Online Measurement of Mechanical Properties of Cold Rolling Strips Based on Multiple Magnetic Parameters

The real-time and accurate assessment of product performance is crucial to ensure production stability and product quality in steel production. In this study, an electromagnetic nondestructive testing (ENDT) system for the mechanical properties (yield strength, tensile strength, elongation) of cold-rolled strip is developed, which incorporates four types of electromagnetic detection techniques, including magnetic Barkhausen noise (MBN), tangential magnetic field (TMF), magnetic incremental permeability (MIP), and multifrequency eddy current (EC). A network-based model is proposed, which consists of a self-adaptive convolutional channel attention mechanism (SACAM) module and an extended channel depthwise separable convolution (EDSC) module to predict the mechanical properties of strips based on the extracted magnetic features. SACAM-EDSC takes advantages of channel attention mechanism in feature enhancement and depth-separable convolution in model lightness. An actual steel coil dataset collected from the ENDT system was used to evaluate the performance of the proposed approach, which consisted of 41 input dimensions and 1521 sample numbers. Three machine-learning (ML) algorithms, namely decision tree (DT) regression, support vector regression (SVR), and neural network (NN) and three deep-learning (DL) algorithms, namely ShuffleNet, GhostNet, EDSC, were implemented to conduct comparative studies on the same dataset. The experimental results show that the proposed method performs well in terms of accuracy and efficiency compared with the other approaches.