Condition Monitoring for Spent Nuclear Fuel Storage

Condition monitoring for spent nuclear fuel storage requires reliable methods to assess the structural health of hollow cylinders made of welded plates, with sensors deployed by robots. Reducing the number of sensing points is essential to streamline the process and maintain efficiency.

3D Defect Localization Using Defect-Induced Waves

Deep Learning, Bayesian Framework, Pattern Recognition, Segmentation, Time-series Data [Journal Article]

• Task: Locate defects by analyzing naturally occurring disturbances within a system, rather than actively introducing external inputs, for real-time monitoring.
• Method: Utilized a Variational Autoencoder (VAE) to identify physics-based propagation patterns of defect-induced waveforms and modeled localization variables using Bayesian frameworks.
• Novelty: Achieved 93% accuracy defect localization using only two sensors, demonstrated the ability to identify and localize defects with minimal data, improving practicality and efficiency for real-world applications. More Details

Generative model for Wavefield Simulation

Physic-Based ML, WaveNet Generative AI, Surrogate model

• Task: Assess structural health by applying controlled signals and analyzing the structure’s response.
• Method: Designed a WaveNet-based generative model to surrogate wavefields using sparse sensor readings.
• Novelty: Enabled baseline-free monitoring and practical high-resolution wavefield data collection on curved surfaces where traditional scanning is not feasible.