Authors: E.V. Astrashab, A.N. Grigorchik, V.A. Kukareko, M.А. Belotserkovsky, A.V. Sosnovskiy

Title of the article: Structure, phase composition and tribotechnical properties of annealed (Ni-Cr)-Al composition coating

Year: 2024, Issue: 3, Pages: 99-108

Branch of knowledge: 2.6.1 Metal science and heat treatment of metals and alloys

Index UDK: 621.793

DOI: 10.26730/1999-4125-2024-3-99-108

Abstract: Ensuring the reliability and safety of steel structures is a key challenge for engineers and maintenance professionals. In this context, non-destructive testing (NDT) becomes an integral part of the process of monitoring the condition of metal structures. At the same time, difficulties arise when choosing a criterion for assessing performance and interpreting the values obtained as a result of calculations, which affects the quality of decision-making on performance and assessing the possibility and method of extending the resource. The paper discusses the process of developing an API (application program interface) for machine learning in the Python programming language, designed to solve the problem of predicting the current state of metal structures based on non-destructive testing (NDT) data. The developed API allowed users to create more complex and informative features based on raw data, which can improve the accuracy and ability of machine learning models to predict the condition of metal structures. Such updating and improvement of models becomes important in the context of constant evolution of the state of structures and NDT data. As a result of the research conducted to assess the influence of the characteristics of the applied ML algorithms on the quality and speed of predicting the state of metal structures made of structural steel and 0.12С-1Cr-1Мo-1Va steel according to non-destructive testing (NDT) parameters, a classification of topologies of ML models was made based on the quality of prediction and decision making. The scientific foundations of the algorithmic support of the decision-making preparation system for assessing the performance and extending the service life of technical devices of hazardous production facilities have been developed based on the use of artificial neural networks; they represent a sequence of operations common to different classes and grades of steel, as well as for equipment that is operated in different conditions (except drums).

Key words: non-destructive testing microdamage neural networks residual life metal structures machine learning

Receiving date: 16.04.2024

Approval date: 29.05.2024

Publication date: 28.06.2024

CITE DOWNLOAD