Individual Project | MSC ADAMS
Before the main project, the course began with a tire modeling review, comparing empirical, physical, and machine-learning approaches, and fitting a simplified Magic Formula model to tire data to understand where its limitations lie.
The Brief
This project used reduced-order MSC ADAMS models to study the handling and suspension of the OBR26 Formula Student car. This is the kind of early-stage analysis used to guide setup decisions before moving to higher-fidelity models or track testing.
Handling Model
I built a linear 2-DOF bicycle model from OBR26 parameters and applied a step steer input sized for the Formula Student skidpad radius. I then swept the center of gravity ±10% of the wheelbase about its midpoint to see how mass distribution changes the car's response.
Handling Results
After the steering input, the yaw rate rises to a steady value, and how it gets there depends heavily on CG position. Moving the CG rearward produced an overdamped rise with higher yaw gain and more lateral force carried by the rear axle, trending toward oversteer. Moving it forward produced a lightly underdamped, understeering response with lower yaw gain, with steady-state yaw rate nearly halving from the rearmost to the most forward position. The baseline OBR26, with its CG slightly behind the midpoint, sat between the two: a near-critically damped response that rose smoothly to steady state without overshoot. The car's low mass and high tire cornering stiffness kept overshoot small even in the forward CG positions.
Suspension Model
Starting with hand calculations and a 1-DOF model to establish natural frequencies and damping behavior, I extended the analysis to a 2-DOF quarter-car model of the front corner. Using dynamometer data for the team's Multimatic DSSV corner dampers, I modeled their nonlinear behavior and swept three valve configurations, each with 12 settings, across springs from 125 to 375 lb/in under a 0.5 to 20 Hz swept-sine road input.
Suspension Results
Rather than relying on a single number, I assessed each setup on both RMS tire force, for grip consistency, and peak tire force, for load spikes. Valve configuration 1 at setting 5 gave the lowest RMS tire force and stayed robust to changes in spring rate, while lower settings of configuration 2 best limited peak forces, suiting rougher surfaces. Configuration 2 at setting 4 offered the best balance of the two, keeping RMS force consistently low across the spring sweep without excessive peaks. The third configuration was outperformed on both measures by the other two and was ruled out.
Next Steps
The natural progression is a twin-track model with nonlinear, load-dependent tires, so that cornering stiffness changes with CG position and load transfer rather than staying fixed, and predicted cornering forces stay realistic. On the suspension side, frequency-response (Bode) analysis with a finer simulation time step would sharpen the natural frequency and peak-load results, before moving to a half- or full-car model including pitch, roll, and aero load. Ultimately, a combined model applying steering and road inputs together, correlated against damper dyno, rig, and track data, would bring both studies together.
Team Project | My Role: Front Splitter and Canards | Simcenter STAR-CCM+, Siemens NX
Before the main project, the course built up from vehicle performance calculations, comparing a 1972 Datsun 240Z with the aero-modified 240ZG, to wind tunnel boundary layer measurements and a NACA 0015 aerofoil study, where I correlated STAR-CCM+ simulations against our own wind tunnel results.
The Brief
As a team, we developed a full aerodynamic package for a full-scale DrivAer Fastback to suit a modified LMGT3 rule set. The goal was not simply maximum downforce, but the best balance of drag, efficiency, and a rear-biased aerodynamic balance of around 55%, within the specified drag area limit. I was responsible for the front of the car, the splitter and canards, and I also led the team's CAD assembly, CFD setup, and simulation runs.
Method
I prepared the DrivAer geometry in Siemens NX, then we ran steady RANS simulations in STAR-CCM+ with moving ground and rotating wheels to represent on-track conditions. Before any development, our baseline model agreed with published DrivAer data to within 10%. After the first package exposed weak resolution beneath the car, we refined the mesh around the underbody and wheel wells. For the yaw cases, we questioned whether a ground moving only in the car's direction was physically valid with a crosswind, compared it against a slip floor, and adopted the slip floor for the yaw study. We removed the wind-tunnel mounting features from the supplied wheels and, for the aero packages, simplified them to solid rotating wheels, closer to a covered LMGT3 wheel, though a real design would also need to account for brake cooling.
Front Splitter and Canard Development
Between full-package iterations, I ran 16 independent splitter variants, deliberately testing both very different concepts and small parameter sweeps. A flatter splitter with a thicker leading edge outperformed the more heavily cambered designs, endplates were essential, and strakes added less than expected, since much of the flow passed over them rather than being turned. The canards progressed from single plates to a triple-element stack, which gave the strongest front loading.
