Sequential projection pursuit for optimal transformation of autoregressive coefficients for damage detection in an experimental wind turbine blade

Abstract

The performance, and with it, the utility of structural health monitoring systems depends strongly on the efficiency of damage sensitive features (DSFs) for describing the state of a structure. Several approaches are available for extracting DSFs from acceleration response signals, but they are often high dimensional. This affects significantly data processing and storage demands. Therefore, reducing DSF dimensions while maintaining or even improving damage detectability is desired. The present study explores the use of sequential projection pursuit for identifying low-dimensional DSF transformations optimized for structural damage detection. Here, transformation vectors are obtained sequentially using an advanced evolutionary optimization technique. A statistical objective function is employed to facilitate making decisions about the structural state with the help of statistical hypothesis testing. Optimal numbers of transformation vectors are found by fast forward selection. The approach is demonstrated using initial DSFs defined as autoregressive coefficients from acceleration response signals of an experimental wind turbine blade. Wind-like excitations were applied with the help of a pedestal fan, and damages were simulated non-destructively by adding small masses. The results demonstrate that the proposed methodology can considerably reduce DSF dimensionalities without deteriorating the damage detection performance. Conversely, the detectability of some damages could be improved in comparison to using selected original DSFs. This is promising for future developments of efficient vibration-based structural health monitoring methods.