Comparison of Data-driven Residual Life Prediction Methods for Rolling Bearings in NPPs

With the continuous optimization of maintenance strategies for nuclear power plants, the remaining life prediction of critical components becomes more and more important. In this paper, we take 6312/C3 deep furrow ball bearings, which have the largest installed capacity in a nuclear power plant, as an example, and collect the condition monitoring data of four bearings throughout their lifetime through the platform test. On this basis, we compare three data-driven remaining life prediction methods, including Hidden Markov Model(HMM), Long Short Term Neural Memory networks (LSTMs) and particle filter (PF). The comparison of prediction accuracy and robustness shows that HMM is more consistent with the actual degradation of bearings, and can obtain a more accurate prediction of the remaining life of bearings. Finally, the scalability of the HMM method is verified by the network open data set. The result shows data-driven method can calculate the expected service life of these equipment/spare parts in nuclear power plants based on the condition monitoring data.