kucherenko/petropowers/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)"

Question 2 - Lithology:

"What lithology? (A) Sandstone (B) Shale (C) Carbonate (D) Limestone (E) Dolomite (F) Mixed/varies"

Question 3 - Visual Features (if user wants detail):

"Any specific visual features? (A) Default/random (B) With fractures (C) With oil staining (D) Specific bedding angles (E) Let me specify"

Question 4 - Full Context (if user wants geological accuracy):

"Need full geological context? If yes, specify: depth range, formation name, field name, well naming convention"

Usage

from synthetic_data.core_photos import CorePhotoGenerator
generator = CorePhotoGenerator(api_key="YOUR_GOOGLE_AI_API_KEY")
result = generator.create_record(
    well_name="Test-Well-001",
    lithology="sandstone",
    depth_range=(1000.0, 2000.0),
    output_dir="./output"
)

Options:

• well_name: Well identifier
• lithology: sandstone | shale | carbonate | limestone | dolomite
• depth_range: (start, end) in meters
• core_length: Length of core sample in meters (default: 1.0)
• field_name: Field name for metadata
• formation: Formation name for metadata
• output_dir: Directory to save images and metadata

Example Assets

Reference examples available at synthetic_data/assets/:

• core-photos.png - Real core sample photographs showing proper format
• core-photos.webp - WebP version

These show proper format: depth markers, well identification, lithology variations, and scale bars.

Validation

Files validate against domain libraries:

• LAS → lasio
• SEG-Y → segyio
• DLIS → dlisio

OSDU Integration

Generate manifests for OSDU:

from synthetic_data.osdu import LASMapper
mapper = LASMapper()
manifest = mapper.to_manifest(
    las_path,
    legal_tags={"legaltags": ["test-license"]}
)

Example Usages

User says:

"Generate 5 LAS files for testing"

→ Uses LASGenerator.create_dataset(n_records=5)

User says:

"Create realistic well log with GR, RHOB, NPHI for carbonate formation"

→ Uses lithology="carbonate"

User says:

"Generate well log data with OSDU manifests for demo deployment"

→ Creates LAS files + OSDU-compliant JSON manifests

User says:

"Generate core photos for testing"

→ Ask: "How many?" then "What lithology?" then generate

User says:

"I need 5 sandstone core photos with visible fractures"

→ Confirms count (5), lithology (sandstone), feature (fractures) → generate

User says:

"Generate synthetic data for my reservoir project"

→ Ask: "What type of data? (well logs, seismic, core photos, production data)" → If core photos selected, proceed with mandatory questions

Software Tasks

If user requests:

• Full OSDU ingestion pipeline → invoke petropowers:oil-gas-delegation
• Real-time data generation → Explain this is batch-only
• Proprietary format support → Explain only open standards (LAS, SEG-Y, DLIS) supported

Dependencies

pip install lasio welly dlisio segyio numpy pandas scipy

Reference Documentation

See references/physical-constraints.md for detailed petrophysical relationships.