(202m) Distributed Sensor Placement Algorithm for a Water Gas Shift Reactor in An IGCC Plant

Mobed, P., Texas Tech University
Maddala, J., Texas Tech University
Bhattacharyya, D., West Virginia University
Turton, R., West Virginia University
Rengaswamy, R., Texas Tech University

Coal-fed integrated gasification combined cycle (IGCC) is a promising technology for electricity generation with zero emissions. In the catalytic water gas shift reactors (WGSRs), as part of an IGCC plant with carbon dioxide capture,  the carbon monoxide present in the syngas reacts with steam to generate carbon dioxide and hydrogen. The carbon dioxide is captured in an acid gas removal unit and the hydrogen-rich syngas is used as a fuel for power generation. The efficiency and performance of the IGCC plant is significantly affected by the WGSRs, which in turn, are affected by a number of disturbances and faults. For instance, disturbances in flow, temperature or molar compositions of the feed affect the conversion and, consequently the efficiency of the reactor. Additionally, the catalyst inside the WGSR is vulnerable to faults due to poisoning, fouling, or thermal cycling resulting in deactivation, porosity reduction, and/or surface area reduction. Therefore, an early detection of disturbances and faults in the WGSRs can ensure safe and efficient operation of the reactor. A first-principles model of the WGSR can be used to predict the states of the reactors and detect the faults. Since process models are never completely accurate and the external disturbances are not predictable, a series of measurements, such as measurements from temperature, pressure and concentration sensors, along with the process model is helpful in estimating the unknown states of the reactor. However, the challenging task is to find the optimal number and location of these sensors for estimation of faults in the presence of disturbances and uncertainties in the process model and measurements.

A first-principles model of the WGS reactor has been developed in our previous work. Since the resulting DAE system is non-linear, an extended Kalman filter (EKF)  is tailored for estimation in DAE systems.  Possible faults, which are identified with prior process knowledge, are incorporated as states and estimated along with other states of the system. An accurate estimation of the states can help in identifying the disturbances and detecting the faults. Therefore, minimizing the error in the estimated states requires finding the best measurement model that contains the optimal number and location of the sensors.

The measurement model is a discrete set of binary values for all measureable states, representing 1 if a sensor is placed and 0 if no sensor is placed for the specific discrete variable. In order to efficiently detect all the faults, a combinatorial optimization problem is solved to find the best measurement model while minimizing the error between the actual and estimated states of the reactor. In this work, a genetic algorithm which can handle discrete optimization problems is proposed, where at each generation, a population of measurement models is used in the EKF with genetic operations performed between each generations, until the best solution is found.

In this presentation, we will present the SP algorithm and the optimal measurement model that can detect faults with reasonable accuracy in the presence of disturbances, measurement noises and uncertainties in the process model.