Artificial Intelligence for EM Problem Set (D&D)

Structural Health Monitoring of D&D Facility to Identify Cracks and Structural Defects for Surveillance and Maintenance

Structural health monitoring is imperative to the ongoing surveillance and maintenance (S&M) across the DOE complex. As these facilities await decommissioning, there is a need to understand the structural health of these structures. Many of these facilities were built over 50 years ago and, in some cases, these facilities have gone beyond operational life expectancy. In any of these scenarios, the structural integrity of these facilities may be compromised, so it is imperative that adequate inspections are performed with the minimal intervention of the human need to be performed on a continuous and ongoing basis.

Machine Learning and Deep Learning are state-of-start technologies capable of facilitating the assessment of structural integrity in aging nuclear facilities by detecting cracks using a Convolutional Neural Network (CNN) deep learning approach.

The overall objective of this task is to investigate specific applications to solve the DOE-EM problem sets in challenging areas, including potential applications of existing state-of-the-art technologies such as imaging, robotics, sensors, big data, and machine learning/deep learning, to assess the structural integrity of aging facilities in support of ongoing surveillance and maintenance (S&M) across the DOE complex.

Benefits:

Identifying the structural defects with the help of the AI systems will minimize the manual inspection of the nuclear facilities.
Early detection of structural defects will avoid the health risk to the project personnel working in the site.
Developed models can detect the cracks with high accuracy and can run inference in Real-Time from remote sensors.
AI models can be deployed for the various surveillance maintenance activities at multiple facilities.

Accomplishments:

Implemented CNN deep learning approach for the surveillance & maintenance of the nuclear infrastructure.
Implemented a Convolutional Auto Encoder (CAE) for anomaly detection which can visually highlight the cracks in imagery data.
Implemented a One Class Classifier (OCC), which is a hybrid approach with a combination of Auto Encoder (AE) and CNN.
Ability to generate 3D point clouds from mesh files that are similar to LiDAR scans.
Designed an AE to reconstruct 3D point clouds from synthetic and LiDAR data capable of detecting cracks.
YOLOv3 (You Only Look Once V3) neural network architecture has been implemented, which is a real-time object detection mechanism suitable for streaming videos, static videos, and static images.