Fault picking

Fault picking software

Two routes to the same fault network: a neural-network wizard that runs a 3D volume from detection through to clustered fault sticks, and hand picking on every section, map and 3D view — in time or in depth, with the discontinuity attributes on screen while you work.

Download free trial How AI detection works Watch Demos

Detect → geometry

Neural network to fault sticks

Sticks or surfaces

Two meshing modes per fault

Time & depth

Converted through the velocity model

Interpreted fault planes in a 3D seismic view beside a crossline section carrying picked fault sticks and horizons in g-Space

Fault attribute volume rendered in the g-Space 3D view with its transfer-function editor

Discontinuity attribute on a depth slice showing a fault network as clear lineaments

In view

Fault sticks & fault planes

Fault picking is horizon picking's harder half. Instead of following a reflector you are chasing where it breaks — so the same correlation principles apply, but the emphasis moves to discontinuity, to the structural pattern the faults follow, and to the stratigraphic units they cut. g-Space attacks that from both ends: automatic detection turns a 3D volume into fault geometry while manual picking keeps the interpreter in charge of every stick, and the two produce the same kind of object — a fault in the project tree, in the time or depth domain, that the maps, the fault model and the structural framework all read.

AI fault detection

From a seismic cube to fault objects.

The AI Faults Wizard runs a complete cycle in three stages, each one feeding the next automatically. It needs a 3D SEG-Y cube in the project — two-dimensional data is not supported — and executes on CPU or on an NVIDIA GPU. Run the stages one at a time to inspect each result, or press Calculate All and let it work through them; either way the settings can be saved into a project workflow and re-run later.

Stage 1

AI Fault Attribute

A trained network analyses the fracture pattern and writes a fault-probability volume. Two models are available — a baseline that takes any window size, and a transformer fixed at 128 — with the analysis window, its overlap and a vertical range that can follow a horizon instead of a fixed time.

Stage 2

Attribute processing

Planarity measures how continuous the reflections are; from it come faultness, which highlights the fault centreline, and orientation, which records its strike in map view. A sharp, balanced or smooth preset covers most data, and a custom mode exposes the smoothing and continuity scales.

Stage 3

Sticks & faults

Sticks are traced along the centrelines, each extending while the fault stays pronounced and its strike stays inside the orientation tolerance. Sticks are then grouped into faults only where they are mutual neighbours — the rule that stops branches from being merged into one fault by mistake.

Filter

Dip, azimuth and size

Set dip, strike and size ranges on a compass before the run and only the faults that pass are ever created; after it, every fault of the run appears as a marker you can point at to read its dip, azimuth and size.

Restrict

Polygon and interval

Confine detection to a polygon and a vertical interval — useful for trying parameters on a representative area before committing the whole volume to a run that can take hours.

Output

Ready to interpret

Faults land in the Data Manager under Time or Depth, meshed as sticks or as surfaces, and are displayed in the 3D view, on the map and on sections — then edited, extended and exported like any hand-picked fault.

Interactive picking

The interpreter still owns the structure.

Automatic detection is a starting point, not a verdict. Everything below is the hand work around it — picking where the network is quiet, correcting where it is wrong, and giving the result the geological meaning a probability volume cannot.

Pick anywhere

Faults are defined on the map, on 2D lines, on inline and crossline sections, along an arbitrary line and in the 3D view — created in the Data Manager, or straight on the section with the picking tool active.

Sticks, then surfaces

Pick the fault as vertical traces section by section, then switch the meshing mode to turn those sticks into a continuous surface and see the whole fault plane inside the volume.

Reassign a stick

Drag a stick into another fault's folder when it turns out to belong there, or park it in Free Sticks while the structural picture is still forming and assign it later.

Attributes on the fly

Calculate a discontinuity attribute — coherence, variance and the rest — from the Attributes bar and it appears on the open section immediately; set the palette and pick against a sharpened image.

Fault prediction

A separate, non-AI tool opened on a time section: it correlates traces over a time and bin window with a maximum shift, and its result becomes a single stick, or sticks for every prediction on the section at once.

Faults from polygons

Draw a polygon on the map where you expect a fault and convert it into one — single polygon or several at once, with the number of sticks per break under your control.

Time and depth as one

Depth sections carry the same tools as time ones. Each stick keeps its own domain, and a click made in the other domain is converted through the current velocity model, so a fault can be extended either way.

Type and its symbol

Classify a fault as normal, reverse, strike-slip or a combination and the map draws it with the right symbol — ticks, teeth or paired half-arrowheads — on the correct side of the line, at a spacing you set.

Workflow

Six steps from a raw cube to a fault framework.

Sharpen the image

Bring a discontinuity attribute onto the section you are working, set its palette for maximum contrast, and see where the breaks actually are before picking anything.

Try the detection small

Run the wizard inside a polygon on a representative area, compare the two models and the sharp, balanced and smooth presets, and settle the parameters before a full-volume run.

