Suraj Bhardwaj

Implicit 3D model generation

• Implemented an embedding function to perform vector quantization and learned a codebook on the bottleneck of the encoder network of the NDF model to collect feature information in both continuous and discrete modes
• Trained the modified NDF model on the processed Shapenet Cars dataset on the GPU partition of the OMNI cluster for five days
• Generated and evaluated the 3D model using generation metrics, and latent diffusion models.

Performance evaluation of Resnet18 and ConvNeXt-tiny on Out of Distribution Robustness using AugMix

• Added Resnet18 and ConvNeXt-tiny model using torchvision library, adamW optimizer, cosine annealing learning rate scheduler using torch.optim and python in the original Augmix code
• Trained variants of “Resnet18 and ConvNeXt-tiny” models on the CIFAR-10 dataset
• Loaded the saved models from the workspace created on the Omni cluster of Uni-Siegen respectively and evaluated them on CIFAR- 10, CIFAR-10-C, and CIFAR-10-P datasets
• Used fundamental knowledge of Transfer Learning and Hyperparameter tuning on the ConvNeXt-tiny model to beat the OOD Robustness performance of the Resnet18 pretrained model.

• Led a team of four, on a fully funded project entitled “Design, Development, and Testing of a Novel Multi-Stage Parabolic Solar Collector" towards its successful completion.

Exerimental Investigations on Performance Evaluation of two-stage Solar Receiver System

• Designed a novel double-stage receiver system and fabricated an experimental facility to meet steam parameters pressure, mass flow rate, and temperature for power generation
• Determined the receiver's collection efficiency in terms of water temperature and observed the double-stage receiver system to be 11.646% more efficient than the single-stage system at a flow rate of 1.3 liters per minute

Improvement in 9S (9-Speed) Gear Box Assembly Line Direct Pass

• Identified factors causing rejection of assembled gearbox and suggested remedies which improved the efficiency of direct pass operation of 9-speed Gearbox Assembly Line by 1.5% (96.235% to 97.825%).