Abstract:A study is presented to compare the performance of crack detection using neural network(NN) and support vector machine (SVM) based on natural frequencies. The SVM is a machine learning algorithm based on statistical learning theory, and it is also a class of regression method with the good generalization ability. Firstly, the basic theory of the back-propagation neural network and support vector regression is briefly reviewed. Then the feasibility of the crack detection using these methods are investigated by locating and sizing cracks in supported beams for which a few natural frequencies are available. It is observed that crack's location and depth can be estimated with a relatively small size error. The results show that the SVM is a powerful and effective method for crack identification.

Crack detection in supported beams based on neural network and support vector machine