Bridge condition monitoring approach based on frequency domain system identification and support vector machine

As the long span suspension and cable-stayed bridges are widely used in the world, the highway department has become increasingly interested in developing the bridge condition monitoring system to enhance the bridge security and reduce the cost for maintenance. Due to the difficulty in acquiring damage case, the bridge condition monitoring is regarded as a one class learning problem in pattern recognition. A bridge condition monitoring method based on frequency domain system identification and Support Vector Machine (SVM), which is a novel kernel-based machine learning algorithm, is developed in this paper. The method can not only gain precise decision function for alarming, but also solve the problem between the sensitivity and generalization. The feature extraction procedure for bridge condition monitoring is a typical output only system identification problem. In order to use the method online, a frequency domain system identification method, CMIF algorithm which is much easy to ran automatically, is used to extract modal parameters. The algorithm in this paper is verified by the field data from Ting Kau Bridge in Hong Kong. The results show that the method is reliable and useful in the bridge monitoring system design.