基于图像深度学习的零件加工特征信息提取方法

针对各类基于模型定义（MBD）的零件模型加工特征的信息集成问题，提出一种多层次提取架构的加工特征全息信息提取方法。通过对零件的结构特点进行分析，以具有制造语义且无法拆分的最简化特征实现加工特征的分类；在阐述提取策略的基础上，构建了基于深度学习图像识别技术的加工特征分类器；依据MBD模型信息标注的特点快速定位、抽取加工特征的拓扑结构，并利用多角度捕捉降维方法将加工特征彩色图像化，在此基础上设计了加工特征多角度图像识别综合分析方法；基于查询视图对MBD模型的标注信息进行过滤，构建了加工特征几何信息双层过滤式提取方法。最后开发了加工特征全息信息提取软件，并以船用柴油机关键件为实例验证了该方法的有效性。

Information Extraction Method of Part Machining Features Based on Image Deep Learning

Aiming at the information integration problems for machining features of various part models based on model definition（MBD）， a holographic information extraction method of machining features was proposed based on multi-level extraction architecture. Through the analysis of structural characteristics of parts， the machining features were classified with the simplest features that had manufacturing semantics and could not be split. Based on elaborating the extraction strategy， a machining feature classifier was built by deep learning image recognition technology. Based on the characteristics of MBD model information annotation， the machining feature topology structure was quickly located and extracted. A multi-view capture dimensionality reduction method was used to make the machining feature color image. And then a comprehensive analysis method for multi-angle image recognition of machining features was designed. Based on the query views， the annotation information of the MBD models was filtered， and a double-layer filtering extraction method for machining feature geometric information was constructed. Finally， a holographic information extraction software for machining features was established， and experimental results of key parts of marine diesel engines show the effectiveness of the method.