"EJUST-PdM-1"

"We introduce EJUST-PdM-1, a novel video-based dataset for predictive maintenance of rotating machinery, designed to support research in non-contact condition monitoring and fault diagnosis. The dataset comprises high-resolution (4K, 60 FPS) video recordings collected from two industrial test rigs\u2014Gunt PT500 and Gunt TM170\u2014under multiple operating conditions, including normal operation, imbalance scenarios with varying severities (10g and 37g), and bearing faults in the outer ring. Recordings are captured across different rotational speeds (250\u20131000 RPM) and diverse camera configurations, varying in distance, height, and viewing angles to enhance real-world variability.EJUST-PdM-1 enables the development and evaluation of machine learning and deep learning models for tasks such as fault classification, vibration signal extraction, and spatio-temporal representation learning. The dataset is particularly suited for video-based techniques, including motion magnification, optical flow analysis, and 3D convolutional neural networks. By providing a comprehensive and structured collection of labeled video data, this dataset addresses the limitations of traditional sensor-based approaches and synthetic datasets, offering a realistic benchmark for advancing non-contact predictive maintenance systems in industrial applications."