Accurate remaining useful life prediction is crucial for the maintenance and operation of complex systems. Despite the significant progress made by deep learning methods in this field, the potential relationships between arbitrary measured parameters in topological space are ignored. To overcome this issue, GraphSAGE-LSTM assisted by the improved osprey optimization algorithm (IOOA-GraphSAGE-LSTM) is designed in this study. In the IOOA-GraphSAGE-LSTM, the Spearman correlation coefficient is used to describe the relationships of measured parameters in topological space within the constructed graph data. Then, GraphSAGE and LSTM are used to obtain the potential relationships from Euclidean space and topological space. To improve the prediction precision of the GraphSAGE-LSTM, an IOOA is developed inspired by tent mapping and the energy-consuming process during the hunting process so that the inhomogeneous initialization and poor post-development can be mitigated. The competitive results of different aero-engines in the CMAPSS dataset indicate the superiority of the developed IOOA-GraphSAGE-LSTM, i.e., R2 higher than 0.97 while RMSE is within 10. This demonstrates the engineering applicability of the proposed algorithm.
Wang et al. (Sun,) studied this question.