System performance tracking is critical to state identification, fault detection and subsequent remaining useful life (RUL) prediction. It is not uncommon that dynamic systems degrade gradually in a non-linear manner, occasionally accompanied with abrupt faults, before and after which the degradation rates vary. In addition, uncertainty coming from sensor measurements poses more challenges to solving this problem. Existing state estimation techniques that are eligible for uncertainty evaluation, such as the extended or unscented Kalman filter and particle filter (PF), can only be applied to non-linear system state tracking excluding abrupt faults, in which case they cannot: 1) detect transient changes of the system states; and 2) track the degradation with varying rate. To address these challenges, a total variation (TV) filter synthesized with an advanced particle filter is developed in this paper. The transient changes of the noisy system states due to abrupt faults are first filtered out by the TV filter. The outputs of the TV filter and the noisy states are then routed to a local search particle filter (LSPF), which is proposed in this paper that can dynamically accommodate its resampling to track the state variation and combat the sample impoverishment problem associated with conventional PFs. The corrected transient changes and estimated gradual degradation of the system states, as well as the uncertainty quantification can be obtained from the LSPF. The proposed approach is presented and demonstrated for automatic fault detection and performance tracking at a component level in heating, ventilation, and air conditioning (HVAC) systems, such as the heat exchanger. To evaluate the developed method, a finite element model for the heat exchanger is established to simulate both normal and faulty cases. The simulation results exhibit the effectiveness of proposed method in fault detection and degradation estimation in heat exchanger.
展开▼
