Real-Time Neural Classifiers for Sensor Faults in Three Phase Induction Motors

Induction motors can be modeled in different ways for correct operation and control, one of these is the ? - ? representation, this model has six state variables that monitored: rotor position, speed, flux, current current. Usually, only three measured directly with sensors. These sensors are subject to long periods work stress, so a failure cannot ruled out. Sensor cause problems control motor, instability or motor performance degradation. That why fault tolerant controllers proposed maintain stability induction despite sensor failure, assuming error classified correctly short period time. This paper concerned detection classification faults: currents, real time, considered faults occur by disconnection, degradation, connection damage, among other hardware software phenomena. Different neural networks compared real-time classification, are: Multilayer perceptron, convolutional network, unidirectional Long short-term memory (LSTM) bidirectional LSTM. The results show CNN network presents best methods, but LSTM shorter time high accuracy classify true class. used corresponds simple configuration convolution layer 20 filters 2×1, followed pooling two dense layers. above 99% an average per sample 4.6236e-08 s. For its part, shows approximately 3.1298e-09 s, MLP 97.96% 5.5 e-10 while BiLSTM 98% 4.47e-4