| OpendTect Workflows Documentation version 4.2 |
|---|
| Prev |
|
Next
|
Purpose: Create a Fault "probability" Cube for fault interpretation. A Fault cube is typically used to visualize the larger scale faults. Detailed faults are better visualized by Similarity on Fault enhancement filtered seismic data.
Theory: Several attributes in OpendTect can be used as fault indicators e.g. Similarity, Curvature, and Energy. A Fault Cube is a new seismic volume that highlights faults by combining the information from several fault indicators into a fault "probability". This is done by training a neural network on two sets of attributes extracted at example locations picked by the human interpreter: one set representing the fault class and the other representing the non-fault class (i.e. normal seismic response).
Software: OpendTect + Dip-Steering + Neural Networks
Workflow:
- Scan your data set for obvious faults.
- Create two New Picksets: one for Faults and one for Non-faults (right-mouse click in the tree).
- Pick examples for faults and non-faults (Select the Pickset in the tree and then left-click in the scene on the element at the position you want to add; Control-click to remove a pick). Try to pick a representative set for both faults and non-faults. This means: pick different faults, pick as many points as possible (several hundred picks for each is typical); for non-faults pick both low- and high energy zones, also pick noisy zones that are not faulted.
- Open the attribute set window and open the default set called: NN Fault Cube. Select seismic and steering cube and Save the attribute set.
- Open the Neural Networks window. Select Pattern recognition (Picksets). Select Supervised, the input attributes, the Picksets (Faults and Non-faults) and the Percentage to set aside for testing (e.g. 30%).
- Train the neural network. Stop where the test set has reached minimum error (beyond that point overfitting occurs: the network learns to recognize individual examples from the training set but looses general prediction capabilities). Store the trained neural network.
- Apply the trained neural network to the seismic data in batch: Processing - Create seismic output, or on-the-fly: right-click on the element in the tree (e.g. part of an inline). You can choose between 4 outputs (Faults_yes, Faults_no, Classification, Confidence): Choose Faults_Yes to create the Fault "probability" Cube.
Tips:
- The default attribute set can be tuned to your data set by changing parameters, and adding or removing attributes.
- The colors in the neural network indicate the relative weight attached to each attribute (ranging from white via yellow to red). White nodes indicate low weights meaning the attributes are not contributing much and can be removed to speed up processing time.
- Display the Mis-classified points (Pickset tree) to evaluate why these are mis-classified. If you agree with the network you may want to remove some of these points from the input sets and retrain the network. This will improve the classification results but the process is dangerous as you are working towards a solution.