Abstract
The complexity of large-scale systems has greatly increased and the tools for monitoring complex systems and ensuring their reliability have also been vastly studied. This study presents a tool to improve the understanding of the state-of-health of complex systems with several subsystems and components and partial observability. Many real systems can have issues with monitoring due to the cost of monitoring or physical limitations for constant monitoring. The proposed method in this study tackles this issue by utilizing all available, limited data to better predict system performance. A Multi-scale Multi-Fidelity Bayesian Learning framework is proposed to provide an online tool for component failure prediction as well as system prognostics. A Dynamic Bayesian Network is used to model the dependencies between components in a system and their impact on system failure. Multi-Fidelity Networks is then used to continuously update the parameters in the network to leverage online data to improve the predictive results. A case study of the leakage failure mode of an offshore production well is conducted to highlight the applicability as well as the benefit of the proposed approach. The results show the advantage of accounting for high-fidelity data to inform low-fidelity data for complex systems.
Original language | English (US) |
---|---|
Pages (from-to) | 555-570 |
Number of pages | 16 |
Journal | Applied Mathematical Modelling |
Volume | 125 |
DOIs | |
State | Published - Jan 2024 |
Keywords
- Bayesian optimization
- Leakage
- Multi-fidelity
- Production well
- System reliability
ASJC Scopus subject areas
- Modeling and Simulation
- Applied Mathematics