Quality Assessment and Consistency Evaluation of Hydrocarbon PVT Data

Abstract

Hydrocarbon PVT data are used for a wide range of reservoir engineering applications including early volumetric assessment, well test evaluation, material balance calculations, and equation of state fluid characterization. High quality, accurate PVT data can reduce uncertainty in reservoir fluid properties, provide a sound foundation for reservoir engineering studies, and improve investment efficiency. Poor quality PVT data can result in lost time due to rework and additional studies, inadequate or overly aggressive development plans, and inefficient investment. While acquiring sufficient volumes of representative reservoir fluid samples is the first step in obtaining reliable PVT data(1), understanding and identifying the quality of the PVT data once obtained from the laboratory is essential to ensuring that data are applied appropriately. A number of techniques exist that can be applied to evaluate PVT data quality and significant experience in using these techniques has been gained over the years. Examples and illustrations of evaluating data quality for fluid compositions and flash data, oil PVT data, and gas condensate PVT data are provided.

Introduction

The techniques described have primarily been used to assess PVT data quality for equation of state fluid characterization, but these techniques have much broader application such as ensuring an appropriate PVT basis for in place volume estimates, analytical reservoir engineering calculations, and regulatory submittals. When preliminary laboratory PVT data are provided to the reservoir engineer, these techniques can be used to screen that data. The reservoir engineer can then use the results to provide feedback to the laboratory and identify when measurements or laboratory calculations may need to be redone.

Equation of state parameters can sometimes be tuned beyond reasonable bounds to match PVT measurements that are not physically realistic. When an equation of state is overly tuned to bad data, at least two issues arise. First, the quality of the match to the remaining good quality data is often worse. Second, and more importantly, the ability to predict properties outside the range of the measurements is compromised. By using these quality and consistency evaluation techniques, discrepancies can usually be resolved thus enabling the reservoir engineer to improve the quality of the fluid characterization and property predictions.

Often reservoir fluid samples and laboratory PVT data are acquired over a period of time. Comprehensive assessment of the data is often performed after data have been acquired from many laboratories using different techniques. This often leads to differences in measurements between laboratories and/or datasets that need to be resolved. When differences in the data are observed, the techniques described can be useful for resolving these differences and in identifying data most appropriate for reservoir engineering applications.

Discussion

The methods described in detail include but are not limited to:

- Material balance checks and Hoffman plots(2) to assess consistency of compositional and flash data

- Graphical techniques to assess the consistency between constant volume depletion and constant composition expansion data for gas condensates with methods applied to liquid dropout data, Z-Factor data, and density data

- Reference volume translation techniques for formation volume factor and solution GOR measurements to assess data quality at low pressure and ensure that differential liberation data for oils are used appropriately

- Cross-plot techniques for comparing data from a variety of sources

Composition and Flash Data

One of the first steps in evaluating laboratory PVT data is to assess the consistency of compositions. Historically, a simple consistency test was to sum the compositions to ensure they added to one. This was particularly important in the days when reports were typewritten and typographical errors could lead to erroneous compositions. Today, however, when reports are generated using spreadsheets, summation calculations are normally done automatically and typographical errors are now unusual. Discrepancies in composition sums from modern PVT reports are usually small and related to significant figures rather than errors.

Two other techniques for evaluating consistency of flash and composition data have also been applied for a long time, but are still very useful today. These methods are material balance tests and Hoffman plots. The material balance plot can be used as a rigorous criterion for consistency while the Hoffman plot is considered a qualitative assessment of data quality. Each of these two tests are discussed in more detail below Material Balance Tests:

The material balance test is often used as a rigorous test to evaluate compositional consistency between feed composition and separator vapor and liquid compositions. The test is derived from component material balance criteria:

Fzi = Lxi +Vyi

Where:

F = Total moles of feed

L = Total moles of separator liquid

V = Total moles of separator vapor

zi = mole fraction of component i in the feed

xi = mole fraction of component i in the liquid

yi = mole fraction of component i in the vapor

This equation can be rearranged to the following:

yi/zi = F/V – (L/V)(xi/zi)

A plot of yi/zi versus xi/zi should yield a straight line with a slope of –L/V and an intercept of F/V. Any deviations from a straight line reflect material balance discrepancies. Reservoir fluid compositions are typically calculated from a mathematical recombination of flashed vapor and liquid compositions. When feed compositions are determined from a mathematical recombination of the separator vapor and separator liquid compositions, a perfectly straight line should be observed in the plot.

In this case, the plot is sometimes useful for identifying discrepancies in the reported compositions. The reciprocal of the slope of the line can then be used to calculate the GOR and compare that with the reported GOR. The liquid density and molecular weight must be provided so that the conversion from moles to barrels can be made. When the feed composition is truly independent from the vapor and liquid compositions, the material balance test provides a good check on the overall consistency of the analyses.

Table 1 provides compositions for an example gas condensate and Figure 1 is the resulting material balance plot when yi/zi is plotted against xi/zi. In this case, the wellstream composition was the result of a mathematical recombination of the separator vapor and liquid so the straight line in the plot should be no surprise. The slope of the line in the figure is -0.622056, which can be converted to a GOR of 16,708 SCF/Sep Bbl by taking the negative reciprocal of the slope and converting it to the measured units. This calculated GOR is about 0.5% higher than the reported GOR of 16616.8.

