Open Access Open Access  Restricted Access Subscription or Fee Access

Selecting Features for Data Based Damage Detection



Data based damage detection methods rely on the identification of features which are sensitive to damage. Many damage-sensitive features (DSFs) have been proposed in the literature, but the question of which are most suitable for a give application remains. In this study we use state-of-the art machine learning feature selection algorithms to compare different DSFs, based on fourier transform coefficients. Experimental data from a steel frame structure is used for this analysis. The dataset includes variations in the excitation type, including random excitation, harmonic excitation and free vibration. Damage scenarios include loosened connections, yielded elements and reduced cross sections. For each damage scenario we identify the most informative frequencies and excitation types using elastic net regularisation. The results of this study serve as a guideline for real depolyments of SHM systems, and are a step towards answering the question of which DSFs are best for a give application.

doi: 10.12783/SHM2015/370

Full Text: