Muhammet Emir Fil

Vehicle dynamics, CFD and structural dynamics in one

You drive a car around a track in the browser, and three simulations run under it at once: a four-wheel car with Pacejka tyres, a 2D fluid that I run around the car body to get drag and downforce, and a bridge that bends when you drive over it. I wrote each model in Python and checked it, then ported it to JavaScript so it runs live; the two versions agree to about 1e-13 for the car and 1e-9 m for the bridge. This one ties together the three projects below.

An inverse problem with quantified error bars

If you put strain sensors on a bridge, you can weigh the trucks driving over it by working backwards from how the bridge bends. I built the forward simulation (a beam under a moving, bouncing vehicle) and then the inverse that recovers the axle weights. The inverse is ill-posed when two axles are close together, so I added Tikhonov regularisation, and a Bayesian version for a calibrated error bar. I checked the forward model against Frýba's closed-form solution, and tested the monitoring idea on 16 months of actual measurements from the KW51 bridge.

Trajectory optimisation

Instead of drawing the fastest line around a track by hand, I set it up as an optimal-control problem and let a solver find it. I used arc length along the track as the variable, discretised it with direct collocation, worked out the Jacobians by hand, and solved it with SciPy. On a 2.4 km track the optimised line comes out 8.1 percent faster than following the centreline, and I checked the lap time against the closed-form skidpad result.

Computational fluid dynamics

I wrote a 2D incompressible Navier-Stokes solver from scratch (staggered grid, projection method, an immersed cylinder). It reproduces the standard Ghia lid-driven-cavity benchmark and shows the von Kármán vortex street behind the cylinder. Then I ran it backwards as a vortex flow-meter, reading the flow speed off the frequency at which the wake sheds vortices.