Identification of Structural Defects Using Computer Algorithms

One of the numerous methods recently employed to study the health of structures is the identification of anomaly in data obtained for the condition of the structure, e.g. the frequencies for the structural modes, stress, strain, displacement, speed, and acceleration) which are obtained and stored by various sensors. The methods of identification applied for anomalies attempt to discover and recognize patterns governing data which run in sharp contrast to the statistical population. In the case of data obtained from sensors, data appearing in contrast to others, i.e. outliers, may signal the occurrence of damage in the structure. The present research aims to employ computer algorithms to identify structural defects based on data gathered by sensors indicating structural conditions. The present research investigates the performance of various methods including Artificial Neural Networks (ANN), Density-Based Spatial Clustering of Applications with Noise (DBSCAN), Manhattan Distance, Curve Fitting, and Box Plot in the identification of samples from damages in a case study using frequency values related to a cable-support bridge. Subsequent to the implementation of the methods in the datasets, it was shown that the ANN provided the optimal performance.