Electromagnetic-Acoustic Sensing-Based Multi-Feature Fusion Method for Stress Assessment and Prediction

Manufacturing and online service of ferromagnetic materials easily induce local stress concentrations and then generate cracks. Research on in-service inspection of stress status is an important criterion for healthy monitoring in steel components and structures. There are inherent limitations for stress analysis by using a single feature from a single sensor source. In this work, a multisensor feature fusion network based on combining principal component analysis (PCA) and the XGBoost algorithm is proposed to analyze the Barkhausen noise sensor and magneto-acoustic emission sensor for assessing and predicting the stress state in ferromagnetic materials. PCA combined with feature correlation analysis is conducted for feature selection by eliminating redundant information and reducing the dimensionality of the dataset. In addition, a machine learning service was used to create an XGBoost model to predict the stress state. Compared with other single sensor feature fusion methods, our proposed electromagnetic-acoustic sensing-based multi-feature fusion network outperforms other models in terms of accuracy and repeatability. Specifically, we discuss why the proposed model is superior to others from the physical mechanism of the stochastic behavior of magnetic domain wall dynamics. Experimental studies on pure iron are further carried out to verify the effectiveness and robustness of our proposed method.