Keywords: nonlinear characteristics, kernel principal component analysis, support vector machine, particle swarm algorithm, damage identification.

A number of large-scale complex structures have been built promoted by the development of economy, science and technology. It is an interesting issue how to monitor and manage these large structures so that an alarm will warn before various accidents occur, ensuring that the disaster loss can be decreased to a minimum. Numerous large long-term structural health monitoring systems have been developed and installed in China. However, another problem arises gradually how to effectively deal with huge and abundant measured information from a structural health monitoring system, thus to assess structural condition states. In view of this, intelligent information processing, such as neural networks, genetic algorithms, particle swarm optimization, and data fusion, are employed to process such data and develop efficient damage detection methods to assess the state of complex structures.

In order to extract the nonlinear features from structural response effectively and to develop practical methods for structural damage identification, a support vector machine (SVM) model optimized by particle swarm optimization (PSO) is proposed to detect the damage of a structure. In this model presented, PSO is used to optimize the kernel parameters of kernel principal component analysis (KPCA) firstly, and then non-linear features are extracted from the structural responses by using the KPCA, and a SVM model is employed to classify and detect damage. Finally the detection results are output.

To verify the effectiveness of the model presented, both single- and multi-damage patterns from a twelve-storey reinforced concrete frame were identified. Some important factors, such as measurement noise, feature extraction methods and a neural network model, were investigated. The results show that, (1) penalty factor C and nuclear parameters sigma of the SVM have a great influence on the generalization capability; (2) the identification and generalization capabilities of an adaptive acceleration particle swarm optimization algorithm (CPSO) combining with the SVM are better than those of the SVM; (3) the searching capability and optimizing effect of the CPSO&SVM are better than the PSO&SVM and GA&SVM; (4) the running time of the PSO&SVM is the shortest, then the CPSO&SVM is next, and the GA&SVM is the longest.

purchase the full-text of this paper (price £20)

go to the previous paper
go to the next paper
return to the table of contents
return to the book description
purchase this book (price £70 +P&P)