Prediction of wind turbine blade deterioration impact on aerodynamic performance using proper orthogonal decomposition and deep reinforcement learning

• Wind turbine blade degradation due to environmental factors reduces aerodynamic efficiency and power output. • Surface roughness parameters are analysed to assess the influence of skin friction on wind turbine blades. • POD extracts key surface pressure features, highlighting dominant patterns and structures within the flow field. • Deep reinforcement learning algorithm predicts aerodynamic features of rough surface blades with minimal R 2 error. • Proposed model improves predictive maintenance and extends system lifespan by reducing downtime. Wind turbine blade deterioration due to environmental factors such as precipitation, sleet and biological contaminants significantly impacts aerodynamic performance leading to reduced power output. This study aims to quantify the effect of blade surface roughness on aerodynamic efficiency by analyzing skin friction variations and pressure distribution on the NACA0015 airfoil at Reynolds number of 2.05 × 10 5 . A comprehensive Proper Orthogonal Decomposition analysis was conducted to assess the impact of different surface roughness grit on aerodynamic characteristics. The study leverages both experimental wind tunnel testing and advanced numerical analysis to evaluate discrepancies in flow behavior. A Deep Reinforcement Learning framework was developed to forecast aerodynamic coefficients and enhance predictive modeling, addressing the computational limitation of conventional computational fluid dynamics and experimental methods. This research introduces an AI-driven aerodynamic prediction model using Deep Reinforcement Learning, offering a computationally efficient alternative to traditional numerical simulation. Further, the application of reduced order models for the rough surface flow decomposition enhances the understanding of unsteady aerodynamic behavior and leads to significant advancements in data-driven wind turbine performance evaluation. Results demonstrate that the proposed learning strategy accurately predicts aerodynamic coefficients of rough surface blade with high correlation (R 2 = 0.99708) significantly reducing the computational cost. The study provides critical insight into flow separation and change in pressure distribution induced by surface roughness. The findings support proactive predictive maintenance strategies, enabling real-time anomaly detection, adaptive forecasting, and operational optimization, thereby reducing downtime and extending wind turbine lifespan.