While not an exact match, the calculated GOR in this example is close enough that liquid molecular weight, liquid density, and the recombination ratio all appear to be reasonably consistent with the reported GOR.

The material balance test can be more useful when the feed composition is measured independently of the separator or flash compositions being evaluated for consistency. When this is the case, deviation of the data from the straight light trend is indicative Liquid Gas Wellstream of errors or uncertainty in the measurements rather than a problem with significant figures or mathematical consistency when the compositions are dependent. Material balance plots can be used for analyzing results from constant volume depletion data or differential liberation data as well.

Hoffman Plots:

A qualitative test to evaluate consistency of Kvalue (y/x) data is often referred to as the Hoffman plot. This technique was published in 1953 by Hoffman, et al.(2) and is based on a technique developed in the 1930’s by S. E. Buckley of Humble Oil and Refining Company. This method utilizes a log-linear plot of K-value versus Hoffman factor, F. F is defined as follows:

F = (log(Pc)-log(14.7))(1/Tb-1/T)/(1/Tb-1/Tc)

Where:

K = K-value (y/x)

P = separator pressure (psia)

T = separator temperature (R)

Tb = boiling temperature (R)

Tc = critical temperature (R)

Pc = critical pressure (psia)

When log(K) is plotted versus F for each component, the result should be a nearly straight line for light hydrocarbons. Light non-hydrocarbons should be close to the behavior of light hydrocarbons, but not necessarily on the same line. Iso- and normal- butane and pentane often fall on either side of a straight line. Additionally, some curvature might occur for heavier hydrocarbons, but extreme curvature can be indicative of potential data issues, often losses in the vapor phase heavy components.

The compositions from Table 1 representing the Example 1 gas condensate were also used to generate the Hoffman plot shown in Figure 2. The normal boiling point, critical temperature, and critical pressure were from standard values for the light components and from standard correlations for the C6+ components. K-values generated using an equation of state are also plotted with the reported data. In this case, the reported data fall close to a straight line with no obvious data issues. Another set of compositions from a second example, in this case for an oil, is provided in Table 2. A material balance plot for these compositions is shown in Figure 3. As expected for mathematically recombined compositions, all of the points fall on a straight line and no obvious discrepancies are identified from the plot. While the plot itself appears consistent, the calculated GOR from the slope of the line is about 5% higher than the reported GOR (604 SCF/BBL vs 572 SCF/BBL). Discrepancies of this nature are often due to misreported or incorrect liquid molecular weights or densities and often result in difficulties when attempting to match predictions with an equation of state.

The Hoffman plot shown in Figure 4 was also made for the Example 2 oil from Table 2. In this case, the points for methane and ethane appear to fall on a very different trend from the rest of the hydrocarbon components. Nitrogen and carbon dioxide also deviate more significantly than expected. Equation of state generated values are also plotted to provide a reference for comparison purposes. The reservoir engineer initially had greater difficulty calibrating equation of state predictions to this and other data. Once the material balance and Hoffman plots were made, the discrepancies in composition became apparent and the data from this particular well was discarded in favor of data from other wells that passed consistency checks.

Differences between experimental data and the expected trend in the Hoffman plot could be due to unstable separator conditions during the experiment or incorrect compositions of the separator products.

Unstable separator conditions during experiment lead to compositions that are not in equilibrium with each other. In this case, the computed feed composition may still be correct but the reported values for GOR, liquid density, and liquid molecular weight should be discarded. If the separator product compositions are measured incorrectly, then the feed composition should be discarded.

Consistency Tests for Gas Condensate Data

Other authors have written on the topic of evaluating gas condensate data, but most have been focused on evaluating compositional data such as Whitson(3) and/or usage of the data, Moses(4). An area that has not been given as much attention is evaluating the bulk property data from constant composition expansion (CCE) and constant volume depletion (CVD) tests. The data from these laboratory tests are often used to calibrate or tune equation of state models and understanding the quality of the bulk data is an important aspect of the tuning process. Significant insight can be gained by simultaneously plotting various properties from the two experiments together.

One of the properties typically given important consideration during equation of state tuning is liquid dropout data. Liquid dropout data are obtained during both CCE and CVD experiments. While CVD liquid dropout data are nearly always reported relative to the dew point volume, CCE liquid dropout data might be reported relative to the dew point volume, the total volume, or both. Simultaneously plotting liquid dropout data relative to the dew point from both experiments has little value, but plotting liquid dropout data from CCE relative to the total volume with the liquid dropout data from the CVD relative to the dew point volume turns out to provide significant insight to the consistency of the data from the two experiments.

When plotted in this manner, the following characteristics should be exhibited:

• Dew points should be identical

• CVD liquid dropout should be greater than CCE liquid dropout (on a total volume basis)

• CVD maximum liquid dropout should occur at a lower pressure than the CCE maximum liquid dropout.

