Large-eddy simulation of the flow in a direct injection spark ignition engine using an open-source framework

Ribeiro, Mateus Dias; Bimbato, Alex Mendonça; Zanardi, Maria Cecília; Balestieri, José Antônio Perrella; Schmidt, David P.

doi:10.1177/1468087420903622

Cited by 10 publications

(5 citation statements)

References 46 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…A well-resolved and validated Large Eddy Simulation dataset [26] is used with over 10 million data points, taken across four time-steps from a simulation of a compressible flow within the cylinder of an advanced propulsion system generated using OpenFOAM C++ libraries [43]. 20% of the dataset is set aside for a-priori testing.…”

Section: Resultsmentioning

confidence: 99%

“…Since some form of data-fitting is needed to optimize the subgrid scale model parameters even for this simplified approach, one can envisage a purely data-driven method to optimally approximate this changing constant based on large scale resolved terms, motivating our approach. More details about the LES implementation as well as the accuracy of results is reported in [26].…”

Section: Physics Of the Problemmentioning

confidence: 99%

See 1 more Smart Citation

On the Effectiveness of Bayesian AutoML methods for Physics Emulators

Mitra¹,

Santo²,

Haghshenas³

et al. 2020

Preprint

View full text Add to dashboard Cite

The adoption of Machine Learning (ML) for building emulators for complex physical processes has seen an exponential rise in the recent years. While neural networks are good function approximators, optimizing the hyper-parameters of the network to reach a global minimum is not trivial, and often needs human knowl- edge and expertise. In this light, automatic ML or autoML methods have gained large interest as they automate the process of network hyper-parameter tuning. In addition, Neural Architecture Search (NAS) has shown promising outcomes for improving model performance. While autoML methods have grown in popularity for image, text and other applications, their effectiveness for high-dimensional, complex scientific datasets remains to be investigated. In this work, a data driven emulator for turbulence closure terms in the context of Large Eddy Simulation (LES) models is trained using Artificial Neural Networks and an autoML frame- work based on Bayesian Optimization, incorporating priors to jointly optimize the hyper-parameters as well as conduct a full neural network architecture search to converge to a global minima, is proposed. Additionally, we compare the effect of using different network weight initializations and optimizers such as ADAM, SGDM and RMSProp, to explore the best performing setting. Weight and function space similarities during the optimization trajectory are investigated, and critical differences in the learning process evolution are noted and compared to theory. We observe ADAM optimizer and Glorot initialization consistently performs better, while RMSProp outperforms SGDM as the latter appears to have been stuck at a local minima. Therefore, this autoML BayesOpt framework provides a means to choose the best hyper-parameter settings for a given dataset.

show abstract

Section: Resultsmentioning

confidence: 99%

Section: Physics Of the Problemmentioning

confidence: 99%

On the Effectiveness of Bayesian AutoML methods for Physics Emulators

Mitra¹,

Santo²,

Haghshenas³

et al. 2020

Preprint

View full text Add to dashboard Cite

show abstract

“…Following the previously developed zonal methodology, [41][42][43]45 the full computational domain can be divided into several sub-domains with the aim to a priori choose the mode of operation (RANS, DES, LES) in each sub-domain. This is achieved by simple combinations of two Boolean parameters, called C z1 and C z2 , as shown in equations (7,8) and table 1.…”

Section: Consistent Detached Eddy Simulation (Cdes) Formulation and Z...mentioning

confidence: 99%

“…Among the available SRS methodologies, engineering-grade Large Eddy Simulation (LES) has the longest track record for industrial applications 2 and is currently employed for a variety of specific engine development phases, including: generic single-cylinder cold flow multi-cycle analyses, [3][4][5][6][7][8][9] generation of large cold flow numerical data sets with parallel perturbation methods, 10 combustion and combustion noise optimization, [11][12][13][14][15][16][17][18][19] impact of flame kernel growth on combustion instability, 20,21 cycle-resolved abnormal combustion, 22,23 liquid and gaseous fuel injector development. [24][25][26][27][28][29] On the other hand, alternative SRS approaches, based on hybrid URANS/LES formulations, are also becoming popular in the engine modeling community.…”

