Uma abordagem baseada em histograma de gradientes orientados para a detecção de barras rompidas em motores de indução com rotor gaiola de esquilo

Abstract

The three-phase induction motor, and especially the one built with the squirrel cage rotor, is one of the most widely used electric machines in the industrial environment around the world today, due to its robustness and low cost when compared to other machines oriented to the same application. However, induction motors are subject to failure in both stator and rotor. In this sense, early failure detection on these machines is an important contributor not only to reducing the number and time of interruption of their operation, but also to reducing maintenance costs and especially avoiding their replacement, which would entail even higher expenditures. Among the most recurring faults in induction motors are the partial and/or total breakage of the bars at the junction with the short circuit ring of the rotor. This causes additional damage to the machine, such as an increase in its vibration, a rise in temperature above its specifications and a reduction in its performance. Unlike traditional techniques such as stator current signature analysis, among other metrics found in the literature that require relatively high slip, this research paper proposes an alternative computational method to detect the total disruption of adjacent rotor bars in an induction motor operating in low slip condition and powered by a frequency inverter. The proposed methodology, as in some technique, is noninvasive and allows to detect the defect from the reading of a motor phase current, using a conventional current transformer and a digital filter stage in order to eliminate noise in the measured signal. As a theoretical contribution to this field of knowledge, the work employs a signal processing method called Oriented Gradient Histogram (HGO), commonly used in the image processing area, which has been particularly modified for the treatment of the stator current signal. The histograms created by the HGO technique allowed to generate a knowledge base with data from a healthy and damaged rotor, and the use of an artificial neural network allowed to classify the defect with good accuracy, for the motor operating at low slip and fed with voltage that is not purely sinusoidal. Experimental results performed with a 7.5 kW motor show the good accuracy achieved with the present study compared to other fault diagnosis techniques.