Chi-squared smoothed adaptive particle-filtering based prognosis

详细信息    查看全文

• 作者：Christopher P. Ley^a ; ^b ; ^{christopher.ley@csiro.au" class="auth_mail" title="E-mail the corresponding author} ; Marcos E. Orchard^a ; ^c ; ^{morchard@ing.uchile.cl" class="auth_mail" title="E-mail the corresponding author}
• 关键词：Particle filtering ; Monte-Carlo filter ; Failure prognosis ; Chi-square smoothing ; Artificial evolution ; Oil degradation
• 刊名：Mechanical Systems and Signal Processing
• 出版年：2017
• 出版时间：1 January 2017
• 年：2017
• 卷：82
• 期：Complete
• 页码：148-165
• 全文大小：1973 K

文摘

This paper presents a novel form of selecting the likelihood function of the standard sequential importance sampling/re-sampling particle filter (SIR-PF) with a combination of sliding window smoothing and chi-square statistic weighting, so as to: (a) increase the rate of convergence of a flexible state model with artificial evolution for online parameter learning (b) improve the performance of a particle-filter based prognosis algorithm. This is applied and tested with real data from oil total base number (TBN) measurements from three haul trucks. The oil data has high measurement uncertainty and an unknown phenomenological state model. Performance of the proposed algorithm is benchmarked against the standard form of SIR-PF estimation which utilises the Normal (Gaussian) likelihood function. Both implementations utilise the same particle filter based prognosis algorithm so as to provide a common comparison. A sensitivity analysis is also performed to further explore the effects of the combination of sliding window smoothing and chi-square statistic weighting to the SIR-PF.