In bearings fault detection application, To solve the problems of difficultly obtaining labeled samples and exploiting a large amount of unlabeled samples, a novel semi-supervised Support vector machine fault detection model based on Laplacian regularization is presented in this paper. A smoothness penalty is introduced into the optimization function of regularization network which can exploit the clustering and manifold information of unlabeled samples. The comparisons with other Support vector machine,Fuzzy Support vector machine and Transductive Support vector machine fault detection algorithm are performed. The experiments show that the proposed approach can efficiently utilize the information provided by unlabeled samples to improve the performance of fault detection with labeled training samples of different sizes. The proposed fault detection methods with test samples and without test samples are compared. The results illustrate the investigated techniques with test samples as unlabeled samples can outperform the one without test samples as unlabeled samples.