20 min (MAX)

Thesis presentation

Good afternoon, as professor Amilcar Soares presented my name is Bruno Melo and I am here
today to present my thesis project about the prediction of formation fluid composition through gas
while drilling analysis.
Before starting I would like to leave a word of gratitude to Geolog for making this research
possible and for all the technical support. I also want to thank everyone that was somehow involved in
this project, with a special thanks to Mr. Gionata Ferroni, Professor Maria Joao and Professor Costa e
Silva for their guidance and availability during this project.

(Change to summary)
The presentation is divided in three major sections: in the first one I will do a little introduction
about the project and stress the main goal. In the second part I will review in detail the workflow
followed in this project along with the presentation of some results. To finish I will draw the final
conclusions, as well as, give some suggestions for further research about this subject.

(Change to objective)
Objective
The scope of this project is studying the relationship between the composition of mud gas
obtained by the mud logging companies, in real-time while drilling, and the downhole fluid samples
collected during well tests and then analysed in the laboratories.
The goal was to recognize consistent relationships, and understand the data uncertainty in the
comparisons between mud gas and bottomhole fluid data. The purpose of this study was the desire to
build a model able to predict, in near real time while drilling, an approximation of the real reservoir fluid
composition.
This is an important topic because in the case of borehole instability or any other hostile
circumstances that prevent the realization of wireline logging, mud gas analysis may be the only formation
evaluation tool available to provide hydrocarbon type information.
It is also relevant to stress the advantage of mud gas analysis in the decision of the testing tool
string design, in the depth selection for the sampling points, and it can also help focusing the formation
evaluation program on any spotted anomaly.

(Change to summary)
Workflow
After this overview about the project, we will now review the followed workflow.

(Change to workflow)
The fluid datasets used in this project are from eight different wells drilled under different
conditions around the world. The datasets were manipulated in order to remove companies name, depth
information and geographical coordinates.
For each well was available the gas log recorded by the mud logging company, and the downhole
fluid information acquired during well tests. It is important to stress that different collect and analyse
techniques were used for each fluids. In this way it is crucial to start by doing a pre-analysis check on the
datasets focusing the data collection and analysis procedures.
The next step was a quality control check on the raw mud gas data, on downhole fluid data and gas
equipment data. If the well is adequate for further studies then we proceed to the main stage of this
research where the datastets were treated in order to become comparable measurements. The next step
was the selection of data for fluid composition comparison and the associated data analysis. The ultimate
goal, after this analysis, was the creation of a model able to predict, in near real time while drilling, the
downhole fluid composition based on the mud gas shows at surface.
 (Change to Pre-Analysis)
So let’s start by analysing the previously mentioned pre-analysis stage. As you can see the
information for both datasets is summarized in these two tables.
In the mud gas dataset, it is highlighted in green the common characteristics for all the wells. So for
this project, the mud gas data was acquired with the constant volume degasser and analyzed the DualFID
chromatograph, which analysis components from methane to pentane. In yellow it is highlighted some
additional features present in some wells as the use of heater or the heavy gases chromatograph. In red it
is stressed the missing extraction efficiency coefficients for all the wells.
The downhole fluid data characteristics are presented in this table. It is possible to see that some
wells have bottomhole sample, others have surface sample, and different types of fluid analysis. An
important factor here is that some fluid samples are from adjacent wells and other from the same well that
was logged at surface during drilling.
Since it is desired to have a common starting point for further studies, the key information here is
that we will only use raw mud gas data from methane to pentane, and downhole recombined fluid
information, which is considered as the best approximation of the real reservoir fluid composition.

(Change to QC assessment)
As mentioned the second step is the QC assessment. As you can see in this image, the quality
control check was performed on the gas data, and on the gas equipment. Specifically it was checked the
constant physical extraction conditions, for example flows and temperatures.
The QC of gas data was performed through a gas quality ratio, which compares the readings of the
chromatograph against the readings of the total gas. The total gas instrument is a FID sensor without a
separation column in this way considers all the organic matter in the sample. The ratio should be between
0,8 and 1,2 if not, the problem can be related to a low concentration of gas, or to the wrong calibration of
gas equipment, or even due to contaminants in the sample.

(Change to Data selection & Treatment)

Now if the well is adequate for further studies, the next step is the data selection and
treatment. The bottomhole fluid composition for each well was extracted from the PVT reports. The
compositional information, expressed in terms of molar percent, considers hydrocarbons from C1 to
C27+. Once the comparison with mud gas will be carried out using a range from methane to pentane,
the next step is to recalculate the downhole fluid molar percentages. The new values in this table were
obtained applying the equation presented here.
In the mud gas datasets, the first objective consisted in identifying the gas peak associated to
the depth of the downhole test. In this case two peaks can be observed, however the first peak has
almost the double of gas concentration compared with the second one. Consequently, the first gas
show was selected for this comparison. To allow a proper comparison the mud gas concentration in
parts per million (PPM) was transformed in percentage values.

