Switching local search particle filtering for heat exchanger degradation prognosis

Abstract

Sequential Monte Carlo (SMC) or particle filtering (PF) has demonstrated effectiveness in non-linear and non-Gaussian system estimation, due to its unique approach for posterior probability density function estimation. However, classical PF techniques suffer from particle degeneracy and sample impoverishment. This paper proposes a new scheme for joint state and parameter estimation, based on the sequential importance resampling (SIR) particle filter. First, a local search strategy is proposed as a resampling strategy to overcome particle impoverishment. Second, a switching multiple modes filter is adopted to handle sudden changes of parameters in the state evolution model due to faults, which cannot be processed by conventional PF that assumes gradual parameter variations over a long period. The proposed estimation method is applied to degradation and remaining useful life (RUL) prediction of dynamic systems, such as heat exchanger in the heating, ventilation and air conditioning (HVAC) systems. Both natural and transient degradations are evaluated, while parameters dominating the degradation models are assumed to change before and after transient decay. The developed method is evaluated using simulation data, and results demonstrate the effectiveness of proposed method in state estimation and degradation prediction in heat exchanger.