A machine-learning approach for structural damage detection using least square support vector machine based on a new combinational kernel function

Health assessment and monitoring of engineered systems have become one of the fastest growing multi-disciplinary research areas over the last two decades. One of the largest concerns in structural health monitoring is how to infer structural conditions from the measurements and the data collected by sensors. The ultimate aim is to detect the structural damages with a high level of certainty and hence to extend the life of structures. In this study, a new strategy for structural damage detection is proposed using least square support vector machines based on a new combinational kernel. Thin plate spline Littlewood-Paley wavelet kernel function introduced in this article is a novel combinational kernel function, which combines thin plate spline radial basis function kernel with local characteristics and a modified Littlewood-Paley wavelet kernel function with global characteristics. During the process of structural damage detection, a social harmony search algorithm optimizes the parameters of least square support vector machine and the thin plate spline Littlewood-Paley wavelet kernel. The results obtained by this method are compared with least square support vector machine based on the other combinational and conventional kernels. These results show that the accuracy of damage detection based on least square support vector machine with thin plate spline Littlewood-Paley wavelet kernel is higher than other methods that utilize conventional kernels under similar conditions. In comparison with other combinational kernels, least square support vector machine with thin plate spline Littlewood-Paley wavelet kernel possesses a better dissemination and learning ability by incorporating the advantages of radial basis function kernel and wavelet kernel functions.