SVDD-based Mechanical Fault Diagnosis for Fiberboard Gluing System

This paper studies the problem that it is difficult to collect the fault information when diagnosing the gluing system of fiberboard. A novel binary-classification method based on support vector data description (SVDD) for fault diagnosis is proposed. Vibration signals are used as data for fault diagnosis, kernel principal component analysis (KPCA) is employed for feature extraction of the normal and fault examples, and SVDD algorithm as classifier is used for fault diagnosis. Experiments show that SVDD algorithm is practical and efficient, and has better identification rate than artificial neural network when lack of unknown fault training examples.