Deep Learning Aided Classification of Ageing Condition of Natural Ester Oils Using FTIR Analysis

In power transformers, mineral oil has been used as an insulating fluid for ages. Alternative insulating fluids, such as synthetic and natural ester liquids, have received a lot of interest in recent years. Natural ester oil is not only made from ecofriendly ingredients, but it also offers additional benefits like being more ecologically friendly, and vulnerable to high temperatures than mineral oil and also promotes green technology. However, for the practical use of natural ester oils, it is essential to monitor the ageing state of these oils, which is the primary goal of this research. In this paper, a deep learning model had been developed with a convolution neural network (CNN), to classify different ageing stages of natural ester oil, while used as insulating oil in a power transformer. For this purpose. Fourier transform infrared spectroscopy (FTIR) data of natural ester oil representing different degrees of thermal ageing has been obtained from experiment. The 1-D FTIR data was converted to 2-D images using an image encoding technique. The 2D FTIR images were fed as inputs to the CNN model. In this study, three types of deep learning algorithms were used i.e., ResNet50, VGGNet19 and AlexNet. Three statistical parameters i.e. accuracy, sensitivity and precision were used as parameter indicators for performance analysis and comparison. Results showed that ResNet50 gave the best results among the other two. the accuracy, sensitivity and precision were 96.68 %, 95.98 % and 96.79 % for ResNet50, which showed the efficacy of the model.