Support Vector Machines for Condition Monitoring of Bearings in a Varying-Speed Machinery

This paper deals with the diagnostics of ball bearings in direct-drive motors by means ofSupport Vector Machines (SVM). For the new generation of packaging machines,direct-drive motors have substituted parts of the mechanical transmission chains asgears train or cams between the motor and the load. In fact, a position feed-back is usedby the control system to perform polynomial profiles of motion involving continuouschanges of the shaft speed and a cyclic inversion of its rotating direction.The presenceof complex shaft dynamics provides a vibration signal extremely noisy, where theevents due to malfunctions or damages disappear and are not periodic, so the classicmethodologies for the monitoring of the support conditions based on the signal analysisin the frequency domain are ineffective. In this paper the use of Support VectorMachines overcomes the constant-speed limits and the SVM are proven to be apowerful and reliable tool to diagnose the health of ball bearing even in variable-speedapplications, based on experimental tests conducted on an industrial packaging machine