Automation of the startup transient analysis of induction motors using a predictive stage

This paper proposes a new method to study the time-frequency diagrams used to diagnose induction motors. The complex pattern of evolving harmonics and interferences obtained from the analysis of the stator current during a high-inertia direct-on-line startup, such as the one needed for an acceleration test, is analyzed by a non-Markovian Particle Filtering state estimation approach. The introduction of a biasing in the system equations, based on the regularity with respect to previous states, and a prediction stage yields an automation procedure that, with reduced computational needs, allows collecting the most important information present in the diagram for diagnosis purposes: the energy content of significant rotor related harmonics during the transient, as well as the torque developed by the machine. An initial validation of the approach is carried out for two types of motors under different fault conditions.