KEPLER Data Factory (KDF)
Dynamic Pipeline Trace Analysis & Scenario Synthesis
Dataset Synthesis & Active Simulation Context
The integrated pipeline execution visualized below streamlines the development and validation of constellation tasking solutions by connecting deterministic orbital physics with semantic text generation to reconcile operational bounds and strategic mission objectives.
Dataset Identifier: View Repository Data
Active Fleet Nodes (Nsat): 5 Spacecraft entities (Weather Group Pool)
Ground Station Assets (Ngs): 2 Operational Segment Downlinks
Strategic Mission Requests: 15 Point/Polygon Target Tasks
Baseline Simulation Seed: 2026 (Incremented across scenarios)
Active Strategic Layer Model: Local LLM with Ollama
Model Hyperparameters: Temperature: 0.4
Natural Language Layer: Enabled
Sensor Capabilities & Generation Rates
Probabilistic assignment of payload characteristics dictating the Physical Simulation Layer constraints.
| Sensor Pool | Weight | Data Rate (Rgen) | Max Look Angle |
|---|---|---|---|
| VIS (Visible Optical) | 30.0% | 40.0 MB/s | 30.0° |
| SAR (Synthetic Aperture Radar) | 25.0% | 80.0 MB/s | 40.0° |
| TIR (Thermal Infrared) | 15.0% | 12.0 MB/s | 35.0° |
| VNIR (Visible-Near Infrared) | 15.0% | 35.0 MB/s | 30.0° |
| NIR (Near Infrared) | 15.0% | 20.0 MB/s | 30.0° |
Hardware Payload Storage Profiles
Solid-state recorder onboard data capacities allocated from the configuration specification to provide planning algorithms with strict data management criteria:
1 Component: Data Collector (DC)
The Data Collector (DC) synthesizes the spacecraft constellation context, downloads empirical Two-Line Element (TLE) sets, samples geopolitical ground station assets, and probabilistically allocates payloads, communications bands, and observation task parameters.
| Active Constellation Satellites Fleet Context | |||
|---|---|---|---|
| Spacecraft Designation | Onboard Sensors | Downlink Bands | Memory Threshold |
| Loading satellite configuration profiles... | |||
| Ground Station Segment Networks |
|---|
|
| Generated Scenario Tasks Parameters | |||
|---|---|---|---|
| Task ID | Region Tag | Priority | Requirements |
| Loading scenario planning criteria... | |||
2 Component: Physics Engine (PE)
The Physics Engine (PE) propagates the orbit of each satellite using Skyfield and the SGP4 algorithm. It applies maximum off-nadir angle and minimum elevation thresholds to identify valid physical capture and communication windows. Expected satellite attitude is computed in the Local-Vertical Local-Horizontal (LVLH) reference frame.
| Window Type | Associated Entity / Target Cluster | Payload Parameter | Start Interval (UTC) | End Interval (UTC) | Data Volume |
|---|---|---|---|---|---|
| Calculating geometric contact windows... | |||||
Note: The table above displays a subset of the first 15 computed pass windows for brevity.
3 Component: Request Generator (RG) & Validation
The Request Generator (RG) acts as the mission strategy layer, prompting an LLM with reverse-geocoded coordinates and relative planning horizons to synthesize natural language operational requests. To ensure strict operational fidelity, an automated evaluation pipeline converts the generated text into semantic embeddings to perform validation checks against the physical simulation's geometric ground truth.
Enriched Prompt Ground-Truth Context
Retrieving semantic validation categories...Generated Natural Language Operational Request
Generating natural language request expression...
Linguistic Alignment Performance
Validation via cosine similarity vector comparisons against reference control vocabularies (sensor types, priority tiers, planning days, and diurnal blocks):
Evaluation and Visualizations
Dynamic communication and imaging tracking windows generated during the initial 24 hours of the active constellation simulation scenario. Visualizations are represented as Google Earth projects rendered from calculated spatial-temporal window segments.
This is the result of exporting to KML and rendering in Google Earth Pro. The videos above are generated from the KML files for each satellite and ground station. Used code can be find in the export_to_kml.py script.