(314f) Development of an Online Health Monitoring Framework for High Temperature Boiler Components By Using Hybrid First Principles-Data-Driven Models

As the integration of renewables in the current energy grid is becoming prevalent worldwide to achieve low-carbon emissions, existing fossil-fired power plants need to operate flexible by rapidly changing their load and starting and stopping more frequently than intended. The current power generation fleet has been mainly designed for base load operations and therefore is facing several operational challenges due to these flexible operations. These severe operations are causing substantial damage to the critical boiler components, compromising these plants' reliability and driving up operational costs. To ensure efficient plant operations, development of advanced condition monitoring tools which are adaptive and can be generalized for different plant configurations can be instrumental in understanding the impacts of load-following. These tools can eventually help these plants to operate with increased safety and reliability and prevent undesired forced outages and unexpected economic losses. Advanced monitoring tools based on data-driven models utilizing extensive operational datasets have been well developed and utilized in industrial plants for rotary equipment items such as gas turbines and steam turbines. However, for static equipment such as boiler systems limited tools are available for their health monitoring. Modeling of these systems can be carried out either using first principles models based on conservation laws or data-driven models based on operational measurement data. While first-principles physics-based models can be predictive, their development for such complex non-linear dynamic systems is time-consuming and computationally expensive for online adaptation. Furthermore, developing first-principles models for systems whose physics is complex and not well understood can be challenging. Conventional artificial intelligence (AI) or data driven models developed using operational data although have faster convergence and relatively simpler structures, they can lead to significantly higher errors when used for systems with high nonlinearity and inadequate and noisy measurement data. In addition, these data-driven models may not be predictive, especially when they are extrapolated and/or if the data used for developing these models suffer from an information gap. For example, high-temperature boiler components such as steam superheater systems have complex dynamics with limited measurements available for key state variables like metal wall temperatures due to harsh operating conditions in which the sensors cannot survive for long. However, predicting these critical variables such as through-wall temperature profile of a tube in the high temperature superheater section of a boiler can provide important indication of scale formation or fouling. Modeling such systems using either physics-based or data-driven models may not be sufficient to provide accurate results. Hence, in this work we have developed a health monitoring framework, that is adaptive and real-time, by synergistically hybridizing physics-based models with data-driven models.

In the existing literature various hybrid modeling approaches have been proposed for boiler systems and its associated components using linear state space and detailed dynamic model of the physical system along with closed-loop data for unknown parameters estimation using either artificial neural networks or genetic algorithms ^{[1], [2]}. A hybrid model approach was proposed to characterize the performance of a steam boiler system ^[3] and a feedwater heater system ^[4] using a nonlinear differential and algebraic equations model of mass and energy balances along with nonlinear least-square optimization formulation for parameter estimation. Detailed mechanistic hybrid models for an electric hot water boiler system ^[5] and a high speed and high temperature airflow system ^[6] have also been investigated. More recently a few hybrid modeling approaches have been implemented for modeling thermal systems of in-service power plants ^{[7], [8]} or for controlling fouling in biomass boilers ^[9]. Although, these hybrid model approaches have been useful for specific applications in particular systems but there is a lack of work in the open literature utilizing a comprehensive hybrid modeling approach for health monitoring of boiler systems with validation using operational data. Oxide scale growth inside the steam tubes is one of the key damage mechanisms for the tubes. Since it is impractical to measure the oxide scale thickness inside the tubes, mathematical models are very useful for predicting the spatio-temporal profile of these oxide scales. Recenlty, some studies have been conducted by measuring the amount of oxide scale growth formation using lab grown oxide scales and also by charcaterizing the damaged tubes from operating plants in order to estimate health of boiler components ^{[10], [11]}. Furthermore, some authors have developed high fidelity CFD models for superheater system for estimating remaining useful life by using parabolic growth law for oxide evolution ^[12]. However, to the best of our knowledge, existing literature lacks studies on development of a tube damage monitoring approach that is validated with the real-life data.

In order to address these challenges, in this work we have developed a comprehensive health monitoring framework for high temperature boiler components using hybrid modeling approaches. The hybrid model is based on a dynamic distributed first-principles 3-D differential-algebraic equation (DAE) model considering differential mass and energy balances equations with detailed heat-transfer models through the tube walls. For the data-driven models, two types of AI models are considered in this work, namely the all-nonlinear static-dynamic neural networks ^[13] and models obtained from a Bayesian machine learning ^[14] based approach. Two different hybrid model structures are developed – one where the data-driven models represent some phenomena or mechanisms , which are then integrated in series with the first-principles model, another where a parallel arrangement of physics-based and data-driven models are used to model the spatial variation in temperature profile. Formation of the oxide scale leads to higher heat transfer resistances leading to local temperature rise that can accelerate creep damage. The oxide scale growth is calculated using a parabolic growth rate law for different material specifications and then Larson-Miller Parameter (LMP) method is used to predict damage accumulation in the system for estimating creep failure. Prediction of damage accumulation based on oxide scale formation is compared with actual oxide scale measurements obtained using metallography studies performed on tube specimens installed in a real-life operating system for a period of two years of plant operation. Performance of the framework as an online health-monitoring tool is studied for a power plant boiler superheater system using two years’ worth of operational data with varying load-following conditions.

