MODELLING AND SIMULATION


 Oxygen gas volume fraction in the anode channels in a polymer electrolyte membrane electrolyzer. The model is used to compute the two-phase flow field in the bipolar plate channels


g


side because we need to compile a lot of different models while maintaining efficiency in HPC facilities.’ Lehmkuhl added: ‘We also have an extra challenge where, usually soot formation modelling is done with gas [engines] but, with kerosene, we have to take into account multi-phase conditions.’ There is also an industry-led


requirement for this work, as Daniel Mira, propulsion technologies group leader at BSC, explained: ‘The aerospace industry has clear targets for CO2 emissions but now the ones for soot emissions are becoming increasingly important. The


“We do not have a good model to predict combustion and turbulence together with soot formation and evolution”


prediction of particle size distribution is a major challenge in the aeroengine industry, where standard prediction tools have massive errors between 200 and 300 per cent. We would like to see how these advanced soot models can predict soot formation at aeroengine-relevant conditions’ The team is making good progress but there are also many experimental considerations to take into account, where the team will eventually need to validate the model with experiments that successfully mimic real-world conditions. Lehmkuhl explained: ‘We must be able to correctly measure the soot formation and distribution in both laminar and turbulent flames, and this is very complex.’ The validation and assessment of the


32 Scientific Computing World Autumn 2020


 Fluid-structure interaction on the surface of a solar panel unit. The wind forces the displacement of the panels, and the risk for failure due to fatigue can be estimated using a coupled CFD and structural mechanics analysis


models will come from the definition of dedicated experiments with an increased level of complexity, going from pre- vaporised counterflow flames, kerosene spray flames up to single sector and full annular rigs. The project includes the development of


reaction mechanisms for the fuel oxidation and soot precursors, and the development of a phenomenological model for primary breakup of airblast atomisers. ‘We have a lot of different physics that we need to put together, and we need to validate as much as is possible,’ Lehmkuhl added.


AI Potential


Could CFD benefit from an AI-enabled upgrade? Polish software specialist byteLake is developing algorithms to optimise existing CFD workflows and lower the cost of such solutions. The byteLake CFD Suite features a


collection of AI models, some replace traditional numerical solvers with equivalent AI models, while others accelerate CFD solvers with AI algorithms. This software suite promises to expedite the analysis process dramatically, providing ‘immediate results’, according to Marcin Rojek, co-founder at byteLake, who added: ‘Artificial intelligence is the next step for CFD, unlocking fast simulations.’ The firm is currently developing its CFD suite using steady-state simulations, before moving to transient simulations. ‘For steady-state, we have not seen any limitations where AI cannot be used,’ Rojek added. The company recently completed its


first accuracy tests with 24 different simulations using OpenFoam. A steady- state icoFoam simulation was carried out with a mesh size of 400 cells. The predicted results achieved a mean


squared error of less than 0.001 and a Pearson correlation coefficient of more than 0.95, with more than 99 per cent relative accuracy. byteLAKE is now investigating larger mesh sizes and accelerating other OpenFoam solvers. Mariusz Kolanko, co-founder at


byteLake, said: ‘The key challenge is to accelerate CFD simulations beyond what is currently possible. We are also working to ensure the AI models are easy to use and compatible with existing workflows.’ The CFD Suite is compatible with


OpenFoam and uses TensorFlow- compatible algorithms to boost its interoperability, where TensorFlow is a leading open source machine learning platform. byteLAKE is working towards providing a plug-and-play interface, where CFD engineers should not need to change data types, replace toolchains, etc, while providing cross-platform compatibility, with no hardware or infrastructure upgrades required to run the software. Ease-of-use is another important


deliverable for the CFD Suite, where engineers should just prepare their input data in the same way that they would for a traditional CFD solver but, instead, send it across to an AI-enhanced CFD model. ‘We want to make a tool that will be


compatible with all the major CFD players and is also easy to use,’ Rojek said. The CFD Suite is scheduled for release in November 2020. While challenges remain for today’s CFD


solutions in terms of usability, validation and performance for a given accuracy, the inclusion of AI technologies is an interesting, new direction. This specialist work and other innovations are now more important than ever to address an increasing range of real-world problems in the CFD space.


@scwmagazine | www.scientific-computing.com


Page 1  |  Page 2  |  Page 3  |  Page 4  |  Page 5  |  Page 6  |  Page 7  |  Page 8  |  Page 9  |  Page 10  |  Page 11  |  Page 12  |  Page 13  |  Page 14  |  Page 15  |  Page 16  |  Page 17  |  Page 18  |  Page 19  |  Page 20  |  Page 21  |  Page 22  |  Page 23  |  Page 24  |  Page 25  |  Page 26  |  Page 27  |  Page 28  |  Page 29  |  Page 30  |  Page 31  |  Page 32  |  Page 33  |  Page 34  |  Page 35  |  Page 36  |  Page 37  |  Page 38