Simultaneous quantitative prediction of tensile stress, surface hardness and case depth in medium carbon steel rods based on multifunctional magnetic testing techniques

The simultaneous quantitative prediction of three target properties (tensile stress, surface hardness and case depth) in medium carbon steel rods based on magnetic nondestructive testing techniques was experimentally investigated. A multifunctional sensor is proposed to simultaneously measure the magnetic Barkhausen noise, tangential magnetic field and magnetic hysteresis curve of the tensioned steel rods with different depths of surface hardened layer. Mean impact value evaluation algorithm is introduced to select the optimal combination from the sixteen candidate feature parameters, which are extracted from the measured signals, as the input nodes of BP neural network. For the two types of rods of 45 steel rods and 42CrMo steel, the established BP prediction model possesses good accuracy in simultaneous quantitative prediction of the three target properties. The maximum prediction error for surface hardness is less than 1.6%. The average prediction error for tensile stress case is around 5.1% in 45 steel rods and 5.3% in 42CrMo steel rods. Under the tensile stress higher than 40 MPa, the average prediction error is less than 3.4%. Although the BP prediction model may lose resolution in distinguishing small changes in the case depth around 0.07 mm, the average prediction error of case depth in both types of medium carbon steel rods is lower than 5%.