A generic approach for detection of wear-out situations in machine subsystems

Active condition monitoring of automation systems is a strongly emerging topic. In this paper a purely data-driven approach is presented which allows for detection of wear-out or faulty situations in machine subsystems, each consisting of a servomotor driving different parts of the machine. The approach is based on empirical analysis of the motor's electric currents, but can be applied to the tracking errors of the servo controller as well. The features gained from the spectra of the motor currents or tracking errors, respectively, are reduced by singular value decomposition and classified by a kernel density based algorithm. Because of the purely data-driven approach the method can also be applied to a variety of different systems.