37. саветовање CIGRE Србија (2025) СИГУРНОСТ, СТАБИЛНОСТ, ПОУЗДАНОСТ И RESILIENCE ЕЛЕКТРОЕНЕРГЕТСКОГ СИСТЕМА МУЛТИСЕКТОРСКО ПОВЕЗИВАЊЕ У ЕНЕРГЕТИЦИ И ПРИВРЕДИ – C4-11
АУТОР(И) / AUTHOR(S): Balša Ćeranić, Slobodan Vukojičić, Marija Novičić, Goran Kvaščev, Leposava Ristić
DOI: 10.46793/CIGRE37.C4.11
САЖЕТАК / ABSTRACT:
In recent years, machine learning has proven to be a solution for many engineering and scientific challenges across a wide range of fields. It has applications in all facets of human activity, and its potential in industrial processes has become evident. Since industry continuously strives to improve process efficiency, cost-effectiveness, and product quality, conventional control strategies are increasingly struggling to meet the growing demands of both manufacturers and consumers. Developing advanced monitoring and control strategies is of vital importance for all branches of industry. Such strategies often rely on knowledge of the parameters of the devices and materials involved in the process. Some parameters are easily measurable, such as speeds, voltages, and currents, while others require complex measurement equipment, use of which may not be justified in certain applications. The aim of this paper is to investigate the capability of neural networks in estimating unmeasurable mechanical parameters in an industrial electric drive system, using available measurements. The approach aims to develop a neural network that, using measurements of the drive machine’s speed, estimates viscosity and stiffness coefficients of the shaft and load inertia, which can vary over time in an industrial process. These mechanical parameters are part of two-mass model used to describe the mechanical resonant system and are essential for predictive control. Various neural network architectures have been developed and tested, including: multilayer perceptron, convolutional neural network, and recurrent neural network. Performance was further improved by employing different data acquisition techniques for training, and the results of the proposed approach are presented in the paper.
КЉУЧНЕ РЕЧИ / KEYWORDS:
Neural networks, Machine learning, Industrial processes, Parameter estimation, Electrical drives
ПРОЈЕКАТ / ACKNOWLEDGEMENT:
ЛИТЕРАТУРА / REFERENCES:
- B. Jeftenić, M. Bebić, S. Štatkić, Višemotorni električni pogoni, Akademska misao, 2011
- S. Vukojičić, L. Ristić, G. Kvaščev, „Comparation Between PI and Model Predictive Control of Two Mass Resonant Mechanical System“, 2022 7th International Conference on Environment Friendly Energies and Applications (EFEA), Bagatelle Moka MU, Mauritius, 2022, pp. 1-6
- I. Goodfellow, Y. Bengio, and A. Courville, Deep Learning. Cambridge, MA: MIT Press, 2016. [Online]. Available: http://www.deeplearningbook.org