Research Projects
Physics-Based Radiography Data Simulator Powered by GPU for Non-Destructive Evaluation Applications
A CAD component with artificial defects inserted is placed in between the virtual flat panel X-ray detector (cyan color) and point X-ray source (red color on the right most side of image). The X-rays are projected from the source to the detector.
Developed a physics based radiography data simulator for NDE applications powered by GPU.
The simulator models the experimental scenario to generate radiograph using any CAD component and is capable of inserting desired defects in the component.
It can automatically produce large amount of data, which can replace the experimental data for the training of Deep Learning models.
We proposed a unique noise modelling technique which significantly improves the image texture and the histograms.
A part of this work has been published in the Journal of Nondestructive Evaluation 2021 (JNDE), DOI: 10.1007/s10921-021-00750-4
GPU Based Raycasting Operation To Calculate Distance Travelled by Rays Inside the Surface Triangulated CAD Component
GPU powered ray cast operation to calculate the distance traveled by the rays inside the CAD geometry with triangulated surfaces.
Implemented in both C++ and python.
Upto 16x speedup compared to multi-threaded CPU implementation.
The time in the above curve is for the processing of 2048*2048 rays.
High-throughput and real-time feature extraction of crystal microstructures from high resolution TEM data.
We extended a graph based crystal detection framework focusing on performance optimization using efficient tree based data structures for near real time crystal characterization in TEM data.
The approach extracts a large corpus of crystal features in high throughput manner using HPC nodes.
We also propose a Wasserstein distance based data sufficiency determination method.