A full-triangle C++20 solar-car irradiance simulator intended for a high-performance Windows or Linux workstation. It loads the complete binary STL, builds an Embree BVH, calculates per-cell direct, diffuse-sky, and recursively reflected irradiance, exports CSV results, and shows the result in a native OpenGL desktop GUI. The GUI can fetch current DNI, DHI, temperature, and cloud metadata for any latitude/longitude from Open-Meteo.
There is one implementation in this repository. The former Python/Tk and voxel-preview path has been removed.
full binary STL -> transformed triangle buffer
|-> upper-shell candidates -> 2x3/1x3 layout
| |
|-> Embree BVH -> irradiance solver -> per-cell CSV
`-> OpenGL VBO --------------------> native GUI
latitude/longitude -> Open-Meteo DNI/DHI + UTC time -> solar position
vehicle heading -------------------------------------> car-relative sun vector
OpenGL displays the original transformed triangles. It does not feed pixels back into the physics. Embree intersects the same triangles directly, so camera position and window resolution cannot alter the numerical answer.
- 64-bit Windows or Linux
- C++20 compiler (Visual Studio 2022 or a current GCC/Clang)
- CMake 3.24+
- Ninja
- vcpkg
- Current OpenGL driver
- Recommended: 32 GB RAM and 8 GB GPU memory for the 14.2-million-triangle source STL
Dependencies are declared in vcpkg.json: Embree 4, GLFW, GLEW, GLM, Dear ImGui, CPR,
and nlohmann/json.
Install the Desktop development with C++ workload in Visual Studio Installer and Git for Windows. Then open the repository in VS Code and run this one command in its terminal:
.\run_windows.batThe launcher finds Visual Studio and automatically installs missing MSVC/Windows SDK, CMake,
Ninja, Git, vcpkg, Embree, GLFW, GLEW, GLM, Dear ImGui, CPR, and JSON dependencies. Windows
may show an administrator prompt while system build tools are installed. It then builds the
project and opens the GUI using
%USERPROFILE%\Downloads\_24-000.stl. A different STL can be supplied as the first argument:
.\run_windows.bat "D:\path\to\car.stl"The initial build downloads and compiles the dependencies, so it is much slower than later runs. The launcher prints a specific Visual Studio Installer instruction if a required build component is missing.
Windows PowerShell:
git clone https://github.com/microsoft/vcpkg.git "$HOME\vcpkg"
& "$HOME\vcpkg\bootstrap-vcpkg.bat"
$env:VCPKG_ROOT = "$HOME\vcpkg"Linux:
git clone https://github.com/microsoft/vcpkg.git "$HOME/vcpkg"
"$HOME/vcpkg/bootstrap-vcpkg.sh"
export VCPKG_ROOT="$HOME/vcpkg"Keep VCPKG_ROOT set in the terminal that launches VS Code or CMake.
From the repository root:
cmake --preset release-vcpkg
cmake --build --preset release-vcpkg
ctest --preset release-vcpkgThe first configure downloads and builds the manifest dependencies. Later builds reuse them. VS Code exposes the same steps as Configure release build, Build simulator, and Run C++ tests tasks.
Linux:
./build/release/solar_whitted_cpp \
--stl /path/to/_24-000.stl \
--rays 64 \
--output results/cell-irradiance.csvWindows PowerShell:
.\build\release\solar_whitted_cpp.exe `
--stl "D:\path\to\_24-000.stl" `
--rays 64 `
--output results\cell-irradiance.csvOr select Run full-mesh simulator from Terminal > Run Task in VS Code and paste the absolute STL path when prompted.
The supplied _24-000.stl defaults are already encoded. Its vertex coordinates are in
metres (the measured transformed bounds are approximately 1.61 m wide, 6.20 m long, and
1.25 m tall):
- input scale
1.0; - rotation
X=90 degrees; and - translation
(0, -0.6563, 0.0153)metres.
Use --help to see all transform, sun, integration, cell-grid, thread, and output options.
Use --headless for CSV-only execution on a compute node.
Candidate placement uses a configurable world-space upper-shell mask. The default inclusion
box is x=[-0.68,0.68], y=[-2.90,2.80], z=[0.12,1.00] metres, requires a local
upward-normal component of at least 0.75, and removes a central canopy prism
x=[-0.32,0.32], y=[-1.45,1.35]. CLI --shell-*, --canopy-*, and
--shell-normal-z options tune these provisional cutoffs. The filtered mesh is used only to
place candidate cells; the complete STL remains in Embree and continues to cast shadows.
Every candidate then requires a complete 3x3 support sample across its 125 mm footprint.
Those points must remain within 6 mm of the locally fitted tangent plane and have consistent
normals. This accepts smooth top-shell curvature while rejecting openings, sharp hub/body
geometry, and quads that would bridge unrelated surfaces. The largest smooth connected
candidate region is retained as the primary shell, removing disconnected islands such as
wheel hubs. Accepted cells follow the local shell tangent with 1 mm clearance. The planarity
tolerance is adjustable with --max-cell-plane-error.
