کنفرانس بین المللی ماشین و محرکه های الکتریکی

صفحه اصلی / The 5th International Conference on Electrical Machines and Drives

Multi-Filter Deep Learning Framework for Partial Discharge Localization in Generator Stators

نویسندگان :

Mehdi Nourollahi¹ Mohammad Hamed Samimi² Arash Abyaz³ Ahmad Kalhor⁴

1- university of tehran 2- university of tehran 3- university of tehran 4- university of tehran

کلمات کلیدی :

Partial Discharge (PD)،Generator Stator،Deep Learning،PD Localization،Transformer Networks

چکیده :

Partial discharges (PDs) are early indicators of insulation degradation in generator stator windings, and their timely localization is essential to ensure reliable operation and reduce costly outages. This paper presents a deep learning-based framework for PD localization, leveraging a high-fidelity finite element (FEM) simulation model of a 12-bar stator. A multi-filter acquisition strategy is employed to capture complementary spectral components of PD signals prior to feature learning. Five neural network architectures are investigated: multi-layer perceptron (MLP), convolutional neural network (CNN), temporal convolutional network (TCN), CNN combined with long short-term memory (CNN+LSTM), and Transformer. Results demonstrate that the Transformer model achieves the highest classification accuracy (87.1%) for PD localization, benefiting from its self-attention mechanism to capture long-range dependencies and reduce misclassification. The CNN+LSTM hybrid model also provides strong performance, while simpler models such as MLP and CNN show limited capability in handling nonlinear, overlapping PD features. The findings confirm that attention-based architectures, combined with multi-filter strategies, offer a powerful approach for practical condition monitoring of high-voltage rotating machines, paving the way for reliable AI-assisted maintenance in industrial applications.