Index similarity assisted particle filter for early failure time prediction with applications to turbofan engines and compressors

The particle filter (PF) has been widely studied in the prognostics’ field due to its ability to deal with nonlinear and non-stationary systems. However, there is no update of the model parameters during the prediction, preventing PF to work in its traditional way to generate accurate long-term predictions. In order to solve this problem, we put forward an improved PF that is based on a novel health index (HI) similarity matching method. This method is employed to search for similar HIs in the training library and construct an optimal “similar HI” for the system under study. Finally, the obtained HI is consistently fed into the PF to deliver precise state-of-health (SoH) estimates. The effectiveness of the proposed PF was validated on the C MAPSS datasets as well as data collected from an operational reciprocating compressor. We observed that the new similarity matching method demonstrated excellent performance in finding suitable HIs for failure time prediction. We also observed that the proposed PF framework had a superior prognostics performance over the standard PF. We obtained an averaged predictive accuracy of 96% (C-MAPSS data) and 92% (compressor data) when only the first 10% of the degradation data were used. This work highlights the promise of combining index similarity, Procrustes analysis and PF for complementing existing prognostic methods.