Pub Date : 2022-06-10DOI: 10.1080/10618562.2022.2113520
R. McConkey, E. Yee, F. Lien
Several machine learning frameworks for augmenting turbulence closure models have been recently proposed. However, the generalizability of an augmented turbulence model remains an open question. We investigate this question by systematically varying the training and test sets of several models. An optimal three-term tensor basis expansion is used to develop a model-agnostic data-driven turbulence closure approximation. Then, hyperparameter optimization is performed for a random forest, a neural network, and an eXtreme Gradient Boosting (XGBoost) model. We recommend XGBoost for data-driven turbulence closure modelling owing to its low-tuning cost and good performance. We also find that machine learning models generalize well to new parametric variations of flows seen in the training dataset, but lack generalizability to new flow types. This generalizability gap suggests that machine learning methods are most suited for developing specialized models for a given flow type, a problem often encountered in industrial applications.
{"title":"On the Generalizability of Machine-Learning-Assisted Anisotropy Mappings for Predictive Turbulence Modelling","authors":"R. McConkey, E. Yee, F. Lien","doi":"10.1080/10618562.2022.2113520","DOIUrl":"https://doi.org/10.1080/10618562.2022.2113520","url":null,"abstract":"Several machine learning frameworks for augmenting turbulence closure models have been recently proposed. However, the generalizability of an augmented turbulence model remains an open question. We investigate this question by systematically varying the training and test sets of several models. An optimal three-term tensor basis expansion is used to develop a model-agnostic data-driven turbulence closure approximation. Then, hyperparameter optimization is performed for a random forest, a neural network, and an eXtreme Gradient Boosting (XGBoost) model. We recommend XGBoost for data-driven turbulence closure modelling owing to its low-tuning cost and good performance. We also find that machine learning models generalize well to new parametric variations of flows seen in the training dataset, but lack generalizability to new flow types. This generalizability gap suggests that machine learning methods are most suited for developing specialized models for a given flow type, a problem often encountered in industrial applications.","PeriodicalId":56288,"journal":{"name":"International Journal of Computational Fluid Dynamics","volume":"14 1","pages":"555 - 577"},"PeriodicalIF":1.3,"publicationDate":"2022-06-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"72726940","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-05-28DOI: 10.1080/10618562.2022.2138361
K. Mizuno, M. Asahara, T. Kamiya, T. Miyasaka
For the direct simulation of liquid atomization processes affected by surface tension, the level-set method was suitable, owing to the ease of estimating normal vectors and interface curvatures. However, the mass is not conserved in the original method because of errors in the interfacial advection evaluation. Therefore, previous studies have developed the interfacial advection equation and the interfacial evaluation method for reinitialization. The reinitialization method can be organized in terms of control parameters. Therefore, we modified and optimized the control parameters of the high-order constrained reinitialization scheme developed previously. Furthermore, the difference method in the convection term was adapted to the interfacial advection equation to evaluate the best method for simulating the droplet atomization process. Consequently, the fifth-order accuracy weighted essentially non-oscillatory scheme was observed to provide the best mass conservation for both the reinitialization and the advection equations.