References

[1] D. Santos et al., A grey-box Neural Network Composite Model for an Industrial Heating Furnace. 2019.

[2] A. Chaibakhsh, “Modelling and long-term simulation of a heat recovery steam generator,” Mathematical and Computer Modelling of Dynamical Systems, vol. 19, no. 2, pp. 91–114, Apr. 2013, doi: 10.1080/13873954.2012.698623.

[3] T. Barszcz, P. Czop, and J. Bednarz, “Model-based evaluation of a power plant steam boiler system,” Problemy Eksploatacji, no. 2, pp. 7–19, 2011.

[4] T. Barszcz and P. Czop, “Estimation of feedwater heater parameters based on a grey-box approach,” International Journal of Applied Mathematics and Computer Science, vol. 21, no. 4, pp. 703–715, Nov. 2011, doi: 10.2478/v10006-011-0056-4.

[5] A. A. Farooq, A. Afram, N. Schulz, and F. Janabi-Sharifi, “Grey-box modeling of a low pressure electric boiler for domestic hot water system,” Applied Thermal Engineering, vol. 84, pp. 257–267, Jun. 2015, doi: 10.1016/j.applthermaleng.2015.03.050.

[6] Y. Xue, C. Cai, and H. Xu, “Grey Box Modeling of Gas Temperature for a High-Speed and High-Temperature Heat-Airflow Test System,” Applied Sciences, vol. 12, no. 24, Art. no. 24, Jan. 2022, doi: 10.3390/app122412883.

[7] J. Yu, N. Petersen, P. Liu, Z. Li, and M. Wirsum, “Hybrid modelling and simulation of thermal systems of in-service power plants for digital twin development,” Energy, vol. 260, p. 125088, Dec. 2022, doi: 10.1016/j.energy.2022.125088.

[8] P. Wang, F. Si, Y. Cao, Z. Shao, and S. Ren, “Prediction of superheated steam temperature for thermal power plants using a novel integrated method based on the hybrid model and attention mechanism,” Applied Thermal Engineering, vol. 203, p. 117899, Feb. 2022, doi: 10.1016/j.applthermaleng.2021.117899.

[9] L. M. Romeo and R. Gareta, “Hybrid System for fouling control in biomass boilers,” Engineering Applications of Artificial Intelligence, vol. 19, no. 8, pp. 915–925, Dec. 2006, doi: 10.1016/j.engappai.2006.01.019.

[10] R. Shukla and D. Pareek, “Condition Assessment of ReheaterSuperheater Tubes using Internal Oxide Scale Thickness Measurement The Right Way,” e-Journal of Nondestructive Testing, vol. 26, no. 03, Mar. 2021, Accessed: Apr. 08, 2024. [Online]. Available: https://www.ndt.net/search/docs.php3?id=25725&msgID=0&rootID=0

[11] J. Qi, K. Zhou, J. Huang, and X. Si, “Numerical simulation of the heat transfer of superheater tubes in power plants considering oxide scale,” International Journal of Heat and Mass Transfer, vol. 122, pp. 929–938, Jul. 2018, doi: 10.1016/j.ijheatmasstransfer.2018.01.119.

[12] L. Sun and W. Yan, “Prediction of wall temperature and oxide scale thickness of ferritic–martensitic steel superheater tubes,” Applied Thermal Engineering, vol. 134, pp. 171–181, Apr. 2018, doi: 10.1016/j.applthermaleng.2018.01.127.

[13] A. Mukherjee and D. Bhattacharyya, “Hybrid Series/Parallel All-Nonlinear Dynamic-Static Neural Networks: Development, Training, and Application to Chemical Processes,” Ind. Eng. Chem. Res., vol. 62, no. 7, pp. 3221–3237, Feb. 2023, doi: 10.1021/acs.iecr.2c03339.

[14] S. Adeyemo and D. Bhattacharyya, “Optimal nonlinear dynamic sparse model selection and Bayesian parameter estimation for nonlinear systems,” Computers & Chemical Engineering, vol. 180, p. 108502, Jan. 2024, doi: 10.1016/j.compchemeng.2023.108502.