Remaining useful life prediction integrating working conditions and uncertainty quantification based on multilayer graph neural networks

Predicting the remaining useful life (RUL) is crucial in the technology of predictive health management. However, two major obstacles to RUL prediction are processing multi-dimensional sensor data under varying working conditions and establishing comprehensive degradation trends across different levels. This research advances the precision of RUL forecasts by proposing a new framework that employs a multilayer graph fusion network with uncertainty quantification (MGCAL-UQ) method for RUL prediction. Firstly, a spatiotemporal graph is constructed based on the correlation of different spatial position sensors of mechanical equipment. Secondly, a spatiotemporal feature extraction module has been developed to investigate the possible degradation patterns of samples across various levels and to further extract temporal data. Finally, the RUL and its confidence interval are estimated by parametric method. To verify the effectiveness of the proposed MGCAL-UQ, two different turbofan engine simulation datasets from the Prognostics Center of Excellence at NASA Ams Research Center are used for modeling and testing. The experimental results show that this method outperforms other existing methods.