Sistema neuronal-difuso aplicado al reconocimiento y evaluación del daño en aceros al carbono apoyado en estadística descriptiva

Abstract

This paper describes the development of an intelligent integrated system able to estimate the damage to carbon steel; the system is integrated by a fuzzy logic architecture developed from descriptive statistics and an artificial neural network multilayer perceptron applied to the recognition of metallographic patterns. The images obtained were characterized from the analysis of textures images using first, second and third order statistical.

The patterns studied were associated to the microstructure of carbon steel (SA 210 Grade A-1). The proposal allowed estimating the damage present in the material from the determination of the physical states of the material. Steel samples were tested in real conditions of operation, such as high temperatures and humidity, suffering deterioration that it was difficulty detected by standard metallographic methods.

The patterns studied in the microstructure of the material were: laminate perlite, spheronization and graphitization. The microstructure was revealed from images obtained by an inverted metallographic microscope (Olympus - GX71) in the Testing Equipment and Materials Laboratory of the Federal Electricity Commission in Mexico. (LAPEM - CFE). The results showed that the damage estimation and pattern recognition in the material were correctly predicted with the developed system compared to the human expert.