This paper proposes an enhanced particle swarm optimization (EPSO)-based support vector classifier (SVC) that extracts the support vector from databases, in order to diagnose vibration faults in steam turbine-generator sets (STGS). SVC has been successfully applied to the classification of data with linear or nonlinear features, because it allows generalization. However, the design of the best SVC model for the acquisition of the best hyperplane is often difficult and depends heavily on the operators' experience or on trial-and-error experiments. In this paper, an EPSO algorithm is used to automatically tune the control parameters of an SVC. Since EPSO is an excellent optimization tool, it is easily sufficient for the design of an optimal SVC model. The proposed approach is applied to an STGS, to test its diagnostic accuracy. The test results demonstrate that the proposed EPSO-based SVC method has a higher diagnostic accuracy and a shorter learning time than classical neural network-based methods. This study also provides advice on handling a loss of data features for unknown reasons.