Concept learning constructs visual representations that are connected to linguistic semantics, which is fundamental to vision-language tasks. Although promising progress has been made, existing concept learners are st...Concept learning constructs visual representations that are connected to linguistic semantics, which is fundamental to vision-language tasks. Although promising progress has been made, existing concept learners are still vulnerable to attribute perturbations and out-of-distribution compositions during inference. We ascribe the bottleneck to a failure to explore the intrinsic semantic hierarchy of visual concepts, e.g., {red, blue,···} ∈“color” subspace yet cube ∈“shape”. In this paper, we propose a visual superordinate abstraction framework for explicitly modeling semantic-aware visual subspaces(i.e., visual superordinates). With only natural visual question answering data, our model first acquires the semantic hierarchy from a linguistic view and then explores mutually exclusive visual superordinates under the guidance of linguistic hierarchy. In addition, a quasi-center visual concept clustering and superordinate shortcut learning schemes are proposed to enhance the discrimination and independence of concepts within each visual superordinate. Experiments demonstrate the superiority of the proposed framework under diverse settings, which increases the overall answering accuracy relatively by 7.5% for reasoning with perturbations and 15.6% for compositional generalization tests.展开更多
The transition in the boundary-layer flow affects the hydrodynamic performances of hydraulic machineries,as the key components in the ship propulsion system.The shear stress transfer(SST)γ-Re_(θt) transition model i...The transition in the boundary-layer flow affects the hydrodynamic performances of hydraulic machineries,as the key components in the ship propulsion system.The shear stress transfer(SST)γ-Re_(θt) transition model is an important prediction tool in the boundary layer simulation for hydrofoils.The present paper improves the prediction accuracy of the SST γ-Re_(θt) model for the boundary layers along a curved hydrofoil.The SST γ-Re_(θt) transition model for the flows along a curved hydrofoil is improved by introducing a correction to the transition onset Reynolds number Reθt.First,the transition onset locations for the flows along the hydrofoils of different curvatures are obtained by the large eddy simulation and by using the SST γ-Re_(θt) model.Then,the transition onset Reynolds numbers Reθt in the SST γ-Re_(θt) model is modified to ensure that the predicted boundary layer parameters are consistent with the large eddy simulation(LES)results.The correlation function between the curvature ratio and the modified transition onset Reynolds number is obtained and subsequently used as a correction function in the original SST γ-Re_(θt) model.Three test cases are used to evaluate the performance of the improved SST γ-Re_(θt) model.For the NACA0035 hydrofoil with a large curvature,the predicted results obtained by using the improved SST γ-Re_(θt) model are quite consistent with the experimental data,which indicates the advantages of the improved model in predicting the boundary layer transition along a hydrofoil.In the test cases of the NACA0016 hydrofoil with a mild curvature and the NACA66(mod)-312 hydrofoil,the prediction results of the improved model are in good agreement with the experimental results in terms of the wake region and the boundary layer parameters,which indicates that the improved SST γ-Re_(θt) model can serve as a powerful tool in the design and the optimization of hydraulic machineries such as the waterjet pumps or the naval propellers.展开更多
The internal flow in an axial flow rotating machinery is affected by the rotating characteristics, often accompanied by a strong rotating separation under small flow conditions. At present, the very large eddy simulat...The internal flow in an axial flow rotating machinery is affected by the rotating characteristics, often accompanied by a strong rotating separation under small flow conditions. At present, the very large eddy simulation (VLES) model commonly used for the separation flow simulation still has certain limitations in simulating such rotating separation flow: (1) The Reynolds stress level is overestimated in the near-wall region. (2) The influence of the rotating effect cannot be effectively considered. The above two limitations affect the simulation accuracy of the VLES model for the rotating separation flow under small flow conditions in the axial flow rotating machinery. The objective of this paper is to provide a new hybrid unsteady Reynolds average Navier-Stokes/large eddy simulation (URANS/LES) model suitable for the simulation of the rotating separation flow in an axial flow rotating machinery. Compared with the original VLES method, the modifications are as follows: (1) A Reynolds stress damping function in the near-wall region is introduced to reduce the overestimation of the Reynolds stress caused by the near-wall Reynolds average Navier-Stokes (RANS) behavior of the VLES model. (2) A control function driven by the vortex is introduced to reflect the influence of the rotating effect. Three typical cases are used to verify the calculation accuracy of the modified model. It is shown that the modified model can capture more turbulent vortices based on the URANS grids, and the prediction accuracy of the rotating separation flow is effectively improved. Compared with the original VLES model, the modified model can accurately predict the head change in the hump region of the axial flow pump.展开更多
