Damage detection using principal component analysis applied to temporal variation of natural frequencies

For preserving existing structures to the extent possible, vibration-based damage detection techniques are gaining more attention. In specific, tracking the changes in a structure’s natural frequencies employing the Principal Component Analysis (PCA) seems to be a promising technique. This paper presents an on-going research for developing an approach for damage detection based on identifying the natural frequencies of the structure (via dynamic identification or monitoring) followed by applying PCA on the temporal variation of the identified natural frequencies. The processing of the dynamic monitoring data was carried out using a proposed semiautomatic algorithm able to identify and track the natural frequencies of the monitored structures and their changes over time. The algorithm was applied on one historical structure and two modern ones. PCA will to be carried out. The so far developed approach as a whole was applied on a steel structure model tested in lab. It was subjected to dynamic identification tests with and without damage. It was found that PCA applied to frequency variation seems to be a possible tool to know damages in structures. Nevertheless, the simplified point of view correlating directly the number of PCs with the damage importance seems to be not accurate. First results seem to indicate that the variance percentage explained by each of the principal components could be correlated with the kind of damages and the shake used to excite the structure.Postprint (published version