Feature extraction and evaluation for quantitative prediction of hardness in bearing steel based on magnetic Barkhausen noise

Abstract

In this paper, magnetic Barkhausen noise (MBN) is employed to quantitatively predict the hardness of GCr15 bearing steel. Firstly, to thoroughly investigate the relationship between MBN signal features and material hardness, a comprehensive study is conducted on multi-feature extraction methods for MBN signals based on time domain, frequency domain and time–frequency domain. Secondly, a novel feature evaluation algorithm is proposed that considers the correlation, stability and discriminability (CSD) of MBN features. This algorithm selects MBN features that are relevant to material hardness, remain stable under the same hardness level, and can distinguish between different hardness levels. Finally, linear regression models and multilayer perceptron models are established for the relationship between MBN features and material hardness. The models built using the features selected by the CSD feature evaluation algorithm demonstrate superior accuracy, with the root mean square error of 1.04 HRC for predicting unknown hardness values.