Identificação de falhas em motores de indução empregando redes neurais

Abstract

Three-phase induction motors (TIM) are very often found in the industry. Yet, in several situ-ations, they are exposed to improper conditions such as high pressure and temperature, load variations and continuous vibration. Such conditions leave them vulnerable to failure (from exter-nal or internal sources) unwanted on industrial processes. In this context, predictive maintenance plays a relevant role, where the diagnosis and detection of failure in a timely manner allows an increase of the motor working life and enables the decrease of costs. Thus, it's important that methods of failure detection are developed to perform predictive maintenance. The traditional methods of failure acquisition are based on deterministic models, that lead to the elaboration of complex softwares and makes its management and maintenance difficult. Therefore, the failure acquisition based on conventional computerized techniques have been replaced for systems based on computer intelligence. This work aimed at developing a method of rotor failure identification in squirrel cage induction motors using Artificial Neural Networks. The use of Neural Networks could substantially decrease the number of sensors that monitor the operational conditions of a TIM. Such sensors contribute directly to the increase in the traditional technique implementation costs. To achieve this, characteristic features extracted from the current signature from motors with and without broken bars in the rotor were studied. These features were applied in Feed Forward Neural Networks to classify motors with and without broken bars and also identify the number of broken bars.