,由于以形状分布直接检索复杂铸件三维模型的效果差,造成的历史成熟工艺复用难等问题,提出了一种形状特征与领域知识驱动的多特征多领域铸件模型检索方法。该方法基于D2算子、模数算子等6个特征算子提取三维模型形状、热节等结构特征与铸造工艺知识特征,建立多个特征向量描述三维模型。计算并比较特征向量相似度大小进行模型的筛选,检索出工艺设计方案相似的三维模型。开展航空、航天、轨道交通、工程机械等领域铸造企业产品三维模型库的相似检索实验,在多领域多类别模型测试集、流体机械单领域多类别模型测试集中,Top-3模型检索准确率分别为85.06%、95.83%。结果表明,相较于传统形状分布算子,该方法对于铸件三维模型描述能力强、检索性能高,为后续基于三维模型检索的铸件历史工艺复用提供技术支撑。

A multi-featured and multi-domain casting model retrieval method driven by shape features and domain knowledge is proposed for casting typical application areas, due to poor direct retrieval of casting 3D models by shape distribution and the resulting difficulty in reusing historical mature processes. The method is based on 6 feature operators such as D2 operator and modulus operator to extract 3D model shape, hot spots and other structural features and casting process knowledge features, establish multiple feature vectors to represent 3D models;calculate and compare the similarity of feature vectors, carry out the “multi-level elimination”of the model, and retrieve the three-dimensional model with similar process design scheme. The similarity retrieval experiments of the product 3D model libraries of foundry enterprises in the fields of aviation,aerospace, rail transportation and engineering machinery were carried out, and the accuracy of Top-3 model retrieval was 85.06% and 95.83% in the multi-domain and multi-category model test set and the singledomain and multi-category model test set of fluid machinery respectively. The results show that, compared with the traditional shape distribution operator, the method has strong capability for describing the 3D models of castings and high retrieval performance, which provides technical support for the subsequent reuse of casting history processes based on 3D model content retrieval.