AI for Generator Health Monitoring: Separating Partial Discharge from Electrical Noise

Large electrical generators are critical assets in the power grid, and their lifetime is primarily limited by the condition of the stator insulation. As insulation degrades, localized electrical discharges—known as partial discharges (PD)—occur within the insulation system. These events generate high-frequency electrical pulses that can be measured using capacitive sensors and high-speed data acquisition (100 MHz sampling rate). The challenge is to develop AI or signal processing methods that distinguish true partial discharge activity from electrical interference and improve the quality of generator health diagnostics.

Partial discharge measurements are among the most powerful diagnostic tools for assessing the condition of generator stator insulation. The measured pulses are commonly visualized as Phase-Resolved Partial Discharge (PRPD) diagrams, where pulse activity is related to the electrical phase angle of the generator voltage. The challenge is that the measurements also contain pulses originating from excitation systems, switching electronics, and external electromagnetic disturbances. 

These unwanted signals often dominate the measurements and make the PRPD diagrams difficult to interpret—even for experienced experts. Participants will work directly with raw measurement data sampled at 100 MHz from a real industrial generator. The objective is to develop methods that separate genuine partial discharge pulses from interference using signal processing, machine learning, deep learning, or hybrid approaches. 

The final goal is to generate cleaner PRPD diagrams that improve the reliability of generator insulation diagnostics. There is no predefined solution. We encourage participants to explore both classical signal processing and modern AI techniques to improve the separation of partial discharge and electrical interference. Novel ideas and creative approaches are highly encouraged.

Stator insulation degradation is one of the primary life-limiting failure mechanisms in large electrical generators. Early detection of insulation defects enables utilities to prevent unexpected failures, reduce maintenance costs, and improve system reliability. Many generators currently in operation are more than 40 years old. 

Accurate partial discharge diagnostics make it possible to assess the true condition of ageing insulation systems, allowing operators to safely extend asset lifetime while avoiding unnecessary maintenance or premature replacement. This challenge addresses a real industrial problem where advances in AI can directly contribute to more reliable and sustainable power generation.

• Artificial Intelligence and Machine Learning
• Digital Signal Processing
• Time-Series Analysis
• Pattern Recognition and Feature Engineering
• Python (NumPy, SciPy, scikit-learn, PyTorch or TensorFlow)