Qiang He
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Qiang He has a total of 66 co-authors. Here are the top co-authors by number of shared publications:
Weifeng Huang
15 shared publicationsYang Hu
12 shared publicationsDecai Li
9 shared publicationsGreg Zhu
8 shared publications- Weifeng Huang
Tsinghua University
5 shared publications - Yang Hu
School of Mechanical, Electronic and Control Engineering, Beijing Jiaotong University 1 , Beijing 100044, China
5 shared publications
Latest Publications
- Open Access
- 8 September 2026
- Published by Elsevier in SSRN Electronic Journal
Algebraic Loss Augmentation: Steering Conflicting Residuals in Physics-Informed Neural Networks for Coupled Partial Differential Equations
Physics-informed neural networks (PINNs) provide a powerful framework for solving partial differential equations (PDEs), yet their training often suffers from gradient conflicts in coupled systems. We propose Algebraic Loss Augmentation (ALA), a lightweight regularization method that constructs auxiliary constraints through algebraic combinations of the original PDEs. These terms generate gradient directions consistent with the original residuals while amplifying their magnitude, effectively guiding the optimization toward physically consistent solutions. ALA mitigates the decoupling between loss minimization and solution error, reducing error without additional computational cost. We introduce a systematic approach for constructing such auxiliary terms and demonstrate the method on different PDEs, including the Navier–Stokes and Gray–Scott equations. By steering the optimization along algebraically compatible directions, ALA promotes convergence to high-fidelity solutions, providing a robust and efficient tool for scientific machine learning.
- 8 September 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessAlgebraic Loss Augmentation: Steering Conflicting Residuals in Physics-Informed Neural Networks for Coupled Partial Differential Equations
Physics-informed neural networks (PINNs) provide a powerful framework for solving partial differential equations (PDEs), yet their training often suffers from gradient conflicts in coupled systems. We propose Algebraic Loss Augmentation (ALA), a lightweight regularization method that constructs auxiliary constraints through algebraic combinations of the original PDEs. These terms generate gradient directions consistent with the original residuals while amplifying their magnitude, effectively guiding the optimization toward physically consistent solutions. ALA mitigates the decoupling between loss minimization and solution error, reducing error without additional computational cost. We introduce a systematic approach for constructing such auxiliary terms and demonstrate the method on different PDEs, including the Navier–Stokes and Gray–Scott equations. By steering the optimization along algebraically compatible directions, ALA promotes convergence to high-fidelity solutions, providing a robust and efficient tool for scientific machine learning.
- Open Access
- 6 September 2026
- Published by Elsevier in Materials & Design
- Vol. 270
Non-equilibrium Partitioning-Driven interface stabilization and microstructure transition in k > 1 Ti–Nb alloys during laser additive manufacturing
- 6 September 2026
- Published by Elsevier in Materials & Design
- Vol. 270
Open AccessNon-equilibrium Partitioning-Driven interface stabilization and microstructure transition in k > 1 Ti–Nb alloys during laser additive manufacturing
- Open Access
- 23 July 2026
- Published by Elsevier in SSRN Electronic Journal
Multi-Dimensional Training-Priority Weighting Based on Physical Information Propagation Paths: A Unified Residual-Weighting Framework for Physics-Informed Neural Networks
Physics-informed neural networks (PINNs) have shown considerable promise for solving partial differential equations (PDEs); however, their synchronous optimization mechanism treats the residuals of different regions and different constraints with equal weight and in parallel, which is often inconsistent with the process by which the solution forms progressively ”from source to response” along the physical information propagation path, thereby degrading training stability and solution accuracy. Existing causal training methods focus mainly on the temporal dimension and lack a unified characterization of the spatial and boundary dimensions. To address this, we define a unified class of training priorities according to the propagation path of physical information: along the PDE’s physical information propagation path, premise regions should be learned before the regions that depend on them; the temporal, spatial, and boundary priorities are precisely instances of this principle on different propagation paths. On this basis, using neural tangent kernel (NTK) training dynamics, we theoretically analyze why the standard PINN does not obey this priority: its residual convergence order is governed by the NTK spectrum and is independent of the propagation path, and the spectral bias does not actively favor premise regions, so it does not establish a definite convergence order. Accordingly, we propose a unified multi-dimensional priority-constraint framework that, by partitioning the domain in order along the propagation path and constructing negative-exponential residual weights, converts the physical propagation order into a training priority at the loss level. For cases in which multiple priorities coexist, we introduce a directional compatibility coefficient to clarify the applicability boundary that ”orthogonal directions can be coupled multiplicatively in synergy, whereas coaxial opposite directions cannot.” Multiple benchmark cases show that the proposed method consistently improves the convergence behavior and prediction accuracy of PINNs on problems with a clear propagation path or a constraint-dominated structure, without modifying the network architecture and with controllable additional computational cost.
