A reliable approach for detection of incipient faults of short-circuits in induction generators using machine learning. (October 2018)

• Record Type: Journal Article
• Authors: Rebouças Filho, Pedro Pedrosa; Nascimento, Navar M.M.; Sousa, Igor R.; Medeiros, Cláudio M.S.; de Albuquerque, Victor Hugo C.
• Is Part Of: Computers & electrical engineering. Volume 71(2018)
• Page Start: 440
• Page End: 451
• Publication Date: 2018-10
• Subjects: Fault detection -- Induction generator -- Machine learning -- Neural networks -- Reliability -- Short-circuit -- Wind turbine; Computer engineering -- Periodicals; Electrical engineering -- Periodicals; Electrical engineering -- Data processing -- Periodicals; Ordinateurs -- Conception et construction -- Périodiques; Électrotechnique -- Périodiques; Électrotechnique -- Informatique -- Périodiques; Computer engineering; Electrical engineering; Electrical engineering -- Data processing; Periodicals; Electronic journals; 621.302854
• DOI: 10.1016/j.compeleceng.2018.07.046
• Languages: English
• ISSNs: 0045-7906

Abstract

Highlights: An approach for detection of incipient faults of short-circuits in wind turbine induction generator using machine learning. This research address a solid approach to detect incipient stator winding inter-turn short-circuits in SCIG applied in wind turbines. A experimental setup for electrical current acquisition of different types of faults is proposed for further applications. The results driven to the neural network, Multi-layer Perceptron (MLP), and the Fourier transform as the better approach for our problem. Evaluation of several classifiers Machine Learning. Abstract: This research contributes with a reliable approach to detect incipient stator winding inter-turn short-circuits in induction generators applied in wind turbines. Using a wind turbine test-bench, we inserted different types of short-circuit in the generator. The electrical current is acquired to build a fault database. We propose the use of four feature extraction techniques with three classifiers. The MLP identified 100% of the generator's Normal conditions with less than 1% false positives and negatives. Using different topologies of MLP, it was possible to identify incipient short-circuits in 1.41% turns with 99.33% accuracy. The combination Fourier-MLP is more useful for fault detection, since it obtained 84.48% of accuracy, with 99.98% of Normal conditions correctly classified.