An Improved Similarity Matching Bidirectional Gated Recurrent Unit Autoencoder Scheme for Remaining Useful Life Prediction

Remaining useful life (RUL) prediction is an important task of prognostic and health management (PHM), which can ensure the safe and reliable operation of the degradation system. In the previous similarity matching prediction methods, the similarity measure is calculated based on the Euclidean distance at the same time point. This single calculation method ignores the shape similarity of the health indicator curve. In addition, it is difficult to solve the similarity matching problem for testing instances with a short running time and rapid degradation trend. To address the above drawbacks, this paper proposes a similarity matching scheme based on the dynamic time warping (DTW) algorithm for mechanical RUL prediction, which is a companion article to our previous work on prediction schemes using BiGRU network, attention mechanism and skip connection (BiGRU-AS). The DTW algorithm measures the similarity between relative time points to consider the shape similarity of health indicator curves. In the end, the proposed method is validated on two datasets: the milling datasets and the railway wagon wheel wear dataset. The results show that the proposed scheme has good generalizability and achieves state-of-the-art prediction performance.