Abstract:

In the vibration-based damage detection of bridges , it is difficult to avoid the uncertainty of the dynamic behaviour caused by changing environmental conditions. In order to overcome this difficulty , a novel damage detec-tion methodology for bridges is proposed based on the nonlinear principal component analysis. In this methodology ,first , an auto-associative neural network is constructed to caπY out a nonlinear principal component analysis for the element stiffness samples identified from healthy structure in various environmental conditions , and a nonlinear transformation mode that reveals the environmental influences is created. Then , the element stiffness samples ob-tained from an unknown state in various environmental conditions are inputted into the constructed network. Thus ,damages can be detected and located via the statistical analysis of the residual errors between the network outputs and the targets. The proposed methodology is finally verified by the numerical simulation of a simply-supported bridge.

Key words: damage detection, environmental influence, nonlinear principal component analysis, neural network