The projected locations are grouped into vertical modules before simulation. A module always contains three consecutive longitudinal cells. At each feasible row band, adjacent columns are taken as a 2x3 module first; any supported column left over becomes a 1x3 module. The generator evaluates all three possible longitudinal row phases and selects the one with the most 2x3 modules, breaking ties with the number of 1x3 modules. Each cell retains its own validated local shell tangent; a 1x3/2x3 group is not treated as one rigid planar plate, so valid groups may follow the curved nose. This is a deterministic prioritized layout for the current regular grid, not yet the final route-energy set-packing optimizer.
The executable opens a GLFW/OpenGL window after loading the STL and completing the initial simulation. Its left control panel provides:
- azimuth, elevation, DNI, and DHI controls;
- latitude, longitude, vehicle heading, and asynchronous Open-Meteo current-weather fetch;
- direct-shadow footprint density, hemisphere-ray, and reflection-depth controls;
- a Run Simulation button that recomputes Embree results;
- current irradiance, incident optical watts, active area, selected-cell count, and 2x3/1x3 module counts;
- direct/sky/reflected contribution breakdown;
- live OpenGL shell irradiance preview and color legend;
- shell, cells, ground-grid, sun-vector, and wireframe toggles;
- CSV export and camera reset.
The right viewport renders the complete STL plus the generated cell layout. The shell heat map retains subtle form shading so its curvature remains visible. Cyan outlines identify cells in 2x3 modules and orange outlines identify cells in 1x3 modules. Cell irradiance fill is optional and off by default so the layout cannot look like a detached opaque sheet. Left-drag orbits, the mouse wheel zooms in small bounded increments, and Escape closes the application. The panel occupies its own screen area, so it does not cover the car. The shell heat map evaluates the local triangle orientation using DNI plus an isotropic diffuse-sky estimate and responds immediately while the sun and irradiance sliders move. Hovering over the shell shows the same preview value in W/m² beneath the legend. This live shell coloring intentionally omits self-shadowing and reflections; the generated-layout result and CSV use the complete Embree visibility and recursive integrator after Run Simulation is pressed.
Enter WGS84 latitude/longitude and the vehicle heading clockwise from true north, then click Fetch Current Weather. The application requests the current model time step from Open-Meteo, applies its DNI and DHI, computes geographic solar azimuth/elevation from the returned UTC timestamp, and rotates the sun azimuth into the vehicle frame. Click Run Simulation to recompute the exact per-cell result.
The solar-position calculation follows the equation-of-time, declination, hour-angle, zenith, and clockwise-from-north azimuth conventions documented by NOAA Global Monitoring Laboratory.
Cloud cover is displayed as provenance metadata but is not multiplied into the irradiance: Open-Meteo's DNI and DHI already include modeled atmospheric and cloud effects. The weather request runs outside the rendering thread, so a slow network response does not freeze camera or GUI interaction. A 15-second request timeout is enforced.
The public endpoint is intended for eligible non-commercial use and requires attribution; review Open-Meteo's current terms before deployment. A production/commercial deployment should use their customer endpoint and API key.
For cell normal n and direction to the sun s:
E_direct = DNI * max(0, n dot s) * visibility
visibility is the fraction of a configurable N x N shadow-ray grid across that cell's
footprint. The default 3 x 3 grid distinguishes fully lit, fully shaded, and partially
shaded cells. Each CSV row includes direct_visibility_fraction. Diffuse sky and scene
reflection are currently sampled from the cell center.
Diffuse sky uses deterministic cosine-weighted hemisphere samples with isotropic sky
radiance DHI / pi. Shell and ground hits return Lambertian plus normalized Phong radiance;
their specular component launches a Whitted reflection ray to the configured depth.
Each selected Maxeon Gen III cell uses a 125 mm footprint and 0.0153 m2 active area. Output
is optical irradiance and incident sunlight—not electrical power.
CMakeLists.txt build definition
CMakePresets.json reproducible release build
vcpkg.json C++ dependencies
include/solar/ public types and interfaces
src/stl_mesh.cpp streaming binary-STL loader and transform
src/trace_scene.cpp Embree triangle scene and ray queries
src/simulation.cpp projection and irradiance integration
src/weather.cpp Open-Meteo client and UTC solar-position calculation
src/viewer.cpp OpenGL + Dear ImGui desktop application
src/main.cpp CLI and application startup
tests/core_tests.cpp deterministic core checks
data/cells/ Maxeon reference inputs
docs/ equations, status, and roadmap
.vscode/ build/test/run tasks
- Real cell/component transforms from the SolidWorks assembly
- Globally optimal, route-energy-weighted 2x3/1x3 set-packing and mechanical validation
- Per-component measured material BRDFs
- Weather/route time-series ingestion
- Temperature and electrical conversion
- Strings, mismatch, bypass diodes, wiring, and MPPT