synthetic-data-generation

Synthetic Data Generator

Natural language interface for generating realistic oil & gas domain data.

Purpose

Generate synthetic well logs (LAS/DLIS), seismic (SEG-Y), core photos, and time-series data with proper physical constraints for testing, demos, and training.

Capabilities

Generate data with realistic relationships:

• Well logs (GR, RHOB, NPHI, RT, DT) with Archie equation constraints
• Seismic volumes with proper geometry
• OSDU-compliant metadata
• Core photos using AI image generation (requires API key, expensive operation)

Invoking

from synthetic_data.well_log import LASGenerator
from synthetic_data.seismic import SEGYGenerator

Well Log Generation

Create realistic LAS files:

generator = LASGenerator(seed=42)
las_path = generator.create_record(
    well_name="Test-Well-001",
    curves=["GR", "RHOB", "NPHI", "RT"],
    depth_range=(1000.0, 2000.0),
    sample_interval=0.15,
    lithology="sandstone"
)

Options:

• well_name: Well identifier
• curves: Log curves to generate (GR, RHOB, NPHI, RT, DT, CALI)
• depth_range: (start, end) in meters
• sample_interval: Sampling interval in meters
• lithology: sandstone | shale | carbonate
• seed: Random seed for reproducibility

Seismic Generation

Create SEG-Y volumes:

generator = SEGYGenerator(seed=42)
segy_path = generator.create_record(
    survey_name="Test-Survey",
    n_inlines=100,
    n_crosslines=100,
    n_samples=500,
    sample_interval=4000  # microseconds
)

Core Photo Generation

IMPORTANT: Expensive Operation

Core photo generation uses AI image generation APIs which are costly. Before generating:

1. Always ask for count - Never generate without explicit number confirmation
2. Ask about aspects based on detail level needed

Mandatory Questions

When user requests core photos, ask these in order:

Question 1 - Count:

"How many core photos do you need? (Image generation is expensive - each photo costs API credits)"