- 23 July 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessMulti-Dimensional Training-Priority Weighting Based on Physical Information Propagation Paths: A Unified Residual-Weighting Framework for Physics-Informed Neural Networks
Physics-informed neural networks (PINNs) have shown considerable promise for solving partial differential equations (PDEs); however, their synchronous optimization mechanism treats the residuals of different regions and different constraints with equal weight and in parallel, which is often inconsistent with the process by which the solution forms progressively ”from source to response” along the physical information propagation path, thereby degrading training stability and solution accuracy. Existing causal training methods focus mainly on the temporal dimension and lack a unified characterization of the spatial and boundary dimensions. To address this, we define a unified class of training priorities according to the propagation path of physical information: along the PDE’s physical information propagation path, premise regions should be learned before the regions that depend on them; the temporal, spatial, and boundary priorities are precisely instances of this principle on different propagation paths. On this basis, using neural tangent kernel (NTK) training dynamics, we theoretically analyze why the standard PINN does not obey this priority: its residual convergence order is governed by the NTK spectrum and is independent of the propagation path, and the spectral bias does not actively favor premise regions, so it does not establish a definite convergence order. Accordingly, we propose a unified multi-dimensional priority-constraint framework that, by partitioning the domain in order along the propagation path and constructing negative-exponential residual weights, converts the physical propagation order into a training priority at the loss level. For cases in which multiple priorities coexist, we introduce a directional compatibility coefficient to clarify the applicability boundary that ”orthogonal directions can be coupled multiplicatively in synergy, whereas coaxial opposite directions cannot.” Multiple benchmark cases show that the proposed method consistently improves the convergence behavior and prediction accuracy of PINNs on problems with a clear propagation path or a constraint-dominated structure, without modifying the network architecture and with controllable additional computational cost.
- Open Access
- 21 May 2026
- Published by Elsevier in SSRN Electronic Journal
Cavitation suppression in shallow-groove mechanical face seals using a novel downstream deep recess design
Cavitation occurring in shallow grooves on mechanical seal faces can significantly affect sealing stability, load-carrying capacity, and dynamic characteristics. In this study, cavitation phenomena in shallow-groove mechanical seals are investigated, taking fan-shaped shallow grooves as a representative configuration. A novel cavitation suppression strategy is proposed by introducing a deep recess at the leeward side of the shallow groove. A numerical model based on the finite volume method is established, in which a homogeneous boiling cavitation model is employed to simulate vapor–liquid phase transition within the seal film. The pressure distribution, vapor volume fraction, and overall sealing performance are systematically compared between configurations with and without the recess structure. To verify the numerical predictions, a visualized mechanical seal test rig is employed to directly observe vapor–liquid phase distributions within the grooves. The results indicate that the recess effectively suppresses cavitation by providing a buffer space for vapor accumulation and by promoting pressure recovery in the downstream region of the groove. A systematic parametric study reveals that the geometric parameters of the recess, particularly its length, play a critical role in determining cavitation behavior and sealing performance.Under optimal conditions, the opening force increases by approximately two times, while the fluid film stiffness is enhanced by up to eight times, indicating a significant improvement in system stability. The introduction of the recess also alleviates the negative stiffness phenomenon that may occur at low rotational speeds due to cavitation, thereby improving the operational stability of the mechanical seal.
