Title: A Constrained Neural Classifier for Pulsed Eddy Current based Flaw Detection in Industrial Pipes
Authors: Gilvan Silva, Edmar Souza, Paulo Farias, Eduardo Simas Filho, Maria Albuquerque, Ivan Silva, Cláudia Farias
Abstract: Decision support systems are important to improve the efficiency of nondestructive evaluation, specially for industrial equipment. Pulsed eddy-current is a magnetic method used for evaluation of metallic equipment. In this paper, is proposed the combination of pulsed eddy current evaluation, digital signal processing, and neural networks to detect flaws in industrial pipes. A novel method using particle swarm optimization is proposed for imposing performance constraints during neural classifier training process. Results obtained for experimental signals acquired from composite-insulated metallic industrial pipes presenting internal and external corrosion areas are used to validate the proposed method. A comparison to neural networks trained from the traditional back-propagation algorithm was presented.
Key-words: Artificial Neural Networks, Particle Swarm Optimization, pulsed-eddy current evaluation, signal processing
DOI code: 10.21528/CBIC2019-127
PDF file: CBIC2019-127.pdf
BibTeX file: CBIC2019-127.bib