Optimization of wind turbine gearbox reliability in Danish offshore wind farm operations

Thesis Abstract

The reliability of wind turbine gearboxes (WTGs) is a critical determinant of offshore farm performance, cost of energy, and serviceability in Danish offshore wind operations, where extreme sea states, remote locations, and high uptime requirements amplify the impacts of gearbox failures. This study addresses the gap in integrative reliability optimization by combining condition-monitoring data, failure mode analysis, and design-end diagnostics to enhance gearbox availability and lifespan. The aim is to develop a data-driven framework that optimizes gearbox reliability through predictive maintenance, component-level design improvements, and operational strategy adjustments that are specifically tailored to Danish offshore wind farm contexts characterized by 10–15 MW turbines, 40–60 adjacent units per park, and publicly accessible SCADA and vibration monitoring datasets. The objectives are to (i) identify dominant failure modes and their causal factors using fault-tree analysis and Bayesian updating, (ii) quantify the influence of operational variables (wind speed, temperature, rotor load, and oil condition) on gearbox degradation via multivariate regression and survival analysis, (iii) develop a predictive maintenance schedule using machine learning models trained on 5-year historical data from 25 turbines across three Danish farms, and (iv) propose a design- and operation-level optimization framework that minimizes expected downtime and maintenance costs under Danish regulatory and environmental constraints. The methodology adopts a mixed-methods research design combining quantitative durability modeling with qualitative engineering expert evaluation. The population comprises operational WTGs from three Danish offshore wind farms with available condition-monitoring data, diagnostic logs, and maintenance records. A stratified sample of 60 turbines is selected, ensuring representation across turbine ratings (8–15 MW), gearbox vintages, and gearbox brands. Data collection instruments include monthly SCADA time-series (wind speed, rotor speed, torque), oil-quality analytics (viscosity, moisture, acidity), vibration signatures, and monthly maintenance and failure records. The study employs survival analysis (Cox proportional hazards model) to model time-to-failure distributions, multiple linear regression and LASSO for degradation predictors, random forest and gradient boosting for predictive maintenance risk scoring, and Bayesian networks to update failure probabilities as new data become available. Model validation uses 10-fold cross-validation and out-of-sample testing on the most recent two-year dataset. An expert elicitation workshop is conducted to refine the model inputs and to validate practical maintenance implications, guided by the Theory of Planned Behavior to interpret management decision-making for maintenance scheduling. The analytical framework incorporates reliability-centered maintenance concepts and the Bayesian decision-theoretic approach for optimal maintenance timing. Ethical considerations address data privacy of operator records and transparent communication of uncertainty in maintenance recommendations. Expected findings include (i) identification of high-impact failure modes such as planetary gear tooth wear, bearing lubrication degradation, and seal leaks, with quantified hazard ratios; (ii) robust predictors of gearbox degradation, including oil moisture content, bearing temperature, and high-frequency vibration features; (iii) a proven predictive maintenance schedule achieving a targeted 15–20% reduction in unplanned downtime and a 7–12% decrease in maintenance cost per turbine over a 5-year horizon; (iv) an integrated optimization framework that balances maintenance interval extension with risk-based intervention thresholds under Danish regulatory constraints and environmental operating envelopes. The study anticipates that ensemble learning models provide superior predictive accuracy (AUC > 0.85) compared with traditional regression approaches and that Bayesian updating significantly improves decision quality in scenarios with limited or noisy data. The contribution to knowledge lies in delivering a transferable, Denmark-specific reliability optimization framework for WTGs that integrates condition-monitoring signals, detailed failure analyses, and a practical maintenance planning tool within a design-operational life-cycle perspective. The recommendations include guidelines for oil-management practices, vibration-based fault detection enhancements, and a decision-support toolkit for operators to implement adaptive maintenance strategies that align with Danish offshore industry standards and safety requirements. The study concludes that proactive, data-driven maintenance informed by real-time condition indicators materially improves gearbox reliability, reduces lifetime maintenance costs, and strengthens grid-supportive offshore wind operations in Denmark.