Loading…
A novel approach based on pattern recognition techniques to evaluate magnetic properties of a non-grain oriented electrical steel in the secondary recrystallization process
•A new method for evaluate magnetic losses of NGO steels using pattern recognition.•The classification problem was defined from measurements of hysteresis curves.•The proposal also involves class definition, feature extraction and over-sampling.•The LSSVM classifier achieved the best results with 88...
Saved in:
Published in: | Measurement : journal of the International Measurement Confederation 2021-01, Vol.167, p.108135, Article 108135 |
---|---|
Main Authors: | , , , , |
Format: | Article |
Language: | English |
Subjects: | |
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!
|
Summary: | •A new method for evaluate magnetic losses of NGO steels using pattern recognition.•The classification problem was defined from measurements of hysteresis curves.•The proposal also involves class definition, feature extraction and over-sampling.•The LSSVM classifier achieved the best results with 88.9% accuracy in test phase.
This paper proposes a new automatic approach based on machine learning strategies to associate the microstructural conditions of Non-Grain Oriented (NGO) steels, during secondary recrystallization, with their magnetic losses, which were determined from hysteresis loops. These hysteresis curves and the states of the secondary recrystallization enabled us to establish the feature extraction and the labels for the classification problem. We also applied a specific methodology to create synthetic samples to overcome the available database, which was too small, and its imbalance among the classes. As far as the authors know, this is the first time that a study has treated this issue in this manner. Normally, magnetic losses of NGO steels are analyzed through expensive and laborious tests. We evaluated our proposal through computer experiments with several state-of-the-art classifiers. The Least Squares Support Vector Machine (LSSVM) achieved the best results with 88.9% accuracy using real samples. |
---|---|
ISSN: | 0263-2241 1873-412X |
DOI: | 10.1016/j.measurement.2020.108135 |