L2,s-norm and L2,p-norm distance metrics regularized deep belief network for fault diagnosis

An accurate and efficient fault detection model based on deep belief network (DBN) in this paper is proposed for rotating machinery. Due to the overlapping problem of different class features extracted by traditional networks, a L 2 , s -norm and L 2 , p -norm distance metrics regularized DBN for intelligent fault diagnosis of rotary equipment is proposed in this paper. In the proposed model, we adopt the joint L 2 , s -norm and L 2 , p -norm regularized DBN to extract the original data features with an unsupervised manner. Specifically, the L 2 , p -norm optimizes the spatial distance of features of same-class data, and the L 2 , p -norm penalizes the spatial distance of features of different-classes data, which can make same-class data more compact and different-classes data more identifiable. Therefore, it can produce more discriminant features and improve the ability of classification. At the same time, two different optimization methods are used to optimize the non-differentiable L 2 , s -norm and L 2 , p -norm. Finally, two different kinds of rotating machinery experiments are constructed to verify the effectiveness and superiority of the proposed method. The results show that the proposed method has more obvious advantages in terms of feature extraction ability and feature clustering. In addition, this method is also superior in the fault diagnosis accuracy of rotating machinery.