A machine learning based model for classifying incipient faults and degree in power transformer windings using voltage current diagram technique

Power transformers are important components of electrical systems. Their failure or malfunction can have serious consequences, affecting the overall functionality or safety of the electrical system. Rapid and accurate diagnosis of transformer internal faults are key factors of efficient and safe operation. in the literature, several techniques of power transformers windings diagnosis exist. The voltage current diagram is a promising and powerful technique acknowledged to be efficient and quick in winding faults diagnosis. However, two major limitations of this technique reported in the literature are: its inability to detect incipient faults and the method of faults classification which is manual. In this study, two machine learning algorithms (SVM and linear regression) are combined with voltage current diagram to diagnose internal incipient faults by analysing data collected during fault simulations on a layer type power transformer. Three main faults responsible of power transformer failure are considerate: turn to turn short-circuit, buckling stress and axial displacement. For each type of fault, a dataset is generated and the model is trained. The SVM algorithm is used to identify the type of fault (classification), and the linear regression algorithm is used to determine its degree of severity. The highest performance of classification was obtained using the RBF kernel (82%) and the determination of the degree of severity using R-squared gave a score of 99,9%.