Integration
The front of the car could not be designed in isolation. Every splitter change altered underbody mass flow, diffuser performance, and the overall balance. The planned front fenders added some downforce, but at a large drag cost, and they clashed with the lowest canard, creating a pocket that trapped flow. We chose to run without modified-fenders and with the triple canards, trading peak downforce for better overall efficiency.
Result
The final package reached a downforce-to-drag ratio of 2.24, up from 2.05 for the modified-fender version, with a 56.1% rear aerodynamic balance. It held its performance with little loss in yaw up to 5°, and we confirmed its behavior with additional simulations at 300km/h.
Key Lessons
Peak local downforce does not define the best vehicle performance. The right front-end design depends on how its airflow interacts with the rest of the car.
Individual Project | AVL Drive Race
The Brief
Using AVL Drive Race simulation data for a lap of the Circuit de la Sarthe, I compared a Le Mans Hypercar (LMH) with an LMDh prototype. Both race in the WEC Hypercar class under a Balance of Performance, but the LMH allows a bespoke chassis and aerodynamics, while the LMDh uses a spec chassis and a standardized hybrid system. The aim was to find how those different rule sets still produce measurably different cars.
Method
I analyzed more than 20 data channels covering fuel and energy use, hybrid deployment, aerodynamic and mechanical balance, handling, and tire loads and temperatures. I then set the findings in the context of the regulations and the 2025 Le Mans 24 Hours, where the Ferrari 499P (LMH) beat the Porsche 963 (LMDh) by just 14 seconds after 387 laps. In all graphs, the LMH is shown in red and the LMDh in green.
Findings
The LMH lapped 0.825 s faster despite a slightly lower top speed. Its hybrid system deployed 6.8 times more electrical energy per lap, driving the front axle for all-wheel-drive traction out of corners and adding regenerative braking at the front. The most counterintuitive result was balance: the LMDh is more front-biased mechanically than the LMH, yet it understeers more, because its spec chassis leaves a larger gap between its mechanical and aerodynamic balance. The LMDh also worked its rear tires harder, building more lateral energy and heat, while the LMH spread the load more evenly, an advantage that compounds over 24 hours.
Limitations and Next Steps
The analysis was based on a single lap driven by an optimal simulation controller rather than a human driver, so tire degradation could only be inferred. Multi-lap stint analysis, a mini-sector breakdown of where the time is gained, and wet-weather simulations would be the natural next steps.
Team Project | My Role: Powertrain Lead | AVL VSM RACE, AVL Drive Race, Excel
The Brief
As a team, we developed a 2026 Formula One car setup for qualifying at Silverstone in AVL VSM RACE under the 2026 FIA regulations, with work split across aerodynamics, vehicle dynamics, powertrain, and race strategy. I led the powertrain setup: engine scaling, MGU-K configuration, gearbox ratios, and energy deployment strategy.
Power Unit
The 2026 regulations remove the MGU-H, cap fuel energy flow at 3,000 MJ/h, and raise MGU-K power from 120 kW to 350 kW, moving toward a roughly 50/50 split between engine and electric power. I scaled the 2025 engine and motor torque curves to the new limits, implemented the regulated taper on MGU-K power at high speed, and checked compliance across the full operating range. To validate the drivetrain behavior outside AVL, I also built an independent straight-line acceleration model in Excel.
Gearbox
F1 rules fix gear ratios for the whole season, so I kept the 2025 ratio spread and optimized the final drive ratio instead: the fixed reduction between the gearbox and the wheels, set by the final and bevel gear pairs. Within the range where all eight gears were still used, the shortest option was 0.385 s faster than the 2025 equivalent. The seventh-to-eighth upshift now happened before MGU-K power began to taper, and corner-exit engine speeds rose about 7%.
Energy Strategy
Without the MGU-H, energy recovery relies heavily on "super clipping": staying at full throttle while the MGU-K diverts engine power into recovering energy instead of driving the wheels. I placed six 100 m clipping windows at the ends of straights, just before braking zones, where giving up that last bit of acceleration cost the least lap time, and confirmed 100 m as the optimum through a sensitivity study. To spend that recovered energy, I then distributed deployment across eight sectors, prioritizing slow corner exits, where each kilojoule bought the most lap time.
Result
The final car set a qualifying lap of 89.534 s, deploying about 7.0 MJ and recovering about 3.5 MJ per lap while staying within the regulated energy limits throughout.
