Public demands for renewable energy generation are the driving factor for advancements in wind energy, with ever larger wind turbines erected in remote sites. However, regular physical inspections of these structures, as defined by international standards and guidelines, contribute a significant share to the total operation and maintenance expenditures. Efficient structural health monitoring (SHM) systems are key technologies for reducing these costs by enabling maintenance actions according to the true structural state and preventing dramatic failures. This paper presents an enhanced methodology for classifying structural damages using optimally projected multivariate damage sensitive features (DSFs) extracted from vibration signals. Sequential projection pursuit is employed for obtaining low-dimensional transformations of DSFs with the help of an advanced evolutionary strategy. The classification algorithm is based on Bayes’ theorem and an advanced multivariate statistic. A stochastic objective function is defined according to this algorithm. The optimal number of transformation vectors is found using a fast-forward selection. The approach is applied to DSFs defined by the coefficients of vector autoregressive models, which are estimated from multivariate acceleration response signals of an experimental wind turbine blade. Small masses were added to the blade to simulate different damage scenarios non-destructively. Wind-like excitations were applied using a pedestal fan. The results demonstrate that the proposed procedure can reduce DSF dimensionality and, at the same time, preserve damage classification accuracies with respect to the original DSFs. The outcomes are promising for future developments of enhanced vibration-based SHM techniques.
|