Traditional physics-based models, while predictive, are time-consuming to develop and computationally expensive, especially for complex, nonlinear systems. Data-driven AI models, which are faster to develop and simpler in structure, have been widely used in industrial plants for rotary equipment like gas and steam turbines. However, they often struggle with accuracy in nonlinear systems and are less commonly applied to static equipment like boiler systems. High-temperature boiler components, such as steam superheaters, have complex dynamics with limited measurements for critical variables like metal wall temperatures [2]. Predicting these variables is crucial for estimating metallurgical changes such as oxide scale formation and estimating remaining useful tube life for the system in operation. Since direct measurement of oxide scale thickness inside tubes is impractical, hybrid mathematical models are useful for predicting the spatio-temporal profile of oxide scales and estimating the time to rupture due to creep fatigue and tube damage accumulation. In the existing literature, numerous studies have utilized linear state-space models with neural networks or genetic algorithms for parameter estimation in boiler systems [3], as well as nonlinear differential-algebraic equation (DAE) models coupled with optimization techniques for steam boilers[4]. However, there remains a gap in the literature in developing and utilizing hybrid models for real-time health monitoring of boiler systems validated with operational data.
In this work, we have developed two hybrid model structures: one where data-driven AI models represent specific phenomena integrated in series with first-principles models and another where a parallel arrangement of models captures spatial temperature variations. The hybrid FP model is based on a dynamic distributed 3-D DAE framework incorporating detailed heat-transfer mechanisms [4]. For the data-driven AI component, static-dynamic neural networks and Bayesian machine learning techniques are used to enhance model accuracy and adaptability [6]. A key focus of this work is on prediction of oxide scale growth in the inner wall of the tubes. They lead to low heat transfer coefficient and localized heating, which, in turn, increases scaling thus leading to a spiraling damage mechanism. Oxide scales not only reduce the structural strength, but they can eventually spall thus leading to thinning of tube radius. There is currently no measurement technology to measure the oxide scale thickness without off-line destructive measurements. In this work, the oxide scale growth is modeled using a parabolic growth rate law, with subsequent creep damage estimated via the Larson-Miller Parameter (LMP) method. These predictions are validated against oxide scale measurements from tube specimens removed from a coal boiler superheater after 22 months of commercial operation. The framework is rigorously tested on a power plant boiler superheater system using operational data under load-following conditions. The results demonstrate the framework's effectiveness as an online tool for monitoring component health, providing predictive insights for efficient maintenance, and enhancing the reliability and safety of power plant operations.
References
[1] V. Saini, Development of Dynamic Modeling and Estimation Techniques for Condition Monitoring of Advanced Energy Systems, Graduate Theses, Dissertations, and Problem Reports (2024). https://doi.org/10.33915/etd.12673.
[2] D. Beahr, V. Saini, D. Bhattacharyya, S. Seachman, C. Boohaker, Estimation-based model predictive control with objective prioritization for mutually exclusive objectives: Application to a power plant, Journal of Process Control 141 (2024) 103268. https://doi.org/10.1016/j.jprocont.2024.103268.
[3] A. Chaibakhsh, Modelling and long-term simulation of a heat recovery steam generator, Mathematical and Computer Modelling of Dynamical Systems 19 (2013) 91–114. https://doi.org/10.1080/13873954.2012.698623.
[4] T. Barszcz, P. Czop, J. Bednarz, Model-based evaluation of a power plant steam boiler system, Problemy Eksploatacji (2011) 7–19.
[5] V. Saini, D. Bhattacharyya, D. Purdy, J. Parker, C. Boohaker, Nonlinear state estimation of a power plant superheater by using the extended Kalman filter for differential algebraic equation systems, Applied Thermal Engineering 251 (2024) 123471. https://doi.org/10.1016/j.applthermaleng.2024.123471.
[6] A. Mukherjee, V. Saini, S. Adeyemo, D. Bhattacharyya, D. Purdy, J. Parker, C. Boohaker, Development of hybrid first principles – Artificial intelligence models for transient modeling of power plant superheaters under load-following operation, Applied Thermal Engineering (2024) 124795. https://doi.org/10.1016/j.applthermaleng.2024.124795.