[OpendTect_Users] fault detection by attributes

[Duan, Taizhong (MRO)]

Friso,

 

Thanks for your help. I will try some your ideas. I have another
question. In OD algorithms, how is the steering cube used? as a
constraint? Or something else. In which OD documents, I can find the
detailed explanation? For example, how is the dip steering cube applied
in calculating the similarity?

Regards,

Taizhong

 

From: Friso Brouwer [mailto:support-americas at dgbes.com] 
Sent: Tuesday, May 31, 2011 10:26 AM
To: Duan, Taizhong (MRO)
Cc: users at opendtect.org
Subject: Re: [OpendTect_Users] fault detection by attributes

 

Dear Taizhong,

That is a non-trivial issue and it may be impossible achieve a complete
removal of stratigraphic features. I do not think the filtering of the
steering cube will make much of a difference, I advise to use the
background steering cube a defined in our standard workflow
<http://www.dgbes.com/images/stories/PDF/effectivedipsteeringworkflowusi
ngbgsteering_primerodata.pdf> . 

What could help, or at least mitigate your problem are the following
solutions, in order of complexity (try them in order) :

1.	Increase the time window of the similarity attribute, this will
tune out vertical small features, which are often the stratigraphic
features, not the faults.
2.	Do post-attribute filtering on the similarity volume, using the
volume statistics.
3.	Use an completely other attribute for fault detection, curvature
is a good candidate, here an article
<http://www.dgbes.com/images/stories/PDF/fault_attributes_niger_delta_aa
pg.pdf> .
4.	If the fault system and stratigraphic features have different
directions in the horizontal plane, one can separate them using
directional decomposition of the attribute. This is not trivial, and if
this is an option for you, let me know, so I can explain you the
details.
5.	Use a second attribute that detects the stratigraphic features,
but not faults (e.g. a small window energy). Then use a cross-plotting
approach to identify the domain in the crossplot where the faults are
"clean" and use the mathematics attribute (conditional statements) to
pass only attribute combinations within this domain.
6.	Extending on point 5 one can choose an user-driven
multi-attribute neural network approach. This relies on there being a
multiple attributes that each can partially separate faults and
stratigraphic features. Based on user picked examples the neural network
will then "learn" how to combine the different attributes to arrive at a
more complete separation of faults and stratigraphic features. Here is
an article
<http://www.dgbes.com/images/stories/PDF/eage_2000_meldahl_obj.pdf>  and
product information
<http://www.dgbes.com/index.php/neural-networks.html> .

All in all it remains a difficult problem as faults and stratigraphic
features often have there main characteristic/attributes, such as
discontinuity (similarity) and curvature, in common. Still with the
above methods you should be able to at least achieve an improvement. I
gladly help you with the further details if you choose to experiment
with one of the methods.

Kind regards,

Friso

-- 
Friso Brouwer
Support Americas
dGB Earth Sciences

+1 281 240 3939 (o)
support-americas at dgbes.com

On Thu, May 26, 2011 at 9:30 PM, Duan, Taizhong (MRO)
<tduan at marathonoil.com> wrote:

Hi all,

 

What is the way to filter out all stratigraphic discontinuities in
similarity calculation? The dip cube steering is supposed to do so?
(bkg: I tried with dip steering cubes filtered with i-j stepout from
3-9, but it seems make not much difference, i.e., there is still
significant stratigraphic discontinuities left on the volume. Of course
I want to keep all non-vertical faults/fractures).

 

 Anyone gets a better workflow?

Thanks.

 

Taizhong Duan

 

Marathon Oil      


_______________________________________________
Users mailing list
Users at opendtect.org
http://lists.opendtect.org/mailman/listinfo/users





-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.opendtect.org/pipermail/users/attachments/20110602/20214460/attachment.html>