- 21 May 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessCavitation suppression in shallow-groove mechanical face seals using a novel downstream deep recess design
Cavitation occurring in shallow grooves on mechanical seal faces can significantly affect sealing stability, load-carrying capacity, and dynamic characteristics. In this study, cavitation phenomena in shallow-groove mechanical seals are investigated, taking fan-shaped shallow grooves as a representative configuration. A novel cavitation suppression strategy is proposed by introducing a deep recess at the leeward side of the shallow groove. A numerical model based on the finite volume method is established, in which a homogeneous boiling cavitation model is employed to simulate vapor–liquid phase transition within the seal film. The pressure distribution, vapor volume fraction, and overall sealing performance are systematically compared between configurations with and without the recess structure. To verify the numerical predictions, a visualized mechanical seal test rig is employed to directly observe vapor–liquid phase distributions within the grooves. The results indicate that the recess effectively suppresses cavitation by providing a buffer space for vapor accumulation and by promoting pressure recovery in the downstream region of the groove. A systematic parametric study reveals that the geometric parameters of the recess, particularly its length, play a critical role in determining cavitation behavior and sealing performance.Under optimal conditions, the opening force increases by approximately two times, while the fluid film stiffness is enhanced by up to eight times, indicating a significant improvement in system stability. The introduction of the recess also alleviates the negative stiffness phenomenon that may occur at low rotational speeds due to cavitation, thereby improving the operational stability of the mechanical seal.
- 15 May 2026
- Published by American Chemical Society (ACS) in Langmuir
- Vol. 42 (21) , 14658-14672
Three-Dimensional Lattice Boltzmann Simulations of Droplet Wetting States on Bi-Gaussian Layered Surfaces
Real engineering surfaces commonly exhibit multiscale, hierarchically layered roughness; however, the independent contributions of the upper and lower roughness components in bi-Gaussian layered surfaces (bi-Gaussian surfaces), as well as their coupled effects on wetting behavior, remain insufficiently understood. In this study, a numerical investigation of droplet wetting behavior on bi-Gaussian surfaces is conducted based on bi-Gaussian surface theory. Gaussian and bi-Gaussian surfaces with well-controlled statistical characteristics are generated using a moving average–fast Fourier transform (MA–FFT) algorithm. A color-gradient lattice Boltzmann multiphase model is then employed to systematically simulate the quasi-static wetting dynamics of droplets, with particular emphasis on elucidating how morphological parameters of bi-Gaussian surfaces govern wetting behavior. The results indicate that, for Gaussian-distributed rough surfaces (Gaussian surfaces), droplet wetting is predominantly governed by the global root-mean-square roughness Sq, exhibiting a clear monotonic dependence. In contrast, for bi-Gaussian surfaces, the predictive capability of the global Sq deteriorates significantly. Further decoupled analysis reveals a triple-dominant mechanism underlying layered wetting behavior. The interlayer separation Zu serves as the primary driving factor that induces the transition of the effective supporting layer from the lower substrate to the upper roughness layer, thereby controlling wetting state transitions. The spiky morphology of the upper Gaussian structure provides the dominant mechanical support for the droplet, while the roughness of the lower Gaussian layer regulates the sensitivity of the wetting transition. Specifically, the lower-layer roughness modulates the sensitivity of the apparent contact angle to the interlayer separation by altering the amplitude of substrate height fluctuations. These findings clarify the wetting behavior of rough surfaces commonly encountered in engineering applications and provide fundamental guidance for understanding complex wetting phenomena on real surfaces, as well as for the rational design of surfaces with tailored wetting properties.
