1D-SCRN: a novel approach for industrial machinery performance degradation trend prediction

Performance degradation trend (PDT) prediction is a commonly used technique for assessing the risk of potential failure modes of industrial machinery in advance. However, the complex working environment makes the traditional deep learning technologies lack generality, which limits its application in different scenarios. To overcome this issue, this paper proposes a new intelligent PDT estimation approach so-called 1D-separable convolutional recurrent network (1D-SCRN). Firstly, a new health indicator (HI) is obtained by multiple statistical features of vibration signals and probabilistic principal component analysis (PPCA). Furthermore, exponentially weighted moving average (EWMA) is used to reduce the local random fluctuation of HI. Finally, the revised HI is fed into the proposed 1D-SCRN to estimate the PDT of industrial machinery. The experiment results show that this method is capable of predicting the PDT, and its superiority is verified by comparing with other baselines.