University of Tehran Researchers' Achievement with AI Assistance:
Identification of Power Generator Faults with 84% Accuracy
According to the Public Relations Office of the University of Tehran, in this research, the performance of several AI models was examined for identifying and determining the location of "partial discharge" in generator windings, as well as estimating the severity of this phenomenon.
Partial discharge is one of the early signs of damage to the electrical insulation of generators, and its timely detection can prevent the spread of damage and the occurrence of serious failures. However, changes in the signals caused by this phenomenon along the winding path make accurate detection of the location and actual severity difficult.
In this study, researchers used a finite element simulation model of the generator and generated data related to partial discharge, evaluating five AI models for two main objectives: identifying the fault location and estimating partial discharge severity.
The results showed that the Transformer model performed better than the other examined models in identifying the location of partial discharge, successfully detecting the fault location in the generator winding with approximately 84% accuracy.
In another part of the research dedicated to estimating electrical discharge severity, the CNN model showed the lowest error among the examined models and was able to estimate the actual value more accurately from the data recorded at the measurement site.
Based on the findings of this research, the use of AI models can make the analysis of generator insulation condition smarter and provide experts with more accurate information about the location and severity of faults. This capability can pave the way for predictive maintenance and identification of problems before they become serious failures.
The results of this research have been published in an article titled "Deep Learning Approaches for Partial Discharge Localization and Apparent Charge Prediction in Generator Stator Windings" by Mehdi Nourollahi, a graduate of the University of Tehran; Dr. Mohammad Hamed Samimi, a faculty member of the Electrical and Computer Engineering, University of Tehran, and an IEEE member; and Dr. Arash Abyaz, a PhD graduate of the Faculty of Electrical and Computer Engineering, University of Tehran, in the journal IEEE Transactions on Dielectrics and Electrical Insulation.
The article is available at the following link:
Your Comment :