Intelligent reverse engineering

Project aim

To develop machine learning techniques that can automate the process of reverse engineering, from point cloud to CAD model, with as little human intervention as possible. 

Project background

Reverse engineering in a mechanical context tends to be a very complicated and laborious task. Historically, this would involve capturing a part's geometry through a series of precise measurements and drawings. Today, surface geometry can be captured using 3D scanning methods and sophisticated software tools. Even with this software, it still requires a lot of work to create an accurate model of a part.

With the success of machine learning in areas such as computer vision, more and more research has been done on geometric deep learning, an area focussed on processing non-Euclidean structures such as graphs and manifolds. The aim is to leverage these techniques to identify features and derive structure from laser scans and generate full history-based CAD models. In a wider context, this will also allow AI agents to more intelligently understand geometry and CAD data.