- 15 May 2026
- Published by American Chemical Society (ACS) in Langmuir
- Vol. 42 (21) , 14658-14672
Three-Dimensional Lattice Boltzmann Simulations of Droplet Wetting States on Bi-Gaussian Layered Surfaces
Real engineering surfaces commonly exhibit multiscale, hierarchically layered roughness; however, the independent contributions of the upper and lower roughness components in bi-Gaussian layered surfaces (bi-Gaussian surfaces), as well as their coupled effects on wetting behavior, remain insufficiently understood. In this study, a numerical investigation of droplet wetting behavior on bi-Gaussian surfaces is conducted based on bi-Gaussian surface theory. Gaussian and bi-Gaussian surfaces with well-controlled statistical characteristics are generated using a moving average–fast Fourier transform (MA–FFT) algorithm. A color-gradient lattice Boltzmann multiphase model is then employed to systematically simulate the quasi-static wetting dynamics of droplets, with particular emphasis on elucidating how morphological parameters of bi-Gaussian surfaces govern wetting behavior. The results indicate that, for Gaussian-distributed rough surfaces (Gaussian surfaces), droplet wetting is predominantly governed by the global root-mean-square roughness Sq, exhibiting a clear monotonic dependence. In contrast, for bi-Gaussian surfaces, the predictive capability of the global Sq deteriorates significantly. Further decoupled analysis reveals a triple-dominant mechanism underlying layered wetting behavior. The interlayer separation Zu serves as the primary driving factor that induces the transition of the effective supporting layer from the lower substrate to the upper roughness layer, thereby controlling wetting state transitions. The spiky morphology of the upper Gaussian structure provides the dominant mechanical support for the droplet, while the roughness of the lower Gaussian layer regulates the sensitivity of the wetting transition. Specifically, the lower-layer roughness modulates the sensitivity of the apparent contact angle to the interlayer separation by altering the amplitude of substrate height fluctuations. These findings clarify the wetting behavior of rough surfaces commonly encountered in engineering applications and provide fundamental guidance for understanding complex wetting phenomena on real surfaces, as well as for the rational design of surfaces with tailored wetting properties.
- Open Access
- 13 April 2026
- Published by Elsevier in SSRN Electronic Journal
Algebraic Loss Augmentation: Steering Conflicting Residuals for Coupled Physics-Informed Neural Networks
Physics-informed neural networks (PINNs) provide a powerful framework for solving partial differential equations (PDEs), yet their training often suffers from gradient conflicts in coupled systems. We propose Algebraic Loss Augmentation (ALA), a lightweight regularization method that constructs auxiliary constraints through algebraic combinations of the original PDEs. These terms generate gradient directions consistent with the original residuals while amplifying their magnitude, effectively guiding the optimization toward physically consistent solutions. ALA mitigates the decoupling between loss minimization and solution error, reducing error without additional computational cost. We introduce a systematic approach for constructing such auxiliary terms and demonstrate the method on different PDEs, including the Navier–Stokes and Gray–Scott equations. By steering the optimization along algebraically compatible directions, ALA promotes convergence to high-fidelity solutions, providing a robust and efficient tool for scientific machine learning.
- 13 April 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessAlgebraic Loss Augmentation: Steering Conflicting Residuals for Coupled Physics-Informed Neural Networks
Physics-informed neural networks (PINNs) provide a powerful framework for solving partial differential equations (PDEs), yet their training often suffers from gradient conflicts in coupled systems. We propose Algebraic Loss Augmentation (ALA), a lightweight regularization method that constructs auxiliary constraints through algebraic combinations of the original PDEs. These terms generate gradient directions consistent with the original residuals while amplifying their magnitude, effectively guiding the optimization toward physically consistent solutions. ALA mitigates the decoupling between loss minimization and solution error, reducing error without additional computational cost. We introduce a systematic approach for constructing such auxiliary terms and demonstrate the method on different PDEs, including the Navier–Stokes and Gray–Scott equations. By steering the optimization along algebraically compatible directions, ALA promotes convergence to high-fidelity solutions, providing a robust and efficient tool for scientific machine learning.
