Design and implementation of novel structural health monitoring algorithms for wind turbines on sensor nodes

Abstract

The progressive ageing of wind turbine fleets, together with the increasing need for safe and economically viable life-extension strategies, has intensified the demand for structural health monitoring solutions capable of operating under real conditions. In this context, this doctoral thesis addresses one of the main limitations of vibration-based monitoring in operating wind turbines: the difficulty of extracting reliable structural information directly on the turbine, under operational conditions and without relying on offline post-processing, in the presence of rotational harmonics, non-stationary excitation, and constrained computational resources. The thesis proposes an integrated framework for online, real-time, lowcomputational- cost structural health monitoring of wind turbines, with special emphasis on its implementation in autonomous embedded sensor nodes. A first contribution of the work is the development of a non-stationary signal modelling framework that jointly estimates structural vibration modes and rotational harmonic components. Unlike conventional approaches, in which harmonics are treated as disturbances to be removed prior to modal analysis, the proposed methodology models both phenomena simultaneously, enabling a more robust separation between structural and rotational effects under operational conditions. This formulation allows the decomposition of a multi-degree-of-freedom problem into several single-degree-of-freedom modal responses, thereby facilitating subsequent modal parameter estimation. Building upon this framework, the thesis presents an ef􀏐icient methodology for the online estimation of modal parameters, particularly natural frequencies and damping ratios. Special attention is devoted to damping estimation, which remains one of the most challenging aspects of operational modal analysis in wind turbines. By combining recursive signal processing strategies with techniques such as Random Decrement, Ibrahim Time Domain analysis, and cascaded linear Kalman filters, the proposed approach achieves accurate and robust estimation while maintaining computational efficiency compatible with real-time embedded implementation. A further contribution lies in the development of synchronous processing techniques based on the rotor angular position. The thesis introduces a real-time azimuth estimation algorithm based exclusively on inertial measurements, avoiding the need for external encoders or SCADA signals. In addition, a methodology is proposed to transform measurements acquired in a rotating reference frame into equivalent 􀏐ixed-frame signals, enabling the application of conventional operational modal analysis tools to sensors installed on rotating components such as the hub. This significantly expands the monitoring capability towards critical rotating subsystems, including blades and pitch-related elements. The proposed methodologies have been validated using both synthetic signals generated with OpenFAST and real measurements acquired from operating wind turbines. Their practical feasibility has been further demonstrated through implementation on prototypes installed in real wind turbines. Finally, the transfer of the developed framework to solar tracking systems confirms its adaptability and broader industrial relevance. Overall, this thesis provides an original and practically deployable contribution to structural health monitoring, bridging the gap between advanced operational modal analysis and scalable industrial solutions for complex energy infrastructures.