THE BLACKBIRD / A PUBLIC-DATA RECONSTRUCTION
Look inside the J58 engine and explore the aircraft around it. Compare the model with NASA’s drawings and inspect the simulation results.
Work in progress — source-based reconstruction, not a validated digital twin. CAD and research datasets are available by request.
Actual CAD geometry, rendered in your browser.
Drag to orbit · scroll or pinch to zoom · select a part
THE INLET AND ENGINE
At high speed, the inlet changes the air before it reaches the compressor. The spike moves, bypass paths manage the flow, and the J58 core keeps running.
PRATT & WHITNEY / J58
The engine core keeps running at high speed. Compressor bleed carries some of the air around the rear core and into the augmentor.
Inlet bypass doors and engine compressor-bleed pipes are separate systems.
DISTANCES IN THE CURRENT CAD MODEL
These distances are measured in the reconstruction. They are not measurements of an installed J58. The connecting duct and ejector interface still need applicable installation drawings.
The engine primary nozzle is represented by fixed lofted petals with no actuator-to-petal linkage. The separate airframe ejector is not the same commanded mechanism.
Inspect the geometry audit ↗SEVEN SHORT FILMS
Choose a film to see the assembly or take a closer look at one component. The motion illustrates the CAD model.
RECORDED EXPERIMENTS / NOT A LIVE SOLVER
These experiments use SymPy equations and JAX derivatives, alongside fluid simulations computed by JAX-Fluids. Open a study to see its configuration, recorded results and original failed checks.
BATCH 6 / AUTHENTICATED FINAL B
These four cases compare the original and characteristic boundary conditions on baseline and expanded domains. Each uses fixed 2D Euler geometry at Mach 0.30 and 24 km. Together they contain 109,985 JAX-Fluids integration steps and 132 saved fields. They are correlated numerical cases, not four validated operating regimes, and all retain their original failed checks.
This film shows the saved fluid fields computed by JAX-Fluids for the original baseline case. Its time coordinate is nondimensional, so playback does not represent real flight time. The case still fails its original acceptance checks. It models 2D inlet flow, without 3D engine behavior, combustion, heat flux or whole-aircraft motion.
The 11,880 diagnostic rows sample cells next to the outlet. They form one evaluation-only family, and every row carries numerical failure code 0. Finite fluid values do not make them accepted performance targets. Access to the raw fields and datasets is available by request.
The batch 2 benchmark tests the boundary operator in the cases described here. It does not validate the whole SR-71 or a combusting J58.
Batch 5 contains 11 running-state trajectories of 0.1 s, totaling 14,500 steps. The calculations track rotor kinetic energy and receiver storage using assumed inertias of 10, 30 and 90 kg m². Measured J58 inertia, starter behavior and ignition remain unresolved.
Batch 4 contains nine 20 ms trajectories, totaling 2,100 steps. The quasi-steady engine equations are solved again at 4,200 integration stages and checked at all nine endpoints. The receiver volume is assumed to be 1 m³ and the maps are synthetic. These runs do not validate long transients or ground starting.
The batch 2 learning experiment includes failed cases and a frozen test set excluded from training. Later prospective-learning results are listed in Evidence. These predictions describe numerical checks, not physical performance; every predicted pass still requires the original solver and acceptance checks.
This earlier batch 1 experiment contains two cases, both of which retain their original acceptance failures. It is separate from the four-case campaign above.
THERMAL AND MECHANICAL FOUNDATIONS
The thermal tests check a synthetic model with constant heat capacity. The rigid-body tests check the motion equations. The completed structural FEM uses a hollow-box cantilever surrogate; it is not mapped to the displayed airframe. Whole-aircraft structural FEM and a calibrated flight controller are still missing.
Further work needs coupled fluid and solid heat transfer, measured material properties and structural joints. Control calibration, startup and ignition are also unresolved.
SAVED MISSION CONDITIONS
Explore saved steady-cycle samples from illustrative mission conditions. Each sample has its own model checks.
Afterburner values are normalized model inputs. All four retained full-inlet CFD cases fail their original acceptance criteria. This view does not show a validated trajectory, ignition sequence or closed-loop flight controller.
SOURCES AND CHECKS
The model uses public documents for its architecture and overall dimensions. Here you can compare its contours with the drawings and see which numerical checks passed or failed.
NASA DRAWING AND MODEL COMPARISON
The drawing is graphic 980065 in NASA’s Six Decades of Flight Research, on printed page 455 (PDF page 464). Its caption says “SR-71” but gives no scale or production tolerance. The low canopy suggests the A configuration; the caption does not confirm it.
Read this page in NASA’s publication ↗
An airframe assembly and a 58-part J58 cutaway. Overall dimensions have public reference anchors; local surfaces and internal details include reconstruction assumptions.
Millimetres. X runs from nose to tail, Y across the aircraft, and Z upward. Browser meshes are tessellated from the CadQuery/OpenCascade components. A full-scale coordinate system does not imply historical completeness.
Reading recorded matching results…
Fixed-geometry inlet calculations expose domain sensitivity and unsteadiness. The coupled aircraft and propulsion system is still being developed.
Solver convergence is numerical evidence. It does not validate compressor maps, physical geometry, hardware or flight.
Reading the current learning dataset…
The dataset contains synthetic numerical labels. Every predicted pass still needs the original solver and all acceptance checks.
SOURCE-FIT DIAGNOSTICS / REFERENCE PIXELS
| View | RMS · px | P95 · px | Maximum · px |
|---|
SR-71 TESTBED / PAYLOAD MASKED
| Observation | RMS · mm | IoU · 0–1 |
|---|
The fin-height check is a post-comparison diagnostic relative to frozen view datums. It exposes how nearest-contour distances can match the wrong feature. It is not untouched validation or a manufacturing correction.
Registration ranges use one-way visible-reference RMS, including top view; that differs from the headline top bidirectional RMS. Analyst perturbations are not confidence intervals and do not bound scan or configuration errors.
PROXY CONTOUR COMPARISON / REGISTERED V2
| View | RMS · mm | IoU · 0–1 |
|---|
RMS is symmetric contour distance; lower means closer here. IoU is projected silhouette intersection divided by union; 1 means identical masks. Neither is an overall aircraft accuracy score.
Registration ranges are 27 analyst box samples, not confidence intervals. Trace sensitivity is separate. Raster detail, configuration differences and datum choices limit interpretation. Internal holes are filled for gross silhouette comparison.
| Stroke · mm | Left · m³ | Right · m³ |
|---|
Full historical surfaces, installed interfaces and a physically qualified engine map remain missing. Downloadable CAD and numerical checks do not establish them.
Request the detailed research files ↗RESEARCH ACCESS
You can explore the models and selected results on this site. Request access if you would like to inspect the private CAD files or raw datasets.
Request CAD or dataset access ↗PUBLIC REFERENCES
CLAIMS, MEASUREMENTS & OPEN GAPS
A software check does not establish that a model matches the aircraft. These records explain what each check establishes and which measurements are still missing.
The detailed qualification register is being assembled. The current exterior shape remains under review.
All selected results retain their original configuration and failure status. Raw records remain request-only.
PRIVATE RESEARCH ACCESS
The repository is private while the work is being developed and reviewed. We plan to make it public.
Ask the research agent about the project, or enter your GitHub username to request access.
Get access or ask the agent ↗Preparing the aircraft geometry…
WebGL could not start in this browser. You can still inspect component descriptions, explore saved conditions, watch the films and request CAD access.
Request CAD access ↗