- 1 February 2026
- Published by AIP Publishing in Physics of Fluids
- Vol. 38 (2)
Numerical analysis of cavitation bubble dynamics at the mechanical seal interface
This study introduces a combined model to explain the microscopic mechanisms that control cavitation in mechanical seals. The model merges a macroscopic analysis of fluid lubrication using the Reynolds equation with a microscopic model of single-bubble dynamics governed by the Gilmore equation. This integration allows for an examination of bubble behavior within the flow field of the fan-shaped groove mechanical seal. An exploration of how cavitation intensity is affected by seal geometric parameters and operational conditions from a micro-bubble dynamics perspective was carried out. The findings indicate that pressure fluctuations cause intense volumetric oscillations in bubbles, leading to the emission of energy in the form of acoustic pressure waves. It is also shown that increasing rotational speed and decreasing groove depth significantly amplify local pressure gradients and flow velocities, thereby enhancing the likelihood of cavitation. The notable cavitation phenomena arise due to the combined effects of shallow groove depths and high rotational speeds. This study establishes a theoretical basis for optimizing mechanical seal design and managing cavitation effects.
- 1 February 2026
- Published by AIP Publishing in Physics of Fluids
- Vol. 38 (2)
Numerical analysis of cavitation bubble dynamics at the mechanical seal interface
This study introduces a combined model to explain the microscopic mechanisms that control cavitation in mechanical seals. The model merges a macroscopic analysis of fluid lubrication using the Reynolds equation with a microscopic model of single-bubble dynamics governed by the Gilmore equation. This integration allows for an examination of bubble behavior within the flow field of the fan-shaped groove mechanical seal. An exploration of how cavitation intensity is affected by seal geometric parameters and operational conditions from a micro-bubble dynamics perspective was carried out. The findings indicate that pressure fluctuations cause intense volumetric oscillations in bubbles, leading to the emission of energy in the form of acoustic pressure waves. It is also shown that increasing rotational speed and decreasing groove depth significantly amplify local pressure gradients and flow velocities, thereby enhancing the likelihood of cavitation. The notable cavitation phenomena arise due to the combined effects of shallow groove depths and high rotational speeds. This study establishes a theoretical basis for optimizing mechanical seal design and managing cavitation effects.
- 1 December 2025
- Published by AIP Publishing in Physics of Fluids
- Vol. 37 (12)
Sharp-enforced physics-informed neural network for inverse hydrodynamic lubrication problem with film thickness discontinuity
Thin film lubrication is crucial in hydrodynamic lubrication, and film thickness measurement is essential for assessing lubrication conditions, optimizing tribological interfaces, and extending the service life of mechanical components. However, directly quantifying film thickness under high-pressure conditions remains a significant challenge. This study proposes an inverse approach based on Physics-Informed Neural Networks to reconstruct film thickness using sparse pressure data. Forward and inverse models based on Kolmogorov–Arnold Networks (KAN) and Multi-Layer Perceptron (MLP) are developed, and their solving performance is compared. MLP-based model and KAN-based model exhibit significant advantages in addressing lubrication problems. In solving the forward problem, both models achieve an accuracy on the order of 10−4 or lower. However, KAN-based model attains higher accuracy with fewer neurons and fewer iterations. Furthermore, both frameworks successfully reconstructed the full-field pressure distribution and accurately predicted the film thickness from sparse pressure measurements. It was found that data from different regions significantly affect the accuracy of film thickness prediction. The study revealed that data from different regions significantly affect the accuracy of film thickness predictions. When the input data originate from within the discontinuous interface, the network fails to provide an accurate solution. Data points located directly on the interface result in the lowest film thickness error, with a prediction error on the order of 0.02%. For data points outside the interface, the film thickness prediction error increases with distance from the interface. However, the error in the pressure field remains comparable to that obtained from points on the interface, also remaining around 0.04%. In conclusion, the hybrid-driven method proposed in this paper, which combines limited data with physical information, provides a new approach for film thickness measurement.
