1A12 Boundary Damage Identification of a Two-Storey Framed Structure Utilizing Frequency Response Functions and Artificial Neural Networks
Ulrike Dachermann,Jianchun Li,Bijan Samali
DOI: https://doi.org/10.1299/jsmemovic.2010._1a12-1_
2010-01-01
The Proceedings of the Symposium on the Motion and Vibration Control
Abstract:For civil engineering structures damage will inevitably occur and continuously accumulate during their service life. As damage develops, both, the physical properties, such as mass, stiffness and damping, as well as the dynamic properties of a structure change. It is therefore feasible to utilise measured dynamic quantities such as time histories, frequency response functions (FRFs) and modal parameters from structural vibration to detect damage. Vibration based damage identification is essentially a form of pattern recognition problem, which looks for the discrimination between two or more signal categories, e.g. before and after a structure is damaged. Artificial neural networks are capable of pattern recognition, classification, signal processing and system identification, and are therefore an ideal tool in complementing dynamic based damage detection methods. Likewise, principal component analysis (PCA) has pattern recognition abilities and is capable of data reduction and noise filtering. Together, these two techniques are able to extract enhanced damage fingerprints from vibrational data and assist in delivering robust and accurate damage identification. This paper presents a dynamic based damage identification method that utilises FRF data to identify boundary imperfections of a two-storey framed structure. The proposed method utilises artificial neural networks to map changes in FRFs to boundary condition damage of the structure. In order to obtain suitable patterns for neural network inputs, PCA techniques are adopted to reduce the size of the measured FRF data. Further, patter recognition and noise filtering capabilities of PCA enhance damage fingerprints in FRF data and thereby improve damage identification results. A hierarchy of neural network ensembles is created to take advantage of individual characteristics measured from different sensors. The method is applied to laboratory and numerical models of two-storey framed structure. Boundary damage is simulated by changing the condition of individual joint elements of the structure from fixed to pinned. In total, ten different types of boundary damage are investigated, i.e. four single and six multiple joint changes. From the numerical and the laboratory structure, FRF data are determined and compressed utilising PCA techniques. The PCA-reduced FRFs are then used as input patterns for training and testing of neural network ensembles predicting the type of boundary damage.