{"title":"Finite Difference and Reinitialization Methods with Level Set to Interfacial Area Transport Equations for Gas–Liquid Two-Phase Flows","authors":"K. Mizuno, M. Asahara, T. Kamiya, T. Miyasaka","doi":"10.1080/10618562.2022.2138361","DOIUrl":"https://doi.org/10.1080/10618562.2022.2138361","url":null,"abstract":"For the direct simulation of liquid atomization processes affected by surface tension, the level-set method was suitable, owing to the ease of estimating normal vectors and interface curvatures. However, the mass is not conserved in the original method because of errors in the interfacial advection evaluation. Therefore, previous studies have developed the interfacial advection equation and the interfacial evaluation method for reinitialization. The reinitialization method can be organized in terms of control parameters. Therefore, we modified and optimized the control parameters of the high-order constrained reinitialization scheme developed previously. Furthermore, the difference method in the convection term was adapted to the interfacial advection equation to evaluate the best method for simulating the droplet atomization process. Consequently, the fifth-order accuracy weighted essentially non-oscillatory scheme was observed to provide the best mass conservation for both the reinitialization and the advection equations.","PeriodicalId":56288,"journal":{"name":"International Journal of Computational Fluid Dynamics","volume":"32 1","pages":"361 - 383"},"PeriodicalIF":1.3,"publicationDate":"2022-05-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"89179548","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-05-28DOI: 10.1080/10618562.2022.2127695
P. Senel, M. Tezer-Sezgin
ABSTRACT Magnetohydrodynamics (MHD) flow in cavities subjected to both the magnetisation and the Lorentz forces due to a point magnetic source is studied. The governing PDEs are derived and iteratively solved by the dual reciprocity boundary elements method (DRBEM) with linear elements. It is shown that the magnetic field decelerates the axial flow around the point magnetic source, and a further increase in Ha causes a reverse flow in the pipe axis direction. An increase in Re, Ha, or Mn reduces the electric potential in magnitude. The planar velocity values decrease at the same rate as the Re increment. The influence of the magnetisation force lessens in high Re cases without alternating the axial velocity and the electric potential within the pipe. This study is the first to give the effects of both the magnetisation and the Lorentz forces on the fluid behaviour in terms of velocity, pressure, and electric potential.
{"title":"Electrically Conducting Fluid Flow and Electric Potential in a Square Cavity Subjected to a Point Magnetic Source","authors":"P. Senel, M. Tezer-Sezgin","doi":"10.1080/10618562.2022.2127695","DOIUrl":"https://doi.org/10.1080/10618562.2022.2127695","url":null,"abstract":"ABSTRACT Magnetohydrodynamics (MHD) flow in cavities subjected to both the magnetisation and the Lorentz forces due to a point magnetic source is studied. The governing PDEs are derived and iteratively solved by the dual reciprocity boundary elements method (DRBEM) with linear elements. It is shown that the magnetic field decelerates the axial flow around the point magnetic source, and a further increase in Ha causes a reverse flow in the pipe axis direction. An increase in Re, Ha, or Mn reduces the electric potential in magnitude. The planar velocity values decrease at the same rate as the Re increment. The influence of the magnetisation force lessens in high Re cases without alternating the axial velocity and the electric potential within the pipe. This study is the first to give the effects of both the magnetisation and the Lorentz forces on the fluid behaviour in terms of velocity, pressure, and electric potential.","PeriodicalId":56288,"journal":{"name":"International Journal of Computational Fluid Dynamics","volume":"17 1","pages":"424 - 439"},"PeriodicalIF":1.3,"publicationDate":"2022-05-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"84818613","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-05-28DOI: 10.1080/10618562.2022.2123108
Surabhi Nishad, R. Bhargava
The current paper is aimed to study entropy generation analysis and natural convective flow of -water based nanofluid filled inside wavy solar collector with right-angled triangular cross sectional area. The flow is assumed to be steady, laminar and incompressible with double diffusive convection. The set of governing equations are the conservation of mass, momentum, energy and concentration which have been solved by implementing the Parallel approach of Element free Galerkin method using parallelism of graphic processor units (GPUs) involving six physical cores (12 logical cores) of Intel® Xeon® E5645 CPUs at 2.40 GHz. The effect of various parameters namely Dufour effect, Soret effect and Schmidt number on the isotherms, iso-concentration, streamlines, average Nusselt number and Sherwood number have been studied. The results show that the waviness of the slant cover plate results in perturbation in the flow which has significant effect on the heat transfer rate because of the increased surface area. The parallel approach of the meshfree technique has improved the computational efficiency which is a novel contribution.
