Experimental ANN-Based Vibration Frequency Model for Transformers Under Different Conditions

In power systems, transformers are fundamental components for ensuring efficient voltage regulation and distribution. Despite their robustness, these electric machines are prone to various faults, with short-circuited turns (SCTs) being one of the most common. Such faults can lead to increased vibrations, overheating, and eventual failure. To accurately diagnose these issues, vibration signature analysis has become a powerful and essential tool, enhancing preventive maintenance and minimizing downtime. Since frequency domain-based vibration signature analysis provides valuable insights into the condition of transformers, developing a model that generates the vibration frequency content of transformers under different operating conditions will be crucial for testing and enhancing diagnostic techniques. In this context, this article advances the state of the art by presenting an artificial neural network (ANN)-based experimental vibration frequency model, tested in a 1.5-kVA dry-type transformer under various conditions: different degrees of SCTs, ranging from 0 (healthy condition) to 10 SCTs, different electrical loads (no load, linear load, and nonlinear load), and various voltage imbalances (0%, 2.5%, 5%, 7.5%, and 10%, with 0% indicates no imbalance). Thus, the proposed model utilizes experimental vibration data to train the ANN, enabling it to identify and interpolate conditions not explicitly covered during training. Key results from this study demonstrate the model’s ability to achieve low root mean square error in both reconstructing known conditions and interpolating unknown ones. This demonstrates the model’s robustness and generalization capability, making it a valuable tool for different applications related to transformer diagnostics.