Improvement of the Damage Detection Performance of a SHM Framework by using AdaBoost: Validation on an Operating Wind Turbine

Abstract

In SHM applications various damage-sensitive features can be used for making decisions regarding damage detection. In all cases, classifiers evaluate the results and make a final decision regarding the state of the structure. Often, there are discrepancies among the decisions of different classifiers, resulting in different detection performances for each damage feature. This is expected as different classifiers may be better suited for different data settings, even in data sets corresponding to the same system. Boosting algorithms combine multiple base classifiers to produce an ensemble, whose joint decision offers a better performance than any of the base classifiers. Adaptive Boosting (AdaBoost) is deployed in this paper to build a strong classifier based on the classifiers of a three-tier modular SHM framework for improving detection performance. The framework consists of three parts: application of machine learning clustering algorithms for data normalization, feature extraction and hypothesis testing (HT). Each connection of damage feature, also referred to as condition parameter (CP), and HT composes a classifier that can be used as a weak classifier in the boosting algorithm. Information from the SHM framework classifiers is used, in order to build a strong classifier that is able to classify the value of any CP and improve the detection performance. The integration of AdaBoost with the three-tier SHM framework is validated on an operating 3 kW wind turbine. The results are demonstrated in receiver operating characteristic (ROC) curves with AdaBoost increasing the performance of damage detection.