Systems and methods for synthesizing a diamond using machine learning

Inventors

MEKALA, Rohan Reddy • Muehle, Matthias • Porter, Adam • Lindvall, Mikael • Becker, Michael

Assignees

Fraunhofer USA Inc • University of Maryland College Park

Abstract

Disclosed herein are systems and methods for synthesizing a diamond using a diamond synthesis machine. A processor receives a plurality of images of a diamond during synthesis within a diamond synthesis machine, each of the plurality of images captured within a time period. The processor executes a diamond state prediction machine learning model using the plurality of images to obtain a predicted data object, the predicted data object indicating a predicted state of the diamond within the diamond synthesis machine at a time subsequent to the time period. The processor detects a predicted defect, a number of defects, defect types, and/or sub-features of such defects and/or other characteristics (e.g., a predicted shape, size, and/or other properties of predicted contours for the diamond and/or pocket holder) of the predicted state of the diamond. The processor adjusts operation of the diamond synthesis machine.

Core Innovation

The invention relates to a method and system for synthesizing a crystal using a synthesis machine by using time series images captured during synthesis. A processor receives a plurality of images of a crystal during synthesis within the synthesis machine and generates a data object based on the plurality of images. The data object indicates a predicted state of the crystal within the synthesis machine at a time subsequent to the capture of the plurality of images.

The processor then detects a predicted defect in the crystal based on the predicted state. The predicted state and predicted defect are used to drive synthesis-machine behavior by adjusting operation responsive to the detected predicted defect, including closed-loop iteration.

In some implementations, the predicted data object includes a predicted image and pixel-based representations with classification labels for defect detection and contour-based defect identification. The prediction is generated using model architectures that support future state prediction from labeled chronological image sequences, and may incorporate reactor operating-parameter schedules as inputs aligned to image capture times.

Claims Coverage

The independent claims cover two parallel aspects: a method that predicts a future crystal state from captured images, detects predicted defects, and adjusts synthesis-machine operation accordingly; and a system configured to perform the same sequence. The independent claims include the core flow of image capture, predicted-state data object generation, predicted defect detection, and defect-responsive synthesis-machine adjustment, with inventive features refined in dependent claims through predicted image/pixel representations, schedule-based prediction inputs, and defect rules based on predicted crystal shape and defect categories.

Across the independent claims, the inventive core is the generation of a predicted-state data object from time series images, followed by predicted defect detection from that predicted state, and using the detected defect to adjust synthesis-machine operation. Dependent claims further narrow these elements to predicted-image/pixel representations, schedule-aligned operating-parameter inputs, and defect categories and contour/shape-based detection using crystal defect rules.

Stated Advantages

Documented Applications