Lasers and Diamonds and Robots, Oh My! - Automated Multimodal Tomography for Flaw Identification in Diamonds

Abstract

This thesis covers my work in improving a multi-modality imaging system used to measure the location of flaws in rough diamonds. This is achieved through refraction correction of OCT data using surface meshes produced by an XCT system. I discuss the steps required to separate OCT and XCT imaging modalities so that they can occur independently to facilitate optimisation of each technique individually. This separation is enabled through the introduction of an adaptive FOV technique which utilises a visual hull to predict the location of the sample as it rotates. This technique improves the average focus of scans through a 37% reduction in average spot size, and 75% reduction in spot size variation. I then present a proof of concept for replacing an XCT system with an additional OCT system for surface profiling. To achieve this, I implemented an adaptive FOV system which enabled capture of OCT images with high axial resolution, before a geometric surface reconstruction was used to combine multiple views into a single surface. The optical surfaces showed promise in replacing XCT surfaces for refraction correction, however some issues remain to be solved such as the presence of excessive noise in the triangulated mesh.

A replacement of the sample mounting system is then shown, with the process of designing, implementing, and validating a vacuum-based mounting system presented. The replacement mounting system is shown to drastically reduce operator time throughout the scanning process, and facilitates the introduction of an automated sample changing system. This sample changing system is discussed, showing the iterations of related hardware such as sample trays and gripper fingers, before quantifying handling success rates using data from a real-world deployment in which more than 9,000 unique stones are scanned and handled using the automated system. These results suggested that 89% of all stones can be handled by the automated system, surpassing the targeted success rate of 80%. Overall, these improvements triple the maximum system throughput while also improving scan quality through adaptive trajectory planning. These improvements have been validated through deployment in a real-world commercial setting in which more than 9,000 rough diamonds have been scanned. Show more