An accurate non-destructive method for determining mechanical properties of plain carbon steel parts using MHL and GRNN. Issue 3 (4th May 2021)

• Record Type: Journal Article
• Authors: Ahadi Akhlaghi, Iman; Kahrobaee, Saeed; Karami Nezhad, Kian
• Is Part Of: Nondestructive testing and evaluation. Volume 36:Issue 3(2021)
• Page Start: 278
• Page End: 296
• Publication Date: 2021-05-04
• Subjects: Non-destructive characterisation -- mechanical properties -- magnetic hysteresis loop -- generalised regression neural network (GRNN) -- heat-treating process; Non-destructive testing -- Periodicals; 620.112705
• DOI: 10.1080/10589759.2020.1740703
• Languages: English
• ISSNs: 1058-9759

Abstract

ABSTRACT: The sensitivity of magnetic non-destructive methods to both chemical composition and microstructure has limited their potential application for determination of mechanical properties in plain carbon steels under conditions of varying carbon content and microstructure. The present paper investigates advantages of applying an artificial neural network (ANN) method to magnetic hysteresis loop (MHL) method for non-destructively measuring mechanical properties of plain carbon steels with unknown carbon and microstructure (resulting from various heat-treating processes). Artificial neural network used in this study is a generalised regression neural network (GRNN), since it has reportedly high performance in estimation and function approximation and could be trained very fast. After it is appropriately trained, the neural network takes one of the four magnetic parameters (or any combination of them) extracted from the measured hysteresis loop to estimate the desired mechanical parameters (hardness, tensile strength, yield strength, and elongation) of the sample under test. The results revealed that the proposed methodology can be a very effective tool to estimate the mechanical properties of the hypoeutectic plain carbon steel sample with unknown carbon content and heat treatment background if appropriate combination of magnetic properties is used as the GRNN inputs.