Oil Quality Index Model Verification and Validation using Total Acid Number and Interfacial Tension Experimental Data
Ugochukwu Elele,Azam Nekahi,Arshad Arshad,Issouf Fofana,Kate McAulay
DOI: https://doi.org/10.1109/tdei.2024.3357665
IF: 2.509
2024-01-01
IEEE Transactions on Dielectrics and Electrical Insulation
Abstract:Transformers are indispensable components in any power networks, facilitating the delivery of generated electricity to consumers at the most secure voltage level. The insulation system of oil-filled transformers is critical for the safe operation of power transformers, although it undergoes consistent degradation over time. Similar to blood in a human body, the insulating fluid serves as condition monitoring medium. Most traditional oil ageing detection methodologies operate offline; thus they are most suitable for planned maintenance activities. However, these methods have their drawbacks including potential safety risks, contamination of samples, loss of productive hours, and the potential risk of overlooking early signs of ageing that could occur beyond the maintenance cycle window. The Myers Oil Quality Index Number (OQIN), derived from the quotient of interfacial tension (IFT) and total acid number (TAN) values, provides a tool for classifying transformer oil into seven distinct categories, expanding the potential for both offline and online oil applications. In this work, eighteen 750ml samples of natural ester oil (NEO) were procured, aged, and analysed using offline TAN and IFT techniques, and their respective values (IFT, TAN, and OQIN) were recorded. These experimental data sets were employed to verify and validate (V&V) an OQIN machine learning model. The model was further validated using existing mineral oil (MIN) data sets. The high-performance metrics, demonstrated in terms of accuracy, precision, sensitivity, specificity and F-Score, confirm the effectiveness of the model for online transformer oil ageing detection and classification. The bagged tree ensemble model showed the best performance for OQIN, NEO, MIN respectively in terms of accuracy (100%, 83.30%, 100%), precision (100%, 90%, 100%), sensitivity (100%, 88%, 100%), specificity (100%, 96%, 100%) and F-Score (100%, 84.76%, 100%). This development proposes the potential for a shift from traditional offline scheduled maintenance ageing detection methods to an online/Internet of Things (IoT) - based prescriptive ageing detection, thereby enhancing the reliability of transformer performance in situ.
engineering, electrical & electronic,physics, applied