Akhil Sukumaran Nair

• Numpy, Pandas, Polars
• Scikit Learn, PyTorch
• Matplotlib, Selenium
• PowerBI, Tableau
• PowerShell
• GitHub, GitLab
• AWS - EC2, Lambda

• Superstructure optimization of circular plastics value chain - Flexibility Analysis (Asian PSE-22, AICHE -23)

(Modeled a cradle-2-gate circular PET plastic superstructure, based on data driven, ML-based surrogates and solved the resulting MO-MILP problem, under epistemic uncertainty to identify optimal frameworks for the PET supply value chain)

• MPC performance monitoring algorithm development and dashboarding (Corning) : Developed an algorithm within the QDMC class using MATLAB to benchmark control model performance and trigger system re Identification for 120+ MPC applications across Asia and US. Deployed real time Grafana dashboards. (16 priorities identified)
• Bi-level Real Time Optimization (RTO) for equipment life enhancement (Corning): Composed a weighted RTO problem to find oxidation compensated temperature SP to feed into lower level MPC loop. Used a CFD inspired PINN model to solve a secondary RTO problem to maintain product quality while avoiding thermal hotspots on the equipment.
• Advanced controls framework for highly nonlinear unit start-up (Corning) : Used an adaptive gain based PID control to replace bang-bang controls and tuned it. Conducted comparison studies between First principles, Mass balance and LSTM based data models to capture model physics and develop a hybrid NMPC controller. (Reduced 2.5% of overall product breakage)
• Rigorous MINLP model for Power (CPP) distributions (CMU) : Ideated a technique to penalize emissions & award carbon credits of operating equipment by revising IBM's Rigorous distillation model using STEAM and created a pareto front for weights optimization. Benchmarked BARON, DICOPT solvers for non linear initialization & finding global optima.
• Increasing efficiency of Transhipment Network models for fleet optimization : Tested efficiency of an Edmond Karp algorithm based formulation against a self-tuned GSA based metaheuristic. (Reduced 2% of median solution time)
• Predicting the Time to coke & Shutdown of Fouling Industrial Heaters (CMU) : Benchmarked shallow & deep ML algorithms in Python for designing coke accumulation on heater tubes, deployed efficient PCA basis to represent 94% information by 5:1 compression, devised PINN's and LSTM RNN models Test accuracy of 93.1%

• Chemical Reaction Engineering
• Machine Learning
• Modern Convex Optimization
• Multi Criterion Decision Making
• Quantum Integer Programming
• Intermediate Deep Learning
• Process Dynamics & Control
• Non-Linear Programming