Fahad KabirMechanical · CUET

Research

Models that stay reliable outside their training conditions.

Scientific machine learning that holds up on unseen operating regimes: detecting when a model is out of its depth, and correcting trained models cheaply. I also bring hands-on experience running GPU infrastructure and building research software.

Interests

  • Machine learning for engineering systems.
  • Physics-informed and physics-guided machine learning.
  • Computational modeling of fluid and thermal systems.
  • Data-driven prediction of how engineering systems behave.
  • AI-assisted engineering analysis, including low-cost intelligent systems.
  • Heat transfer and refrigeration, especially low-GWP refrigerants.

On the island: pitch the thesis airfoil → · open the heat-exchanger stack →

Publications

Localized forgetting in parameter-efficient adaptation of a frozen physics-informed airfoil surrogateUnder review

M. A. K. Fahad and M. A. M. Hossain. 10th BSME International Conference on Thermo-Fluid Engineering, 2026.

Optimizing evaporation heat transfer predictions of low GWP R455A in an industrial plate heat exchanger via machine learningUnder review

International Journal of Refrigeration, 2026.

Machine learning based prediction of the condensation heat transfer of low GWP refrigerant R1336mzz(E) in an industrial plate heat exchangerUnder review

International Journal of Refrigeration, 2026.

An updated residual entropy scaling model for the viscosity of CF₃I (R13I1)In preparation

Built with PropBench.

Experience

Ongoing

Machine learning for refrigerant heat transfer

Advisor: Monowar Wadud Hridoy, Mechatronics and Industrial Engineering, CUET

  • Models of heat transfer for two low-GWP refrigerants, from Saga University experiments.
  • Tested on operating conditions left out of training, showing that standard validation overstates generalization.
  • Measurement uncertainty analysis and data quality checks for the R455A study.

Thesis

Localized forgetting in PEFT of a physics-informed surrogate

Advisor: Prof. Dr. Md. Abu Mowazzem Hossain, Mechanical Engineering, CUET

  • Adapted a released physics-informed airfoil surrogate beyond its training range with low-rank updates.
  • Generated CFD reference data in SU2 and compared eight adaptation methods on accuracy and forgetting.

How I work

  • Rules before results

    Pass/fail criteria are written before the data exists, and applied by code.

  • Every number traceable

    Each figure in a write-up is copied from a saved run, with its ID.

  • Honest scope

    Simulation is labelled as simulation, and missing physics is listed.