Speaker
Description
Particle accelerators, such as the Spallation Neutron Source (SNS), require high beam availability in order to maximize scientific discovery. Recently, researchers have made significant progress utilizing machine learning (ML) models to identify anomalies, prevent damage, reduce beam loss, and tune accelerator parameters in real time. In this work, we study the use of uncertainty aware convolutional neural networks (CNNs) to provide capacitance predictions for the High-Voltage Converter Modulator (HVCM) systems in the SNS. Utilizing the vast amounts of simulated and measured waveforms available, we can estimate these capacitance values using existing monitoring devices in a non-invasive way and inform preventative maintenance to replace worn components before failure. Additionally, by providing uncertainty quantification (UQ) through a Deep Gaussian Process Approximation (DGPA) model, we can evaluate our confidence in the model's predictions and potentially alert beam technicians to anomalous activity that may require further investigation.
| Consider for long presentation | No |
|---|
