Quantitative testing of micro-cracks by the MFL technique based on GA-BP neural network

Magnetic flux leakage (MFL) testing is one of the traditional electromagnetic non-destructive test (NDT) techniques, and the focus of the MFL technique is to predict the sizes of defects, particularly micro-cracks. In this paper, parameters identification of artificial rectangular micro-cracks ranging between 0.1-0.3 mm by the MFL technique is investigated with a BP neural network improved by a genetic algorithm (GA-BP neural network). In order to predict the sizes of artificial rectangular micro-cracks ranging between 0.1-0.3 mm, a MFL system based on anisotropic magneto-resistive (AMR) sensors is developed, and parameters identification is implemented with a GABP neural network. The results show that parameters identification of artificial rectangular micro-cracks can be implemented effectively with the developed MFL system and a GA-BP neural network, which provides a basis for predicting the sizes of the natural cracks. [Received 19 July 2016; Accepted 16 November 2016]