Searcharxiv⌕ Search

arXiv subjects

Md. Abdul Aziz

Publications and source records attributed to Md. Abdul Aziz.

3 recordsLinked to original sources

Data-driven Correction Reduced-Order Stochastic Collocation Method for the Time-Dependent Navier-Stokes Equations

We develop two hybrid data-driven Uncertainty Quantification (UQ) Reduced-Order Models (ROMs) for the non-stationary incompressible Navier--Stokes Equations (NSEs) with random viscosity. The proposed framework combines Proper Orthogonal Decomposition (POD) ROMs, spatial filtering, data-driven closure modeling, and the Stochastic Collocation Method (SCM). A key feature is the approximation of the realization-dependent closure term by the closure associated with the expected viscosity. Consequently, a single data-driven closure model constructed from the full-order solution corresponding to the expected viscosity can be reused across all realizations, thereby avoiding the expensive construction of realization-specific closure models. This leads to the Data-Driven Correction UQ ROM (DDC-UQ-ROM) and its physically constrained variant, the Constrained DDC-UQ-ROM (CDDC-UQ-ROM), in which physical constraints are imposed on the data-driven correction model. The proposed methods are evaluated for a two-dimensional (2D) flow past a circular cylinder problem with a 5D random viscosity field, considering mean Reynolds numbers $Re=100$ and $Re=500$, different correlation lengths, and both in-sample and out-of-sample parameter settings. Numerical results show that the DDC-UQ-ROM generally improves upon the standard Galerkin ROM (G-ROM), while the CDDC-UQ-ROM exhibits the most accurate and robust approximation of the benchmark energy evolution across the cases considered. The robustness and accuracy of the CDDC-UQ-ROM are further assessed through statistical error analyses of the first- and second-order energy moments, reduced-training-data experiments, and an indicative comparison of wall-clock computational costs with the full-order UQ calculations.

physics.flu-dyn↗

Self-adaptive physics-informed neural network for forward and inverse problems in heterogeneous porous flow

We develop a self-adaptive physics-informed neural network (PINN) framework that reliably solves forward Darcy flow and performs accurate permeability inversion in heterogeneous porous media. In the forward setting, the PINN predicts velocity and pressure for discontinuous, piecewise-constant permeability; in the inverse setting, it identifies spatially varying permeability directly from indirect flow observations. Both models use a region-aware permeability parameterization with binary spatial masks, which preserves sharp permeability jumps and avoids the smoothing artifacts common in standard PINNs. To stabilize training, we introduce self-learned loss weights that automatically balance PDE residuals, boundary constraints, and data mismatch, eliminating manual tuning and improving robustness, particularly for inverse problems. An interleaved AdamW-L-BFGS optimization strategy further accelerates and stabilizes convergence. Numerical results demonstrate accurate forward surrogates and reliable inverse permeability recovery, establishing the method as an effective mesh-free solver and data-driven inversion tool for porous-media systems governed by partial differential equations.

physics.flu-dyn↗

Efficient, Accurate, and Robust Penalty-Projection Algorithm for Parameterized Stochastic Navier-Stokes Flow Problems

This paper presents and analyzes a fast, robust, efficient, and optimally accurate fully discrete splitting algorithm for the Uncertainty Quantification (UQ) of parameterized Stochastic Navier-Stokes Equations (SNSEs) flow problems those occur in the convection-dominated regimes. The time-stepping algorithm is an implicit backward-Euler linearized method, grad-div and Ensemble Eddy Viscosity (EEV) regularized, and split using discrete Hodge decomposition. Additionally, the scheme's sub-problems are all designed to have different Right-Hand-Side (RHS) vectors but the same system matrix for all realizations at each time-step. The stability of the algorithm is rigorously proven, and it has been shown that appropriately large grad-div stabilization parameters vanish the splitting error. The proposed UQ algorithm is then combined with the Stochastic Collocation Methods (SCMs). Several numerical experiments are given to verify this superior scheme's predicted convergence rates and performance on benchmark problems for high expected Reynolds numbers ($Re$).

math.NA↗