Anindya Bhaduri

Propagation of uncertainty/variability from input parameters to output quantities in data (simulated or real) is essential for characterizing the underlying system. For complex systems, this requires lots of data, which can be prohibitively expensive. Thus, for effective uncertainty propagation, surrogate modeling is used to construct an efficient mathematical model - characterizing the system with minimal data.
My interest lies at the intersection of machine learning and uncertainity quantification, specifically in developing surrogate modeling approaches using deep neural networks, stochastic collocation methods, etc. coupled with adaptive parameter sampling to tackle problems in fields like solid mechanics, epidemiology, and molecular dynamics.

Deep learning based stress field prediction of fiber-reinforced composite materials

A U-net deep architecture has been trained from scratch for mapping composite material images to stress field images. Sensitivity of different training data combonations has been studied on the prediction results.

Reconstructing random multiphase materials using deep learning

A transfer learning based microstructure reconstruction approach has been implemented using a pretrained VGG19 network. This method is capable of reconstructing large sized images efficiently for large-scale analysis.

A stochastic collocation approach with adaptive mesh refinement

An efficient stochastic collocation method with adaptive mesh refinement (SCAMR) has been developed to deal with high dimensional stochastic systems with discontinuities. The method employs a dimensionality reduction strategy to decompose the original high-dimensional problem to a number of lower-dimensional subproblems.

Composite plate penetration under projectile impact

An adaptive domain-based decomposition and classification method, combined with sparse grid sampling, is used to develop an efficient classification surrogate modeling algorithm for discrete output systems. As an application problem, the probabilistic velocity response (PVR) curve or the V0-V100 curve is generated for S-2 glass/SC-15 epoxy composite plates under ballistic impact.

An adaptive collocation method based on derivative estimates

The proposed methodology tracks discontinuities while also avoiding unnecessary function evaluations in smoother regions of the stochastic space by using a finite difference based one-dimensional derivative estimation technique.

Free energy landscape reconstruction

A modified single sweep approach using space-filled design and weighted reconstruction for generating free energy landscapes.

On the usefulness of gradient information in surrogate modeling

The primary goal is to investigate whether additional gradient information obtained at a relatively small cost helps in generating surrogates of better quality compared to those obtained without any gradient information. The surrogate considered here describes the variation of the homogenized stress at a given input strain as a function of the fiber/matrix interface damage parameters in a multi-fiber reinforced composite model