Application of Generative Topographic Mapping to the Classification of Bearing Fault

In this paper we use bearing data as a test bed for dimensionality reduction methods based in latent variable modeling, in which an underlying lower dimension representation is inferred directly from the data, and apply it to the classification of bearing fault. The optimization of GTM processing experiential parameters, which consist of the number of latent points and basis function, a width parameter of basis function and the weight regularisation parameter, may lead to directly the best performance for classification. Experiments indicate that nonlinear latent variable modeling reveals a low-dimensional structure in the data inaccessible to the investigated linear models. The GTM theory may be successfully employed as a tool for bearing fault detection and diagnosis.