Analysis of Deep Learning-Based Frameworks for Fault Detection in Big Research Infrastructures: A Case Study of the SOLARIS Synchrotron

Abstract

This paper presents an in-depth analysis of multi-modal, deep learning-based frameworks for fault detection within big research infrastructures, with a specific focus on synchrotron facilities. The study investigates various approaches and architectures and their accuracy in identifying anomalies indicative of potential faults or irregularities in operation. Leveraging a case study approach, we check the performance and applicability of these frameworks within the complex environment of big research infrastructures. Through comprehensive analysis and comparison, we highlight the strengths, limitations, and potential challenges associated with deploying deep learning techniques for fault detection in such a large-scale environment. We achieved accuracy at the level of 94.6% with image data input and a high sensitivity of 95.7% with the multi-modal framework. This investigation contributes valuable insights for researchers and practitioners seeking to enhance the reliability and efficiency of fault detection systems in synchrotron facilities and analogous setups.

Keywords

• Anomaly detection
• convolutional neural networks
• signal analysis
• synchrotron radiation system
• transfer learning