Experimental validation of an interval observer-based sensor fault detection strategy applied to a biohydrogen production dark fermenter

Abstract

In this paper, the problem of designing an interval observer applied to a biohydrogen production dark fermenter is addressed, in order to: (i) estimate the glucose and biomass concentrations, and the hydrogen flow rate produced, (ii) reduce the influence of the unknown inlet glucose concentration and the model uncertainties, and (iii) detect the presence of faults in the hydrogen flow rate sensor. Based on literature reports, we propose an interval observer structure, constituted by a Luenberger observer and an interval predictor, for a class of linear systems. The interval observer offers three estimations: the upper and lower estimations provided by the interval predictor, and one estimation given by the Luenberger observer. The upper and lower estimations provide bounds at any instant of time for the state vector of the perturbed/uncertain biohydrogen production bioreactor model from an appropriate initial condition. The design conditions presented in this work are based on control H∞ in combination with pole placement and polytopic parametric uncertainties. The observer design is proposed as a semi-definite optimization problem subject to Linear Matrix Inequalities, which do not depend on the cooperativity property. Based on the proposed observer structure, we also present two fault detection schemes for the same class of uncertain systems to detect the occurrence of sensor faults: one makes use of the logic comparison between the set threshold and the residual produced by the Luenberger observer; the other considers adaptive thresholds for the output signal by using the upper and lower estimates of the interval predictor. Finally, the proposed approaches are validated with experimental data from a biohydrogen production dark fermenter.