The aim of this research project is to use recent advances in image analysis and classification to automate the process of adapting the design of mechanical designs for production by particular manufacturing processes.
 
Every mechanical component has a shape designed for a particular method of production (e.g. machining, casting, forging or fabrication etc) but as the capability of manufacturing technologies evolve and economic factors vary the method of a component’s production is often changed. However, the cost of redesigning a part for manufacture by a new process often cause significant delay in switching to an alternate process. To remove this bottleneck and enable agile production systems that can quickly change from, say, casting to additive layer, an automated approach to “redesign for manufacture” is required. Such a system would enhance the productivity and agility of manufacturers by enabling them to easily adopt processes appropriate to market demands.
 
This project would first investigate the application of Convolutional Neural Networks (CNN) to the recognition of characteristics associated with different manufacturing processes in 2D images. Once established this facility to drive a form of 3D shape optimization that modifies CAD models of components to suit a particular production technology. The ambition is to create a system for manufacture parts that is analogous to the CNN based ‘style transfer’ processes for transforming 2D images of painting from one artistic form to another.
 
The project would suit a mechanical engineering graduate who has an interest in programming and artificial intelligence or a computer science graduate with a background in industrial manufacturing.   
 
The work will contribute to an area of academic research known as geometric reasoning and is connected to ongoing work to develop a public dataset of 3D CAD models of mechanical parts to support industrial applications of Artificial Intelligence technologies.