Sequential state estimation of nonlinear/non-Gaussian systems with stochastic input for turbine degradation estimation. (May 2016)

• Record Type: Journal Article
• Authors: Hanachi, Houman; Liu, Jie; Banerjee, Avisekh; Chen, Ying
• Is Part Of: Mechanical systems and signal processing. Volume 72/73(2016)
• Page Start: 32
• Page End: 45
• Publication Date: 2016-05
• Subjects: Dynamical system -- System identification -- Stochastic input -- State estimation -- Multivariate particle filter -- Turbine degradation; Structural dynamics -- Periodicals; Vibration -- Periodicals; Constructions -- Dynamique -- Périodiques; Vibration -- Périodiques; Structural dynamics; Vibration; Periodicals; 621
• DOI: 10.1016/j.ymssp.2015.10.022
• Languages: English
• ISSNs: 0888-3270

Abstract

Abstract: Health state estimation of inaccessible components in complex systems necessitates effective state estimation techniques using the observable variables of the system. The task becomes much complicated when the system is nonlinear/non-Gaussian and it receives stochastic input. In this work, a novel sequential state estimation framework is developed based on particle filtering (PF) scheme for state estimation of general class of nonlinear dynamical systems with stochastic input. Performance of the developed framework is then validated with simulation on a Bivariate Non-stationary Growth Model (BNGM) as a benchmark. In the next step, three-year operating data of an industrial gas turbine engine (GTE) are utilized to verify the effectiveness of the developed framework. A comprehensive thermodynamic model for the GTE is therefore developed to formulate the relation of the observable parameters and the dominant degradation symptoms of the turbine, namely, loss of isentropic efficiency and increase of the mass flow. The results confirm the effectiveness of the developed framework for simultaneous estimation of multiple degradation symptoms in complex systems with noisy measured inputs. Highlights: We developed particle filter (PF) scheme for dynamical systems with stochastic input. We proposed a bivariate nonstationary growth model to test state estimation methods. We characterized turbine fault symptoms and estimated them by measurable parameters. We proposed a prediction model for turbine degradation using the operating profile. Regularized PF had superior performance than the auxiliary PF in turbine degradation.