Abstract
Robust monitoring is vital in offshore oil and gas exploration, especially for Floating Production, Storage, and Offloading (FPSO) platforms pivotal in subsea hydrocarbon extraction. Motivated by the glaring absence of failure detection systems in mooring lines on FPSOs, this work tackles the broader issue of inadequate monitoring and warning systems in the sector despite their pivotal role in safety. Existing approaches often overlook inter-series correlations in multivariate scenarios. The authors introduce Graph Neural Networks (GNNs), particularly Spectral-Temporal GNN (StemGNN), leveraging recent time series processing advancements to fill this gap. Operating in the spectral domain, StemGNN captures intra-series and inter-series correlations, enhancing accuracy. The innovative approach improves classifying multivariate time series, particularly the motion time series of offshore platforms for mooring line rupture detection, considering platform oscillation. This work develops, implements, and tests an alert system for potential cable breaks on offshore platforms. Beyond operational value, it contributes to technological innovation through machine learning, overcoming conventional monitoring limitations. Results show comparable performance to state-of-the-art models, highlighting the proposed model’s suitability for real-world scenarios with noisy or missing data. Its complexity addresses challenges without sacrificing performance, advancing mooring line failure detection in offshore oil and gas, guiding future research.
