Towards intrusive non-destructive online ageing detection of transformer oil leveraging bootsrapped machine learning models

Transformers play a crucial role in power networks, ensuring that generated electricity is delivered to consumers at the safest voltage level, reducing losses, and enabling metering and grounding. The insulation system is critical for ensuring that the function of the power transformer is carried out safely, at the expense of its gradual deterioration over time. Most conventional oil ageing detection methods are offline and, as a result, are best suited for scheduled maintenance practices which cause risk to life, sample contamination, loss of man-hour, and risk of missing critical incipient ageing responses outside the maintenance cycle window. Oil quality index (OQIN) data sets were bootstrapped to develop robust machine learning models for online ageing classification (class A to class G). Correlation models from existing mineral oil dataset was develop to predict the refractive index (RI), breakdown voltage value (BDV), dielectric dissipation factor (DDF), dissolved decayed products (DDP), total acid number (TAN), interfacial tension (IFT), and oil quality index (OQIN) using the optical fibre sensor transduced output voltage (OF STOV) of intensity modulated optical fibre. The existing mineral oil dataset also validates the developed OQIN machine learning model. The high correlative models presented in this paper have the potential to enable the transition from traditional offline scheduled maintenance ageing detection methods to online/IoT-based prescriptive ageing detection solutions. This paper also lays the groundwork for revolutionising in-situ transformer oil ageing detection to include digital-twinning capability, thereby improving reliability, reducing risks, and reducing operational costs.