arXiv subjects

Alok Warey

Publications and source records attributed to Alok Warey.

4 recordsLinked to original sources

arXiv2026

Adapting Automotive Aerodynamics Surrogates to New Vehicle Families via Transfer Learning

Deploying Scientific Machine Learning surrogates in industrial CFD workflows requires adapting pretrained models to new vehicle families without large datasets; yet whether geometric representations learned by a geometry encoder transfer to topologically distinct shapes remains unvalidated. We address this through leave-one-family-out experiments on a 61.47M-parameter Transformer surrogate (AB-UPT) pretrained on four vehicle families (411 external aerodynamics cases) and adapted to the held-out fifth with only 20 samples. Three strategies are compared: Full Fine-Tuning (FFT), Lightweight Fine-Tuning (LFT), and Low-Rank Adaptation (LoRA). The central finding is that pretrained geometry encoders learn transferable representations, but the adaptation mechanism determines whether they can be exploited. FFT destabilizes as 61.47M unconstrained parameters overfit to 20 samples (R^2=0.40); LFT fails because the frozen encoder cannot represent unseen shapes (R^2<0). LoRA resolves both: rank-constrained adapters injected into all layers regularize the loss landscape while preserving pretrained features, achieving R^2=0.85+/-0.02 across all five families with 50% lower force RMSE than FFT and 28% lower pointwise field errors. LoRA also outperforms from-scratch training using 3x more target-family data, eliminating the need for large per-family datasets. These results recast LoRA from a memory-saving convenience into a convergence enabler for geometry transfer: a shared backbone paired with lightweight per-family adapters trainable in hours from minimal data.

cs.CE↗

arXiv2025

Validation of Quantum Computing for Transition Metal Oxide-based Automotive Catalysis

Quantum computing presents a promising alternative to classical computational methods for modeling strongly correlated materials with partially filled d orbitals. In this study, we perform a comprehensive quantum resource estimation using quantum phase estimation (QPE) and qubitization techniques for transition metal oxide molecules and a Pd zeolite catalyst fragment. Using the binary oxide molecules TiO, MnO, and FeO, we validate our active space selection and benchmarking methodology, employing classical multireference methods such as complete active space self-consistent field (CASSCF) and N-electron valence state perturbation theory (NEVPT2). We then apply these methods to estimate the quantum resources required for a full-scale quantum simulation of a $Z_2Pd$ ($Z=Al_2Si_{22}O_{48}$) fragment taken from the $Pd/2(Al_xSi_{(1-x)})$ catalyst family where x=Si/Al. Our analysis demonstrates that for large Pd zeolite systems, simulations achieving chemical accuracy would require ~$10^6-10^7$ physical qubits, and range that is consistent with the projected capabilities of future fault-tolerant quantum devices. We further explore the impact of active space size, basis set quality, and phase estimation error on the required qubit and gate counts. These findings provide a roadmap for near-term and future quantum simulations of industrially relevant catalytic materials, offering insights into the feasibility and scaling of quantum chemistry applications in materials science.

physics.chem-ph↗

arXiv2022

Classification of Computer Aided Engineering (CAE) Parts Using Graph Convolutional Networks

CAE engineers work with hundreds of parts spread across multiple body models. A Graph Convolutional Network (GCN) was used to develop a CAE parts classifier. As many as 866 distinct parts from a representative body model were used as training data. The parts were represented as a three-dimensional (3-D) Finite Element Analysis (FEA) mesh with values of each node in the x, y, z coordinate system. The GCN based classifier was compared to fully connected neural network and PointNet based models. Performance of the trained models was evaluated with a test set that included parts from the training data, but with additional holes, rotation, translation, mesh refinement/coarsening, variation of mesh schema, mirroring along x and y axes, variation of topographical features, and change in mesh node ordering. The trained GCN model was able to achieve 88.5% classification accuracy on the test set i.e., it was able to find the correct matching part from the dataset of 866 parts despite significant variation from the baseline part. A CAE parts classifier demonstrated in this study could be very useful for engineers to filter through CAE parts spread across several body models to find parts that meet their requirements.

cs.LG↗

arXiv2021

Investigation of Numerical Diffusion in Aerodynamic Flow Simulations with Physics Informed Neural Networks

Computational Fluid Dynamics (CFD) simulations are used for many air flow simulations including road vehicle aerodynamics. Numerical diffusion occurs when local flow direction is not aligned with the mesh lines and when there is a non-zero gradient of the dependent variable in the direction normal to the streamline direction. It has been observed that typical numerical discretization schemes for the Navier-Stokes equations such as first order upwinding produce very accurate solutions without numerical diffusion when the mesh is aligned with the streamline directions. On the other hand, numerical diffusion is maximized when the streamline direction is at an angle of 45{\deg} relative to the mesh line. The amount of numerical diffusion can be reduced by mesh refinements such as aligning mesh lines along the local flow direction or by introducing higher order numerical schemes which may introduce potential numerical instability or additional computational cost. Couple test cases of a simple steady-state incompressible and inviscid air flow convection problem were used to investigate whether numerical diffusion occurs when using Physics Informed Neural Networks (PINNs) that rely on automatic differentiation as opposed to numerical techniques used in traditional Computational Fluid Dynamics (CFD) solvers. Numerical diffusion was not observed when PINNs were used to solve the Partial Differential Equation (PDE) for the simple convection problem irrespective of flow angle. The PINN correctly simulated the streamwise upwinding, which has great potential to improve the accuracy of Navier-Stokes solvers.

physics.flu-dyn↗