MACHINE LEARNING FOR THE DEVELOPMENT OF DATA DRIVEN TURBULENCE CLOSURES IN COOLANT SYSTEMS

This work shows the application of Gene Expression Programming to augment RANS turbulence closure modelling, for flows through complex geometry designed for additive manufacturing. Specifically, for the design of optimised internal cooling channels in turbine blades. One of the challenges in internal cooling design is the heat transfer accuracy of the RANS formulation in comparison to higher fidelity methods, which are still not used in design on account of their computational cost. However, high fidelity data can be extremely valuable for improving current lower fidelity models and this work shows the application of data driven approaches to develop turbulence closures for an internally ribbed duct. Different approaches are compared and the results of the improved model are illustrated; first on the same geometry, and then for an unseen predictive case. The work shows the potential of using data driven models for accurate heat transfer predictions even in non-conventional configurations and indicates the ability of closures learnt from complex flow cases to adapt successfully to unseen test cases.

Assessment of Machine-Learned Turbulence Models Trained for Improved Wake-Mixing in Low-Pressure Turbine Flows

This paper presents an assessment of machine-learned turbulence closures, trained for improving wake-mixing prediction, in the context of LPT flows. To this end, a three-dimensional cascade of industrial relevance, representative of modern LPT bladings, was analyzed, using a state-of-the-art RANS approach, over a wide range of Reynolds numbers. To ensure that the wake originates from correctly reproduced blade boundary-layers, preliminary analyses were carried out to check for the impact of transition closures, and the best-performing numerical setup was identified. Two different machine-learned closures were considered. They were applied in a prescribed region downstream of the blade trailing edge, excluding the endwall boundary layers. A sensitivity analysis to the distance from the trailing edge at which they are activated is presented in order to assess their applicability to the whole wake affected portion of the computational domain and outside the training region. It is shown how the best-performing closure can provide results in very good agreement with the experimental data in terms of wake loss profiles, with substantial improvements relative to traditional turbulence models. The discussed analysis also provides guidelines for defining an automated zonal application of turbulence closures trained for wake-mixing predictions.

Effect of Turbulence Models in Performance Characterization of a Low Reynolds Number UAV Propeller

The advancement of computer technology has given the necessary impetus to perform numerical modelling and simulation in engineering. Turbulence modelling in Computational Fluid Dynamics is characterized by non-physics based modelling and there are several developments in this area that also has contributed to the growing rise in empiricism. Typically, turbulence models are chosen based on expert knowledge and experience. In this paper, the problem of selecting a turbulence closure is addressed for a small Unmanned Aerial Vehicle propeller rotating at a low Reynolds number. Using scientific approaches, verification and validation of performance data against experimental results have been performed for a selected number of turbulence model candidates available in the well-known finite-volume solver Fluent. Modified bivariate plots of performance data error reveal a few numbers of strong candidates of turbulence closures for this problem. After performing a series of checks for consistency, accuracy and computational cost, the two-equation standard k-ω is selected as the preferred model for further propeller simulations.

Numerical experiments on turbulent entrainment and mixing of scalars

Numerical experiments on the turbulent entrainment and mixing of scalars in a incompressible flow have been performed. These simulations are based on a scale decomposition of the velocity field, thus allowing the establishment from a dynamic point of view of the evolution of scalar fields under the separate action of large-scale coherent motions and small-scale fluctuations. The turbulent spectrum can be split into active and inactive flow structures. The large-scale engulfment phenomena actively prescribe the mixing velocity by amplifying inertial fluxes and by setting the area and the fluctuating geometry of the scalar interface. On the contrary, small-scale isotropic nibbling phenomena are essentially inactive in the mixing process. It is found that the inertial mechanisms initiate the process of entrainment at large scales to be finally processed by scalar diffusion at the molecular level. This last stage does not prescribe the amount of mixing but adapts itself to the conditions imposed by the coherent anisotropic motion at large scales. The present results may have strong repercussions for the theoretical approach to scalar mixing, as anticipated here by simple heuristic arguments which are shown able to reveal the rich dynamics of the process. Interesting repercussions are also envisaged for turbulence closures, in particular for large-eddy simulation approaches where only the large scales of the velocity field are resolved.

INTEGRATION OF MACHINE LEARNING AND COMPUTATIONAL FLUID DYNAMICS TO DEVELOP TURBULENCE MODELS FOR IMPROVED LOW-PRESSURE TURBINE WAKE MIXING PREDICTION

This paper presents the development of accurate turbulence closures for low-pressure turbine (LPT) wake mixing prediction by integrating a machine-learning approach based on gene expression programming (GEP), with Reynolds Averaged Navier-Stokes (RANS) based computational fluid dynamics (CFD). In order to further improve the performance and robustness of GEP-based data-driven closures, the fitness of models is evaluated by running RANS calculations in an integrated way, instead of an algebraic function. Using a canonical turbine wake with inlet conditions prescribed based on high-fidelity data of the T106A cascade, we demonstrate that the ‘CFD-driven’ machine-learning approach produces physically correct non-linear turbulence closures, i.e., predict the right down-stream wake development and maintain an accurate peak wake loss throughout the domain. We then extend our analysis to full turbine blade cases and show that the model development is sensitive to the training region due to the presence of deterministic unsteadiness in the near-wake. Models developed including the near-wake have artificially large diffusion coefficients to over-compensate for the vortex shedding steady RANS cannot capture. In contrast, excluding the near-wake in the model development produces the correct physical model behavior, but predictive accuracy in the near-wake remains unsatisfactory. This can be remedied by using the physically consistent models in unsteady RANS. Overall, the ‘CFD-driven’ models were found to be robust and capture the correct physical wake mixing behavior across different LPT operating conditions and airfoils such as T106C and PakB.