A comparison of filter-based approaches for model-based prognostics

Abstract

Model-based prognostics approaches use domain knowledge about a system and its failure modes through the use of physics-based models. Model-based prognosis is generally divided into two sequential problems: a joint state-parameter estimation problem, in which, using the model, the health of a system or component is determined based on the observations; and a prediction problem, in which, using the model, the stateparameter distribution is simulated forward in time to compute end of life and remaining useful life. The first problem is typically solved through the use of a state observer, or filter. The choice of filter depends on the assumptions that may be made about the system, and on the desired algorithm performance. In this paper, we review three separate filters for the solution to the first problem: the Daum filter, an exact nonlinear filter; the unscented Kalman filter, which approximates nonlinearities through the use of a deterministic sampling method known as the unscented transform; and the particle filter, which approximates the state distribution using a finite set of discrete, weighted samples, called particles. Using a centrifugal pump as a case study, we conduct a number of simulation-based experiments investigating the performance of the different algorithms as applied to prognostics.

A Bayesian Approach to Model Calibration and Parameter Estimation in Potential Drop Measuring

“…The modelling of dynamic processes with imperfect or unknown information gives rise to numerous sources of uncertainty and variability that have to be accounted for [13][14][15][16]. These sources contribute to the system's state uncertainty, the modelling uncertainty and -in terms of prognosis -the future uncertainty.…”

“…The unscented Kalman filter (UKF) [26][27][28] as well as the unscented Rauch-Tung-Striebel smoother (URTSS) [19] address the filtering and smoothing problem by utilising the unscented transform as a way of deterministic sampling [14] in a Gaussian framework. Considering two random variables ξ and υ with a nonlinear model function υ = g(ξ), an approximation to the joint distribution can be obtained by estimating the mean µ υ and covariance Σ υυ by means of a minimal set of weighted samples.…”

Distributed Prognostics Based on Structural Model Decomposition

“…6). Therefore, we need only four submodels on which to base our local estimators, M In this paper, we use an unscented Kalman filter (UKF) [25], [26] with a variance control algorithm [27] for the estimation problems. The UKF assumes the general nonlinear form of the state and output equations described in Section II, but restricted to additive Gaussian noise.…”

“…The variance values assigned to the parameters determine both the rate of parameter estimation convergence and the estimation performance once convergence is achieved. Therefore, several heuristic approaches have been developed to tune this value online to optimize performance, e.g., [2], [27]- [29]. We adopt the approach presented in [2], [27], in which the algorithm modifies the variance in order to control the variance of the parameter estimate to a user-specified range.…”

Current/pressure transducer application of model-based prognostics using steady state conditions

“…This is important for prognostics (predicting failure), scheduling maintenance, and triggering automated mitigation actions. This is often done using methods such as a Kalman Filter or Particle Filter [10,11]. The resulting outlet pressure for each fault mode given a high and low input current can be seen in Figure 4.…”