Abnormal Signals Elimination in Hardness Evaluation Using Barkhausen Noise and Tangential Magnetic Field

In this paper, magnetic Barkhausen noise (MBN) and tangential magnetic field (TMF) are employed to quantitatively predict the hardness of bearing steel GCr15. In order to solve the problems that MBN and TMF signals are susceptible to electromagnetic interference (EMI) and sensor vibration during the inspection process, which lead to the decrease of the hardness prediction accuracy, a feature-based abnormal signal elimination algorithm is proposed. The features of MBN and TMF signals are used to determine whether the signals are affected by EMI or sensor vibration. To verify the effectiveness of the algorithm, the multiple linear regression (MLR) and multilayer perceptron (MLP) hardness prediction model are developed based on MBN and TMF features. After removing abnormal signals, the hardness prediction error of MLR model is reduced from 21.39 to 1.25% and the hardness prediction error of MLP model is reduced from 7.75 to 0.13%.