Rotating separation flow(RSF)in hydraulic machinery is characterized by the large flow separations and complex vortical structures induced by the effects of strong rotation,large curvature and multiple wall surfaces,a...Rotating separation flow(RSF)in hydraulic machinery is characterized by the large flow separations and complex vortical structures induced by the effects of strong rotation,large curvature and multiple wall surfaces,and conducting efficient engineering computation and putting forward effective control strategy for the RSF are important topics in the inner flow theory.To meet these engineering requirements,the studies on computational method and control strategy of the RSF are conducted in this paper.In terms of the computational method,the time-scale-driven(TSD)hybrid unsteady Reynolds-averaged Navier-Stokes/large eddy simulation(URANS/LES)modelling strategy is clarified,and an adaptive TSD hybrid model is established based on the RSF characteristics in hydraulic machinery,thereby avoiding the problem of non-monotonic grid convergence and improving the robustness.Besides,a novel vortex-feature-driven idea suitable for the RSF is further developed inspired by it.In terms of the control strategy,the secondary flow generation mechanism in a rotor domain is revealed,and the relationship between natural secondary flows and blade loading distributions is grasped.On the basis of it,an active control strategy with general significance is proposed,and a general alternate loading technique(GALT)is established.Both aspects can provide generalized paradigms with expandable potential,which are of benefit to the efficient computation and effective control of the RSF in hydraulic machinery.展开更多
A novel Omega(Ω)-driven dynamic partially-averaged Navier-Stokes(PANS)model is proposed in this paper.The ratio of the modeled-to-total turbulent kinetic energies fk is dynamically adjusted by the rigid vorticity rat...A novel Omega(Ω)-driven dynamic partially-averaged Navier-Stokes(PANS)model is proposed in this paper.The ratio of the modeled-to-total turbulent kinetic energies fk is dynamically adjusted by the rigid vorticity ratio(the ratio of the rigid vorticity to the total vorticity),the key parameter of the Ω vortex identification method.Three classical flow cases with rotation and curvature are used to test the model.The results show that the turbulent viscosity is effectively adjusted by the new dynamic fk and the LES-like mode is activated,which can help the revelation of more turbulence information and improve the prediction accuracy.The new PANS model does not contain any explicit dependency on the grid size and enjoys good adaptability to the flow fields,and can be used for efficient engineering computations of the turbulent flows in the hydraulic machinery.展开更多
基金financially supported by the Scientific and Technological Innovation Foundation of Foshan,China (No.BK20BE011)the Fundamental Research Funds for the Central Universities,China (No.FRF-GF-20-10B)。
基金supported in part by the Australian Research Council(ARC)(Nos.FL-170100117,DP-180103424,IC-190100031 and LE-200100049).
文摘Concept learning constructs visual representations that are connected to linguistic semantics, which is fundamental to vision-language tasks. Although promising progress has been made, existing concept learners are still vulnerable to attribute perturbations and out-of-distribution compositions during inference. We ascribe the bottleneck to a failure to explore the intrinsic semantic hierarchy of visual concepts, e.g., {red, blue,···} ∈“color” subspace yet cube ∈“shape”. In this paper, we propose a visual superordinate abstraction framework for explicitly modeling semantic-aware visual subspaces(i.e., visual superordinates). With only natural visual question answering data, our model first acquires the semantic hierarchy from a linguistic view and then explores mutually exclusive visual superordinates under the guidance of linguistic hierarchy. In addition, a quasi-center visual concept clustering and superordinate shortcut learning schemes are proposed to enhance the discrimination and independence of concepts within each visual superordinate. Experiments demonstrate the superiority of the proposed framework under diverse settings, which increases the overall answering accuracy relatively by 7.5% for reasoning with perturbations and 15.6% for compositional generalization tests.
基金Projects supported by the National Science Foundation of China(Grant Nos.51836010,51779258 and 51839001)This work was supported by the Nature Science Foundation of Beijing(Grant No.3182018)the China Scholarship Council(CSC)Fund(Grant No.201806350195).
