| Abstrakt: |
In this paper, a combination of non-destructive magnetic technique and artificial neural networks is introduced, firstly, to ensure that the heat treatment process applied to a given API X65 steel sample results in desired microstructure and mechanical properties and secondly, to determine thickness variation which may occurs as a result of corrosion effect. To evaluate the effects of microstructure/mechanical properties and thickness variations on magnetic parameters, the magnetic hysteresis loop method has been applied on API X65 steel specimens with the thicknesses of 1–4 mm, each subjected to four different heat-treating cycles (austenitised samples were cooled in furnace, air, oil and water). It was found that the magnetic parameters extracted from hysteresis loop are strongly dependent on both the cooling rate of the applied heat treatment (which varies the morphology and grain size of ferrite phase), and thickness of the sample. In the proposed method, probabilistic and radial-basis function neural networks have been used to simultaneously determine the microstructure, mechanical properties and thickness with high reliability and accuracy. Experimental results show that using a simple probabilistic neural network, the type of heat-treatment process applied to the sample under test could be perfectly determined. Moreover, thickness estimation of the sample, with a radial basis neural network, has an error less than 0.05 mm, which is actually outstanding. [ABSTRACT FROM AUTHOR] |