(Change to EEC correction)

The next step was the application of the extraction efficiency coefficient to the selected mud
gas data. The extraction efficiency coefficient is necessary because not all gas in the mud is extracted
in the degassing process. By applying a field procedure, in different wells around the world, Geolog
found three different sets of EEC values, dependent on the mud properties.
The correction of the previously selected mud gas composition was performed for each well
analysing the mud type and temperature. The proper EEC set was selected, the correction performed
and the relative percentages recalculated one more time.
(Change to Comparison of data)
 At this point the gas peaks associated to the downhole fluid samples are identified and
characterized.
Now the first target when comparing these two different types of information will be the
validation of the selected gas show. In this figure it is possible to observe that mud gas has almost a
perfect compositional match with downhole fluid collected at surface, therefore, it proves that both
fluids have the same origin. This can be easily explained once both samples were collected at similar
conditions.
Once confirmed the same origin for both fluids, the next step was the behavior study of the
bottomhole sample flashed gas (BHS FG), the bottomhole sample recombined gas (BHS RG), the mud
gas, and the mud gas corrected with the extraction efficiency coefficient. It is important to remember
that for this project it is considered that bottom hole sample recombined gas concentration is similar
to the real reservoir fluid concentration.
When observing the Pixler plots it is clear that mud gas composition becomes much closer to
the BHS RG composition once the extraction efficiency coefficients are applied. When analyzing all the
pixler plots it is also safe to infer that mismatches get worse when the gas component concentration
becomes smaller. That is the reason why the ratios like C1/C4 or C1/C5 have large discrepancies
when comparing the corrected mud gas with the downhole recombined gas.

(Change to Data analysis)

The next step was the data analysis, which started with the calculation of the correlation
between the corrected mud gas and the downhole-recombined gas. When comparing in this table the
correlation factors for C3 , nC4 and nC5 , it is possible to notice a decreasing trend with the increase
in molecular weight. A feasible explanation for this fact can be that a lower gas concentration is
associated to the heavier components, which will make them more susceptible to be masked due to
errors or contaminations. Another interesting information that can be drawn from this table is the
different correlations values between iso-molecules compared with normal-molecules, as the case of
iC4-nC4 and iC5-nC5. Nonetheless the iso-butane has a bigger correlation factor than iso-pentane
following the above conclusion.
To better understand these results it was decided to calculate the relative errors between the
compared data. The information was displayed in box-plots to clearly visualize the results. The
common characteristics are the small range for errors in lighter gases (from C1 to C3) and the wider
interval of error values when analyzing butane and pentane fractions. It is also valid to say that iso-
molecules have a wider dissemination of errors than the normal molecules. This distribution of errors
supports the previous conclusions.

(Change to Model analysis)

After the comparison of the data, the next step was de development of the desired model. The
first step was the representation of the bottomhole recombined gas sample versus the mud gas
concentration corrected with the extraction efficiency coefficient. Observing the distribution of the
points, it was decided to test a linear regression fit. Where m is the coefficient measuring the
relationship between the recombined gas values and the corrected mud gas composition, and epsilon i
(𝜀i) is a random error term. In these regressions the constant (a) is zero once a null mud gas
composition value implies a zero recombined gas concentration. The model parameters were
estimated with the least square method.
The proposed regressions have a confidence interval of 95%. Therefore, if P-values are lower
than 0,05 the null hypothesis is rejected, proving the statistical validation of this model. Observing the
P-values in the table is possible to confirm the statistical significance of all the proposed regressions.
The R2 is the coefficient of determination, and gives information about the quality of the model
to represent the desired variable. Taking methane as an example, the R2 of 0,994 means that 99.4%
of the variation on the bottomhole recombined gas is explained by the corrected mud gas values.
Analyzing the residual plot, the points appear to have a random pattern around the horizontal
axis, being symmetrically distributed above and bellow the zero line, indicating a good fit for the
linear model.
 (Change to summary)

Final Considerations
Lets now review the main conclusions and give some possible paths for further research about
this subject.
(Change to Conclusions)
Lets start by stressing that the regressions used for the predictive model only have seven
points. Another point of concern is the heterogeneity of the mud gas data, which is introducing some
additional uncertainty.
Nevertheless, in this project we exposed some interesting facts, as the composition similarity
between raw mud gas data and downhole fluid sample collected at surface. As well as the fact that the
corrected mud gas follows the bottomhole sample recombined gas compositional behavior. We also
built a statistically valid predictive model that is also robust to forecast concentrations from C1 to C3.
(Change to Recommendations)
As mentioned the main achievements in this project were the verification of the relationship
between mud gas and downhole fluid, and the development of a successful procedure to build a model
able to estimate an approximation of the reservoir fluid composition from mud gas.
The recommendations for further studies are divided in three categories. The first one is the
expansion of the model limits by extending the research until heavy gases; the second is the confirmation
of the model behaviour by updating it with more data. And the most important step is the error mitigation,
which can be improved by doing a mud fingerprinting procedure and by only feed homogeneous mud gas
data to the model.