{"title":"Entropy Generation Parallel Analysis and Optimised Efficiency of a Wavy Solar Collector","authors":"Surabhi Nishad, R. Bhargava","doi":"10.1080/10618562.2022.2123108","DOIUrl":"https://doi.org/10.1080/10618562.2022.2123108","url":null,"abstract":"The current paper is aimed to study entropy generation analysis and natural convective flow of -water based nanofluid filled inside wavy solar collector with right-angled triangular cross sectional area. The flow is assumed to be steady, laminar and incompressible with double diffusive convection. The set of governing equations are the conservation of mass, momentum, energy and concentration which have been solved by implementing the Parallel approach of Element free Galerkin method using parallelism of graphic processor units (GPUs) involving six physical cores (12 logical cores) of Intel® Xeon® E5645 CPUs at 2.40 GHz. The effect of various parameters namely Dufour effect, Soret effect and Schmidt number on the isotherms, iso-concentration, streamlines, average Nusselt number and Sherwood number have been studied. The results show that the waviness of the slant cover plate results in perturbation in the flow which has significant effect on the heat transfer rate because of the increased surface area. The parallel approach of the meshfree technique has improved the computational efficiency which is a novel contribution.","PeriodicalId":56288,"journal":{"name":"International Journal of Computational Fluid Dynamics","volume":"338 1","pages":"384 - 403"},"PeriodicalIF":1.3,"publicationDate":"2022-05-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"80678833","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-05-28DOI: 10.1080/10618562.2022.2146676
Songxiang Tang, Jie Li, Zi-Qiao Wei
The transonic characteristics of a flow over a hemispherical turret with a freestream Mach number M∞ of 0.8 are numerically studied. The flow field is simulated using the improved delayed detached eddy simulation (IDDES) with a modified sub-grid scale. The results using the adopted medium and fine grids, including for the mean pressure coefficients and the incoming boundary layer, match experimental data very well. It is observed that the wake fluctuates periodically as the shock moves upstream and downstream. The dominant flow characteristics are identified along with their frequencies. The first two proper orthogonal decomposition (POD) modes show a lateral shift of the wake, fluctuating at a Strouhal number (St) of 0.15–0.2. Modes 3 and 4 show a wall-normal fluctuation of the wake with St of 0.35. A dynamic mode decomposition (DMD) gives St values of the lateral shift and wall-normal fluctuation of 0.1813 and 0.3467, respectively.
{"title":"A Numerical Investigation of the Dominant Characteristics of A Transonic Flow Over A Hemispherical Turret","authors":"Songxiang Tang, Jie Li, Zi-Qiao Wei","doi":"10.1080/10618562.2022.2146676","DOIUrl":"https://doi.org/10.1080/10618562.2022.2146676","url":null,"abstract":"The transonic characteristics of a flow over a hemispherical turret with a freestream Mach number M∞ of 0.8 are numerically studied. The flow field is simulated using the improved delayed detached eddy simulation (IDDES) with a modified sub-grid scale. The results using the adopted medium and fine grids, including for the mean pressure coefficients and the incoming boundary layer, match experimental data very well. It is observed that the wake fluctuates periodically as the shock moves upstream and downstream. The dominant flow characteristics are identified along with their frequencies. The first two proper orthogonal decomposition (POD) modes show a lateral shift of the wake, fluctuating at a Strouhal number (St) of 0.15–0.2. Modes 3 and 4 show a wall-normal fluctuation of the wake with St of 0.35. A dynamic mode decomposition (DMD) gives St values of the lateral shift and wall-normal fluctuation of 0.1813 and 0.3467, respectively.","PeriodicalId":56288,"journal":{"name":"International Journal of Computational Fluid Dynamics","volume":"39 1","pages":"404 - 423"},"PeriodicalIF":1.3,"publicationDate":"2022-05-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"74580334","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-04-21DOI: 10.1080/10618562.2022.2108020
A. Amicarelli, Emanuela Abbate, A. Frigerio
The 3D SPH code SPHERA (RSE SpA) simulates the triggering and runout stages of a full-scale earth-filled dike failure with coastal flood. The code formulation is consistent with the Kinetic Theory of Granular Flow KTGF and requires no tuned input. The main Landslide Sliding Surface LSS within the dike is detected from the field of the quadratic invariant of the strain-rate tensor. The comparison with the experimental LSS position shows Maximum Gross Errors of 0.90m, in terms of depth difference at fixed planimetric point, and of 0.41m, in terms of LSS distance (i.e. 22% and 10% of the initial dike-top height over the excavation bottom). Inter-comparisons consider calibrated FEM results for the triggering stage. The LSS detection criterion is defined and verified via analytical solutions and code validations on spherical Couette flows. The impact on a virtual electricity pylon is assessed. The code improvements and tutorials are available at github.com.