Section: Introductionmentioning

confidence: 99%

A wall-adapted zonal URANS/LES methodology for the scale-resolving simulation of engine flows

Iacovano

D'Adamo

Fontanesi

et al. 2021

International Journal of Engine Research

View full text Add to dashboard Cite

In the present paper, a comprehensive, wall-adapted zonal URANS/LES methodology is shown for the multidimensional simulation of modern direct-injection engines. This work is the latest update of a zonal hybrid turbulence modeling approach, specifically developed by the authors for a flexible description of in-cylinder turbulent flow features with an optimal balance between computational costs and accuracy. Compared to the previous developments, a specific near-wall treatment is added, in order to preserve full-URANS behavior in the first near-wall cells, having in mind typically available mesh resolution in this part of the fluid domain. The updated methodology is applied to the multi-cycle simulation of a reference single-cylinder optical engine, which features a twin-cam, overhead-valve pent-roof cylinder head, and is representative of the current generation of spark-ignited direct-injection thermal power units. Results based on phase-specific flow field statistics and synthetic quality indices demonstrate the consistency and effectiveness of the proposed methodology, which is then qualified as a suitable candidate for affordable scale-resolving analyses of cycle to cycle variability (CCV) phenomena in direct-injection engines.

show abstract

“…Focusing on gasoline engines, specific consumption reduction is mainly obtained promoting knock mitigation by means of different techniques, such as improved cooling [3], variable compression ratio [4], bore reduction [5], alternative valve strategies [6,7], cooled EGR, water or water/methanol injection [8][9], split injections [10] and increased injection pressures (up to several hundred bar to promote evaporation and mixing) [11]. In this panorama, a key-role is played by CFD simulations whose capabilities to predict knock onset, to understand its origin and to investigate other in-cylinder process (such as complex turbulent flow [12][13][14][15][16], injection [17,19], flame kernel development [20] and combustion [21]) are fundamental to guide design, thus saving costs and time [22][23][24]. Focusing on the field of internal combustion engine research, numerical investigations under motored conditions represent a common practice in order to evaluate the agreement between 1D/3D models and experiments [25].…”

Section: Introductionmentioning

confidence: 99%

A Preliminary 1D-3D Analysis of the Darmstadt Research Engine Under Motored Condition

et al. 2020

View full text Add to dashboard Cite

In the present paper, 1D and 3D CFD models of the Darmstadt research engine undergo a preliminary validation against the available experimental dataset at motored condition. The Darmstadt engine is a single-cylinder optical research unit and the chosen operating point is characterized by a revving speed equal to 800 rpm with intake temperature and pressure of 24 °C and 0.95 bar, respectively. Experimental data are available from the TU Darmstadt engine research group. Several aspects of the engine are analyzed, such as crevice modeling, blow-by, heat transfer and compression ratio, with the aim to minimize numerical uncertainties. On the one hand, a GT-Power model of the engine is used to investigate the impact of blow-by and crevices modeling during compression and expansion strokes. Moreover, it provides boundary conditions for the following 3D CFD simulations. On the other hand, the latter, carried out in a RANS framework with both highand low-Reynolds wall treatments, allow a deeper investigation of the boundary layer phenomena and, thus, of the gas-to-wall heat transfer. A detailed modeling of the crevice, along with an ad hoc tuning of both blow-by and heat fluxes lead to a remarkable improvement of the results. However, in order to adequately match the experimental mean in-cylinder pressure, a slight modification of the compression ratio from the nominal value is accounted for, based on the uncertainty which usually characterizes such geometrical parameter. The present preliminary study aims at providing reliable numerical setups for 1D and 3D models to be adopted in future detailed investigations on the Darmstadt research engine.

show abstract

Large-eddy simulation of the flow in a direct injection spark ignition engine using an open-source framework

Cited by 10 publications

References 46 publications

On the Effectiveness of Bayesian AutoML methods for Physics Emulators

On the Effectiveness of Bayesian AutoML methods for Physics Emulators

A wall-adapted zonal URANS/LES methodology for the scale-resolving simulation of engine flows

A Preliminary 1D-3D Analysis of the Darmstadt Research Engine Under Motored Condition

Contact Info

Product

Resources

About