(147n) Hybrid Grey-Box Modeling and Estimation for Condition Monitoring of Complex Dynamics Energy Systems

My research focuses on the development of dynamic modeling and estimation approaches applicable to operating plants, offering valuable benefits for process monitoring and control applications. The growing integration of intermittent renewables into the existing energy grid has compelled power plants to frequently and rapidly adjust their loads. This practice has resulted in significant damage to critical components, compromised reliability, and increased costs. To tackle these challenges, my research aims to create advanced monitoring tools that enable plants to better comprehend the effects of load-following. The specific objectives of my work are to provide utility plants with tools that facilitate the creation of efficient preventive maintenance plans, prevent forced outages, and establish advanced control strategies. These efforts aim to enhance operational flexibility while prioritizing safety and reliability.

Research Interests:

• Dynamic Process Modeling

The task of developing advanced and adaptive monitoring tools for industrial processes is challenging due to several factors, including the presence of high nonlinearity and complex dynamics. Furthermore, there is a limitation in the availability of measurement data, which is often restricted to specific variables associated with these processes. Throughout my work, I have dedicated my efforts to constructing comprehensive process models that accurately depict the system's geometry and real-time behavior in various operational scenarios. My expertise lies in developing detailed first principles models based on ordinary/partial differential equations (ODEs or PDEs) to capture the dynamic behavior of energy systems. These models have undergone rigorous validation using industrial datasets obtained from a wide range of operating plants, encompassing various energy systems such as coal-fired and combined cycle systems. By performing sensitivity studies utilizing two years' worth of operating data, the models have shown a favorable alignment between their predictions and the actual real-time behavior. Furthermore, the models have successfully captured the inherent non-linearities present in the system.

To summarize, my research has revolved around developing first principles models capable of capturing the dynamic behavior of energy systems in industrial processes. These models have been rigorously validated using industrial datasets and subjected to comprehensive sensitivity studies, demonstrating their robust agreement with real-time behavior and accurate representation of system non-linearities. This expertise serves as a solid foundation for the development of advanced and adaptive monitoring tools tailored to industrial processes

• Hybrid Grey- Box Modeling

Process systems can be represented using two types of models: white-box models, which are based on first principles, and black-box models, which utilize measurement data. While first-principles models provide predictive capabilities, they may struggle to capture complex phenomena with unknown physics, like flow maldistribution and unsteady heat transfer. Additionally, they can be computationally demanding and challenging to adapt in real time. On the other hand, black-box models based on data-driven machine learning techniques are simpler to construct and adapt online. However, they may lack predictive abilities, particularly when extrapolated or when the data used for their development have information gaps.

To address these challenges, my research focuses on developing models that synergistically combine both the data-driven and first-principles approaches to effectively monitor complex dynamic process systems. I am particularly interested in developing innovative hybrid grey-box modeling approaches that integrate rigorous physics-based first-principles models with data-driven black-box models. This hybrid approach proves valuable in modeling systems that involve unknown phenomena or have poorly understood physics A significant portion of my work involves exploring first-principles process models based on distributed differential-algebraic equation (DAE) models. These models are employed in analyzing the performance of high-temperature boiler components, such as superheaters and reheaters, under load-following conditions. The proposed models incorporate mass and energy balances based on first principles, along with robust properties models to ensure accurate heat transfer calculations and geometrical characterization. A modular design approach has been adopted, making these models versatile for various systems. In addition, I have expertise in developing data-driven black-box models using machine learning techniques that utilizes operational data and incorporate straightforward online adaptation formulations

I am particularly interested in applying these hybrid grey-box approaches to establish a health monitoring framework for energy systems. By leveraging extensive plant data alongside physics-based models, we can enhance the accuracy and reliability of the monitoring process. Furthermore, I aim to develop novel grey-box architectures that integrate serial, parallel, or integrated structures for multi-scale process system modeling. These advancements hold potential applications in areas such as equipment life consumption analysis and unconventional process control.

• State and Parameter estimation

In my research, I also explore the development of joint state and parameter estimation techniques using dynamic models for energy systems. Given the inherent incompleteness and inaccuracy of both the dynamic models and measurement data obtained from operating plants, an optimal quantitative estimation algorithm can be employed to leverage these resources and establish an adaptive, real-time, and accurate condition monitoring framework. Specifically, I have gained expertise in utilizing various dynamic models in conjunction with Kalman filter-based estimation techniques for state and parameter estimation in operating plants.

Recognizing that the standard algorithms available in the literature provide suboptimal results within the developed modeling framework, I have devised modified estimation algorithms capable of handling uncertainties present in both differential and algebraic equations, while duly accounting for noise in both differential and algebraic states. Additionally, I am keen on exploring the domain of constrained state estimation algorithms to ensure compliance with mass and energy balances in generic process systems. This aspect is crucial as the measurement data obtained from sensors often violate fundamental principles and thermodynamics laws. Directly employing such data in the estimation framework may lead to inaccuracies. Therefore, it is essential to incorporate algorithms that account for these errors when utilizing such datasets alongside the process models within the comprehensive model-estimator framework. These algorithms can be further validated using literature and industrial data.

The field of hybrid grey box modeling and estimators presents significant promise for monitoring energy systems, particularly considering the wealth of available measurement data. These approaches create ample opportunities for advancements in this area, allowing for more accurate and comprehensive monitoring techniques.