A Feature Selection-based Approach for the Identification of Critical Components in Complex Technical Infrastructures: Application to the CERN Large Hadron Collider. (September 2020)

• Record Type: Journal Article
• Authors: Baraldi, Piero; Castellano, Andrea; Shokry, Ahmed; Gentile, Ugo; Serio, Luigi; Zio, Enrico
• Is Part Of: Reliability engineering & system safety. Volume 201(2020)
• Publication Date: 2020-09
• Subjects: Complex technical infrastructure -- Critical components -- Feature selection -- Wrapper -- Classification -- Support vectors machines -- Binary differential evolution -- Large Hadron Collider; Reliability (Engineering) -- Periodicals; System safety -- Periodicals; Industrial safety -- Periodicals; Fiabilité -- Périodiques; Sécurité des systèmes -- Périodiques; Sécurité du travail -- Périodiques; 620.00452
• DOI: 10.1016/j.ress.2020.106974
• Languages: English
• ISSNs: 0951-8320

Abstract

Highlights: This work proposes a method for critical components identification in CTIs. The method is based on the analysis of the monitoring data. The critical components identification is addressed as a feature selection problem. The method combines wrapper feature selection and binary classification techniques. The method is validated using data collected from the CERN Large Hadron Collider. Abstract: Complex Technical Infrastructures (CTIs) are large-scale systems made of tens of thousands of interdependent components organized in complex hierarchical architectures. They evolve in time in a way that at one point their functional logic may be more complex than originally designed, and, therefore, traditional reliability/risk importance measures cannot be used for identifying the critical components on which the protection and recovery efforts should be primarily allocated. We propose an approach for identifying the most critical components based on the large amount of operational data collected from the CTI monitoring systems over long time periods and under different operational settings. The underlying idea is to develop binary classifiers to associate different combinations of measured signals to the CTI operating or failed state. The critical CTI components are those whose condition monitoring signals allow optimally classifying the CTI state. To identify the signals and to build the classifier, we consider a feature selection wrapper approach based on the combined use of Support Vector Machine classifiers and the Binary Differential Evolution algorithm for optimization. The approach is successfully applied to a real dataset collected from the CERN ( European Centre for Nuclear Research ) Large Hadron Collider, a CTI for experiments of physics.