For offshore wind turbines (OWTs), effective maintenance decision-making depends on the timely and intelligent anticipation of developing faults. Without requiring the installation of additional sensors, failure-related information can be extracted from the widely available Supervisory Control and Data Acquisition (SCADA) system. This thesis presents an integrated deep learning framework designed to interpret high-dimensional, unlabeled, and often low-quality SCADA data for fault diagnosis and Remaining Useful Life (RUL) estimation.

The framework identifies historical failure events through reconstruction-based anomaly detection and the construction of a health indicator. By clustering detected anomalies, associated failure modes are inferred, allowing classification of future fault types. Using the estimated moments of failure as guidance, it then learns degradation trends in the reconstruction feature space and performs RUL prediction.

Given the complexity of offshore environments and the unpredictable nature of wind turbine faults, the framework is first validated in a controlled setting using NASA's C-MAPSS simulated aircraft engine dataset. The results are competitive and align well with those reported in related studies. Subsequent application to real-world OWT SCADA data demonstrates the practical feasibility of the approach. However, challenges such as data imbalance, obscured features due to SCADA data quality issues, and propagation of errors between model components complicate implementation and reduce prediction reliability.

Despite these challenges, the proposed framework successfully extracts health-relevant insights, enabling predictive maintenance optimization and contributing to more informed data-driven decision-making in offshore wind operations.