Digital condition monitoring of complex (bio)chemical reaction systems in the presence of model uncertainty: Application to environmental hazard monitoring

Abstract

A new approach to the state estimation and condition monitoring problem for complex nonlinear (bio)chemical reaction systems in the presence of model uncertainty is proposed. In particular, a new robust nonlinear state estimation method is developed that can be digitally implemented with the aid of a computer. The proposed method explicitly incorporates and processes all the available pertinent information associated with two sources: (i) a dynamic process model which is inevitably characterized by various degrees of uncertainty, and (ii) a set of available sensor measurements through which the values of certain variables are recorded. Based on the above information, a robust digital state estimator is designed capable of dynamically (over time) reconstructing all other key physicochemical variables that cannot be measured online (due to physical and/or technical limitations), while remaining critical from a process condition monitoring point of view. A set of conditions are derived that ensure the existence of such a state estimator, whose algorithmic implementation can be readily realized via a simple MAPLE code. Furthermore, the convergence of the estimation error or the mismatch between the actual unmeasurable states and their estimates is analyzed and characterized in the presence of model uncertainty. Finally, the performance of the proposed digital estimator is evaluated in an environmental hazard monitoring case study, where a simple bioremediation model describing the degradation of an organic hazardous pollutant is considered that exhibits nonlinear behavior coupled with parametric uncertainty. The estimation objective is, through the use of the proposed estimator, to reliably reconstruct the dynamic profile of the unmeasurable organic pollutant (substrate) concentration using microorganism (biomass) concentration measurements, as well as the available process model. The performance characteristics of the proposed estimator in the presence of model uncertainty are assessed by conducting simulation studies.