Software sensor designs for urban wastewater systems

• Main Author: Benazzi, Farid
• Published: University of Strathclyde 2006
• Subjects: 628.240285
• Online Access: http://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.435151

The objective of this thesis is the design of software sensors for an urban wastewater system. The dynamic behaviour of urban wastewater systems are typically described by highly non-linear deterministic models. These non-linearities are mainly caused by important variations in the influent flow rate. In addition, there is a clear lack of instrumentation and automation systems. Therefore, this thesis contains a thorough discussion of software sensors, with applications to urban wastewater treatment systems and a special emphasis on the wastewater treatment plant. Since the original activated sludge process model utilised in this work is not observable, three reduced order models, based on the activated sludge model no. 1, are proposed. A linear piece-wise observability analysis based on the Kalman rank theory is investigated on each of the reduced models, prior to non-linear observability analyses based on the Lie derivative. Furthermore, a procedure to remove the unobservable modes and to design software sensors in the presence of disturbances is proposed. The main objective of the work on state observers and parameter estimators design is to achieve on-line estimation of non-measurable concentrations and parameters based on extended Kalman filters. Initially, on-line monitoring of abnormal substrate concentrations is proposed. The designed state observer can detect substrate shock loads with a reasonable response time. Then, as on-line measurements of the biomasses concentrations are not available in real wastewater treatment plants, online monitoring of the heterotrophic biomass and autotrophic biomass concentrations is proposed. Finally, a joint state and parameter estimation application is presented, where the reduced model is augmented with an additional state variable. The main objective aims at demonstrating the parametric estimation difficulties when designing software sensors on such complex augmented non-linear model. The work on robust non-linear filtering is motivated by the fact that the extended Kalman filters presented significant drawbacks. Applications based on H∞ filtering, in which the model describing the activated sludge process is corrupted by significant process noise is presented. A comparative study between both types of software sensors is performed. The final contribution of this work is toward software sensing applications on an integrated urban wastewater system. A software sensor is implemented on the treatment plant and the sewer network effect on the estimated concentrations is demonstrated through simulation studies.