Machine learning and multiresolution decomposition for embedded applications to detect short-circuit in induction motors

•Statistical features may be combined with machine learning to detect short-circuits.•The inclusion of multiresolution analysis improves the detection of short-circuits.•The inclusion of multiresolution analysis minimizes the occurrence of false alarms.•Signal acquisition parameters may be reduced w...

Full description

Saved in:
Bibliographic Details
Published in:Computers in industry 2021-08, Vol.129, p.103461, Article 103461
Main Authors: Guerreiro Carvalho Cunha, Rebeca, da Silva, Elias Teodoro, Marques de Sá Medeiros, Cláudio
Format: Article
Language:English
Subjects:
Discrete wavelet transform
Embedded systems
Fault detection
Induction motor
Machine learning
Citations: Items that this one cites
Items that cite this one
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:•Statistical features may be combined with machine learning to detect short-circuits.•The inclusion of multiresolution analysis improves the detection of short-circuits.•The inclusion of multiresolution analysis minimizes the occurrence of false alarms.•Signal acquisition parameters may be reduced without drastic performance reduction. Due to the relevance of induction machines (IM) in industrial applications, the development of solutions to predict and detect incipient faults in such equipment is an important field of study. Despite the variety of solutions already proposed by other investigations, most of them do not take into consideration aspects of its execution in the field. This paper describes an algorithm combining the discrete wavelet transform (DWT) for multiresolution analysis (MRA), statistical features and machine learning (ML) techniques to detect incipient short-circuit faults (ISCF) in IM using voltage signal induced by axial leakage flux signal. The most important result is the true negative rate (normal class) of 100%, eliminating the occurrence of false alarms. Additionally, an accuracy of 99.23% was achieved for normal versus defective classification. With the progress of the (Internet of Things) IoT in the industry, intelligent fault-detection solutions must seek to reduce their computational cost to become pervasive. Therefore, further analysis was carried out to decrease the computational cost of the proposed approach without significantly compromising the accuracy of the model.
ISSN:0166-3615
1872-6194
DOI:10.1016/j.compind.2021.103461