Deepak Kumar Pokkalla

I'm an engineer and researcher specializing in scientific machine learning (SciML) at Dassault Systèmes in Boston. Previously, I contributed to AI/ML applications in material design and manufacturing at ORNL and honored with the 2023 R&D100 Technology Award, often regarded as the "Nobel Prize of Engineering".

I earned my PhD in Computational Mechanics from the National University of Singapore, where I was advised by Prof. Poh Leong Hien and Prof. Quek Ser Tong. In 2020, I received the ASCE Engineering Mechanics Institute Best Paper Award for my doctoral research. I hold a Bachelor's degree from the Indian Institute of Technology (IIT) Varanasi, where I was awarded the IIT (BHU) Varanasi Gold medal.

Email  /  LinkedIn  /  Google Scholar

I am interested in applying AI/ML to computational sciences to address real world challenges across various domains, including engineering simulations, physical AI, climate modeling, and healthcare. My current focus is on accelerating engineering simulations and product design by integrating AI/ML into CAD/CAE workflows, with an emphasis on physics-based machine learning and geometric deep learning. I am also exploring large language models (LLMs) for scientific documentation processing and diffusion models for generative design applications.

I have hands-on expertise in designing, training, and deploying AI models on large computational science datasets and using distributed deep learning for surrogate models to enable real-time predictions and reducing reliance on computationally expensive high-fidelity simulations. Additionally, I develop and apply nonlinear simulation and optimization techniques for multiscale and multiphysics engineering problems.

I built a GPT and GPT-2 from scratch and trained the transformer network on a Shakespeare dataset following Andrej Karpathy Neural Networks: Zero to Hero video series and "Attention is all you need" paper. Other popular videos from the series include State of GPT that talks about various stages in building a chat assistant and Deep Dive into LLMs, which provides an overview of LLMs.

Developed a novel isogeometric shape optimization framework for the design of auxetic materials with prescribed nonlinear mechanical responses. It involved developing code for nonlinear isogeometic analysis and gradient-based/gradient-free optimization frameworks. The capability of these frameworks is illustrated through additive manufacturing of optimized materials/structures and experimental validations.

We introduce a deep residual networks-based design framework for rapid inverse design of mechanical metamaterials with desired nonlinear mechanical responses. Our framework rapidly explores the topological design space through efficient numerical analysis and leverages the learned structure-property relationships. We illustrate the applicability of our framework by identifying previously unknown designs with enhanced mechanical performance.

We introduce an efficient deep neural networks based design framework for functionally graded lattices with enhanced mechanical responses. We demonstrate the capability and versatility of our framework by designing multiple lattice structures and experiments on additively manufactured optimized lattices.

We introduce symmetry breaking in bistable kirigami to access geometric frustration and anisotropic morphing, enabling arbitrarily scaled deployment in planar and spatial bistable domains. Our approach involves nonlinear numerical simulations (finite element analysis), design optimization, and experiments on elastic kirigami sheets to unlock anisotropic bistable deployment in planer and flat-to-3D configurations for deployable space structures, wearable technologies, and soft machines.

We employ a gradient-free genetic algorithm based isogeometric optimization framework to design soft network materials with desired nonlinear mechanical responses and fabricate them using a liquid additive manufacturing for biomedical applications.

We present an isogeometric shape optimization framework using genetic algorithm to design metamaterials with prescribed nonlinear mechanical responses. Through a combination of nonlinear numerical simulations and experiments, we demonstrate the design and manufacturability of soft flexible materials under a range of loading conditions.

We present a novel gradient-based isogeometric shape optimization framework for designing metamaterials with prescribed deformation over large strains, considering evolving geometric shapes and manufacturability constraints. We illustrate the capability of our framework by designing auxetic materials under different loading conditions and by reproducing the deformation behaviour of cat's skin.

We present the first numerical analysis upon the effect of heterogeneity on the stresses in fretting fatigue problems. We analyze the stress distribution in heterogeneous material under fretting fatigue loading condition based on Direct Numerical Simulations (DNS) in solid mechanics.

Template from here