Data-driven prediction of energy loss and heat transfer over rough surfaces

  • contact:

    Jiasheng Yang, Alexander Stroh

  • funding:

    Friedrich und Elisabeth Boysen-Stiftung (BOY-151)

  • Partner:

    Prof. Dr.-Ing Willy Dörfler

  • startdate:


Turbulent flows over rough surfaces cause higher amounts of energy loss (frictional loss) and surface heat transfer compared to smooth surfaces. In a variety of industrial applications, e.g. turbomachinery, maritime transport and aviation, prediction of this effect is a key to an optimized designed and enhanced energy-efficiency. The predictive correlations currently available in the literature have limited ranges of applicability and cannot deliver reliable predictions for any ‘new’ rough surface due to the highly non-linear mapping of roughness effects. Therefore, the aim of the present project is to develop an ‘universal’ predictive tool by means of Machine Learning (ML) which is inherently suitable for this non-linear multi-variate system. The machine learning model is trained using a database generated through Direct Numerical Simulation (DNS) of fluid flows over mathematically generated rough surfaces. In addition, this project explores the physics revealed by the machine learning model in order to cultivate trust in the model as well as to convey an approachable message to users.