Empirical Analysis of Requirements for a CAD Assembly Search System

The efficient reuse of existing assemblies is a critical factor for reducing development time and effort in mechanical engineering. Despite the potential benefits, searching for suitable CAD assemblies in existing databases remains a significant challenge due to inconsistent data structures and limited retrieval functionalities. This study investigates the requirements and boundary conditions for a digital search system for CAD assemblies based on an empirical survey with 92 mechanical engineers with different levels of experience. The objective was to evaluate engineers’ perceptions of the benefits of improved reuse and to determine which types of data they consider essential as system inputs and outputs. The results reveal a high demand for improved reuse support. Participants expressed a strong preference for using natural language prompts as search input to make the system more intuitive and accessible. They also emphasized the need for outputs that go beyond CAD geometry, including relevant metadata such as material numbers, status information, and organizational attributes. Furthermore, participants suggested implementing a pre-classification of assemblies and providing standardized functional descriptions to improve search efficiency and relevance. Based on these findings, user-centered requirements for a machine learning assisted CAD assembly search system are defined and a strategic concept is derived. This work empirically consolidates user-desired characteristics of CAD query systems and forms a basis for designing intelligent retrieval tools that align with the workflows and expectations of engineering practitioners.