A condition monitoring dataset based on electrical signals for a squirrel cage induction generator

The squirrel cage induction generator (SCIG) is still used in variable-speed wind turbines, although many other generator topologies are employed in renewable energy systems. Despite its mechanical robustness, low maintenance requirements, and the reduced complexity of control strategies typically employed in SCIG-based systems, the SCIG remains susceptible to internal faults, making early detection crucial for preventing severe damage and unexpected shutdowns. Monitoring critical components is essential. This dataset provides high-resolution electrical measurements from a SCIG under healthy and faulty conditions to support the development and validation of fault detection techniques. The experimental setup consists of a laboratory test bench with an SCIG designed as a scaled-down version of a real wind turbine generator. Internal faults, including inter-turn and inter-winding short-circuits, were introduced in a controlled manner using a script that commanded a contactor to close the short-circuit for 400 ms. The faults used a single resistance of 2.6 Ω, and the number of affected turns was varied to represent different fault severities. The tests covered multiple steady-state operating points, with rotor speeds of 1200, 1500, and 1800 rpm and mechanical torques of 5.2, 6.4, and 8.0 Nm. Signals were sampled at 20 kHz and recorded during three-second intervals. The dataset contains raw voltage, current, torque, and speed measurements from 24 distinct short-circuit scenarios plus one healthy condition, resulting in 225 .mat files. A Python interface supports visualization and analysis of the time-domain signals. The dataset can support signal processing studies aimed at enhancing short-circuit detection, serves as a resource for generator monitoring in wind turbine research, and assists in the development and testing of machine learning algorithms for time series classification. Although collected independently, this dataset complements another previously published by the same authors. The differences in machine topology and control approach justify the development of this new dataset.