Integration of deep learning and Bayesian networks for condition and operation risk monitoring of complex engineering systems. (June 2022)

• Record Type: Journal Article
• Authors: Moradi, Ramin; Cofre-Martel, Sergio; Lopez Droguett, Enrique; Modarres, Mohammad; Groth, Katrina M.
• Is Part Of: Reliability engineering & system safety. Volume 222(2022)
• Publication Date: 2022-06
• Subjects: Complex engineering systems -- Condition monitoring -- Risk assessment -- Deep learning -- Bayesian networks -- Bayesian neural networks -- Uncertainty; 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.2022.108433
• Languages: English
• ISSNs: 0951-8320

Abstract

Abstract: A challenging problem in risk and reliability analysis of Complex Engineering Systems (CES) is performing and updating risk and reliability assessments on the whole system with sufficiently high frequency. The challenge stems from both operational data complexity and systems' complexity. The data complexity calls for novel and advanced data-driven methods, e.g., Deep Learning (DL). However, the systems' complexity cannot be addressed only using black-box models; engineering knowledge and systems modeling methods must also be considered. In this paper, a novel mathematical architecture for operation condition and risk monitoring of CES is presented. In this architecture, a Bayesian Network (BN) is used to model the system, subsystems' relations, scenarios leading to adverse events, and to fuse subsystem-level information. Further, Bayesian DL models are trained for subsystems' diagnostics based on condition monitoring data, and their outputs are integrated into the root nodes of the designed BN. This integration enables addressing both the data and systems complexity in a single architecture that provides system-level insight. The proposed architecture also has the capacity to incorporate human inputs and qualitative information. We demonstrate the effectiveness of our proposed approach by performing a case study on a real-world Vapor Recovery Unit at an offshore oil production platform. Highlights: The proposed architecture is capable of monitoring the operation condition and risk simultaneously. Bayesian network is used for system and scenario modeling. Bayesian deep learning is used to train online condition monitoring models. A case study on a real-world vapor recovery unit with 188 monitoring sensors is performed.