文摘The transition in the boundary-layer flow affects the hydrodynamic performances of hydraulic machineries,as the key components in the ship propulsion system.The shear stress transfer(SST)γ-Re_(θt) transition model is an important prediction tool in the boundary layer simulation for hydrofoils.The present paper improves the prediction accuracy of the SST γ-Re_(θt) model for the boundary layers along a curved hydrofoil.The SST γ-Re_(θt) transition model for the flows along a curved hydrofoil is improved by introducing a correction to the transition onset Reynolds number Reθt.First,the transition onset locations for the flows along the hydrofoils of different curvatures are obtained by the large eddy simulation and by using the SST γ-Re_(θt) model.Then,the transition onset Reynolds numbers Reθt in the SST γ-Re_(θt) model is modified to ensure that the predicted boundary layer parameters are consistent with the large eddy simulation(LES)results.The correlation function between the curvature ratio and the modified transition onset Reynolds number is obtained and subsequently used as a correction function in the original SST γ-Re_(θt) model.Three test cases are used to evaluate the performance of the improved SST γ-Re_(θt) model.For the NACA0035 hydrofoil with a large curvature,the predicted results obtained by using the improved SST γ-Re_(θt) model are quite consistent with the experimental data,which indicates the advantages of the improved model in predicting the boundary layer transition along a hydrofoil.In the test cases of the NACA0016 hydrofoil with a mild curvature and the NACA66(mod)-312 hydrofoil,the prediction results of the improved model are in good agreement with the experimental results in terms of the wake region and the boundary layer parameters,which indicates that the improved SST γ-Re_(θt) model can serve as a powerful tool in the design and the optimization of hydraulic machineries such as the waterjet pumps or the naval propellers.
基金the National Natural Science Foundation of China(Grant Nos.51836010,51779258).
文摘The internal flow in an axial flow rotating machinery is affected by the rotating characteristics, often accompanied by a strong rotating separation under small flow conditions. At present, the very large eddy simulation (VLES) model commonly used for the separation flow simulation still has certain limitations in simulating such rotating separation flow: (1) The Reynolds stress level is overestimated in the near-wall region. (2) The influence of the rotating effect cannot be effectively considered. The above two limitations affect the simulation accuracy of the VLES model for the rotating separation flow under small flow conditions in the axial flow rotating machinery. The objective of this paper is to provide a new hybrid unsteady Reynolds average Navier-Stokes/large eddy simulation (URANS/LES) model suitable for the simulation of the rotating separation flow in an axial flow rotating machinery. Compared with the original VLES method, the modifications are as follows: (1) A Reynolds stress damping function in the near-wall region is introduced to reduce the overestimation of the Reynolds stress caused by the near-wall Reynolds average Navier-Stokes (RANS) behavior of the VLES model. (2) A control function driven by the vortex is introduced to reflect the influence of the rotating effect. Three typical cases are used to verify the calculation accuracy of the modified model. It is shown that the modified model can capture more turbulent vortices based on the URANS grids, and the prediction accuracy of the rotating separation flow is effectively improved. Compared with the original VLES model, the modified model can accurately predict the head change in the hump region of the axial flow pump.
基金Project supported by the National Natural Science Foundation of China(Grant Nos.51836010,U22A20238 and 52209117)the China Postdoctoral Science Foundation(Grant No.2021M703516).
文摘Rotating separation flow(RSF)in hydraulic machinery is characterized by the large flow separations and complex vortical structures induced by the effects of strong rotation,large curvature and multiple wall surfaces,and conducting efficient engineering computation and putting forward effective control strategy for the RSF are important topics in the inner flow theory.To meet these engineering requirements,the studies on computational method and control strategy of the RSF are conducted in this paper.In terms of the computational method,the time-scale-driven(TSD)hybrid unsteady Reynolds-averaged Navier-Stokes/large eddy simulation(URANS/LES)modelling strategy is clarified,and an adaptive TSD hybrid model is established based on the RSF characteristics in hydraulic machinery,thereby avoiding the problem of non-monotonic grid convergence and improving the robustness.Besides,a novel vortex-feature-driven idea suitable for the RSF is further developed inspired by it.In terms of the control strategy,the secondary flow generation mechanism in a rotor domain is revealed,and the relationship between natural secondary flows and blade loading distributions is grasped.On the basis of it,an active control strategy with general significance is proposed,and a general alternate loading technique(GALT)is established.Both aspects can provide generalized paradigms with expandable potential,which are of benefit to the efficient computation and effective control of the RSF in hydraulic machinery.
基金Supported by the National Natural Science simulating the unsteady eddying motions⑴.Foundation of China(Grant Nos.51836010,51779258 and 51839001)the National Key Research and Development Program of China(Grant No.2018YFB0606103)the Nature Science Foundation of Beijing(Gmat No.3182018).
文摘A novel Omega(Ω)-driven dynamic partially-averaged Navier-Stokes(PANS)model is proposed in this paper.The ratio of the modeled-to-total turbulent kinetic energies fk is dynamically adjusted by the rigid vorticity ratio(the ratio of the rigid vorticity to the total vorticity),the key parameter of the Ω vortex identification method.Three classical flow cases with rotation and curvature are used to test the model.The results show that the turbulent viscosity is effectively adjusted by the new dynamic fk and the LES-like mode is activated,which can help the revelation of more turbulence information and improve the prediction accuracy.The new PANS model does not contain any explicit dependency on the grid size and enjoys good adaptability to the flow fields,and can be used for efficient engineering computations of the turbulent flows in the hydraulic machinery.