Реферат: Gas porosimetry became widespread nowadays as never due to worldwide availability of numerous instruments for adsorption-desorption isotherms measurements from different manufacturers. More and more researchers with insufficient background in adsorption science apply this method in various fields. There are two ways to meet this wave: the scientific popularization of the essential features and complications of this method through webinars, teaching programs, etc., and, the second, employment of prior analysis of raw experimental data into modern software prompting inexperienced researchers the most reliable ways for the proper analysis. Both ways are important. In this contribution we discuss the possibility of automatic recognition of the adsorption-desorption hysteresis loop type by means of the machine learning based on numerous data accumulated in the Boreskov Institute of Catalysis for dozens of years. The proper attribution of the hysteresis loop type is important for the successive choice of the method for the pore size distribution, prompting which branch reflects more reliable information on porous structure, presence or absence of evaporation induced cavitation, pore blocking effects, and their deconvolution. There are six known canonical types of the hysteresis loops. But, our practice shows that intermediate shapes are also important. E.g., frequently we observed the loops very similar to the case of incomplete pore filling, as it is observed when the hysteresis loop scanning technique is applied (Figure 1). In this case the traditionally recommended desorption branch cannot be used for the pore size distribution evaluation, since only the unknown part of pores is involved into desorption. This and some other illustrative examples are discussed in the presentation, showing the significance of the problem. According to our cumulative knowledge inspection of only one mathematical parameter, such as the slope, inflection, widening, narrowing towards both axis of adsorption-desorption curves is not sufficient for the robust attribution of the loop. The particular values of all these parameters show a fairly wide spread, and only machine learning approach allow establishing the reliable correlations that allow the proper attribution of the loop type with the reliable probability. These aspects are discussed in the presentation. Acknowledgements This work was supported by the Ministry of Science and Higher Education of the Russian Federation within the governmental order for Boreskov Institute of Catalysis (project АААА-А21-121011390054-1).

Библиографическая ссылка: Mel’gunov M.S.
Machine Learning Exercise for the Adsorption-Desorption Hysteresis Loop Recognition
В сборнике Физико-химические проблемы адсорбции, структуры и химии поверхности нанопористых материалов: всероссийская конференция с международным участием (к 120-летию со дня рождения М.М. Дубинина), 18 - 22 октября, 2021, Москва, Россия : сборник тезисов докладов. – ИФХЭ РАН., 2021. – C.35-36. – ISBN 9785446534074. РИНЦ

Переводная: Mel’gunov M.S.
Machine Learning Exercise for the Adsorption-Desorption Hysteresis Loop Recognition
В сборнике All-Russian conference with international participation devoted to the 120th birth anniversary of Academician M.M. Dubinin PHysicochemical problems of adsorption, structure, and surface chemistry of nanoporous materials Book of Abstracts. – IPCE RAS., 2021. – C.34-35. РИНЦ

Идентификаторы БД:

РИНЦ:

47824666

Цитирование в БД: Пока нет цитирований