|About this Abstract
||2020 TMS Annual Meeting & Exhibition
||Characterization of Minerals, Metals and Materials
||Autonomous Light Optical Microscopy for Quality Control Screening
||Andrew R. Kitahara, Elizabeth A. Holm
|On-Site Speaker (Planned)
||Andrew R. Kitahara
Quality control is a key initiative in manufacturing processes, and automating this process has desirable returns. We present a system that builds on automated stage control for conventional light optical microscopes along with recent successes in applied computer vision for materials characterization. Image acquisition utilizes reflected light optical microscopy with user-selected objectives. We approach the image analysis task with supervised fine-tuning of a convolutional neural network (CNN) to integrate expert knowledge of quality control engineers with an existing framework for image analysis. Image acquisition and analysis are run in parallel for autonomous defect detection. In this, the main goal is to identify and classify anomalous microstructural features for further expert inspection and action. The system we present addresses these key initiatives and offers a step towards autonomous data collection and curation for quality control applications.