{"title":"SPH Modelling of a Dike Failure with Detection of the Landslide Sliding Surface and Damage Scenarios for an Electricity Pylon","authors":"A. Amicarelli, Emanuela Abbate, A. Frigerio","doi":"10.1080/10618562.2022.2108020","DOIUrl":"https://doi.org/10.1080/10618562.2022.2108020","url":null,"abstract":"The 3D SPH code SPHERA (RSE SpA) simulates the triggering and runout stages of a full-scale earth-filled dike failure with coastal flood. The code formulation is consistent with the Kinetic Theory of Granular Flow KTGF and requires no tuned input. The main Landslide Sliding Surface LSS within the dike is detected from the field of the quadratic invariant of the strain-rate tensor. The comparison with the experimental LSS position shows Maximum Gross Errors of 0.90m, in terms of depth difference at fixed planimetric point, and of 0.41m, in terms of LSS distance (i.e. 22% and 10% of the initial dike-top height over the excavation bottom). Inter-comparisons consider calibrated FEM results for the triggering stage. The LSS detection criterion is defined and verified via analytical solutions and code validations on spherical Couette flows. The impact on a virtual electricity pylon is assessed. The code improvements and tutorials are available at github.com.","PeriodicalId":56288,"journal":{"name":"International Journal of Computational Fluid Dynamics","volume":"11 1","pages":"265 - 293"},"PeriodicalIF":1.3,"publicationDate":"2022-04-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"90716659","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-04-21DOI: 10.1080/10618562.2022.2104262
Tianbai Xiao
ABSTRACT Considerable uncertainties can exist between the field solutions of coarse-grained fluid models and the real-world flow physics. To study the emergence, propagation, and evolution of uncertainties poses great opportunities and challenges to develop both sound theories and reliable numerical methods. In this paper, we study the stochastic behaviour of multi-scale gaseous flows from molecular to hydrodynamic level, especially focussing on the non-equilibrium effects. The theoretical analysis is presented based on the gas kinetic model and its upscaling macroscopic system with random inputs. A newly developed stochastic kinetic scheme is employed to conduct numerical simulation of multi-scale and non-equilibrium flows. Different kinds of uncertainties are involved in the gas evolutionary processes. New physical observations, such as the synchronous travel pattern between mean fields and uncertainties, sensitivity of different orders of uncertainties and the influence of boundary effects from continuum to rarefied regimes, are identified and analysed.
