Top

FNO GraphNet Model for Salt Spray Corrosion Prediction of Offshore Transformers Based on Marine Climate and Inspection Data

Authors

Andrea Conti
Department of Electrical, Electronics and Computer Engineering, University of Catania, Catania, 95125, Italy
Davide Lombardi
Department of Electrical, Electronics and Computer Engineering, University of Catania, Catania, 95125, Italy

Files

Abstract

Offshore transformers operate under persistent humidity, wind-driven salt aerosol, chloride deposition, rainfall wash-off, ultraviolet exposure, thermal cycling and repeated maintenance disturbance. These coupled conditions accelerate coating degradation, bushing contamination, cabinet-joint leakage, fastener corrosion, cable-terminal deterioration and radiator surface damage, yet the deterioration process is difficult to predict because marine climate records and inspection images are collected at different temporal intervals. This paper introduces an FNO-GraphNet model for salt-spray corrosion prediction of offshore transformers based on marine climate, operating state, inspection evidence and component metadata. Wind speed, wind direction, relative humidity, sea-salt deposition, wetting duration, rainfall, temperature, ultraviolet level, operating temperature, load rate, component material, coating age, image quality tag, visible rust grade and maintenance history are aligned into component exposure windows. A transformer component graph represents the enclosure, bushing, radiator, junction box, cable terminal, cabinet joint and fastener regions as connected structural nodes. The Fourier Neural Operator branch learns continuous marine exposure fields and long-range climate effects, while the GraphNet branch propagates corrosion evidence among physically adjacent and functionally related components. Experiments use a simulated offshore transformer dataset with unseen seasons, unseen platforms and missing-inspection cases. The model outputs corrosion score, risk level, confidence, dominant evidence, vulnerable component and suggested inspection priority. The framework is designed to support preventive maintenance planning by linking environmental stress, structural vulnerability and reviewable risk evidence rather than producing an isolated hidden prediction.

Details