Clasificación del tratamiento térmico de aceros con ensayos no destructivos por corrientes inducidas mediante redes neuronales

Abstract

Eddy current-nondestructive techniques are

increasingly present in industry because of the

growing quality control demand. In addition

to the classical crack detection, eddy currents

permit physic and metallurgic properties

detection in steels. The impedance values of

one eddy-current-generating coil permit to

differentiate steel pieces with different heat

treatments. These impedance values can be

processed with Artificial Neural Networks

(ANNs) to implement automatic and efficient

classifiers.

In this article two ANN classifiers that

processed monofrequency impedances and one

ANN classifiers that processed multifrequency

impedances are compared. The impedances

were extracted from two steel samples sets with

different heat treatments. The predominant

microstructure in the first set was martensite

and bainite and in the second one was perlite.

As experimental results, the monofrequency

classification accuracy rate was 90% while the

multifrequency classification reached 99.9%,

the theoretical computing workload of the best

multifrequency classifier was between 33%

and 50% lower than the best monofrequency

classifier, and the multifrequency classification

execution time was 22% smaller than the

measured time using the monofrequency

methods.