Remaining useful life prediction considering multiple uncertainty information via Bayesian BiGRU-based method

The prediction of remaining useful life (RUL) of manufacturing equipment is a critical task in prognostics and health management (PHM). There is a large amount of uncertain information in the prediction, which can lead to unreliability in the RUL prediction results. Existing research lacks a comprehensive analysis of these uncertainty sources. To this end, a novel prediction method considering four types of uncertainty information via Bayesian Bidirectional Gated Recurrent Unit (BF-BiGRU) model is proposed in this paper. Four types of multi-source uncertainties, including aleatoric uncertainty, model uncertainty, data missing uncertainty, and semantic uncertainty, are comprehensively considered and quantified using a Bayesian neural networks (BNN) framework. The advantages of BiGRU in modeling complex time series data are fully leveraged to reduce overall errors and improve the accuracy of prediction results. Ultimately, the RUL distribution is determined by combining failure thresholds and the degradation trajectories. A case study on grease dataset and an electric spindle degradation dataset verify that the proposed method exhibits excellent predictive performance, effectively enhancing the accuracy and reliability of the predicted RUL.