AN ARTIFICIAL NEURAL NETWORKS APPROACH TO STATOR CURRENT SENSOR FAULTS DETECTION FOR DTC-SVM STRUCTURE
 
 
More details
Hide details
1
Wrocław University of Technology
CORRESPONDING AUTHOR
Kamil Klimkowski   

Wrocław University of Technology, Smoluchowskiego 19, 50-372 Wrocław, Poland
 
Power Electronics and Drives 2016;1 (36)(1)
KEYWORDS
TOPICS
ABSTRACT
Background: During last few years, fault-tolerant control systems (FTCS) [1] for electric motor drives became a very active research field for many research groups. The proper choice of the fault tolerant control algorithm and topology depends on the drive system requirements and used components. To ensure the proper work of complex systems, it is necessary to take account of diagnostic techniques, that within a reasonable period of time will allow to detect a failure and an appropriate response of the control structure [2, 8, 9]. Material and methods: The main goal of the paper is selection of the most appropriate neural net-work intended for the detection and identification of stator current sensor faults. Different network structures and learning algorithms were tested. Results: Detectors based on neural networks with various learning methods and structures were presented and tested during different drive conditions. Conducted tests and obtained results may be used in designing process of advanced detection algo-rithm for Fault Tolerant drives. Conclusions: It was proved that the lowest error values and quickest response of diagnostic unit were achieved for neural network with two hidden layers learnt by Leven-berg-Marquardt algorithm. Similar results were obtained for the same network but trained by Quasi-Newton method with the BFGS algorithm.
eISSN:2543-4292
ISSN (print):2451-0262