This paper analyzes the partial discharge (PD) magnitude and phase- resolved PD (PRPD) patterns in 13.8 kV generator stator end-windings to evaluate the impact of surface contamination. The ingress of ambient air for cooling inevitably introduces atmospheric contaminants, leading to the fouling of winding surfaces. This accumulation of foreign matter, such as dust, increases the dissipation factor (tanδ) and PD magnitude, eventually triggering arcing and catastrophic insulation failure. Diagnostic insulation tests were conducted during a scheduled outage on an air-cooled generator (144 MVA, 13.8 kV) after 11 years of operation. While the AC current and dissipation factor remained within acceptable limits, the PD magnitude indicated poor and cautionary conditions. Specifically, at the rated phase voltage (8 kV), the PD magnitude in Phase A reached 163,080 pC, exhibiting a complex pattern of surface and arc discharges. The amplification ratio (AR) was redefined to compare the PD magnitude at 8 kV with that at 1.25 times the rated phase voltage (10 kV). A critical finding revealed a non-linear increase in the AR of the PD magnitude by 12.5 to 28 times, signifying severe insulation degradation induced by phase-to-phase arc discharge within the end-windings. Despite the significantly higher PD magnitude in Phase A during testing, a subsequent insulation failure occurred in Phase C after 18 months of operation. This suggests that the specific defect location and interfacial conditions are as critical as the absolute PD magnitude for predicting dielectric breakdown. The results highlight the necessity of analyzing PD AR under overvoltage conditions for reliable insulation health monitoring.
Voltage Stability Margin (VSM) analyses are predominantly done offline because of the size and complexity of the analysis itself. Further, operator actions impact VSMs which may cause system insecurity. Thus, improving system observation capabilities beyond state estimation and security constrained generation dispatch is beneficial. To bridge this gap, this work leverages popular Artificial Intelligence (AI) models namely: Gradient boosting regression trees, neural networks, and state-of-the-art Transformer networks to quickly estimate VSMs. In this paper, training data is synthesized using Monte Carlo sampling of the IEEE 39-bus and 118-bus test systems. Theoretical feasibility is enforced using power flow analysis and static security checks. Feature selection for training is done by filtering inputs according to Pearson Correlation, Random Forest Importance, and Mutual Information rankings. The best-performing individual configuration, a reduced-feature Transformer variant, achieves an RMSE as low as 63.7 MW (4.7
PMSM drives are central to electric mobility, but electromagnetic NVH remains a persistent barrier. Purely data-driven AI models generalize poorly and lack certification guarantees. This paper proposes a trustworthy physics-informed AI (PI-AI) framework embedding governing electromagnetic, structural, and acoustic equations into a neural surrogate. The framework includes Bayesian uncertainty quantification, SHAP-based explainability, and OOD robustness monitoring. A composite loss combines data fidelity, PDE residuals, and boundary constraints. Simulation results on a representative interior PMSM show 68–74
This paper proposes a surrogate-model-based design framework for interior permanent magnet synchronous motors (IPMSMs) considering irreversible demagnetization (ID) characteristics under high-temperature operating conditions. To reduce the computational burden associated with repetitive demagnetization analysis, a demagnetization evaluation index was derived using no-load and load finite element analysis (FEA) data obtained during surrogate model construction. The permanent magnet region was divided into multiple segments, and the influence of local demagnetization on motor performance was quantified through sensitivity analysis based on back electromotive force (BEMF) variation. Using the estimated operating points and segment sensitivities, an ID coefficient representing the overall demagnetization vulnerability was formulated and incorporated into the multi-objective optimization process together with conventional electromagnetic performance indices. A random forest-based surrogate model and genetic algorithm were employed to derive the optimal design. The optimization results demonstrated that the proposed method effectively improved demagnetization robustness while simultaneously reducing BEMF total harmonic distortion and torque ripple without requiring additional demagnetization-specific analyses.
High-voltage motors are critical assets in industrial facilities, and unexpected insulation breakdown failures—particularly in essential sectors, such as power generation and steel manufacturing—can result in significant economic losses. Consequently, effective management of insulation deterioration conditions in high-voltage motors has become increasingly important. In Korea, however, assessment criteria for insulation diagnostic tests differ significantly among manufacturers, asset owners, and testing organizations, resulting in inconsistency and confusion across the industry. Given that insulation characteristics vary according to motor design, manufacturing processes, and insulation materials, the application of a single absolute criterion to all high-voltage motors is technically unfeasible. This study undertook a comprehensive review of international standards, Japanese guidelines, and domestic evaluation criteria and analyzed the statistical distribution of insulation diagnostic test results from approximately 3,300 high-voltage motors in operation at power plants. Furthermore, accelerated thermal aging tests were performed to monitor variations in insulation diagnostic parameters, such as partial discharge characteristics and dielectric dissipation factor (tan δ), in alignment with actual insulation deterioration processes. Drawing on the review of assessment criteria, analysis of field diagnostic data distributions, and findings from accelerated aging experiments, trend-based management criteria were developed to more accurately reflect the progression of insulation deterioration in high-voltage motors. These criteria were established through collaboration with experts from domestic manufacturers, utilities, research institutes, and testing organizations. The proposed approach offers practical and realistic guidance for insulation condition management in high-voltage motors, addressing the current challenges posed by the coexistence of multiple, and often impractical, diagnostic criteria in the domestic industrial landscape.
