Use of FFT and ANN techniques in monitoring of transformer fault gases

This paper reports on continuing work undertaken by the authors in developing a portable monitor to detect multigases especially H₂ and CO dissolved in oil. A significant improvement in reliability of detection has been achieved using FFT and ANN signal processing techniques. The monitor uses the membrane and forced diffusion techniques to extract the dissolved gases in oil, and senses the gases by two thin film semiconducting gas sensors. Stability in sensor detection especially in a practical varying ambient conditions is achieved by a cyclic heating technique. And ac conductance measurement technique. Sensitivity and quantitative selectivity are achieved by signal processing using FFT and ANN techniques. The FFT resolves the frequency components of the sensor response up to 3rd harmonics and the data are used to train a back propagation ANN network to map the fingerprints. Detection up to 20 ppm has been achieved by just using two sensors. Detection for other gases like C₂H₂ and CO₂ has been done by using different membranes using the same two sensors and the measured response will be reported. A portable microcontroller based monitor was developed by incorporating in software the derived weights of ANN and the measuring techniques. The results are reported in this paper