The study is dedicated to enhancing the efficiency of the automated monitoring and control system of technological processes in the ethanol industry, particularly in the production of ethanol and bioethanol, through the use of virtual control tools.
In the first chapter, based on an analysis of scientific literature, the features of technological processes in ethanol and bioethanol production, key automation challenges, and prospects for implementing Industry 4.0 innovations are examined. According to the identified optimization challenges, the research objectives are formulated, including the development of mathematical models, the synthesis of virtual sensors, and the creation of an automated intelligent control system for ethanol and bioethanol production processes.
In the second chapter, a systemic analysis of technological units in ethanol production is conducted, and an applied ontology of mathematical models is developed, within which the use of virtual sensors for monitoring technological variables is studied, taking into account the suitability of existing models for the research objectives. An algorithm based on the Gaussian Mixture Model was proposed for real-time detection of indicators of technological process instability.
In the third chapter, NARX models are used to predict unmeasurable parameters, allowing for the explanation of 99% of the variation in the target variable when forecasting ethanol concentration. The impact of NARX model hyperparameter variations on their performance is investigated. A comparison is made between XGBoost, Decision Tree, Gradient Boosting, RNN, and NARX methods for monitoring system operation and compensating for sensor failures. Specifically:
• modeling the mash flow rate based on other sensor readings achieved a coefficient of determination R² = 0.96;
• forecasting pressure in the concentration column reached R² = 0.98;
• temperature sensor prediction yielded R² = 0.972.
Algorithms have been developed for optimizing the operating modes of the fermentation section using regression tree models to find optimal process settings according to the defined optimization task, enabling an average increase in ethanol yield by 8.4% or a reduction in fermentation time by 12.2%.
In the fourth chapter, an intelligent control system based on Reinforcement Learning methods is implemented. Among various agents, the Soft Actor-Critic (SAC) agent with Entropy weight 1 demonstrated superior performance, ensuring high control accuracy (<0.1%), low dynamic error (21–26% of the target value), and the lowest total error according to the integral squared error criterion.
The architecture of an automated monitoring system for ethanol and bioethanol production was proposed, integrating virtual sensors and implemented using software and hardware platforms such as Unity Pro, Node-RED, and cloud-based tools (InfluxDB, Grafana, Grafana Machine Learning). This architecture enables the incorporation of virtual instruments into the production system and facilitates real-time exchange of production data.
The scientific novelty of the study lies in the development of a new concept for automated monitoring and control of ethanol and bioethanol production processes based on virtual control tools. For the first time, an ontological structure of mathematical models for technological processes has been formulated, incorporating the virtual sensor as a core element. The study advances methods for creating virtual sensors using machine learning to estimate and predict key parameters in dynamic systems, allowing for fault detection, data loss compensation, and improved control reliability.
The practical significance of the research is reflected in the creation of decision support tools and the increased reliability of technological systems in ethanol and bioethanol production. The proposed solutions enable effective compensation for sensor failures, optimization of operating modes, and intelligent control of non-stationary processes. The results have been practically implemented in industrial conditions at Marylivsky Distillery and in educational practices at the Department of Automation and Computer-Integrated Control Systems named after Prof. A.P. Ladanyuk.
