Real-Time Fault Isolation in Multiphase Multilevel NPC Converters Using Active Semi-supervised Fuzzy Clustering Algorithm with Pairwise Constraints
Conference Paper
Overview
Research
Identity
Additional Document Info
Other
View All
Overview
abstract
2017 IEEE. A new methodology to improve the performance of real-Time open switch fault isolation scheme in multilevel multiphase converters has been proposed in this paper. Within the presented framework, the fault is detected by a model-based method using self-recurrent wavelet neural network. Afterward, a semi-supervised fuzzy clustering technique is applied to locate the failed switch while multiresolution wavelet analysis is used for feature extraction. Presented method offers the ability to detect the fault online and minimizes the isolation time without deploying extra hardware. In comparison with supervised classifiers to locate the failed switches, the proposed scheme requires significantly fewer data collected under fault operation to train the clustering model. The developed scheme has been validated by different fault scenarios on MATLAB simulations as well as on a lab prototype including a five-phase induction motor fed by a 3-level 5-phase neutral point clamped inverter. The results have demonstrated that the method is a promising fault diagnosis scheme for multilevel multiphase drives in less than 500 microsecond isolation time which is about 1% of the fundamental period of currents.
name of conference
2017 IEEE International Electric Machines and Drives Conference (IEMDC)