Nikola Tesla Institute
Uroš Radoman
Permanent monitoring of partial discharges: experiences in complex generator diagnostics / Towards data-driven predictive maintenance of hydro turbine-generator units
ABSTRACT: Permanent Monitoring of Partial Discharges: Experiences In Complex Generator Diagnostics
Permanent monitoring of generators during operation is essential for effective facility management and energy production control. The most commonly monitored variables include temperature and vibrations of various generator components. Partial discharge (PD) monitoring is particularly important for assessing the condition of the generator stator’s insulation system. Various stress and aging factors contribute to increased PD activity over the generator’s operational lifespan. If left unchecked, PDs can lead to insulation defects and potential failures. However, they also provide valuable insights into the condition of the insulation system and, indirectly, the overall state of the generator. The advantages of implementing a PD monitoring system are numerous. Firstly, it ensures high-quality data collection, including a large volume of PD data under different operating conditions, synchronized online data from multiple sources, and multi-channel data acquisition. This allows for trend analysis and correlation identification, facilitating the early detection of potential issues due to data coherence. One key benefit of PD monitoring is the automatic alarm notification system, which alerts operators when PD activity reaches or exceeds predefined thresholds. Additionally, remote access allows for configuration adjustments, real-time data review, and trend analysis. The system also enables comprehensive data collection for post-event analysis. By analyzing PD data alongside complex diagnostics results, the risk of failure can be assessed, and maintenance strategies can be optimized. This paper presents practical experiences in diagnosing the condition of generator stator insulation systems, with a significant focus on PD monitoring. The study examines correlations between PD activity and other variables such as vibration, temperature, active and reactive power, among others. Identifiable or confirmable generator conditions include loose coil rods, poor coil head connections, accelerated insulation aging (due to overheating), contamination of insulation surfaces, and other dielectric disturbances.
Keywords: partial discharges, correlations, diagnostics, generator
ABSTRACT: Towards data-driven predictive maintenance of hydro turbine-generator units
In modern power systems, hydropower plants play a crucial role. Considering their irreplaceable functions in providing fast system response, balancing variable energy sources, and integrating an increasing share of renewable energy, their importance is expected to continue growing in the future. The central element of every hydropower plant is the hydro unit, a complex electromechanical system comprising the generator, turbine, and auxiliary equipment, whose unplanned failures can lead to significant economic losses, reduced system reliability, and disruptions in electricity supply. In addition to direct losses, such failures may also cause secondary effects, including increased balancing costs and additional stress on other generation capacities. The development of modern information and communication technologies and advanced analytical methods represents a key prerequisite for the modernization of hydropower plant operation and the transition toward intelligent control and maintenance systems. The integration of diverse data sources, their centralization, and advanced processing enable a deeper understanding of system behavior and support real-time decision-making. Improving maintenance strategies for hydro units and associated equipment requires a transition from traditional approaches, based on periodic inspections or reactive maintenance, to more advanced concepts such as Condition-Based Maintenance (CBM). This approach relies on the analysis of large volumes of data collected from various sources, including SCADA systems, monitoring systems, dedicated sensors, and results of periodic testing, to continuously assess the current condition of the equipment (data-driven approach). This enables the timely identification of deviations from normal operation, early detection of degradation indicators, and data-driven decision-making for planning and executing maintenance activities, thereby reducing the risk of unplanned failures. A specific subset of CBM strategies is Predictive Maintenance (PdM), which utilizes advanced algorithms based on machine learning and artificial intelligence to forecast degradation trends and estimate the remaining useful life of hydro units. These models enable not only anomaly detection but also the prediction of potential failures before they occur, significantly reducing the risk of unplanned outages, increasing system availability, and optimizing overall maintenance costs. Within the poster session, the conceptual design and key functionalities of a dedicated predictive maintenance solution for hydro units (Ægir), developed by the US-based company Elder Research, will be presented. The Ægir platform enables advanced analytics, early diagnostics, and decision support in real-world operational environments. Based on previous experience with the application of the Ægir platform, its implementation is expected to contribute to increased reliability and availability of hydro units, as well as to the overall efficiency improvement of the power system.
Keywords: hydrogenerator, diagnostics, predictive maintenance, AI
Biography of the presenter
Dr Uroš Radoman was born in Belgrade, Serbia. He received his PhD in Electrical Engineering from the University of Belgrade, School of Electrical Engineering. He specializes in thermal modeling, monitoring, and diagnostics of power electrical equipment, with a particular focus on oil-immersed power transformers. Currently a Research Associate at the Center for Electrical Measurements, Electrical Engineering Institute Nikola Tesla, Dr Radoman contributes to projects related to condition monitoring, diagnostics, and modeling of power equipment. From 2013 to 2022, he worked as a Research Assistant at the School of Electrical Engineering, University of Belgrade. His work focused on the development of a thermal-hydraulic network model of liquid-immersed power transformers, including its software implementation and validation using field data. In addition, he was involved in advanced numerical modeling of coupled physical processes, including FEM and CFD analyses. During this period, he contributed to a national technological development project and participated in commercial projects in collaboration with international partners from the power and transformer industry. His research interests include digital twins, thermal processes, and data-driven approaches for condition assessment and predictive maintenance of power equipment.