Insulators are critical components in the transmission of electrical energy and their correct operation is essential to guarantee the safety and continuity of the electrical supply. However, their constant exposure to environmental conditions such as pollution can affect their performance and lead to costly power outages. Proper monitoring and maintenance of this equipment are essential to avoid unexpected failures. Although significant progress has been made in the implementation of new monitoring methodologies and instruments, a solution that enables accurate and reliable online monitoring remains a challenge. This work presents a novel insulator flashover alert system that uses an acoustic sensor and an ultra-high frequency antenna to detect partial discharge activity and generate alerts in three different states. The system uses hybrid deep learning models that combine a Convolutional Neural Network and a Long Short-Term Memory network that process the information provided by the sensors and generate local alerts to produce the general decision. This system can indicate whether the monitored insulator is in a normal state, attention state or alarm state, which could help specialists take preventive measures and avoid sudden insulator failures.
30 January 2026 in IEEE Transactions on Instrumentation and Measurement
Pollution on the surface of an insulator, combined with conditions such as fog, dew, or moisture, directly influences partial discharge (PD) activity, a key indicator of an insulator’s aging process. Using an artificial pollution chamber, three classes of insulators were subjected to saline dew across multiple test runs, during which radiofrequency emissions were recorded by an antenna. The contamination process was monitored and characterized using three indicators derived from the Cumulative Energy per Interval, the Cumulative Average Time Between Signals, and the Cumulative Distributed Energy. By employing segmented regression analysis on these variables, it was possible to detect and quantify changes in PD activity as measured in the ultra-high-frequency (UHF) band by a nearby antenna. Compared with galvanic, leakage-current based approaches and artificial intelligence (AI)-based models, the proposed scheme has lower computational cost and provides operator tunable alert states for training-free, non-contact, trend-based early warning. In every test run, the developed algorithm identified at least one trend change by 80% of the test duration. Analyzing each trend change with the three indicators enabled graded alert states, demonstrating flexibility for assessing flashover (FO) risk. In all experiments conducted on the different types of insulators, FO consistently occurred during the highest alarm state.