Development of an analysis tool utilizing robotic image acquisition for early detection of thermal anomalies in HVDC converter equipment

Voltage Direct Current (HVDC) technology a critical component of modern transmission systems. These assets operate in controlled environments that are inaccessible during operation, making conventional thermal inspections impractical. At the same time, converter valves and

DC equipment account for a significant share of HVDC outages, highlighting the need for early anomaly detection to improve reliability and minimize unplanned downtime.

This paper introduces “SARTA” (System for Automated Recognition of Thermal Anomalies), an automated decision-support tool developed by Elia Group and designed to detect thermal anomalies in HVDC-converter equipment. The system leverages robotic image acquisition combined with cloud-based processing to transform large volumes of inspection data into actionable insights. An autonomous ground-based robot performs daily inspection missions, capturing thermal and optical images at predefined points of interest. These datasets, enriched with operational load and environmental metadata, are processed through a structured workflow that includes plausibility checks, comparison image selection, image alignment using computer vision techniques, and temperature delta calculations. Detected anomalies are clustered within predefined regions of interest while excluding irrelevant areas, and results are communicated through automated reports to operational teams.

Tests were conducted over several months using nearly 23,000 thermal images from operational

HVDC-converter stations. The analysis confirmed the robustness of the approach: 83% of images showed no anomalies, 15% lacked suitable comparison data, and only 0.04% were flagged as potential hotspots. Manual review revealed these cases were caused by environmental factors such as switched light bulbs and reflections rather than equipment faults.

Although no real thermal defects were detected during testing, “SARTA” demonstrated its capability as a scalable, explainable early warning system that reduces manual effort and supports proactive maintenance strategies.

Future development will focus on integrating AI-based object detection, dynamic parameter tuning, and trend analysis to enhance predictive capabilities. By compressing complex datasets into clear, actionable information, “SARTA” provides a flexible foundation for digitalized asset management in high-voltage networks, helping to improve operational reliability and minimize unplanned outages.