About Me

Hi, I am Nithin Somasekharan. I am pursuing my PhD in Aeronautical Engineering at Rensselaer Polytechnic Institute, Troy, New York. I work at the Computational Scientific Machine Learning Lab with Dr. Shaowu Pan.

I am passionate about scientific computing pursuing research in the field of computational fluid mechanics with focus on data driven moedling. My current work focuses on combining a discontinuos galerkin solver for fluid problems with machine learning for model closure, reduced order modeling etc.

Prior to my doctoral studies, I worked as a data scientist at UnitedHealth Group (Optum) and as a computational scientist at ExxonMobil, with a combined industrial experience of around 4 years. I have worked on developing machine learning and deep learning based models, NLP, solver development for fluid dynamic, Bayesian optimization and mathematical modeling. I have extensive experience in using Python, C/C++, MATLAB for code development.

What I do?

Computational Fluid Mechanics

Machine Learning/Deep Learning

GPU Computing

Mathematical Modeling

Cups of coffee
Projects
Collaborators

My Skills

Softwares:

ANSYS Fluent

90%

MATLAB/Simulink

90%

Abaqus

90%

Xfoil

100%

SQL

100%

GMESH

100%

Languages:

C/C++

80%

Python

100%

Libraries:

Tensorflow

100%

Pytorch

100%

Pandas

100%

Scikit-learn

100%

General:

Microsoft Office Suite

90%

Latex

90%

Linux

80%

Git

80%

Education

Rensselaer Polytechnic Institute (RPI), Troy, NY, USA

January 2023 - December 2027 (Expected)

GPA: 4.00/4.00

Indian Institute of Science (IISc), Bangalore, India

August 2017 - July 2019

GPA : 8.8/10

Amrita School of Engineering (ASE), Coimbatore, India

August 2013 - May 2017

GPA : 9.71/10

Current Research

GPU accelerated discontinuos Galerkin fluid dynamics solver

  • Develop a 3D discontinous galerkin solver of high order on Python for solving fluid dynamic problems.
  • The solver is developed using NVIDIA Warp which is a highly optimized library for GPU computing.
  • Adjoint based optimization methods are used for solving inverse problems in fluid dynamics for design optimization.
  • Machine learning techniques are used for accelerating the time consuming steps of the simulation.

Past Research

Natural Language Processing for call reason identification
Industrial Work
UnitedHealth Group (Optum), India, August 2022 - December 2022

  • Recordings of customer and agent was converted to transcripts using state of the art speech to text libraries (Whisper - OpenAI)
  • Transcripts were summarized using BART based model
  • Clustering algorithms were used based on cosine similarity to group together similar transcripts
  • Latent Dirchlet Allocation (LDA) was used to pick out the topics being discussed within the transcripts
  • LDA was used to curate a labelled data set
  • The labelled data set was used to retrain an exsisting model to predict the topic of discussion for a given summary of the transcript
  • The pipeline for speech to text was deployed in GCP kubernetes and exposed via fast api

Right time to call
Industrial Work
UnitedHealth Group (Optum), India, January 2022 - August 2022

  • Developed a predictive model to find the right time to call a customer
  • Customer demographic and past call patterns were used to curate the dataset for training
  • A four layered neural network was trained using this data
  • Deployed and integrated the model predictions into the business process
  • The estimated additional revenue generated by the model was around $16 Million for 2022

Identify the right medium to contact a customer
Industrial Work
UnitedHealth Group (Optum), India, August 2022 - December 2022

  • Developed a predcitive model to identify the best medium to reach a customer (phone, email, text)
  • Customer demographic and behavioral data were used to curate the dataset for training
  • A four layered neural network was trained using this data
  • Deployed the model on GCP Kubernetes

Characterization and Modeling of Uncertainties in Composites
Advisors - Dr. C.S. Upadhyay and Dr. P.M. Mohite
Research Assistant at Structural Analysis Lab, IIT Kanpur, India, Aug. 2013 - Jul. 2014

  • Conducted over 500 axial recoil tests on single carbon fibers of varying lengths and diameters (order of 10 microns) in order to determine the compressive strength of the fiber. This experimental data was then used to statistically model the scattering in strength due to various parameters.

Detection of Cracks in a Wind Turbine Blade using Modal Analysis
Bachelor’s Thesis, Advisors - Dr. Laxman Vaitla and Dr. V.L. Satheesh
ASE, India (in collaboration with National Aerospace Laboratories), Aug. 2013 - Jul. 2014

  • Examined frequency response of non‑rotating GFRP (Glass Fiber Reinforced Plastic) rectangular beams using finite element analysis for varying delamination sizes and positions.
  • Conducted sinusoidal sweep experimental tests on GFRP beams to validate the simulation results.
  • Performed modal analysis (using finite element method) on rotating composite beams to simulate rotating wind turbine blades.

Past Projects

Backstepping Control for a Variable-RPM Quadrotor Aircraft
RPI, NY, USA, Jan. 2019 - May 2019

  • Demonstrated the applicability of backstepping control design to a hobbyist scale quadrotor vehicle.
  • Examined the performance of a backstepping feedback controller in comparison to a PID controller on the quadrotor.

