About this Abstract |
| Meeting |
2024 TMS Annual Meeting & Exhibition
|
| Symposium
|
Novel Strategies for Rapid Acquisition and Processing of Large Datasets from Advanced Characterization Techniques
|
| Presentation Title |
Real-Time In-Situ Characterization with Web Technologies at Any Scale |
| Author(s) |
Kevin Field, Christopher R.. Field |
| On-Site Speaker (Planned) |
Kevin Field |
| Abstract Scope |
Advances in machine learning (ML) have enabled methods for rapid quantification of large imaging datasets, but many of the proposed methods are post-facto-based workflows. Recently, workflows are being migrated to real-time capabilities on a per-instrument vendor basis. Here, we will present our recent developments of leveraging web technologies, such as WebRTC, Grafana, and Ray, to perform instrument agnostic ML-assisted real-time in-situ characterization. Specifically, we will discuss how leveraging these tools enables an adaptable hardware-software real-time quantification platform that can be tailored for any experiment with or without Internet access using edge, near-edge, and/or cloud computing resources. Results show no loss of detection efficiency and quality compared to post-facto workflows with the highest inference rates of 90 FPS achieved to date using a YOLOv8 model running on a near-edge blade server with NVIDIA A6000 GPUs. We will conclude with a discussion on the current needs to accelerate real-time workflows using ML. |
| Proceedings Inclusion? |
Planned: |
| Keywords |
Machine Learning, Characterization, Nuclear Materials |