An Evolutionary Computation-Based Method for Nondestructive Tensile Force Measurements in Two Materials Using Hysteresis Loop

In traditional methods, features extracted from the hysteresis loop are manually analyzed to establish a model for predicting tensile force in one ferromagnetic material. However, the hysteresis loop is sensitive to material type, thus increasing the difficulty in predicting tensile forces in two materials. Moreover, manual feature extraction is time consuming. It is necessary to complete automatic feature extraction and elimination of the influence of material type on the hysteresis loop in order to automatically establish a prediction model for tensile force prediction in two materials. In this article, an evolutionary computation-based method was proposed to perform the above-mentioned two tasks simultaneously. The proposed method involves three steps. First, hysteresis loops of two materials under different tensile forces were measured. Second, the measured hysteresis loops were transformed into corresponding transformed ascending hysteresis curves through a transformation operation of the curve of hysteresis loop change (CHLC) for eliminating the influence of material type on the hysteresis loop. Third, an improved bat algorithm was proposed to eliminate further the influence of material type on the transformed ascending hysteresis curve. The improved bat algorithm iteratively gave the position of a material type-insensitive feature on the transformed ascending hysteresis curve. The material type-insensitive feature-based neural network was used to predict tensile forces in two materials. Experimental results verified the proposed method. The motivation of obtaining a general tensile force prediction model, the applicability of the proposed method in more than two materials, and the influence of CHLC transformation operation on the improved bat algorithm and neural network are also discussed.