Next Steps
Thermal analysis of the engine and motor under race conditions, clipping window lengths tuned for each zone rather than applied globally, and a global optimization of the sector deployment, which reached a local optimum.
Key Lesson
The 2026 power unit is limited by energy rather than power. At Silverstone, the binding constraint was how much energy the car could physically recover, not the regulatory cap.
Team Project | My Role: Team Lead, Laminate Analysis, Structural Calculations, and CAD | Excel, Siemens NX
The Brief
Design a composite sandwich safety cell for a battery electric vehicle, protecting both the occupants and the high-voltage battery, for two production scenarios: 10 units a year and 10,000 units a year. The cell had to fit a 2,100 × 1,410 × 250 mm envelope while housing 27 battery modules with their cooling and cabling, stay under 110 kg, and meet torsional rigidity, bending rigidity, and impact deflection targets with a safety factor of at least 1.5. Impact was defined as a 1,300 kg mass at 13.86 m/s, with the safety cell allowed to deflect no more than 5 mm after attenuation. The design also had to account for service temperatures from −40 °C to 100 °C, fire and electrical safety, and penetration resistance, with at least 20% natural fiber content for the low-volume version and full recyclability for the high-volume version.
Design
We carried out the Classical Laminate Theory analysis from first principles for both versions, building the laminate stiffness matrices and deriving the effective properties used in the structural calculations. Each version used two laminates: a sandwich enclosure, with carbon fiber skins over a core, carrying the torsion and bending loads; and a solid, mostly unidirectional carbon fiber internal impact structure carrying the impact loads, with woven plies on the outer faces for impact and wear resistance and against the core for better bonding. To keep the analysis conservative, each was sized independently for its own load cases. We then iterated the layups until every target was met. The low-volume design used a toughened epoxy prepreg with an end-grain balsa core, which met the natural fiber requirement at just over 32% while also carrying real structural shear load. The high-volume design used a recyclable thermoplastic with a PEI foam core, selected to withstand the thermoplastic's processing temperature. I also sized the bolted battery cover joint against the shear flow from torsion, and modeled the safety cell in Siemens NX.
Manufacturing and Sustainability
The two volumes called for different processes. Low volume used hand layup and autoclave curing, allowing the enclosure and internal impact structure to be co-cured as a single piece. High volume used press consolidation, with the enclosure and impact structure panels made separately and resistance welded together. An eco audit compared the environmental impact of the two production routes.
Individual project | HyperMesh, LS-DYNA
The brief
With the material constraints no longer applying at this stage, I combined the high-volume safety cell geometry, whose internal spine was expected to give a stiffer structure, with the low-volume composite laminate. I modeled the structure in HyperMesh and LS-DYNA and evaluated foam impact attenuator designs within a 500 mm frontal and 200 mm side crush zone. The goals were to keep safety cell deflection under 5 mm and peak deceleration under 40 g.
Model
The composite skins used a progressive damage material model, the balsa core a crushable plastic model, and the attenuators a crushable foam model capturing elastic, plateau, and densification behavior. I tested six attenuator variants: solid and four-tier stepped geometries with two foam densities, plus two graded designs combining both. Alongside them, I ran unattenuated baselines, an isolated attenuator check, and a side impact case, ten simulations in total. One baseline used 4.38 m/s, the speed remaining if the attenuator absorbs its 90% energy target.
Findings
The lower-density foam consistently outperformed the denser foam, which transmitted crush forces high enough to destroy the safety cell before the impact was complete. The stepped geometry cut peak transmitted force by up to 51% compared with an equivalent solid block, by engaging the foam progressively. The graded stepped design, with soft outer tiers and a stiff inner tier, achieved the lowest peak deceleration at 25.8 g, well within the 40 g limit.
No variant met the 5 mm deflection target, and the isolated check showed why. Against a rigid wall, the attenuator absorbed all of the impact energy within its 500 mm stroke, so the attenuator was working; the safety cell could not resist the crush force. The panels parallel to the impact buckled, so even the unattenuated baseline deflected three times more than laminate theory predicted, pointing to the chassis structure, not the layup, as the limiting factor.
Next steps
Properly defining the connection between the internal spine of the impact structure and the enclosure, developing a stiffer chassis structure around the existing layup, and testing alternative attenuator geometries and materials, such as other foams, aluminum honeycomb, and carbon fiber tube crash structures.
Key lesson
An attenuator has to absorb energy at a crush force the safety cell can withstand. Here, the attenuator did its job but the structure behind it could not, a distinction laminate theory alone could not reveal and the FEA made clear.