A deep-learning reduced-order model for thermal hydraulic characteristics rapid estimation of steam generators. (1st December 2022)

• Record Type: Journal Article
• Authors: He, Shaopeng; Wang, Mingjun; Zhang, Jing; Tian, Wenxi; Qiu, Suizheng; Su, G.H.
• Is Part Of: International journal of heat and mass transfer. Volume 198(2022)
• Publication Date: 2022-12-01
• Subjects: Reduced-order models -- Deep-learning -- POD -- CFD -- Steam generators; Heat -- Transmission -- Periodicals; Mass transfer -- Periodicals; Chaleur -- Transmission -- Périodiques; Transfert de masse -- Périodiques; Electronic journals; 621.4022
• DOI: 10.1016/j.ijheatmasstransfer.2022.123424
• Languages: English
• ISSNs: 0017-9310

Abstract

Highlights: A reduced-order model based on POD and machine leaning method is proposed. Rapid estimation of steam generator was achieved with the surrogate model. The order of void fraction and temperature field decreases by 88.3% and 96.7%. The maximum absolute errors of surrogate model are 0.1 and 0.03 K. Surrogate model reduces simulation time with speedup on the order of 10 4 . Abstract: Model reduction is a method that maps full-order conservation equations into lower-order subspaces or establish a data-driven surrogate model to reduce the complexity of the entire physical system, which has been widely applied in various fields in recent years. Compared with computational fluid dynamics (CFD) simulations, reduced-order model (ROM) can quickly and instantly obtain simulation results at low cost, which provides an economical alternative approach for the research and design process which need large number of repetitive simulations. In this paper, a deep-learning ROM was developed based on the proper orthogonal decomposition (POD) and machine learning (ML) method. The rapid estimation of two significant thermal hydraulic parameters in steam generator (SG), including the void fraction and temperature, was carried out by ROM. By POD mode analysis, the order for void fraction and temperature field was reduced by 88.3% and 96.7%, respectively. An artificial neural network was trained to reflect the implicit nonlinear mapping relationship between the CFD inputs and feature coefficients. The ROM was validated by comparing the predicted results with refined CFD results. The maximum absolute errors of void fraction and temperature are 0.1 and 0.03 K with speedup on the order of 10 4, indicating that the developed ROM can quickly and accurately estimate the thermal hydraulic characteristics of SG under different operating conditions. This work may provide a novel approach for the parameter sensitivity analysis and optimization design of SG and give valuable reference for the digital twin and the real-time online monitoring of the SG.