Digital twin and machine learning for decision support in thermal power plant with combustion engines

Abstract

The reliability and performance of the generating machines in a thermal power plant are crucial to ensure agility and assertiveness in decision-making, maximize economic results, and ensure meeting the electricity sector demands. In this work, a decision support system (DSM) was developed to predict trends and operational deviations in thermal power plants with combustion engines in an automated and reliable way. It is based on digital twin models for thermoelectric generation engines and their subsystems associated with models of machine learning for predictive maintenance, allowing the classification of failures in the generating units of the plant. The models represent the mechanical, thermal, and electrical conditions and parameters of each piece of equipment under normal operating conditions, and the tool generates alerts when deviations from the base model occur. The benefits from event forecasting range from a reduction in operational issues to the company’s strategic objectives due to the reduction in corrective maintenance downtimes, resulting in reduced operation and maintenance costs. Considering the real-time execution character of the models, it is essential for the tool to meet the operation’s decision-making needs; so an on-premises application is necessary. The proposed architecture can be applied to any industrial sector that uses SCADA supervisors and can be adapted, expanded, and evolved to other generation technologies, such as thermal plants that use different fuels and small hydroelectric, wind, and thermonuclear plants. The techniques used in conjunction with the developed architecture can be replicated in other systems and energy sectors, such as distribution and transmission, and can also be applied to industry in general: chemical, petrochemical, oil and gas, and others.