Data-driven condition monitoring of stator winding terminal insulation for inverter-fed machine using enhanced switching oscillation signals

The stator winding terminal insulation of inverter-fed machine is subjected to a higher risk of premature insulation breakdown. To quantitatively evaluate the terminal insulation degradation of the machine, this paper proposes a novel data-driven method using enhanced switching oscillation signals. Different from traditional methods, which require accurate high-frequency modeling, this paper aims to automatically learn fault features and predict the degrees of terminal insulation degradation. First, the original switching oscillation signals are reconstructed by wavelet packet analysis for insulation-sensitive feature enhancement. Then, a one-dimensional convolutional neural network (1DCNN) regression is established to extract state information from the enhanced switching oscillation signals and evaluate the terminal insulation capacitance. The experimental results show that the proposed method can evaluate the winding terminal insulation capacitance with high precision of pF level.