PV-Fault-DS226: A High-Resolution Multivariate Dataset of Physically Induced Photovoltaic Faults in Tropical Coastal Conditions

The scarcity of high-fidelity, real-world datasets capturing photovoltaic failures in tropical environments remains a significant barrier to the development of robust predictive maintenance models. This article presents a specialized experimental methodology and the resulting multivariate dataset, PV-Fault-DS226, collected from a dual-panel outdoor testbed in Douala, Cameroon (4.0511° N, 9.7679° E). The approach focuses on capturing the non-linear dynamics of solar systems under extreme hot and humid operational stress, providing a rigorous foundation for fault diagnosis and prognosis research. Faults including progressive degradation, open-circuit, short-circuit, and partial shading were physically induced using precision potentiometers, circuit breakers, and manual shading, rather than numerical simulation, to capture authentic non-linear failure signatures. A 16-bit acquisition framework was designed to preserve signal integrity and resolve subtle electrical transients under extreme equatorial climate stress, synchronously logging current, voltage, irradiance, and temperature across both photovoltaic subsystems. After cleaning and multi-rate temporal synchronization, the resulting dataset comprises 226,175 labelled, high-frequency observations spanning five operational states, whose physical separability was confirmed through statistical distribution and three-dimensional clustering analyses. PV-Fault-DS226 thus provides an authentic, tropical, hardware-in-the-loop ground truth for developing and benchmarking photovoltaic fault diagnosis and prognosis models under real degradation trajectories rather than synthetic approximations.