Fault Analysis in Power Transformers with Finite Element Analysis and Deep Learning: A Study on Flux Distributions

In this study, a 2000 kVA power transformer with34.5/0.4 kV voltage, 50 Hz frequency, and Delta-Star connection is modeled under nonlinear loading conditions using Finite Element Analysis (FEA). Various fault types (short circuit, voltage imbalance, and harmonic distortions) that may occur under operating conditions of the transformer are analyzed through FEA simulations. For each fault type and for a healthy transformer, magnetic flux distributions are obtained. These magnetic flux distributions reflect the distinctive characteristics of each fault type, illustrating points of divergence in the transformer's internal magnetic behavior. The obtained flux distribution images are used as a dataset for the automatic classification of fault types. For the classification process, deep learning-based Artificial Intelligence (AI) models, namely ResNet18, ResNet50, VGG16, GoogleNet, and MobileNetv2, are utilized. These models are distinguished by their ability to differentiate between various fault conditions based on flux distributions. Results demonstrate that faults can be predicted from flux distribution images with accuracies up to 100%. The conducted classification studies provide an effective method for the early detection of faults in power transformers, thereby enhancing system reliability. In conclusion, this study presents a comprehensive analysis of the use of AI techniques for identifying and classifying different fault conditions in power transformers based on flux distributions. This method has the potential to optimize transformer maintenance and repair processes, thereby improving the reliability of energy systems.