(147p) State Estimation and Cost Optimal Sensor Network Design for Electrochemical Sensor Based Corrosion Monitoring

In the future, it is anticipated that renewables will take precedence as the main source of electricity generation. However, to maintain the stability of the electricity grid, non-renewable sources such as coal will continue to be utilized. As a result of the growing integration of intermittent renewables, coal-fired power plants are being required to implement load-following strategies. The detrimental effects of load following can lead to accelerated equipment damage, resulting in forced outages and potential revenue loss. Any deviation from base load conditions in aging coal-fired power plants has a negative impact on the health of the equipment. One critical component affected by damage is the waterwall, with corrosion being the primary cause. Real-time corrosion monitoring in such harsh environments is challenging, and even if suitable technologies exist, it is impractical to place sensors at every location within the waterwall section. Consequently, my research primarily focuses on corrosion modeling, state estimation using the Kalman filter, and the development of a cost-optimal sensor placement algorithm that incorporates economic models for real-time corrosion monitoring in coal-fired power plants. While my research primarily focuses on addressing a specific gap within the existing literature, I also possess the flexibility and expertise to modify and adapt these techniques to be applicable to relevant systems like power plants, chemical manufacturing facilities or generic process systems where condition-based health monitoring through sensor placement is desired.

Research Interests:

• First-principles, mechanistic and data-driven model building

Fireside corrosion behavior of equipment in coal-fired power plants is complicated and influenced by multiple factors. Through my research I have identified the key factors that have a significant impact on the corrosion rate and have subsequently developed an appropriate mechanistic kinetic model to describe its rate. The distribution of combustion gases and temperature in the waterwall section of a coal-fired power plant was analyzed. Sensitivity analysis was performed for these states to understand the magnitude of the effect they have on corrosion depth, establishing the need for suitable model development. Emphasizing the burner placement along the waterwall section, two types of models were developed: first-principles models relying on mass balance and data-driven models utilizing feedforward neural networks.

During the process of developing these models, I have come across and become acquainted with different techniques in regression analysis, which include exploring interactions among multiple predictors and employing model selection methods based on correlations between variables. I am skilled in formulating empirical models that can replicate the desired trends observed in the data. I also possess a foundational understanding of neural networks and similar machine learning techniques. I possess a strong comprehension of first principles through my hands-on experience in performing mass and energy balance calculations for combustion calculations.

• State estimation

An estimator is employed in situations where there is a lack of complete trust in both the process model and the measurements of a state variable. As a part of my research, I have developed an optimal estimator that will employ the corrosion depth measurements provided by proposed sensors and estimate the corrosion rate along the entire length of the waterwall tubes. I have conducted extensive research on the Kalman filter and its derivative variants, the Extended Kalman filter (EKF) and the Unscented Kalman filter (UKF). I have successfully applied these techniques to diverse non-linear dynamic systems. My expertise lies in utilizing any of these methods for state estimation purposes in both industrial and research domains. I have experience working with systems such as ordinary differential equations, partial differential equations, and differential algebraic equations. Furthermore, I have successfully applied the aforementioned estimators to these complex systems.

• Economic analysis

The corrosion depth estimation enables prediction of the failure time of waterwall due to corrosion. By planning maintenance activities accordingly, forced outages due to corrosion can be avoided, aiding to the overall increase of availability of plant. The forced outage prevented due to sensor placement results in potential revenue gain that is worth evaluating. For investment in corrosion sensors in ailing coal-fired plants, utility owners would like to ensure a positive return on investment. The actual electricity produced and the profit the plant will make depend on similar improvements in the availability of other plants and rapid deployment of renewables, stochasticity in the market demand and price, etc. My research aims to capture this market elasticity through economic analysis conducted using energy system model (ESM) which forecasts energy market. They provide information about the change in electricity production because of the higher availability of the power plant, and other uncertainties in the energy market. By utilizing ESM like The Integrated MARKAL-EFOM System (TIMES), I have identified a range of technologies and their corresponding factors that have an impact on electricity production from different fuel sources within a specified prediction horizon. I possess a fundamental comprehension of the structure of the U.S. energy market and the interplay among various participating agents within it. I am adept at utilizing the any ESM like TIMES, and my experience with it has made me skilled in conducting scenario based economic analysis. My skills include conducting sensitivity analysis, formulating scenarios that simulate potential energy futures, and analyzing the electricity generated from various fuel sources within those scenarios, among other capabilities.

• Optimal sensor placement

Process information is obtained from a given sensor network, and the reliability of sensor network improves as the number and/or the accuracy of the installed sensor increases. To assess the optimality of sensor network design, one can take into account various factors such as estimation error, reliability, observability, efficiency, economic measures, and more. It is essential to consider the uncertainty in the corrosion measurements when such sensors are deployed in real-life. Since the measurement is localized and it is neither required nor possible to deploy these sensors at every possible location, a cost-effective solution is to place these sensors at optimal locations. Hence, my research focuses on developing an optimal sensor network placement that maximizes the net present value (NPV) by considering the cost of the sensors and the incremental increase in the revenue integrated over the plant life. For estimating the improvement in revenue due to increased availability under influence of energy market uncertainties, various scenarios were created in ESM and utilized in the optimal sensor placement algorithm. The sensor network design problem leads to a mixed-integer nonlinear programming challenge, and efforts are underway to address it using optimization techniques such as the branch and bound algorithm. My encounter with solving this problem has exposed me to the concepts of mixed-integer linear programming and familiarized me with combinatorial analysis. My primary expertise lies in incorporating market elasticity into traditional NPV analysis, ensuring a comprehensive and optimal approach to decision-making in sensor network placement.