Optimisation of LESsCOAL for large-scale high-fidelity simulation of coal pyrolysis and combustion


Key Personnel

PI: Dr. Jun Xia - Brunel University London

Technical: Kaidi Wan - Zhejiang University

Relevant Documents

eCSE Technical Report: Optimisation of LESsCOAL for large-scale high-fidelity simulation of coal pyrolysis and combustion

Project summary

This project aims to substantially improve the parallel performance of an in-house code, LESsCOAL (Large-Eddy Simulations of COAL combustion). This has been under active development over recent years by one of the PIís main collaborators, Prof Zhihua Wang, working at the State Key Laboratory of Clean Energy Utilization of Zhejiang University, China. While originally designed to properly predict turbulent pulverised-coal combustion using high-fidelity simulation techniques, the software package has now developed to a stage at which it can be used for a variety of turbulent multiphase reacting flows.

The collaboration between the PI and Prof Wang uses high-fidelity experimental and computational techniques to help the Chinese coal industry achieve cleaner and more efficient coal combustion. Coal currently produces 70% of the electricity in China, so this can have a huge impact on emissions. For the UK, biomass-fired or co-firing of biomass and coal is considered to be a promising way of using renewable solid fuels in existing coal-fired facilities. Other combustion concepts and technologies such as oxy-coal combustion are also under development for reducing carbon emission from these industrial solid-fuel boilers.

Three modules - particle-tracing, radiation, and pressure - had been identified as the bottlenecks affecting the parallel efficiency of the code. By optimising the parallel algorithms of these modules, two types of simulation have now been enabled. The first is direct numerical simulation (DNS) of a laboratory-scale pulverised-coal jet flame. While advanced nonintrusive laser diagnostics are being developed rapidly, measurement of a solid-fuel-particle-laden flame is still difficult. Under this circumstance, DNS data will be invaluable source for numerical verification before detailed validation against experimental data is carried out. The second type is large-eddy simulation (LES) of two-phase reacting flow in a large-scale industrial pulverised-coal burner, for which experimental data is available for comparison. Simulations of both types were not possible before the eCSE project due to the poor parallel efficiency of the code.

The impact of the achievement will be significant in that we have overcome the hurdles due to parallel efficiency of the code and now can efficiently utilise high-fidelity simulation techniques, supported by high-performance computing, to research complex turbulent multiphase reacting flow such as pulverised-coal combustion and many others.

Achievement of objectives

The overall aim of the project was to achieve 80% of the theoretical speedup for LESsCOAL when up to 3,000 compute cores are used on ARCHER. The planned stage objectives were:

  1. Develop and implement a new parallel particle-tracing algorithm to radically improve the load balances among processor cores.
  2. Implement a three-dimensional domain decomposition approach.
  3. Improve the pressure solver, considering both robustness and efficiency.
  4. Improve the initialisation of the iteration in the radiation module.
  5. Implement new MPI and FORTRAN functionalities.

Success Metrics 1: The speedup of turbulent gaseous jet cases with no particles reaches the overall aim set, i.e. achieving 80% of the theoretical value when using 3,000 cores on ARCHER.

Success Metrics 2: The speedup of LESsCOAL achieves 80% of the theoretical value when using 3,000 cores on ARCHER to run a turbulent hot jet laden with coal particles.

Summary of the software

LESsCOAL is now a centralised module on ARCHER and freely available to the ARCHER community. We also welcome other UK academics to use the code. If the code is only used for research purposes, there will be no license restrictions. If industrial users are interested in using the code, e.g. integrating a module into commercial software, please get in touch with Dr Jun Xia at jun.xia@brunel.ac.uk or call 01895 265 433. We will discuss the matter with the Research Support and Development Office of Brunel University London and EPCC to see what is the best way forward. According to experience a Knowledge Transfer project will need to be set up.

A user document together with test cases is available in the code package to facilitate the use of the code. Direct technical support from the PI will be available if necessary to help prospective users effectively use the code.

As the only condition of using the code, new users are required to inform Dr Xia on what case scenarios the code has been tested/used and its performance.

Finally, it should be stressed that sharing the code with a wider scientific community is favourably seen by the PI as an effective means to receive feedback, thus accelerating further development of the software package.