Insulation failure prediction in AC machines using line-neutral voltages

This paper presents a novel technique to detect insulation failure in polyphase AC machines. The machine must be star connected and have the neutral accessible. The mathematical theory of the technique is provided and supported by an experimental validation-turn faults in the stator of an induction machine are detected from the algebraic sum of the three instantaneous line-neutral voltages. However, this voltage sum contains undesirable frequencies that decrease the sensitivity of the scheme. Bandpass filtering the voltage sum (around the fundamental) optimizes sensitivity by removing the harmonics that arise from core saturation, slot harmonics, etc. The design of the scheme makes it practically immune to false alarms attributable to varying load conditions, operating temperatures and source voltage perturbations. It is shown that this technique is simpler in both theory and practice than other techniques based upon accurate calculations of sequence voltages and currents. The simplicity of this technique permits a low-cost implementation to flag a turn fault within a few cycles of the fundamental machine excitation.