Feature Selection - Extraction Methods Based on PCA and Mutual Information to Improve Damage Detection Problem in Offshore Wind Turbines

Abstract

Damage Detection problem in Structural Health Monitoring (SHM) is widely studied by many researchers, therefore lots of damage detection algorithms can be found in the literature. Feature Selection / Extraction methods are essential in the accuracy of these algorithms, they provide the suitable data to be used. The main goal of this work is to improve the input data to be the most representative for the damage detection problem. This is done using different Feature Selection / Extraction methods (PCA, UmRMR, and a combination of both). After taking the representative features, the results are tested using a damage detection method; the NullSpace in this case. The data has been collected from a Laboratory Offshore tower model. The different results are compared (different preprocessing vs Raw data) and these show how the correct preselection of the data can improve damage detection.