An integrated furnace co-simulation methodology based on a reduced order CFD approach

An integrated thermofluid modelling methodology for pulverised fuel fired utility-scale boilers that is computationally inexpensive, fast, and sufficiently accurate would be valuable in an industrial setting. Such a model would enable boiler operators to investigate a range of off-design operating conditions, which includes flexible operation. The aims of this study was: to develop a reduced order computational fluid dynamics (CFD) model of the furnace and radiative heat exchangers that captures all the important particulate effects while using a Eulerian-Eulerian (EE) approach; using the reduced order CFD model to generate a database of results that covers a wide range of operating conditions; to develop a data-driven surrogate model using machine learning techniques; to integrate the surrogate model with a 1-D process model of the complete boiler; and finally to demonstrate the use of the integrated model to investigate flexible operation and off-design operating conditions. The validity of the CFD modelling approach was demonstrated via application to a 2.165 [MWth] lab-scale swirl pulverised fuel burner, as well as to a 620 [MWe] utility-scale subcritical two-pass boiler, both operating at various loads. The results were compared to measured data and a detailed CFD model using the conventional Eulerian-Lagrangian (EL) approach. A computational speed enhancement of 30% was achieved. The data-driven surrogate model uses a mixture density network (MDN) to predict the heat transfer in the furnace and radiative heat exchangers, together with the uncertainty in the predicted values. The integrated model was validated against applicable measured data and then applied to a utility-scale case study boiler to investigate the optimal burner firing positions for low-load operation, as well as to investigate the effects of fuel quality on the overall boiler performance. It was shown that the integrated data-driven surrogate model and 1-D process model can predict the overall thermofluid response of the boiler and the uncertainties associated with it with good accuracy, whilst maintaining a low computational effort when compared to a conventional CFD model coupled to 1-D process model.

Grantor

Department of Mechanical Engineering

Year dc:date.issued

2023

Author dc:creator

• Rawlins, Brad

Advisors dc:contributor.advisor

• Rousseau, Pieter
• R. Laubscher

• Engineering and the Built Environment

Handle dc:identifier.uri

http://hdl.handle.net/11427/38114

OAI identifier oai:identifier

oai:open.uct.ac.za:11427/38114