Fault detection and condition assessment of power transformers using practical signal processing methods

Dissolved gases in transformer oil indicate the occurrence of a fault in the transformer. Many standards use the ratio of dissolved gases in transformer oil to detect faults in transformers. These traditional methods cannot, however, be used in cases in which the transformer needs to be immediately removed from service in case of a serious error such as electric arcs to prevent the error from spreading. For this purpose, this paper uses signal processing methods that analyze the signal online. The paper assumes that CO and CO2 dissolved gases in transformer oil are measured online by a sensor and then some signal processing methods are applied to the measurement data. These methods include Fast Fourier Transform, Discrete Wavelet Transform, Hilbert Transform, Gabor Transform, the combination of discrete wavelet and Hilbert transform, the combination of discrete wavelet and Gabor transform, and spectrogram. These methods are continuously applied to these two gases that are soluble in transformer oil to determine their variations in the transformer oil at a certain time or a certain frequency. The gases are also modified and then the performance of each of the processing methods mentioned in these changes is investigated. Fault detection reference is the CIGER 761-2019 standard. The purpose of this paper is to find out the samples of gases change in a frequency interval or timeframe together irrespective of the faults in the transformer and changes in the volume of dissolved gases in transformer oil, analyze the signal processing methods, and detect the type of fault using CIGRE 761-2019 standard. The fast Fourier transform method analyzes signal power by frequency. The discrete wavelet transform method extracts high-frequency components of gases and detects faults based on the largest components. The Hilbert transform method converts the signal into two real and imaginary parts. Then, it uses the imaginary part that represents the signal phase to detect faults. The Gabor transform method extracts instantaneous frequencies in the time-frequency plane and uses this method to detect faults. In the methods that combine discrete wavelet transform and Hilbert or Gabor transform, high-frequency components are extracted by discrete wavelet transform, and then Hilbert or Gabor transform methods are applied. The spectrogram method also indicates the size of the short-time Fourier transform, which is used to analyze the signal. These signal processing methods are compared in several features, and the discrete wavelet transform method is introduced as the best method for fault detection.