Reduced Order Models for a Pilot Burner and an Industrial Pulverized-Coal Furnace by OpenFOAM

It has been a challenging task to apply CFD in the industrial sector, since excessive computation time limited its use in the design phase, but not in real time operation and maintenance. Recently Reduced Order Model (ROM) was applied for large scale industrial combustion facilities. Once a ROM is built, it can produce 3D fields similar to the CFD results in the operation range within seconds or tens of seconds. A ROM can be constructed through Proper Orthogonal Decomposition (POD) and regression from CFD results at a limited number of sampled operating conditions. In this work we introduce the ROMs built for a pilot scale burner and a large pulverized coal-fired furnace to allow reconstruction of full 3D fields within a few seconds. We also developed an improved ROM that integrates the CFD ROM and measurement data to compensate for any difference of simulation results from actual measurements. We applied the gappy-POD method which was initially developed in face recognition research [1] and has been employed as a common technique to infer missing original data from available measured values. Here the gappy-POD is applied to calibrate the CFD ROM to match measurement data, while maintaining the proper characteristic trends in 2D or 3D fields. We showed that the gappy-POD method could reconstruct the missing original data successfully by using virtual measurement data in the industrial scale pulverized coal-fired furnace. We also developed the coupled gappy-POD method which can calibrate not only for the scalars with corresponding measurement data, but also the other vector or scalar quantities maintaining coupled functional relationships with those scalars with available measurements. For example, temperature data can be used to calibrate not only the temperature field, but also the pressure and gas components related with the temperature. The coupled POD method is validated in the pilot scale furnace with temperatures measured at seventeen sensor locations along the wall. Results are provided for the oxygen concentration field together with the temperature field which are both in close agreement with the original CFD fields. Reference [1] R. Everson and L. Sirovich, Karhunen-Loeve procedure for gappy data, J. Opt. Soc. Am. A. 12 (1996) 1657-1664