Transformer Core Saturation Fault Analysis using Current Sensor Signals and Thermal Image Features

Abstract

Transformer faults are identified and classified using current sensor signals. Transformer core saturation detection is challenging using current sensor signals due to overlapping of load-based faults such as overload, short circuit, grounding faults, in-rush and power supply fluctuations in current sensor signals. Existing methods are unable to differentiate load-based fault and Transformer core-based faults from current sensor signals. In this paper, transformer core-based faults such as overheating, voltage regulation issues due to power fluctuation, increased current draw due to short circuit or overload are differentiated from load-based faults, using current sensor signal energy band and thermal image of current sensor which are acquired simultaneously. In this paper, transformer core-based faults are differentiated from load-based faults after the current signals are processed with Modified- Tunable Q-factor Wavelet Transform and Rational Dilation Wavelet Transform and current sensor thermal images are processed with Multi Resolution wavelet - Deep Convolutional Neural Network. Energy band-based values from current sensor signal and current sensor thermal image Haralick features are used for differentiating transformer core-based and load-based faults. From the experimental and simulation results the transformer core-based and load-based faults are detected with an accuracy of 95% and compared with traditional methods.

Keywords

• condition monitoring
• fault detection
• image processing
• thermal analysis
• feature extraction