A data-driven diagnostic tool for wind turbines under operational variability

Abstract

The need for real-time condition assessment of complex systems relies on implementation of holistic Structural Health Monitoring (SHM) strategies that are capable of tracking structural behavior in a complete operational spectrum of the structure, distinguishing between true system changes and nonthreatening variations. The proposed data-driven framework utilizes an autonomous bi-component tool able to link monitored structural response with random evolution of Environmental and Operational Parameters (EOP) affecting the monitored system. The approach combines the implementation of a Smoothness Priors Time Varying Autoregressive Moving Average (SP-TARMA) method for modeling the temporal variability in structural response, and a Polynomial Chaos Expansion (PCE) probabilistic model for modeling the propagation of response uncertainty. The computational tool is applied on long-term data, collected from an active sensing system installed for four years on a real operating WT structure located in Dortmund, Germany. The twenty one-month tracking of the proposed PCE-SPTARMA diagnostic index, further assessed by means of statistic-based analysis, demonstrates that the proposed symbiotic treatment yields a robust model, capable of separating benign EOP fluctuations from potential pattern alterations due to actual structural damage. The obtained data-driven model verifies the future prospective of the strategy for development of an automated SHM diagnostic tool.