Design of a novel noise resilient algorithm for fault detection in wind turbines on supervisory control and data acquisition system

Wind turbine faults, including electrical, mechanical or aerodynamics-related, can potentially reduce operational efficiency, causing downtimes or, in some cases, leading to severe damage. Hence, timely detection of these operational anomalies is crucial for optimizing performance and reducing maintenance costs. The following study explores the application of Supervisory Control and Data Acquisition (SCADA) data for fault detection and diagnosing different operational states in wind turbines by focusing on environmental and operational factors which affect the performance. An efficient noise-resilient classification framework is proposed which includes a novel Perturbed-Random Forest (P-RF) algorithm to diagnose operational states with high accuracy even under noisy conditions. Seasonal discrepancies in power generation are analyzed using theoretical and actual power analysis. Further, distributions of features are realized via kernel density estimation and efficiency metrics in order to identify performance inefficiencies. The P-RF algorithm achieved 99.72% accuracy in diagnosing the status of the SCADA-based wind turbine although with a slightly higher computational complexity than a baseline Random Forest algorithm. Multiple Evaluation metrics were employed to assess the performance of the proposed model under different signal-to-noise conditions.