Civil structural health monitoring is a key technology to ensure the safe operation of infrastructure, among which damage identification is the most important link. Traditional methods often lack stability when there are few samples and high noise levels. Therefore, this study developed a damage identification method that combines modal parameters with strain energy characteristics, based on support vector machines. By adjusting the model parameters, its generalization ability has been improved. In the experimental section, a finite element model of a reinforced concrete beam bridge was selected, and 200 sets of samples were generated, including 5 types of damage locations and 3 degrees of damage, all of which were added with high-intensity noise. Then, compare the performance of support vector machine with methods such as BP neural network, random forest, and K-nearest neighbor. The results showed that the support vector machine achieved an average recognition accuracy of 92.7% when using all samples. Even with only 50 sets of samples in the training set, its accuracy is still 88.2%, significantly better than other methods. Especially for minor injuries with a degree of 5%, the recognition accuracy of support vector machine is 90.1%, which is 7.8% higher than that of BP neural network. This study demonstrates that support vector machines have strong adaptability to small sample data and good recognition ability for minor damages in civil structure damage identification. Therefore, this method provides a feasible and easy to implement technical solution for practical engineering applications.
Xuyu Cheng (Thu,) studied this question.