Page 95 - Well Logging and Formation Evaluation

Advanced Log Interpretation Techniques 85

• A better measurement of permeability than currently possible using
traditional poroperm-type plots
• In-situ measurement of oil viscosity
• Differentiation of oil/gas zones
• The elimination of the need to run nuclear sources in the hole

Overall, it may be said that some petrophysicists really believe in the
future of NMR logging and see such a tool eventually replacing conven-
tional logs. Others point to the fact that NMR logging has been around
for 15 years and has offered few real advantages in most fields. I have
seen many NMR logs in which the tool shows oil in known water legs
and both gas and oil both above and below the GOC (gas/oil contact). I
have also seen permeabilities differ by a factor of 10 or more when com-
pared with core-calibrated values derived from poroperm relationships.
However, I have also seen the tool explain why some zones, with high
total water saturation, are capable of producing dry oil. Therefore, there
are situations in which NMR can offer real advantages, but running the
tool should be justified on a case-by-case basis, and not just from a need
to be perceived as “high tech.”

5.5 FUZZY LOGIC

“Fuzzy logic” is a technique that assists in facies discrimination, and
that may have particular application in tying together petrophysical and
seismic data. In this chapter, the basic technique will be explained,
together with a worked example to illustrate the principle. Consider a sit-
uation in which one is using a GR (gamma ray) log to discriminate sand
and shale. With the conventional approach, one would determine a cutoff
value below which the lithology should be set to sand and above which
it should be set to shale. To use fuzzy logic, one would do the following:

1. In some section of the well where sand and shale can be identified with
complete confidence, one would generate a “learning set,” that is,
create a new log in which the values are set to 0 or 1 depending on
whether the formation is sand or shale.
2. Over the interval defined by the learning set, one would separate all
the bits of GR log corresponding to sand and shale, respectively.
3. For the sand facies, a histogram would be made of all the individual
GR readings. To this distribution would be fitted a mathematical func-
tion (most commonly a normal distribution) that would capture the