Data supporting the publication: A hybrid Convolutional Autoencoder training algorithm for unsupervised bearing health indicator construction

This dataset contains the frequency-domain representations of vibration signals from two well-known bearing run-to-failure experimental datasets: the IEEE PHM 2012 Prognostic Challenge dataset and the XJTU-SY dataset. The PHM 2012 dataset consists of multi-axis vibration data from bearings tested under varying speeds and loads on the PRONOSTIA platform to monitor gradual degradation. The XJTU-SY dataset features similar run-to-failure vibration profiles that capture accelerated degradation across different operational states. This quantitative research aims to facilitate the development and validation of advanced data-driven predictive maintenance algorithms, specifically targeting the unsupervised construction of bearing health indicators and the prediction of remaining useful life (RUL). Original data collection was performed continuously via accelerometers mounted on physical testbeds, recording raw temporal vibration signals from a healthy state until complete physical failure. The data in this repository consists of preprocessed frequency-domain features of the original vibration signals, prepared for training and evaluating machine learning models.