From CFD Results to Machine Learning Models

Artificial Intelligence (AI) and Machine Learning (ML) are rapidly becoming more and more important for Computer-Aided Engineering (CAE): AI and ML affect the methods and tools used for generating results as well as the postprocessing of these results. In Computational Fluid Dynamics (CFD) AI may be used for pre-processing geometry and generating meshes, to optimize solvers, or to replace slower and unhandy physical models [1]. Using the results computed by CFD as an input to ML, similar data sets can be generated. In this article we will focus on postprocessing CFD results and investigate what additional value can be generated by ML methods.

The application we will use as an example is the impingement of multiple jets as encountered in convective drying applications. The data used was generated by CFD (StarCCM and openFoam) and originally used to derive Nusselt-number correlations for the heat transfer and to optimize the energy efficiency. This data will now be used and postprocessed by ML methods.

The analysis in this article focuses on the application of ML methods, thus “generic methods” are used. A challenge for the application of AI is the comparative scarcity of input data as the CFD computations generating the data are computationally expensive.

T​his article appeared in the January 2023 issue of BENCHMARK