- 1 December 2025
- Published by AIP Publishing in Physics of Fluids
- Vol. 37 (12)
Sharp-enforced physics-informed neural network for inverse hydrodynamic lubrication problem with film thickness discontinuity
Thin film lubrication is crucial in hydrodynamic lubrication, and film thickness measurement is essential for assessing lubrication conditions, optimizing tribological interfaces, and extending the service life of mechanical components. However, directly quantifying film thickness under high-pressure conditions remains a significant challenge. This study proposes an inverse approach based on Physics-Informed Neural Networks to reconstruct film thickness using sparse pressure data. Forward and inverse models based on Kolmogorov–Arnold Networks (KAN) and Multi-Layer Perceptron (MLP) are developed, and their solving performance is compared. MLP-based model and KAN-based model exhibit significant advantages in addressing lubrication problems. In solving the forward problem, both models achieve an accuracy on the order of 10−4 or lower. However, KAN-based model attains higher accuracy with fewer neurons and fewer iterations. Furthermore, both frameworks successfully reconstructed the full-field pressure distribution and accurately predicted the film thickness from sparse pressure measurements. It was found that data from different regions significantly affect the accuracy of film thickness prediction. The study revealed that data from different regions significantly affect the accuracy of film thickness predictions. When the input data originate from within the discontinuous interface, the network fails to provide an accurate solution. Data points located directly on the interface result in the lowest film thickness error, with a prediction error on the order of 0.02%. For data points outside the interface, the film thickness prediction error increases with distance from the interface. However, the error in the pressure field remains comparable to that obtained from points on the interface, also remaining around 0.04%. In conclusion, the hybrid-driven method proposed in this paper, which combines limited data with physical information, provides a new approach for film thickness measurement.
- 1 September 2025
- Published by AIP Publishing in Physics of Fluids
- Vol. 37 (9)
Phase field-lattice Boltzmann model for two-phase flows with near-contact interactions
This study presents a phase field-lattice Boltzmann method for modeling two-phase flows with near-contact interactions. The second-order conservative phase-field equation is coupled with the velocity-based Navier–Stokes equations for phase interface tracking and flow field prediction, respectively. To account for the near-contact interactions between droplets or bubbles in two-phase flows, a continuum repulsive force model is introduced, which is inversely proportional to the cubic power of the interfacial distance. A double-population multiple-relaxation-time lattice Boltzmann scheme is proposed to solve the governing equations. The model is validated through several numerical examples involving near-contact interactions. First, a static concentric droplet problem is quantitatively analyzed, in which the pressure jumps across the phase interfaces due to surface tension and repulsive forces are captured accurately. It can be seen clearly that the maximum relative errors in most cases are below 5% and still meet the requirements under high density ratio. Next, simulations of droplet motion in shear flow and bubble rise under gravity are conducted to investigate the non-coalescence mechanisms of droplets and bubbles in two-phase flows. Finally, as an application example of the present model, the generation of micro-droplets in a microchannel with a T-junction is simulated. The numerical results demonstrate that the coalescence of droplets and bubbles can be effectively prevented under the influence of the repulsive force, and thus, the validity of proposed model is verified.
- 1 September 2025
- Published by AIP Publishing in Physics of Fluids
- Vol. 37 (9)
Phase field-lattice Boltzmann model for two-phase flows with near-contact interactions
This study presents a phase field-lattice Boltzmann method for modeling two-phase flows with near-contact interactions. The second-order conservative phase-field equation is coupled with the velocity-based Navier–Stokes equations for phase interface tracking and flow field prediction, respectively. To account for the near-contact interactions between droplets or bubbles in two-phase flows, a continuum repulsive force model is introduced, which is inversely proportional to the cubic power of the interfacial distance. A double-population multiple-relaxation-time lattice Boltzmann scheme is proposed to solve the governing equations. The model is validated through several numerical examples involving near-contact interactions. First, a static concentric droplet problem is quantitatively analyzed, in which the pressure jumps across the phase interfaces due to surface tension and repulsive forces are captured accurately. It can be seen clearly that the maximum relative errors in most cases are below 5% and still meet the requirements under high density ratio. Next, simulations of droplet motion in shear flow and bubble rise under gravity are conducted to investigate the non-coalescence mechanisms of droplets and bubbles in two-phase flows. Finally, as an application example of the present model, the generation of micro-droplets in a microchannel with a T-junction is simulated. The numerical results demonstrate that the coalescence of droplets and bubbles can be effectively prevented under the influence of the repulsive force, and thus, the validity of proposed model is verified.