Operational Model Updating of Low-Order HAWT Models for Structural Health Monitoring Applications

For the past decade, wind turbines have become the largest source of installed renewable-energy capacity in the United States. Economical, maintenance and operation are critical issues when dealing with such large slender structures, particularly when these structures are sited remotely. Because of the chaotic nature of non-stationary rotating-machinery systems such as the horizontal-axis wind turbines (HAWTs), in-operation modeling and computer-aided numerical characterization is typically troublesome, and tends to be imprecise while predicting the real content of the actual aerodynamic loading. Loading environment under operation conditions is usually substantially different from those driven by modal testing or computer-aided model characterization and difficult to measure directly in the field. In addition, rotational machinery such as HAWTs exhibit complex and nonlinear dynamics (i.e., precession and Coriolis effects, torsional coupling, nonlinear geometries, plasticity of composite materials); and are subjected to nonlinear constrained conditions (i.e., aeroelastic interaction). For those reasons, modal-aeroelastic and computer-aided models reproduced under controlled conditions may fail to predict the correct non-stationary loading and resistance patterns of wind turbines in actual operation. Operational techniques for extracting modal properties under actual non-stationary loadings are needed in order to (1) improve computer-aided elasto-aerodynamic models to better characterize the actual behavior of HAWTs in operational scenarios, (2) improve and correlate models, (3) monitor and diagnose the system for integrity and damage through time, or even (4) optimize control systems. For structural health monitoring (SHM) applications, model updating of stochastic aerodynamic problems has gained interest over the past decades. For situations where optimizing objective functions are not differentiable, convex or continuous in nature that is the case of gradient methods such as Modal Assurance Criterion (MAC), global optimization (metaheurstic) methods based on probability principles have emerged. These search engine techniques are promising suitable to cope with non-stationary-stochastic system identification methods for model updating of HAWT systems. A probability theory framework is employed in this study to update the wind turbine model using such a stochastic global optimization approach. Structural identification is addressed under regular wind turbine operation conditions for non-stationary, unmeasured, and uncontrolled excitations by means of the eigensystem realization theory (ERA). This numerical framework is then tied up with an adaptive simulated annealing (ASA) numerical engine for solving the problem of model updating. Numerical results are presented for an experimental deployment of a small HAWT structure. Results are benchmarked and validated with other empirical mode-decomposition and time-domain solutions.