When a structure is damaged, the dynamic characteristics of the structure will change. Generally, main components of a structural health monitoring system are (1) data collection approach including a network of sensors for collecting the response data and (2) an extraction technique to obtain information on the structural health condition. Data mining (DM) is one of the emerging data extraction techniques. DM can play an important role to find out the hidden patterns in databases. Generally, this sophisticated tool isudemployed to find the relationship between data in datasets. Models and patterns, which are obtained from DM process, are used to make predictions. In this study, frequency response function (FRF) measurements obtained from experimental modal analysis of an intact and damaged composite girder deck are used as inputs for data mining to extract the principal components (PCs) of raw FRF data. Experimental modal analysis of the structure is carried out by exerting incrementally enhanced damage severity at specific location. Totally, 6 damage scenarios are considered with depth of 15 mm to 75 mm with the increment of 15 mm at the mid-span of the structure. In the modelling phase of DM process, principal component analysis (PCA) is employed to train a model. The performance of the model is illustrated by comparing the original FRFs and reconstructed FRFs with first 10 PCs.
展开▼
