Neural Network-Based Classification of Current Sensor Failures in Fault-Tolerant Control Induction Motor Drive

Modern drive systems with induction motors (IM) currently use fault-tolerant control (FTC) solutions which increase reliability and safety. Implementation of such systems requires the use of hardware or software redundancy. Due to the much lower cost of algorithmic solutions, they are used in industrial systems. Moreover, it is necessary to use an algorithm that allows for the detection and localization of faults. That makes it possible to isolate the damaged component and replace its signal with the estimated value. However, to make the fault compensation as good as possible, the damage should be classified. Only then is the control structure able to choose the best solution. Because in the vector control algorithm of the IM drive the most important variables are the stator currents, the article focuses on the classifier of current sensor (CS) faults. To fully automate the classification process, shallow neural networks (NN) are used. Thanks to the early detection and isolation of the faulty CS the influence of the malfunctioning control structure does not affect the NN, which can precisely classify the type of damage, based only on the measured and estimated stator currents.