Vibration signal analysis is an effective tool for fault diagnosis in industrial/manufacturing machinery. Gearboxes are a fundamental component of many industrial machines, and their failure can cause significant downtime, production losses, and safety hazards. Analyzing vibration signals makes it possible to detect, classify, and diagnose faults in gearboxes, enabling timely maintenance and preventing catastrophic failures. Vibration signals are sensitive to changes in the operating conditions and internal components of gearboxes, making them a reliable indicator of potential faults. This paper introduces a new vibration signal data set, referred to as VibraFault, which has been the focus of the ICPHM23 data challenge. The dataset contains vibration signals acquired from a test rig consisting of a driving motor, a two-stage planetary gearbox, a two-stage parallel gearbox, and a magnetic brake. The experiments include various operating conditions and focus on common sun gear faults on the planetary gearbox, such as surface wear, chipped, crack, and tooth-missing. For each operating condition, normal and fault vibration signals have been recorded at a sampling frequency of 10 kHz. Vibration signals have been collected in three directions to facilitate more comprehensive research studies on mapping between different types of faults and the system’s vibration response. The dataset has the potential to promote research in fault diagnosis, particularly in the development of advanced solutions based on Machine Learning (ML) and Deep Neural Networks (DNN).