Detect and cluster

Run the whole cube: probability, planarity, faultness and orientation, then sticks traced along the centrelines and clustered into faults inside the dip, azimuth and size ranges you set.

Interpret the result

Review the faults in 3D, on the map and on sections. Extend them by hand where the response was weak, reassign sticks that landed on the wrong fault, and add the ones detection missed.

Give them meaning

Assign each fault its geological type so the map carries the right symbol and the direction of movement reads correctly, and mesh the ones you need as surfaces.

Break the maps and the model

Rebuild the horizon maps honouring the faults, build the 3D fault network, and carry that framework into structural and geological modelling.

Fault objects

A fault is project data, not a drawing.

Every fault sits in the Data Manager with its domain, its sticks, its surface, its type and its colour — which is why the same object can drive a map, a model and an export without being redrawn.

What a fault carries

The parts of a fault object in the project tree

Domain

Time or depth, chosen when the fault is created, with conversion to the other domain adding a copy and leaving the original untouched.

Sticks

The picked traces themselves, each keeping its own domain, organised under the fault or held in Free Sticks until assigned.

Surface

A triangulated mesh through the sticks, with its own visibility so the plane and the traces can be shown apart or together.

Type

Unknown, normal, reverse, strike-slip, or strike-slip with a normal or reverse offset — stored as a property of the fault.

Point sets

Named sets of points kept under the fault and saved with the project, for the control points an interpretation needs.

Colour

Set per fault or assigned across a folder at once, and carried through to every view and to converted copies.

Where faults go next

What the rest of g-Space does with them

Maps

Set map interpolation to honour selected faults or a whole fault model, rebuild the horizon map, and the surface breaks where the structure does.

Map display

Fault plane projections along the current horizon or a depth map, drawn as filled polygons or as fault lines carrying the type symbol.

Fault model

The Fault Modeling wizard builds the 3D fault network from these picks and reads each fault's assigned type automatically.

Structural framework

That network is what structural and geological modelling honour when they build surfaces, layers and grids.

Exchange

Binary and Charisma in both directions, one fault, a selection or the whole folder in a single step, on the current datum.

Surfaces out

A fault's triangulated surface writes to GOCAD TSurf for modelling and simulation packages downstream.

Tools in this group

The parts of g-Space this topic is built from

AI Faults Wizard Fault picks Fault prediction Interpretation bar Faults Free sticks Fault model Attributes bar Import / export faults

More g-Space capabilities

Faults are half of the structural picture.

The other half is the horizons they cut, the wells that date them and the model they are built into. These are the other topics in the g-Space workflow.

g-Space overview Horizon picking Geological modeling Attribute analysis & AVO Well tie Seismic mis-tie Geological interpretation Petrophysics Velocity modeling Reservoir characterization

FAQ

Questions, answered.

What does the AI Faults Wizard actually produce?

Fault geometry, not just an attribute. A trained neural network first estimates fault probability across the cube; that result is turned into a planarity attribute measuring reflection continuity, and from it a faultness attribute marking the fault centreline plus an orientation attribute holding its strike. The last stage traces sticks along those centrelines and clusters them into individual faults, which land in the Data Manager ready to interpret, mesh, visualise and export. It needs a 3D SEG-Y cube in the project — two-dimensional data is not supported — and runs on CPU or on an NVIDIA GPU.

Can faults still be picked by hand?

Yes, and on every view: the map, 2D lines, inline and crossline sections, arbitrary lines and the 3D scene, in the time domain and in depth. You create a fault in the Data Manager or straight on a section, then pick sticks along it. A stick that turns out to belong to a neighbour is dragged to that fault's folder, and one you are unsure about parks in Free Sticks until you decide. Discontinuity attributes calculated on the fly appear on the open section immediately, so the pick is made against a sharpened image.

What is the difference between sticks and surface mode?

Sticks are the fault traces as picked, section by section — the representation to work and QC in. Surface mode reconstructs a continuous mesh through them so the whole fault plane can be seen inside the volume. Every fault carries a Surface node of its own with its own visibility, so the mesh and the sticks can be shown separately or together, and that surface writes out to GOCAD TSurf.

How do picked faults reach the maps and the model?

Give a fault its geological type — normal, reverse, strike-slip or a combination — and the location map draws it with the matching symbol: tick marks, teeth, or paired half-arrowheads for the sense of lateral movement. Set map interpolation to honour selected faults or a whole fault model and rebuild the horizon map, and the surface breaks where it should. The Fault Modeling wizard reads each fault's type automatically when it builds the 3D fault network, and faults convert between time and depth through the current velocity model.

Get started

Run the detection on your own cube.

Point the wizard at a polygon in your survey, compare the presets, and see what the network finds before you commit the whole volume — take g-Space for a trial run, or talk to Geomage about a demo.

Download free trial See pricing