RBPF for residual life prediction and application in bearing degradation assessment

Grasping the states of a running device in real-time and assessing its remaining useful life (RUL) and reliability are of great significance to ensure the security of stable operations of entire production system. The particle filtering (PF) algorithm is commonly used to obtain the optimal estimate of the state of nonlinear and non-Gaussian degenerate system. However the computational efficiency of the algorithm will be seriously reduced when the dimension of the system state space increases. To reduce the filtering computational complexity and improve the performance of the filter, Rao-Blackwellization technology, which dealing with the linear and nonlinear parts of the state vector separately, is applied to form the modified PF algorithm. In this paper, the improved algorithm was used in bearing degradation tests, and a comparison was made between RBPF prediction data and real data. The results showed the evidence that RBPF method has better online performance and filtering accuracy, which is an effective way to handle the issue of the computational complexity in assessment.