Hybrid Electrical Fault Detection Dataset (Real + Synthetic)

This dataset is titled Hybrid Electrical Fault Detection Dataset (Real and Synthetic) and contains 22,000 labeled instances of electrical faults in three-phase power systems. It was created by combining two real-world datasets obtained from Kaggle: the Electrical Fault Detection dataset, which contains measurements of voltage and current with labeled fault types, and the Electric Fault-Line Detection dataset, which provides similar electrical measurements with different fault labels. By integrating these two datasets and generating additional synthetic samples, we produced a larger, more balanced, and standardized dataset suitable for machine learning and research purposes. Each row contains measurements of phase currents (current_A, current_B, current_C) and phase voltages (voltage_A, voltage_B, voltage_C), a fault label describing the type of fault, a timestamp, and the source of the data, indicating whether the row is from one of the original datasets or was synthetically generated. The fault labels include healthy conditions and specific faults such as AG fault, BG fault, CG fault, AB fault, BC fault, and ABC fault.

The synthetic samples were generated using a statistical modeling and signal simulation approach. First, medians and variability measures were calculated from the real datasets to define realistic ranges for voltages and currents. Sinusoidal signals were then simulated with randomized amplitudes, phases, and small added noise to resemble natural fluctuations in electrical systems. Controlled perturbations were applied depending on the fault type. For example, in a single-phase-to-ground fault, the corresponding phase voltage was reduced and the current increased. For double-phase faults, voltages of the affected phases were reduced while currents were amplified. Triple-phase faults were simulated by reducing all phase voltages while adjusting currents to maintain realistic behavior. This approach ensures that synthetic measurements are representative of real-world faults, providing diversity and coverage across all fault types while addressing class imbalance.