Quantification of Ageing State of Insulating Oils Using FTIR Spectra and Deep Learning

Fourier transform infrared spectroscopy (FTIR) is a non-invasive optical diagnostic test method that provides insight into the ageing state of transformer oils. However, correct interpretation and extraction of suitable features from FTIR data are the keys to accurately classifying the ageing stages of oils. This research presents an automated framework employing deep learning (DL) to classify the ageing state of insulating oils using FTIR data. At first, the mineral oil, natural ester oil, and synthetic ester oil samples are subjected to accelerated thermal ageing. After that, FTIR analysis was conducted on these aged samples. The 1-D FTIR spectra are converted to the 2-D images using a Gramian angular field and Markov transition field. A modified convolutional neural network (CNN) is designed to classify the ageing stages of the oil. Finally, the proposed method is compared with various image processing techniques to show its efficacy and generalizability. The findings of this study show that the proposed CNN model delivers exceptionally high accuracy in classifying the ageing states of different insulating oils. The proposed method can be potentially implemented for remotely monitoring the insulation system of power transformers.