Study Of Bridge Damage Localization Based On Probabilistic Neural Networks

Structural health monitoring is a hot research subject in civil engineering recently. Damage identification technique is one of the key issues for structural health monitoring research. The technology of artificial neural networks is an available tool to solve the issues. This paper discusses the theory and application of structural damage identification based on probabilistic neural networks in civil engineering and the main research and conclusions are summarized as follows:The paper starts from the reviews of structure damage identification and damage identification technology based on neural networks. Relatively comprehensive discussions are made on the significances, applications, obtained research productions and problems in the current methods of the structural damage identification. Then, the basic theory of neural networks is introduced and the pattern recognition mechanism of probabilistic neural networks is analyzed, the damage identification method and procedure based on probabilistic neural networks are also studied.Simply-supported beam is a usual structural form. Based on it's dynamic response of many different damage cases, frequencies, modal shapes, curvature modal shapes, and their compounded index are used as probabilistic neural networks' input parameter respectively to study bridge damage localization. The result indicates that probabilistic neural networks can localize the single damage correctly, and the networks with the compounded index show better effectiveness. The training sets size and the training sets with or without noise have little influence on the result; the networks have good extendability and can resist noise effectively, but it is unsuitable for multi-damage cases.Concrete-filled steel tubular bridge is a complex structural system. Probabilistic neural networks and substructure method are used to localize the damage of Beichuan River Bridge with two steps. The result indicates the networks show good performance in the process of two-step damage localization, different input parameters should be chosen effectively when establishing PNN to localize damage of different substructures.