(588b) Optimal Sensor Network Design and State Estimation of Nonlinear Differential Algebraic System: Application to Corrosion Monitoring in a Power Plant Boiler

Coal is one of the primary sources of electricity production. Due to the increased penetration of intermitted renewables during the past ten years, coal-fired power plants have been subjected to load-following. During load following the rapid fluctuations in the operating conditions induce high stress on various equipment resulting in their damage.^[1] The existing supercritical coal-fired power plants are largely designed to operate at base load conditions, making load-following operations a significant threat to their equipment health. The high metal temperature of equipment, coupled with the presence of corrosive gases results in equipment damage primarily by corrosion fatigue, through Type II hot corrosion mechanism.^[2] A novel type of electrochemical sensor has been previously developed in-house for in-situ monitoring of hot corrosion.^[3] Using such sensors, the corrosion depth can be measured in susceptible locations in real-time. Corrosion monitoring can enable damage mitigation steps like pre-planned maintenance to avoid forced outages. However, due to spatial and economic constraints, it is not feasible to place sensors at a large number of locations on the surface of the equipment. Moreover, the spatial distribution of corrosive gases and the structure of equipment can cause uneven distribution of corrosion. Hence, an optimal sensor network algorithm is desired which would facilitate real-time corrosion monitoring.

A model of the hot corrosion mechanism is developed and validated using the data obtained from an industrial boiler. Metal temperature and concentrations of corrosive gases SO₂ and O₂ play a key role in the corrosion process and therefore spatial profiles of these variables are included as candidate measurements. While these variables are represented by algebraic models, the corrosion process is usually well represented by an ODE system leading to a differential-algebraic equation (DAE) system.

For state estimation, the standard Unscented Kalman Filter (UKF) is modified as that standard algorithm is mainly applicable to only ODE systems and not to DAE systems. The UKF algorithm was extended for semi-explicit index 1 DAE systems by Mandela et al and Mercieca et al. ^[4],[5],[6] Their works assumed that the algebraic states are dependent on the differential states and not vice versa, and the algebraic states are noise-free. However, in this work, the propagation of differential states is dependent on algebraic states and not vice versa, with inherent noise present in algebraic states. We also consider that some of the algebraic states are measured. An updated UKF algorithm is developed for updating estimates of algebraic states, and unscented sampling, and by incorporating the improved estimates of the algebraic states for updating the sigma points. An optimal sensor network is synthesized using the normalized a posteriori error covariance from the updated filter algorithm, by varying the number and placement of active sensors from the candidate sensors.

The algorithm is applied to a coal-fired power plant. In our work, we are particularly interested in estimating corrosion depth across the waterwall (WW) section of the boiler, which is one of the most damage-prone equipment. For incorporating inputs as algebraic states, metal temperature, and gas concentrations were modeled by feedforward neural networks. Variation in the optimal sensor network was studied by simulating a number of scenarios including the corrosion characteristics and mechanisms. It was observed that with only a few sensors placed at optimal locations, an accurate estimate of the corrosion profile can be obtained.

References

1. P. Besuner and S. Lefton, “The cost of cycling coal fired power plants,” Coal Power Mag., pp. 16–20, 2006.
2. F. Pettit, “Hot Corrosion of Metals and Alloys,” Oxidation of Metals, vol. 76, pp. 1–21, 2011.
3. N. N. Aung and X. Liu, “High temperature electrochemical sensor for in situ monitoring of hot corrosion,” Corros. Sci., vol. 65, pp. 1–4, 2012.
4. R. V. Mandela, R. Rengaswamy, and S. Narasimhan, “Nonlinear State Estimation of Differential Algebraic Systems,” IFAC Proceedings, vol. 42 (11), pp. 792-797, 2009.
5. R. V. Mandela, R. Rengaswamy, S. Narasimhan, and L. N. Sridhar, “Recursive state estimation techniques for nonlinear differential algebraic systems,” Chemical Engineering Science, vol. 65 (16), pp. 4548-4556, 2010.
6. J. Mercieca, “Estimation of temporal and spatio-temporal nonlinear descriptor systems,” Ph.D. dissertation, Dept. Automatic Control and Systems Engineering, University of Sheffield, 2018.