Efficient reliability analysis for offshore wind turbines: Leveraging SVM and augmented oversampling technique

Abstract

This study develops an efficient reliability assessment method designed to optimize maintenance strategies for Offshore Wind Turbines (OWT), aiming for significant cost savings through reduced maintenance frequency and enhanced efficiency. Effective cost management requires a robust and accurate approach for reliability-based lifecycle management. Therefore, this paper introduces an improved predictive maintenance method, grounded on the reliability-based failure probability of OWT systems. To augment computational efficiency and diminish computational time, a surrogate model is proposed for the estimation of failure probability. This surrogate model integrates the classification strengths of Support Vector Machine (SVM) with an augmented Synthetic Minority Oversampling Technique (SMOTE), specifically adapted for extremely imbalanced data. The study’s contributions are twofold: firstly, it develops a novel reliability-based predictive maintenance method allowing for the quantitative assessment of OWTs’ current conditions; secondly, it presents a surrogate model adept at managing extreme data imbalance, thereby enhancing prediction accuracy. The effectiveness of the surrogate model is validated through a case study under two distinct weather conditions. The proposed predictive maintenance method serves as an efficient and effective tool for improved maintenance planning for OWTs.