Detecção de faltas internas em SCIG utilizando sistemas inteligentes

Abstract

Electrical machines, eventually, will present some failure and within the possibilities that may occur, failures in the stator coils jeopardize their correct operation. These faults can evolve quickly, irreversibly damaging the coil and causing economic losses. In the literature, different ways to solve the problem have already been implemented, such as: use of fuzzy logic, prediction models, current signature analysis, among others. This work consists of the implementation and evaluation of intelligent systems for fault detection in the turn-turn and turn-to-ground stator of the squirrel-cage induction machine operating as a wind generator. Therefore, it is proposed the implementation of two machine learning models, an artificial neural network and a convolutional neural network, with the purpose of learning the characteristics of the stator electrical currents and differentiating a healthy machine from a damaged one. The systems are trained with artificial data from simulations and these are based on models of the main components of a wind generation system such as: control systems, actuation command switch, mechanical model of the turbine and electrical model of the generator with faults. All faults evaluated have a fraction less than or equal to 10% of the damaged windings. To analyze the results, the k-fold method was used, with a better performance of convolutional networks in comparison with artificial neural networks in the order of 15.95 percentage points.