Prediction of the hardness of X12m using Barkhausen noise and component analysis methods

Abstract Barkhausen noise (BN) generated by the stochastic movements of domain walls is one of the most popular non-destructive testing signal. To measure the property of material, the feature(s) extracted from BN signal has been focused by the existing studies. Although the physical characteristic of several BN features could be proven, many features used in the BN-related works are prone to being interfered by the noise, temperature and other measurement conditions. In this paper, to build a stable and unified representation of BN signal, a novel BN feature extraction and hardness prediction method is proposed. The proposed method includes BN-reconstructed AR model, modified slow feature analysis for fusing different AR-order signal and discriminant incoherent component analysis for the hardness prediction. In the experiment, all potential parameters involved in our method were tested to show the relationship between the parameters and hardness prediction accuracy. Then our proposed method was compared with other component-analysis-based methods and self-defined isolated-feature-based prediction methods. The experimental result implies that our proposed method outperforms other methods, including features generated by component analysis methods and the combination of conventional BN features.