Files
Abstract
Managing rapid engineering changes in requirements and complex computer-aided design (CAD) models continue to increase the risk of industrial project failures in smart manufacturing. As products evolve over time, tracking design changes across different domains has become increasingly difficult to operate. Mismanagement incidents can derail industrial product development and result in financial losses. Existing practices often lack connections to cross-domain analysis and rely on domain experts to interpret engineering change propagation. To reduce the burden of this taxing process, this study proposes computational tools as digital threads that assist engineers in understanding the correlations of change propagation. The proposed framework investigates three components of analyzing engineering changes within and across domains. Particularly, the work pertains to (1) a topic modeling approach to narrow down engineering changes within requirements topics, (2) a framework for recognizing mechanical designs based on point clouds representations, and (3) an approach to incorporating joint embedding to learn the correlation between requirements and CAD images. The study makes use of several datasets, including three different heterogeneous industrial requirements documents, ShapeNetCore, and synthetic image datasets. Using this framework, engineers can generate interpretable results and determine the correlations of text-to-text and text-to-images for complex systems. The outcome of this study can contribute to building digital threads and assisting designers to make informed engineering decisions, track change propagation within and across domains, and reduce unanticipated engineering changes.