Field inversion for data-augmented RANS modelling in turbomachinery flows

Ferrero, Andrea; Iollo, Angelo; Larocca, Francesco

doi:10.1016/j.compfluid.2020.104474

Cited by 31 publications

(17 citation statements)

References 39 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…It is possible to enrich the training database by performing the inversion procedure for several working conditions and then joining the databases in order to explore a larger range of input variables and increase the predictive capability of the final data-augmented model. An application of this approach to turbomachinery flows was investigated by Ferrero et al [17]. In the present work the machine learning analysis is not discussed.…”

Section: Machine Learning From Correction Fieldmentioning

confidence: 92%

Uncertainty propagation in field inversion for turbulence modelling in turbomachinery

Ferrero

Larocca

Pennecchi

2020

2020 IEEE 7th International Workshop on Metrology for AeroSpace (MetroAeroSpace)

Self Cite

View full text Add to dashboard Cite

The simulation of turbulent flows in turbomachinery requires to describe a wide range of scales and non-linear phenomena. Since the cost of scale resolving simulations is prohibitive for several configurations, turbulence closure models are still widely used in the framework of Reynolds-averaged Navier-Stokes (RANS) equations. In order to improve the prediction capability of these models, several machine learning strategies have been proposed. Among them, the field inversion approach allows to find a correction field which can be applied to the source term of the turbulence model in order to match experimental data: the correction field can then be generalised and expressed as a function of some flow features in order to extract modelling knowledge from the data. However, the reference experimental data are affected by uncertainty and this propagates to the correction field and to the final data-augmented model. In this work, the uncertainty propagation from the reference experimental data to the correction field is investigated. In particular, the flow field around a low pressure gas turbine cascade is studied in a challenging working condition characterised by laminar separation and transition to turbulence. The original RANS results are improved by the application of the field inversion algorithm in which the required gradients are computed by means of an adjoint approach. A sensitivity analysis is performed in order to provide a linearised propagation of the uncertainty from the experimental wall isentropic Mach number to the correction field.

show abstract

Section: Machine Learning From Correction Fieldmentioning

confidence: 92%

Uncertainty propagation in field inversion for turbulence modelling in turbomachinery

Ferrero

Larocca

Pennecchi

2020

2020 IEEE 7th International Workshop on Metrology for AeroSpace (MetroAeroSpace)

Self Cite

View full text Add to dashboard Cite

show abstract

“…These features along with the augmentation values from different field inversion solutions which are all obtained using the same augmented formulation of the low-fidelity model are then used to obtain optimal ML-model parameters to establish a functional relationship between the features and the augmentation term. This formulation of FIML, hereafter referred to as the classic FIML, has been used by several research groups, with applications including, but not limited to, predictive modeling of adverse pressure gradients flows [23,24], separated flows [19,[25][26][27], bypass transition modeling [28], natural transition modeling [29], hypersonic aerothermal prediction for aerothermoelastic analysis [30], turbomachinery flows [31], shock-turbulent boundary layer interactions [32], etc. Matai et al [24] proposed a zonal version of FIML, where the augmentation field obtained from field inversion was quantized into a set number of clusters, following which a decision tree based architecture was used to classify corresponding features into appropriate clusters.…”

Section: A Field Inversion and Machine Learning (Fiml)mentioning

confidence: 99%

Generalizable Physics-constrained Modeling using Learning and Inference assisted by Feature Space Engineering

Srivastava¹,

Duraisamy²

2021

Preprint

View full text Add to dashboard Cite

This work presents a formalism to improve the predictive accuracy of physical models by learning generalizable augmentations from sparse data. Building on recent advances in data-driven turbulence modeling, the present approach, referred to as Learning and Inference assisted by Feature Engineering (LIFE), is based on the hypothesis that robustness and generalizability demand a meticulously-designed feature space that is informed by the underlying physics, and a carefully constructed features-to-augmentation map. The critical components of this approach are : (1) Identification of relevant physics-informed features in appropriate functional forms to enable significant overlap in feature space for a wide variety of cases to promote generalizability; (2) Explicit control over feature space to locally infer the augmentation without affecting other feature space regions, especially when limited data is available; (3) Maintaining consistency across the learning and prediction environments to make the augmentation case-agnostic; (4) Tightly-coupled inference and learning by constraining the augmentation to be learnable throughout the inference process to avoid significant loss of information (and hence accuracy). To demonstrate the viability of this approach, it is used in the modeling of bypass transition. The augmentation is developed on skin friction data from two flat plate cases from the ERCOFTAC dataset. The augmented model is then applied to a variety of flat plate cases which are characterized by different freestream turbulence intensities, pressure gradients, and Reynolds numbers. The predictive capability of the augmented model is also tested on single-stage high-pressure-turbine cascade cases, and the model performance is analyzed from the perspective of information contained in the feature space. The results show consistent improvements across these cases, as long as the physical phenomena in question are well-represented in the training.

show abstract

“…The SA model was chosen for this work because in the considered test case it provides results which are in the range spanned by more complex turbulence models (like for example the SST k − ω [30] or the k − [31]) as will be shown in Section 5. Furthermore, the SA model was succesfully used in the simulation of aerospace propulsion systems [32] and can be augmented by data-driven corrections to improve its predictive ability [33]. The inlet boundary condition for the SA model is set toν/ν = 3, according to the recommendations of Spalart and Rumsey [34] for high Reynolds number flows.…”

Section: Compressible Reynolds-averaged Navier-stokes (Rans) Equationsmentioning

confidence: 99%

Control of a Supersonic Inlet in Off-Design Conditions with Plasma Actuators and Bleed

Ferrero

2020

Aerospace

Self Cite

View full text Add to dashboard Cite

Supersonic inlets are a key component of present and future air-breathing propulsion systems for high-speed flight. The inlet design is challenging because of several phenomena that must be taken under control: shock waves, boundary layer separation and unsteadiness. Furthermore, the intensity of these phenomena is strongly influenced by the working conditions and so active control systems can be particularly useful in off-design conditions. In this work, a mixed compression supersonic inlet with a double wedge ramp is considered. The flow field was numerically investigated at different values of Mach number. The simulations show that large separations appear at the higher Mach numbers on both the upper and lower walls of the duct. In order to improve the performances of the inlet two different control strategies were investigated: plasma actuators and bleed. Different locations of the plasma actuator are considered in order to also apply this technology to configurations with a diverter which prevents boundary layer ingestion. The potential of the proposed solutions is investigated in terms of total pressure recovery, flow distortion and power consumption.

show abstract

Field inversion for data-augmented RANS modelling in turbomachinery flows

Cited by 31 publications

References 39 publications

Uncertainty propagation in field inversion for turbulence modelling in turbomachinery

Uncertainty propagation in field inversion for turbulence modelling in turbomachinery

Generalizable Physics-constrained Modeling using Learning and Inference assisted by Feature Space Engineering

Control of a Supersonic Inlet in Off-Design Conditions with Plasma Actuators and Bleed

Contact Info

Product

Resources

About