Non-destructive ultrasonic testing and machine learning-assisted early detection of carburizing damage in HP steel pyrolysis furnace tubes

•The early stages of carburization damage in HP steel furnace tubes are evaluated by non-destructive ultrasonic (US) tests and machine learn models.•The US signals are used as a non-destructive diagnostic technique applicable to monitoring structural integrity in pyrolysis furnaces.•Ultrasonic signa...

Full description

Saved in:

Bibliographic Details
Published in:	Measurement : journal of the International Measurement Confederation Vol. 218; p. 113221
Main Authors:	Paz da Silva, Francirley, Matos, Robert S., da Fonseca Filho, Henrique D., da Silva, Mario. R.P., Ţălu, Ştefan, dos Santos, Ygor T.B., da Silva, Ivan C., Martins, Carlos O.D.
Format:	Journal Article
Language:	English
Published:	Elsevier Ltd 15-08-2023
Subjects:	Carburizing HP steel pyrolysis furnaces Machine learning Ultrasound tests Carburizing HP steel pyrolysis furnaces Ultrasound tests Machine learning
Online Access:	Get full text
Tags:	Add Tag No Tags, Be the first to tag this record!

Description
Summary:	•The early stages of carburization damage in HP steel furnace tubes are evaluated by non-destructive ultrasonic (US) tests and machine learn models.•The US signals are used as a non-destructive diagnostic technique applicable to monitoring structural integrity in pyrolysis furnaces.•Ultrasonic signals are converted in fast Fourier transform signals and used as data sources to feed machine learning models that identify early stages of damage in HP steel.•The machine learn-based results are compared with the volumetric fractions of chromium carbides determined by a traditional microstructural analysis. During the operation of HP steel furnace tubes, structural deterioration occurs due to carburization mechanisms. Herein, machine learning models were employed to detect carburization damage in furnace tubes using ultrasonic signals. The microstructural and elemental analysis revealed phases like austenite, chromium carbide, and niobium carbide. Our volumetric fraction analysis showed that the harmful chromium carbide phase increased toward the tube wall thickness. Three machine learning models, namely Gaussian Naive Bayes (GNB), Kernel Naive Bayes (KNB), and Subspace Discriminant (SD), were used to analyze the ultrasound signals. The GNB model demonstrated the highest accuracy rate (99.2%) and high sensitivity for the dataset with 26 features and a K-fold cross-validation with K value = 5, arising as the most effective classifier for detecting carburization damage in HP steel. Our results underscore the efficacy of the combined use of ultrasonic testing and machine learning for detecting carburization in HP steel furnace tubes.
ISSN:	0263-2241 1873-412X
DOI:	10.1016/j.measurement.2023.113221