Complex augmented representation network for transferable health prognosis of rolling bearing considering dynamic covariate shift

Data-driven methods based on deep learning have achieved satisfactory results in the field of health prognosis. However, the current network modeling still stays in the real number field, which makes most of existing network-based methods criticized as black-box models that lack interpretability. In addition, cross-operating prognostic in scenario where target domain knowledge is not available has not been fully discussed. In this paper, complex augmented representation network considering dynamic covariate shift (CARN-CDCS) is proposed to realize cross-operating health prognosis without target domain knowledge. Specifically, our proposed framework introduces the problem of dynamic covariate shift and generalizes the degradation model under the same condition to multiple sub-models within single class. Reasonable transfer metrics are adopted to measure and minimize the intra-class distance of multiple sub-models constructed from single degradation knowledge, thereby reducing the distribution mismatch of multi-domain knowledge. Two case studies proven that CARN-CDCS can effectively achieve cross-operating health prognosis without target domain knowledge. Compared with some state-of-the-art methods, CARN-CDCS has great advantages in terms of prediction accuracy and computational complexity, which provides a new perspective for health prognosis of mechanical equipment.