Extracting, managing, and exploiting the semantics of mechanical CAD models in assembly tasks

The manufacturing of mechanical products is increasingly assisted by technologies that exploit the CAD model of the final assembly to address complex tasks in an automated and simplified way, to reduce development time and costs. However, it is proven that industrial CAD models are heterogeneous objects, involving different design conventions, providing geometric data on parts but often lacking explicit semantic information on their functionalities. As a consequence, existing approaches are mainly mathematics-based or need expert intervention to interpret assembly components, and this is limiting. The work presented in the thesis is placed in this context and aims at automatically extracting and leveraging in industrial applications high-level semantic information from B-rep models of mechanical products in standard format (e.g. STEP). This makes possible the development of promising knowledge intensive processes that take into account the engineering meaning of the parts and their relationships. The guiding idea is to define a rule-based approach that matches the shape features, the dimensional relations, and the mounting schemes strictly governing real mechanical assemblies with the geometric and topological properties that can be retrieved in CAD models of assemblies. More in practice, a standalone system is implemented which carries out two distinct operations, namely the data extraction and the data exploitation. The first involves all the steps necessary to process and analyze the geometric objects representing the parts of the assembly to infer their engineering meaning. It returns an enriched product model representation based on a new data structure, denoted as liaison, containing all the extracted information. The new product model representation, then, stands at the basis of the data exploitation phase, where assembly tasks, such as subassembly identification, assembly planning, and design for assembly, are addressed in a more effective way.