Avaliação da carburização em aços HP por ensaio ultrassônico e processamento de sinais

Abstract

Carburization is a microstructural damage and has been the main factor for reducing the useful life and operational reliability of pyrolysis furnace tubes made of HP steel. HP steel has been employed in the manufacturing of these furnaces for applications in extremely aggressive environments and high temperatures. Traditionally, non-destructive methods based on magnetic measurements have been used to assess the structural integrity of such equipment. In this work, ultrasonic characterization combined with machine learning and signal processing techniques proved to be a promising alternative compared to current magnetic inspection techniques. Thus, the objective of this thesis was to evaluate the influence of frequencies of 2.25 MHz and 5 MHz on the accuracy of ultrasonic processing in machine learning models applied in the classification of specimens of HP steel with 4 different levels of damage by carburization from pyrolysis furnaces. The models used were GNB, KNB, KNN, SD and SVM. The results were validated based on microstructural characterization techniques (SEM, EDS, EBSD) through which the thickness of the carburized layer and the chromium carbide fractions can be determined, using the multi-scenario analysis. The results showed a better adherence of the model in detecting carburization at lower levels, with the GNB and KNB classifiers reaching the F1-score of 100% and 97.44% respectively at 5 MHz The lowest performance was obtained by the KNB-FFT model with 91.94% at the 5 MHz frequency, for the most carburized class. However, its performance for the same class at the 2MHz frequency showed better results.