Broken Rotor Bar Fault Identification using Current Signature Data and Deep Learning

Fazeela  Rattar; Abdul Wahid; Muhammad Zakir Shaikh; Dileep Kumar; Majid Hussain; Junaid Ahmed

doi:10.55447/jaet.05.02.37

Authors

Fazeela Rattar NCRA Condition Monitoring Systems Lab, Mehran University of Engineering & Technology, Jamshoro, Pakistan
Abdul Wahid NCRA Condition Monitoring Systems Lab, Mehran University of Engineering & Technology, Jamshoro, Pakistan
Muhammad Zakir Shaikh NCRA Condition Monitoring Systems Lab, Mehran University of Engineering & Technology, Jamshoro, Pakistan
Dileep Kumar Mehran University of Engineering and Technology Jamshoro http://orcid.org/0000-0002-6211-1078
Majid Hussain NCRA Condition Monitoring Systems Lab, Mehran University of Engineering & Technology, Jamshoro, Pakistan
Junaid Ahmed NCRA Condition Monitoring Systems Lab, Mehran University of Engineering & Technology, Jamshoro, Pakistan

DOI:

https://doi.org/10.55447/jaet.05.02.37

Keywords:

Broken Rotor Bar, Condition Monitoring, Current Signatures, Deep Learning, FPGA

Abstract

Three-phase induction motors (IMs) are used widely used in industrial and transportation applications owing to its cost-effectiveness, simple construction, and high efficiency. However, extensive use of IMs without maintenance causes operation failures and economic losses. The broken rotor bar (BRB) is one of the most occurring faults in rotating machines. To avoid this fault an efficient condition monitoring system is required for diagnosing BRB fault in motor. This paper presents, a BRB fault identification method based on Deep Learning (DL) models. The proposed system acquires the data using non-invasive current sensors through the myRIO board, which offers efficient data acquisition due to its FPGA capabilities in it. The acquired data is used to train DL models and then it is utilized to test the models in order to determine the motor condition. Among the employed DL models, the Long-Short Term Memory (LSTM) achieved best with 100% classification accuracy with raw data. This proposed approach provides an effective and robust BRB fault detection.