About Me
Bing Gao
Hi! 👋 I'm Bing Gao, a Master's student in Modelling & Computational Mechanics at École supérieure d’ingénieurs Léonard de Vinci (ESILV). I am also an aerodynamics team member for our school's Formula Student racing team, Vinci Eco Drive. My responsibilities primarily cover aerodynamic and aero-kit design, cooling-system design, thermal-fluid modelling, numerical validation, and the development of custom simulation tools.
Before moving into engineering, I spent nearly 8 years in mobile software development and team management. As a core developer of Lao You, which means "Old Friend" in Chinese (iOS / Android), I built the live-streaming application from scratch, contributed more than 70% of its core code, and helped scale it from zero to more than 2 million users. That work gave me hands-on experience with audio and video codecs, live-room architecture, complex business logic, cross-platform integration with React Native and Flutter, hot fixing, application security, code obfuscation, and CI/CD automation.
From 2015 to 2023, I was also a core organizer for Google Developer Group Beijing. I organized more than five technical events each year, with 200 to 1,000 attendees per event. I invited developers and technical experts from teams at Google and major technology companies in Beijing, and helped organize knowledge sharing around mobile development, front-end engineering, Google Cloud, TensorFlow, and emerging technologies.
I also contribute to open-source projects on GitHub, mainly around scientific computing, numerical solvers, serialization, and data-processing tools. Every library I contribute to is one I use in my own work, and every pull request starts from a problem I have encountered. I keep regression tests in place for every merged change.
Projects
Paddock CVA sourced career-path atlas of trackside engineering staff across F1, F2, WEC / Le Mans, Formula E, and F1 Academy.
Modeling a 2026 F1 Car from Scratch with GSD in 3DExperienceUsing Generative Shape Design in 3DExperience, I successfully modeled the sidepods and rear wing of a 2026 F1 car by creating points, cross-sections, splines, lofts, and trimmed surfaces, with plans to build the complete 2026 F1 car model in future steps.
OpenFOAM in Practice: Exploring the F1 2026 Aero Window Across 37 CasesConstrained by computing resources, I performed half-car OpenFOAM CFD simulations on a third-party F1 2026 CAD model across 37 usable cases. Through this study, I explored numerical discretization schemes and trade-offs involving front and rear wings, tire contact patches, ride height, yaw, rake, and lap times.
Building an FSAE Cooling System from ScratchI first used a thermo-fluid model to reject the passive E3 architecture under 40 °C ambient conditions, split the system into a dual-temperature loop, and evaluated E7 and E8 using public component data, boundary sweeps, and OpenFOAM surrogate duct models. The results showed that passing numerical checks alone does not justify procurement approval.
Reconstructing a 3D Spacecraft Model in Blender from ESA BlueprintsDuring my 17-week internship in 2026, I reconstructed a 3D model of the Space Rider spacecraft in Blender based on publicly available European Space Agency blueprints and technical documentation.
Flow Field Reduced-Order Modeling: Compressing 480 Snapshots into Dominant ModesDuring unsteady CFD simulations, transient snapshot storage quickly overwhelms disk capacity. I investigated whether hundreds of complex flow snapshots could be compressed into 8 core spatial modes—similar to video compression—and used to forecast future flow dynamics without solving the governing equations.
FlowLab: Rebuilding a Lattice Boltzmann Method SolverTo understand CFD solver internals, I implemented a 2D Lattice Boltzmann Method (LBM) solver in JavaScript and verified it against the classic Re=100 lid-driven cavity benchmark.
Investigating Ground Effect with Image VorticesI built a ground-effect Vortex Lattice Method (VLM) tool.
Simulating the Kármán Vortex Street in FluentIn this CFD coursework, we investigated two regimes of flow past a cylinder: a steady wake at Re=40 and unsteady vortex shedding at Re=150. I set up two Fluent cases, triggered the vortex street via an intentional velocity patch perturbation, and recorded lift and drag histories.
Digital Twin in Action: Reconstructing a Tensile Test in AbaqusCoursework record: from physical tensile testing to Abaqus analysis
Crashworthiness Analysis of an Automotive Front Rail in AbaqusStarting from an automotive front longitudinal rail, we learned to interpret crash curves and compared materials, impact scenarios, and cross-section thicknesses. The results included models that failed to solve, as well as concepts that absorbed energy well on paper but were too heavy or produced excessive peak forces.
Comparative Analysis of Fluent Results Against Exact Solutions, Meshes, and Experimental DataVerifying Fluent pipe flow calculations against the Poiseuille exact solution, followed by mesh, domain, and Re=0.1–20 sweeps for flow past a cylinder compared with experimental data.
Cutting CFD Turnaround with Upfront Sanity ChecksCFD simulations can be computationally expensive, taking hours or even days. Inverted parameters can render days of computation entirely wasted. This is a learning note on low-cost sanity checks: I built a Python demo using thin-airfoil theory and the Hess–Smith panel method to estimate lift and surface pressure distributions in milliseconds, validated against NASA wind-tunnel measurements.
From Melting Snow to a 2D Heat Diffusion SolverOne day while watching snow on cobblestones, I noticed an interesting melting pattern. This sparked my curiosity and led me to build and verify a 2D FTCS heat diffusion solver.
How to Validate a Finite Difference ModelI built a 1D finite difference model of a metal rod with internal heat generation, fixed temperature at the left end, and convective air cooling at the right end.
Drive Cycle Simulation and Energy Consumption Comparison of Four Powertrain ArchitecturesSystematic comparison of energy consumption and CO₂ emissions across ICE, HEV, PHEV, and BEV architectures on the Spa circuit and standard driving cycles using a backward quasi-static model, quantifying PHEV initial SOC sensitivity and deadweight penalty after battery depletion.
How to Estimate Cooling Loop Flow Rate Without Physical HardwareA cooling pump datasheet may state 40 L/min, but that does not mean it will deliver 40 L/min once connected to hoses, a radiator, and engine water jackets. By modeling these head losses segment by segment, I determined the actual operating point at 26.22 L/min and 51.34 kPa, while verifying friction factors, line pressure drops, and curve intersections.
Building My First Potential Flow Model with Complex PotentialExploring the problem of flow past a circular cylinder
A Dimensionless Numbers Tool That Catches MistakesI built a small utility that enforces unit consistency, rejects unphysical inputs, and retains full validation records.