Alok Warey

Seunghwan Keum, Alok Warey

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.

Alok Warey, Rajan Chakravarty

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.