{"title":"An Investigation of Uncertainty Propagation in Non-equilibrium Flows","authors":"Tianbai Xiao","doi":"10.1080/10618562.2022.2104262","DOIUrl":"https://doi.org/10.1080/10618562.2022.2104262","url":null,"abstract":"ABSTRACT Considerable uncertainties can exist between the field solutions of coarse-grained fluid models and the real-world flow physics. To study the emergence, propagation, and evolution of uncertainties poses great opportunities and challenges to develop both sound theories and reliable numerical methods. In this paper, we study the stochastic behaviour of multi-scale gaseous flows from molecular to hydrodynamic level, especially focussing on the non-equilibrium effects. The theoretical analysis is presented based on the gas kinetic model and its upscaling macroscopic system with random inputs. A newly developed stochastic kinetic scheme is employed to conduct numerical simulation of multi-scale and non-equilibrium flows. Different kinds of uncertainties are involved in the gas evolutionary processes. New physical observations, such as the synchronous travel pattern between mean fields and uncertainties, sensitivity of different orders of uncertainties and the influence of boundary effects from continuum to rarefied regimes, are identified and analysed.","PeriodicalId":56288,"journal":{"name":"International Journal of Computational Fluid Dynamics","volume":"49 1","pages":"294 - 318"},"PeriodicalIF":1.3,"publicationDate":"2022-04-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"81466186","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-04-21DOI: 10.1080/10618562.2022.2108807
Yuchen Yang, Xinyu Qi, Zhenming Wang, Jianming Liu, N. Zhao
In this paper, a set of parallelised adaptive hierarchical Cartesian-based immersed boundary methodology is developed for high Reynolds number compressible flow. First, a robust and efficient grid generation method based on the separation axis theorem for arbitrary geometry is presented for automatic Cartesian grid generation. Second, an immersed boundary method (IBM) is presented coupling with wall model for high Reynolds number flow. Third, a parallel strategy is implemented and special treatment is proposed to guarantee large-scale computing. Finally, cell-based adaptive mesh refinement (AMR) technology is used to capture fluid phenomena under different regimes including but not limited to shock waves and vortices. The overall performance of the methodology is tested through a wide range of regimes including transonic and supersonic flow with high Reynolds number in both two and three dimensions. Results are in good agreement with reference data and indicate the capability and robustness of the present methodology.
{"title":"An Immersed Boundary Method Based on Parallel Adaptive Cartesian Grids for High Reynolds Number Turbulent Flow","authors":"Yuchen Yang, Xinyu Qi, Zhenming Wang, Jianming Liu, N. Zhao","doi":"10.1080/10618562.2022.2108807","DOIUrl":"https://doi.org/10.1080/10618562.2022.2108807","url":null,"abstract":"In this paper, a set of parallelised adaptive hierarchical Cartesian-based immersed boundary methodology is developed for high Reynolds number compressible flow. First, a robust and efficient grid generation method based on the separation axis theorem for arbitrary geometry is presented for automatic Cartesian grid generation. Second, an immersed boundary method (IBM) is presented coupling with wall model for high Reynolds number flow. Third, a parallel strategy is implemented and special treatment is proposed to guarantee large-scale computing. Finally, cell-based adaptive mesh refinement (AMR) technology is used to capture fluid phenomena under different regimes including but not limited to shock waves and vortices. The overall performance of the methodology is tested through a wide range of regimes including transonic and supersonic flow with high Reynolds number in both two and three dimensions. Results are in good agreement with reference data and indicate the capability and robustness of the present methodology.","PeriodicalId":56288,"journal":{"name":"International Journal of Computational Fluid Dynamics","volume":"1 1","pages":"319 - 341"},"PeriodicalIF":1.3,"publicationDate":"2022-04-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"82974629","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-04-21DOI: 10.1080/10618562.2022.2107201
Sergio Peralta, Jhon Córdova, Cesar Celis, D. Maza
The main parallelisation related features of a computational tool based on the finite element method (FEM) for the numerical modelling of mineral-slurry like flows are described in this work. In particular, both the domain decomposition method (DDM) and the processes communication strategy employed are discussed in detail. The DD algorithm is based on the iterative update of the boundary conditions imposed on the interfaces between subdomains, the so-called transmission conditions. Due to its versatility in several parallel architectures, the message-passing standard used here is the message passing interface (MPI) one. Since mineral-slurries rheology may change according to the prevailing local flow conditions, Newtonian and non-Newtonian viscous fluids are considered in this work. Indeed, both Newtonian and non-Newtonian laminar flows are numerically studied in two well-known canonical configurations usually found in mineral-slurry transport. The main results show that the parallel FEM based tool is capable of carrying out high-fidelity numerical simulations of mineral-slurry like flows. Finally, in all numerical simulations performed, relatively good speedups were obtained.
