Insulation protection and online stress agent identification for electric machines using artificial intelligence

Abstract

Insulation failures are among the most important problems affecting electric machines. Although standard procedures are widely used to assess the insulation condition of these machines, no alternative is available to allow the identification of the stress agents affecting the insulation system. Another limitation of the standard procedures is the fact that they are based on tests that require the interruption of the operation of the machine. This study presents a technique that allows the online assessment and protection of the insulation of low-and medium-voltage electric machines, along with the identification of the stress agents responsible for the degradation of the insulation. The proposed technique is based on the use of a highly sensitive current transformer, an algorithm to identify the most affected phase, and an artificial neural network to indicate the stress agent affecting the machine. The presented experimental results show that the proposed system can identify thermal degradation, moisture absorption and contamination by oil. The low current levels that can be measured by the developed system enable the application of the proposed technique to low-and medium-voltage machines.

Insulation failures prognosis in electric machines: preventive detection and time to failure forecast

“…As presented in [30,31], it is possible to determine the stress factor affecting the insulation of machines based on the leakage current components. The resistive (I R ) and capacitive (I C ) components of the leakage current can be calculated from I˙L, taking into account the angle (θ 0 ) between this current and the phase voltages, measured through the VT.…”

“…As it was shown, as the leakage current increases, the maximum error in the calculation of the resistive and capacitive components of the current becomes <5%. From the evolution of these quantities over time, compared to their values when the machine insulation was considered adequate, an ANN is used to detect and identify the stress agent affecting the insulation [30,31]. Fig.…”

“…Since the machine does not have any neutral connection, the sum of the phase currents is zero. For this reason, the current measured by the HSCT is the resultant leakage current, given by As presented in [30, 31], it is possible to determine the stress factor affecting the insulation of machines based on the leakage current components. The resistive () and capacitive () components of the leakage current can be calculated from , taking into account the angle () between this current and the phase voltages, measured through the VT.…”

“…From the evolution of these quantities over time, compared to their values when the machine insulation was considered adequate, an ANN is used to detect and identify the stress agent affecting the insulation [30, 31]. Fig.…”

“…Table 2 presents, from the data used for training, an indication of limits for and DF , defining each class. More detail on the separation of classes can be found in [30].…”