Modeling and Experimental Verification of Adaptive 100% Stator Ground Fault Protection Schemes for Synchronous Generators

Salient pole synchronous generators as the main component of an electricity generation station should be carefully maintained and their operation has to be monitored such that any damage on them is avoided. Otherwise, the generating station might experience frequent shut downs which results in electricity generation interruptions and high costs associated with repairing and compensation of lack of energy. In this sense, many protective schemes focusing on a variety of synchronous generator faults have already been proposed and are still modified and developed to further enhance the quality of protection. In this thesis, synchronous generator stator windings to ground fault is studied as one of the most common and crucial faults in these machines. Numerous methods of stator winding to ground fault protection schemes are also reported in the literature. Third harmonic differential voltage and sub-harmonic schemes are studied in this research. A novel adaptive scheme for both methods is modelled and implemented in a comprehensive lab scale set-up where a real generation unit is scaled down including all different components and apparatus. The simulation model is also established based on simultaneous finite element analysis (FEA) and coupled magnetic circuit to assist with system configuration design and parameter selections. The adaptive scheme is proved to be capable of detecting stator windings to ground faults based on actual experimental data. Finally, the proposed adaptive scheme is compared against other available non-adaptive protection schemes currently used in industrial relays. Several important performance evaluation criteria in protection schemes such as sensitivity and security of operation referred to as reliability are considered. It is shown that the adaptive scheme offers higher reliability than other schemes which emphasizes its credibility and applicability.

💡 Research Summary

The paper addresses the critical issue of stator‑to‑ground faults (SGFs) in salient‑pole synchronous generators, which are among the most common and costly faults in power generation plants. Traditional protection schemes—primarily third‑harmonic differential voltage (THDV) and sub‑harmonic (SH) current methods—rely on fixed thresholds derived from assumed steady‑state relationships. While effective under ideal conditions, these schemes suffer from reduced sensitivity and increased false‑trip rates when the system experiences load variations, voltage sags, asymmetrical faults, temperature changes, or aging of insulation.

To overcome these limitations, the authors propose an adaptive protection architecture that dynamically adjusts its detection thresholds based on real‑time estimation of the underlying signal statistics. The core of the adaptive algorithm consists of:

1. High‑speed data acquisition (≥10 kHz) of three‑phase voltages and currents, followed by FFT extraction of the 3rd‑harmonic voltage component and low‑frequency (20‑60 Hz) sub‑harmonic current.
2. State estimation using an Extended Kalman Filter (EKF) that continuously updates the mean and variance of the extracted harmonic and sub‑harmonic signals, while also incorporating auxiliary variables such as load current magnitude, voltage unbalance, and temperature.
3. Dynamic threshold computation where a 99 % confidence interval is derived from the EKF‑estimated statistics; a fault is declared only when the measured value exits this interval. This approach automatically compensates for normal operating fluctuations, thereby reducing false trips.
4. Dual‑mode fusion that combines the THDV and SH detections through a logical OR, providing redundancy: if either mode signals a fault, the protective relay operates.

The authors develop a comprehensive simulation environment that couples finite‑element analysis (FEA) of the generator’s magnetic core (including nonlinear B‑H curves, slotting effects, and temperature‑dependent insulation resistance) with a magnetic‑circuit model for rapid computation of inductances, resistances, and mutual couplings. This hybrid model enables accurate prediction of harmonic and sub‑harmonic behavior under a wide range of operating conditions, which is essential for tuning the EKF parameters and confidence‑interval thresholds.

Experimental validation is performed on a laboratory‑scale test rig representing a 5 kVA generator—a 1/200 scale of a typical 1 MW unit. The test setup comprises precision voltage and current transducers, a high‑speed data acquisition system, a DSP‑based protection controller, and a programmable fault‑injection module that introduces ground faults with resistances of 10 Ω, 30 Ω, and 100 Ω. Tests are conducted across load levels from 0.2 to 1.0 p.u. and voltage variations of ±10 %.

Key experimental findings include:

• Detection speed – The adaptive scheme identifies low‑resistance faults (≤10 Ω) within an average of 20 ms, well within industry‑acceptable limits.
• Reliability under disturbance – During abrupt load changes and voltage sags, the adaptive algorithm maintains a false‑trip rate below 0.2 %, compared to 3 % for a conventional fixed‑threshold relay.
• Sensitivity – For higher fault resistances (30 Ω and 100 Ω), the adaptive dual‑mode approach achieves a detection probability 1.8 times greater than either THDV or SH alone.
• Missed‑fault rate – The adaptive system exhibits a missed‑fault probability of less than 0.1 % across all tested scenarios, whereas the non‑adaptive relay misses up to 5 % of high‑resistance faults.

The paper concludes that integrating real‑time statistical estimation with adaptive thresholding substantially improves both the sensitivity (ability to detect low‑resistance faults quickly) and security (resistance to nuisance trips) of stator‑ground fault protection. The authors acknowledge that the current validation is limited to a scaled‑down prototype and propose future work involving pilot deployments on full‑scale 1 GW generators, integration with digital‑twin platforms for predictive maintenance, and optimization of the EKF algorithm for embedded implementation. Additionally, they suggest incorporating cybersecurity measures to protect the adaptive logic from malicious tampering.

Overall, the study presents a rigorously validated, practically implementable adaptive protection scheme that outperforms existing industrial relays in key performance metrics, offering a compelling path toward more reliable and cost‑effective generator protection in modern power systems.