{"title":"Parallel Domain Decomposition of a FEM-based Tool for Numerical Modelling Mineral Slurry-like Flows","authors":"Sergio Peralta, Jhon Córdova, Cesar Celis, D. Maza","doi":"10.1080/10618562.2022.2107201","DOIUrl":"https://doi.org/10.1080/10618562.2022.2107201","url":null,"abstract":"The main parallelisation related features of a computational tool based on the finite element method (FEM) for the numerical modelling of mineral-slurry like flows are described in this work. In particular, both the domain decomposition method (DDM) and the processes communication strategy employed are discussed in detail. The DD algorithm is based on the iterative update of the boundary conditions imposed on the interfaces between subdomains, the so-called transmission conditions. Due to its versatility in several parallel architectures, the message-passing standard used here is the message passing interface (MPI) one. Since mineral-slurries rheology may change according to the prevailing local flow conditions, Newtonian and non-Newtonian viscous fluids are considered in this work. Indeed, both Newtonian and non-Newtonian laminar flows are numerically studied in two well-known canonical configurations usually found in mineral-slurry transport. The main results show that the parallel FEM based tool is capable of carrying out high-fidelity numerical simulations of mineral-slurry like flows. Finally, in all numerical simulations performed, relatively good speedups were obtained.","PeriodicalId":56288,"journal":{"name":"International Journal of Computational Fluid Dynamics","volume":"10 1","pages":"342 - 360"},"PeriodicalIF":1.3,"publicationDate":"2022-04-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"84985670","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-03-27DOI: 10.1080/10618562.2022.2049258
Matteo Ugolotti, Benjamin Vaughan, P. Orkwis
In gradient-based aerodynamic optimisation, the functional gradient required by the optimiser can be obtained as a product of adjoint-based functional sensitivities to volume grid nodes and the volume mesh sensitivities to the design parameters. For turbomachinery applications, it is desirable to use the actual blade design variables as degrees of freedom for the optimisation process, but this can lead to tedious programming tasks. As an alternative, a Machine Learning (ML) model is created to mimic and differentiate the blade geometry and the mesh generation processes. The typical ML forward pass is followed by a back differentiation operation enabling the computation of the volume mesh derivatives with respect to design parameters. The model is tested by comparing the ML-predicted and reference grid, and the modelled sensitivities are verified through algorithmic differentiation. The modelled mesh sensitivities are successfully employed for adjoint-based reverse engineering and design optimisation problems on a turbine blade.
{"title":"A Conceptual Alternative Machine Learning-Based Method for Mesh Sensitivities Calculation in a Turbomachinery Blades Optimisation Framework","authors":"Matteo Ugolotti, Benjamin Vaughan, P. Orkwis","doi":"10.1080/10618562.2022.2049258","DOIUrl":"https://doi.org/10.1080/10618562.2022.2049258","url":null,"abstract":"In gradient-based aerodynamic optimisation, the functional gradient required by the optimiser can be obtained as a product of adjoint-based functional sensitivities to volume grid nodes and the volume mesh sensitivities to the design parameters. For turbomachinery applications, it is desirable to use the actual blade design variables as degrees of freedom for the optimisation process, but this can lead to tedious programming tasks. As an alternative, a Machine Learning (ML) model is created to mimic and differentiate the blade geometry and the mesh generation processes. The typical ML forward pass is followed by a back differentiation operation enabling the computation of the volume mesh derivatives with respect to design parameters. The model is tested by comparing the ML-predicted and reference grid, and the modelled sensitivities are verified through algorithmic differentiation. The modelled mesh sensitivities are successfully employed for adjoint-based reverse engineering and design optimisation problems on a turbine blade.","PeriodicalId":56288,"journal":{"name":"International Journal of Computational Fluid Dynamics","volume":"17 1","pages":"520 - 537"},"PeriodicalIF":1.3,"publicationDate":"2022-03-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"73957642","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：