Mechanical Engineering - Research Publications
Permanent URI for this collection
Search Results
1 - 4 of 4
-
ItemDEVELOPMENT AND USE OF MACHINE-LEARNT ALGEBRAIC REYNOLDS STRESS MODELS FOR ENHANCED PREDICTION OF WAKE MIXING IN LPTS(AMER SOC MECHANICAL ENGINEERS, 2018-01-01)Non-linear turbulence closures were developed that improve the prediction accuracy of wake mixing in low-pressure turbine (LPT) flows. First, Reynolds-averaged Navier-Stokes (RANS) calculations using five linear turbulence closures were performed for the T106A LPT profile at exit Mach number 0.4 and isentropic exit Reynolds numbers 60,000 and 100,000. None of these RANS models were able to accurately reproduce wake loss profiles, a crucial parameter in LPT design, from direct numerical simulation (DNS) reference data. However, the recently proposed kv2w transition model was found to produce the best agreement with DNS data in terms of blade loading and boundary layer behavior and thus was selected as baseline model for turbulence closure development. Analysis of the DNS data revealed that the linear stress-strain coupling constitutes one of the main model form errors. Hence, a geneexpression programming (GEP) based machine-learning technique was applied to the high-fidelity DNS data to train non-linear explicit algebraic Reynolds stress models (EARSM). In particular, the GEP algorithm was tasked to minimize the weighted difference between the DNS and RANS anisotropy tensors, using different training regions. The trained models were first assessed in an a priori sense (without running any CFD) and showed much improved alignment of the trained models in the region of training. Additional RANS calculations were then performed using the trained models. Importantly, to assess their robustness, the trained models were tested both on the cases they were trained for and on testing, i.e. previously not seen, cases with different flow features. The developed models improved prediction of the Reynolds stress, TKE production, wake-loss profiles and wake maturity, across all cases, in particular those trained on just the wake region.
-
ItemMachine Learning for Turbulence Model Development Using a High-Fidelity HPT Cascade Simulation(American Society of Mechanical Engineers, 2017-01-01)The validity of the Boussinesq approximation in the wake behind a high-pressure turbine blade is explored. We probe the mathematical assumptions of such a relationship by employing a least-squares technique. Next, we use an evolutionary algorithm to modify the anisotropy tensor a priori using highly resolved LES data. In the latter case we build a non-linear stress-strain relationship. Results show that the standard eddy-viscosity assumption underpredicts turbulent diffusion and is theoretically invalid. By increasing the coefficient of the linear term, the farwake prediction shows minor improvement. By using additional non-linear terms in the stress-strain coupling relationship, created by the evolutionary algorithm, the near-wake can also be improved upon. Terms created by the algorithm are scrutinized and the discussion is closed by suggesting a tentative non-linear expression for the Reynolds stress, suitable for the wake behind a high-pressure turbine blade.
-
ItemA Comparative Study of Contrasting Machine Learning Frameworks Applied to RANS Modeling of Jets in Crossflow(American Society of Mechanical Engineers, 2017-01-01)Classical RANS turbulence models have known deficiencies when applied to jets in crossflow. Identifying the linear Boussinesq stress-strain hypothesis as a major contribution to erroneous prediction, we consider and contrast two machine learning frameworks for turbulence model development. Gene Expression Programming, an evolutionary algorithm that employs a survival of the fittest analogy, and a Deep Neural Network, based on neurological processing, add non-linear terms to the stress-strain relationship. The results are Explicit Algebraic Stress Model-like closures. High fidelity data from an inline jet in crossflow study is used to regress new closures. These models are then tested on a skewed jet to ascertain their predictive efficacy. For both methodologies, a vast improvement over the linear relationship is observed.
-
ItemApplying Machine Learnt Explicit Algebraic Stress and Scalar Flux Models to a Fundamental Trailing Edge Slot(American Society of Mechanical Engineers, 2018)A form of supervised machine learning was applied to highly resolved large-eddy simulation (LES) data to develop non linear turbulence stress and heat flux closures with increased prediction accuracy for trailing-edge cooling slot cases. The LES data were generated for a thick and a thin trailing-edge slot and shown to agree well with experimental data, thus providing suitable training data for model development. A Gene Expression Programming (GEP) based algorithm was used to symbolically regress novel nonlinear Explicit Algebraic Stress Models (EASM) and heat-flux closures based on either the gradient diffusion or the generalized gradient diffusion approaches. Following a-priori assessment, the new models were used for steady RANS calculations of both thin and thick trailing-edge slot geometries, testing their performance and robustness. Overall, the best agreement with LES data was found when training the RANS model in the near wall region where high levels of anisotropy exist and using the mean squared error of the anisotropy tensor as cost function. In the case of the thin lip geometry, combining an improved EASM model with the standard eddy-diffusivity model predicted the adiabatic wall effectiveness in good agreement with the LES and experimental data. Crucially, the obtained model was also applied to different blowing ratios of the thin lip geometry and a significant improvement in the predictive accuracy of adiabatic wall effectiveness was observed for those cases not previously seen in the training process. For the thick lip case the match with reference values deteriorated due to the presence of large-scale, relative to the slot height, vortex shedding. The machine-learning algorithm was therefore also used to ‘learn’ an appropriate closure for the turbulent heat flux vector. The constructed scalar flux model, in conjunction with a trained RANS model, was found to have the capability to further improve the prediction of the adiabatic wall effectiveness.