2026, Vol. 13, No. s1. - go to content...

Permanent address of this page - https://resources.today/en/02faor126.html

Метаданные этой статьи так же доступны на русском языке

DOI: 10.15862/02FAOR126 (https://doi.org/10.15862/02FAOR126)

Full article in PDF format (file size: 971.8 KB)

For citation:

Ivanov I.A., Shakhovskoy A.V., Goncharov A.V., Tarakanova V.V. Development of a soft-sensing system based on neural networks for estimating the supersaturation degree of sugar solutions in a vacuum pan. Russian journal of resources, conservation and recycling. 2026; 13(s1). Available at: https://resources.today/PDF/02FAOR126.pdf (in Russian). DOI: 10.15862/02FAOR126

Development of a soft-sensing system based on neural networks for estimating the supersaturation degree of sugar solutions in a vacuum pan

Ivanov Ivan Andreyevich
K.G. Razumovsky Moscow State University of Technology and Management, Moscow, Russia
E-mail: a.goncharov@mgutm.ru

Shakhovskoy Andrey Vladimirovich
K.G. Razumovsky Moscow State University of Technology and Management, Moscow, Russia
E-mail: a.shakhovskoy@mgutm.ru

Goncharov Andrey Vitalievich
K.G. Razumovsky Moscow State University of Technology and Management, Moscow, Russia
E-mail: a.goncharov@mgutm.ru

Tarakanova Valentina Viktorovna
K.G. Razumovsky Moscow State University of Technology and Management, Moscow, Russia
E-mail: walentt@yandex.ru

Abstract. The modern sugar industry of the Russian Federation is characterized by intensive implementation of digital technologies and intelligent process control systems, with particular relevance being attributed to the development of virtual measurement systems (soft sensors) for monitoring critical parameters of sucrose crystallization that are inaccessible to direct measurement by traditional hardware. The subject of the study is the theoretical and methodological foundations for designing soft-sensing systems based on deep neural networks, intended for estimating the supersaturation degree of sugar solutions in batch and continuous vacuum pans. Machine learning methods are considered, including recurrent neural networks with long short-term memory (LSTM), convolutional neural networks (CNN), and hybrid CNN-LSTM architectures that provide extraction of spatiotemporal characteristics of the technological process. Analysis of the evolution of approaches to monitoring crystallization processes demonstrates a transition from traditional methods of measuring electrical conductivity and refractive index to intelligent predictive analytics systems integrated into the digital twin concept. Key results of the study include systematization of architectural solutions for neural network soft sensors, identification of critical factors determining the accuracy of supersaturation prediction, and substantiation of the methodology for integrating virtual sensors into automatic vacuum pan control loops. A conceptual model of a soft-sensing system has been developed that takes into account the nonlinear dynamics of the crystallization process, the multifactorial dependence of supersaturation on technological parameters, and the requirements of real-time industrial operation. The scientific novelty lies in the integration of modern deep learning approaches with the theoretical foundations of physicochemical processes of sugar production, enabling the achievement of supersaturation estimation accuracy with a relative error of less than 3 %. The practical significance is determined by the possibility of implementing the developed solutions at sugar industry enterprises in Russia to improve the quality of crystalline sugar, optimize energy consumption, and reduce sucrose losses to molasses.

Keywords: soft sensor; virtual sensor; neural networks; deep learning; LSTM; sugar crystallization; supersaturation; vacuum pan; digital twin; Industry 4.0; predictive analytics; production automation

Download article in PDF format