Recently, interest in PHM (Prognostics and Health Management), a technology that predicts and preserves failures of assets such as facilities and machines in advance, is increasing. Model-based and data-based approaches exist as PHM approaches. Among ...
Recently, interest in PHM (Prognostics and Health Management), a technology that predicts and preserves failures of assets such as facilities and machines in advance, is increasing. Model-based and data-based approaches exist as PHM approaches. Among them, as the design and structure of assets become more complex, data-based approaches that can effectively predict the life of assets without defining mathematical and physical degradation models of assets are in the spotlight. In particular, among data-based approaches, deep learning-based approaches, which are known to be end-to-end learning and have excellent performance, are the main ones. PHM often utilizes vibrations that best represent the state of the machine. However, extracting significant features from raw-state vibrations is still challenging, especially when learning data from different environmental conditions together does not significantly increase the performance of the model. In this work, we present a continuous predictive coding CPC-based methodology that effectively extracts features for predicting RUL (Residual Useful Life) of machines from vibration data under different environmental conditions. The methodology was verified using bearing vibration data from the PRONOSTIA platform