Extreme Altitude Mountain Rescue Vehicle
RPI, NY, USA, Jan. 2019 - May 2019

  • Proposed a synchropter as a VTOL aircraft design capable of search and rescue missions at the summit of Mt Everest, where only one helicopter has ever landed and taken off again.
  • Developed a 6‑DOF flight dynamics model and analyzed the autonomous dynamics of the proposed rotorcraft design. Design constraints include freezing temperatures, thin air and hostile weather conditions with degraded visual environment all contribute to making rescue work in high‑altitude environments particularly dangerous or impossible.
  • Won third place in the graduate category the vertical flight society’s 36th annual student design competition.

Controller Design for a Quadrotor UAV
RPI, NY, USA, Jan. 2017 ‑ May 2017

  • Nonlinear flight simulation model with PID and LQR controllers were developed for a 2 kg quadrotor helicopter and their performance was compared in simulation.
  • Compared control schemes using controller gains tuned for the hover flight condition and flown for single waypoint and multiple waypoint trajectories.

Design of V/STOL Air Taxi
IIT Kanpur, India, Dec. 2014 ‑ May 2015

  • Worked with a team of seven to propose a design for an air taxi system that operates out of confined urban areas requiring vertical and short takeoff and landing capability.
  • Worked with a team of seven to propose a design for an air taxi system that operates out of confined urban areas requiring vertical and short takeoff and landing capability.

Publications and Awards

  • Vayalali, P., McKay, M., Gandhi, F., “Damage Tolerant Control Allocation for a Compound Helicopter,” CEAS Aeronautical Journal, Submitted,October 2021.
  • Vayalali, P., McKay, M., Gandhi, F., “Redistributed Pseudoinverse Control Allocation for Actuator Failure on a High‑Speed Compound Helicopter,” Journal of American Helicopter Society, (Accepted with revisions), August 2021.
  • Vayalali, P., McKay, M., Krishnamurthi, J., Gandhi, F., “Fault‑Tolerant Control on a UH‑60 Black Hawk Helicopter using Horizontal Stabilator,” CEAS Aeronautical Journal, Vol. 12(1), Jan 2021, pp. 13‑27, DOI:10.1007/s13272‑020‑00476‑5.
  • Vayalali, P., McKay, M., Krishnamurthi, J., Gandhi, F., “Horizontal Stabilator Utilization for Post Swashplate Failure Operation on a UH‑60 Black Hawk Helicopter,” Journal of the American Helicopter Society, Vol. 65, April 2020, pp. 1–13(13), DOI:10.4050/JAHS.65.022009.
  • Vayalali, P., McKay, M., Gandhi, F., “Fault-Tolerant Control Allocation on a Compound Helicopter in Cruise,” Proceedings of the Vertical Flight Society 77th Annual Forum, Virtual, 10-14 May 2021.
  • McKay, M., Vayalali, P., Gandhi, F., Berger, T., Lopez, M. J. S., “Redistributed Control Allocation for Flight Control Failure on a Coaxial Helicopter,” Proceedings of the Vertical Flight Society 77th Annual Forum, Virtual, 10-14 May 2021.
  • Vayalali, P.,McKay, M., Gandhi, F., “Redistributed Pseudoinverse Control Allocation for Actuator Failure on a Compound Helicopter,” Proceedings of the Vertical Flight Society 76th Annual Forum, Virtual, 6-8 Oct 2020.
  • McKay, M.,Vayalali, P.,Gandhi, F., “Post‑Failure Control Reconfiguration for a Lift‑Offset Coaxial Helicopter,” Proceedings of the Vertical Flight Society 76th Annual Forum, Virtual, 6-8 Oct 2020.
  • Vayalali, P.McKay, M., Krishnamurthi, J., Gandhi, F., “Robust Use of Horizontal Stabilator in Feedback Control on a UH‑60 Black Hawk,” Proceedings of the Vertical Flight Society 75th Annual Forum, Philadelphia, PA, 13-17 May 2019.
  • Vayalali, P., McKay, M., Krishnamurthi, J., Gandhi, F., “Swashplate Actuator Failure Compensation for UH‑60 Black Hawk in Cruise Using Horizontal Stabilator,” Proceedings of the 74th American Helicopter Society Annual Forum, Phoenix, AZ, 7‑10 May 2018.
  • Third Place, Vertical Flight Society’s 36th Annual Student Design Competition, Virginia, USA
    2018
  • Runner-up, The American Helicopter Society Northeast Region Robert L. Lichten Award Competition, Stratford, USA
    2017
  • Second Place, RC aircraft modeling event as a part of techfest of Amrita School of Engineering, Coimbatore, India
    2012
  • First Place, SAE College level Aero Modeling Tier I event, Coimbatore, India
    2010

Relevant Courses

Finite Element Programing

Deep Learning

Computational Fluid Dynamics

GPU Computing

Contact

nithinsomu@gmail.com

+1 (518) 308-2829

Computational Scientific Machine Learning Lab,
Rensselaer Polytechnic Institute, JEC 5204,
Troy, NY 12180