Robust fault detection with Interval Valued Uncertainties in Bond Graph Framework

Abstract

This paper describes a novel formalism for modelling uncertain system parameters and measurements, as interval models in a Bond Graph (BG) modelling framework. The main scientific interest remains in integrating the benefits of BG modelling technique and properties of Interval Analysis (IA), for efficient diagnosis of uncertain systems. Structural properties of Bond graphs in Linear Fractional transformation (BG-LFT) are exploited to model interval-valued uncertainties over a BG model in order to form an uncertain BG. The inherent causal properties are exploited to generate interval-valued fault indicators. Then, various properties of IA are used to generate point valued residual and interval-valued thresholds. The latter must contain the point valued residuals under nominal system functioning. A systematic procedure is proposed for passive-type fault detection method which is robust to uncertain system parameters and measurements. The viability of the method is shown through experimental study of a steam generator system. The limitations associated with existing fault detection method based on BG-LFT are alleviated by the proposed approach. Moreover, it is shown that proposed approach generalizes the BG-LFT method. This work forms the initial step towards integrating interval analysis based capabilities in BG framework for fault detection and health monitoring of uncertain systems.