A Predictive Maintenance Strategy for a Single Device Based on Remaining Useful Life Prediction Information: A Case Study on Railway Gyroscope

Predictive maintenance (PdM) strategies constitute an essential component of prognostics and health management (PHM), enhancing device safety and availability while concurrently reducing repair costs. The implementation of PdM typically involves two primary steps: remaining useful life (RUL) prediction and the development of maintenance strategies. However, the limited accuracy of RUL prediction due to the absence of historical degradation information coupled with maintenance decisions normally made independently of the RUL prediction process present significant challenges in developing effective PdM strategies. In this article, a PdM strategy for a single device is presented. First, this article proposes an RUL prediction method based on data augmentation and similarity measures. In data augmentation, Gaussian process regression (GPR) and uncertain simulation are used to generate additional simulated degradation trajectories for RUL prediction. Subsequently, RUL prediction information is obtained using similarity matching. Following that, building on RUL prediction information, we propose a multiobjective optimization PdM strategy. This strategy considers both device availability and maintenance cost to guide maintenance tasks under actual working conditions. Finally, the effectiveness of the proposed PdM strategy is verified by taking a railway gyroscope as an example. The experimental results demonstrate that the proposed RUL prediction method exhibits an average absolute error improvement of at least 10.89% and an average probability value enhancement of at least 7.96% compared with other prediction methods, such as uncertain process simulation. Furthermore, the optimal maintenance times identified by the proposed PdM strategy are all located before the true RUL, indicating that the strategy is effective in preventing device failures.