Browse categories
Explore
Fiverr Pro
English
$
USD
I will build physics informed neural networks and ml surrogates
Taimoor Amin
About this gig
FEA looks convincing even when it is wrong. Wrong mesh, wrong contact, wrong constraint, and the stress plot is fiction. I run full finite element analysis and explicit dynamics work full time: brackets, CubeSat frames, composite panels, aircraft struts, crash tubes, drop tests, blast panels, friction stir welding. Before I write a line of the report, I check: - Reaction forces balance the applied load - Mesh independence across three densities - Deformation shape makes physical sense - Hourglass energy under 10 percent on every LS-DYNA run Scope: static structural analysis, modal, buckling, nonlinear contact, fatigue life, composite layups and failure criteria, explicit crash and impact, drop testing, blast loading, thermal structural coupling. Tools: ANSYS Mechanical, ANSYS ACP for composites, LS-DYNA with LS-PrePost, SolidWorks Simulation, HyperMesh. Part of a practice with 60+ documented CFD, FEA and crash projects across aerospace, automotive and energy work. Send the geometry and load case. I will tell you straight which package fits, and whether the physics you want is solvable in the budget you have.
Get to know Taimoor Amin
Taimoor Amin
Your expert for CFD, FEA, LS DYNA Simulation, ML Surrogates for engineering
- FromPakistan
- Member sinceMar 2025
- Avg. response time1 hour
- Last delivery2 months
Languages
English, Urdu, French, German, Italian, Spanish
I run MachX Dynamics: CFD in ANSYS Fluent and CFX, FEA and LS-DYNA crash analysis, and CAD in SolidWorks, up through Mach 8 aerodynamics. I also build physics informed neural networks and reduced order models that swap a slow solve for a fast, validated prediction, work most CFD and FEA sellers do not cover. Gold Medal for best undergraduate thesis, NUST High Achievers Award among 15,000+ students, and gas turbine experience. Every job gets mesh independence and validation against theory where one exists. A budget or scope mismatch gets flagged before the order starts, not after.
My Portfolio
FAQ
What is a physics informed neural network and why would I want one?
A normal neural network learns only from data. A PINN also has the governing equation built into its loss function, so it is penalised for producing answers that break the physics. In practice that means it needs far less training data and it does not produce nonsense outside the range it
How is this different from hiring a data scientist?
A data scientist will treat your CFD output as a table of numbers. I have run the solver that produced it. I know which boundary condition drives the result, where the mesh was too coarse, and whether an error of 3 percent matters for your design decision or not. That context is the diff
What data do you need?
Either simulation output, a parametric sweep is ideal, or the governing equation and boundary conditions if you want a purely physics-driven model with no data at all. If you have neither yet, I can generate the training data by running the simulations, which is a separate quote.
Will the surrogate actually be faster?
That is the point. A trained surrogate evaluates in milliseconds where the original solve took minutes or hours. The honest caveat is that training costs time upfront, so this pays off when you need many evaluations, for optimisation, sensitivity studies or design space exploration, and