The initial retrograde portion beginning at the dew point and decreasing in pressure can be nearly identical for the two experiments at high pressures, but the CVD should exhibit greater liquid dropout as more liquid condenses. The simple explanation for this behavior is that the composition of the gas phase is leaner than the liquid phase once retrograde condensation begins. The leaner gas is removed during the CVD experiment resulting in a composition that is richer than the original composition that remains in the CCE experiment. Because the total composition for the CVD fluid is richer, it will have greater liquid dropout when compared on a total volume basis.

Figure 5 shows the CCE and CVD data from the Example 1 gas condensate and provides an illustration of consistent liquid dropout behavior for a gas condensate. The liquid dropout is in Figure 5 is plotted relative to the total volume in the cell for both the CVD and CCE experiments. In the case of the CVD liquid dropout, the total volume and the dew point volume are obviously the same. Note that this comparison only indicates that the liquid dropout data from the two tests are consistent with each other. The test does not imply accuracy for either test.

On the other hand, when CCE data and CVD data are inconsistent, at least one of the tests is certainly inaccurate.

Figure 6 presents liquid dropout data for a rich gas condensate. The liquid dropout data from the CCE experiment were originally reported relative to the dew point volume. Little insight can be gained when these data, represented by the open squares labeled as CCE Raw in the figure, are compared with the CVD liquid dropout data, represented by the solid diamonds. By converting the CCE liquid dropout data to be relative to the total volume the comparison between CVD and CCE liquid dropout data becomes much more meaningful. This conversion is accomplished by dividing the liquid dropout relative to the dew point volume by the relative volume, V/Vsat, at each pressure to obtain liquid dropout relative to the total volume and is described in equation form as follows:

Liquid dropout (% of total volume) = Liquid dropout (% of dew point volume)/(V/Vsat)

The solid squares in Figure 6 represent the same CCE liquid dropout data relative to the total volume. In this case, the facts that the CCE liquid dropout data is initially greater than, then crosses the CVD liquid dropout data to become lower in the retrograde region is meaningful and indicates that the liquid dropout data between the two experiments is inconsistent.

While the data from one or the other, or possibly both experiments, are inaccurate, the test does not provide a clear indication of which experiment is likely more accurate. In the case of this particular gas condensate, other samples were available that had consistent data so these particular data were given no weight during equation of state tuning. Once the equation of state was tuned to other available data, comparisons made with the tuned equation of state suggested that first CVD liquid dropout measurement at approximately 3400 PSIA was in

error.

Theoretical predictions with an equation of state provide a clearer picture of the expected behavior. Figure 7 presents theoretical predictions of the same rich gas condensate using a tuned equation of state. The solid line represents the theoretical CVD liquid dropout and the line with squares represents the CCE liquid dropout relative to the dew point volume. While the trend for liquid dropout relative to the dew point volume is typically for the CCE to be greater than the CVD initially, it can also cross during the revaporization portion of the curve as described by Whitson(5) making evaluation of data consistency difficult. When the comparison is made on a total volume basis as represented by the dashed line for the CCE, the behavior is now such that the CVD will consistently have at least as much liquid dropout as the CCE.

Other simultaneous plots of CCE and CVD data can also be very useful in identifying data consistency. Two properties often compared are Z-factors and vapor phase density. Sometimes only Z-factors or only vapor phase densities are reported. It is very common for laboratories to report only Z-factors with CVD data. When this occurs, it is often still useful to calculate the density for comparison with the CCE data. This can be done using the Real Gas Law:

ρ = P/ZRT

Where:

ρ = Molar density

P = Pressure

Z = Z-factor

R = Gas constant

T = Temperature (absolute)

The molar density can then be easily converted to mass density by multiplying by molecular weight. Combining these tests has proven useful over the years. Recently, a reservoir engineer had laboratory measurements made on a gas condensate in support of government regulatory filings. One of the important parameters was the density at initial reservoir conditions.

Once the preliminary data were provided, the consistency test for density was performed. Figure 8 shows the comparison of the density data from the CVD and the CCE. The procedure above was used to convert the CVD vapor phase Z-factors to mass density so that they could be compared with the reported CCE mass densities. The plot demonstrated that the densities from the two experiments were not consistent and supporting equation of state calculations showed excellent agreement with the CVD density values. The laboratory was initially reluctant to revisit their CCE densities solely based on equation of state comparisons, but once the plot of densities from the two experiments was provided to them, they identified that incorrect conditions had been used to calculate the original CCE vapor densities and the corrected values shown in Figure 8 were then provided.

Material balance and composite Hoffman plots are also useful in evaluating compositional data from CVD tests. Whitson(3) provides a good discussion on evaluating compositional data from constant volume depletion experiments. The methods described enable the calculation of liquid compositions from the measured vapor compositions and volumetric data. Once liquid compositions are calculated, composite Hoffman plots can be made to evaluate the consistency of the compositional data. Combining the material balance tests and Hoffman plots provides insight to understanding when the compositional measurements from CVD experiments should and should not be used for equation of state tuning.

Consistency Tests for Oil Data

McCain(6), Al-Marhoun(7), and others have written extensively on analyzing black oil PVT reports for the purposes of combining results from differential liberation data and separator test data to obtain appropriate formation volume factors and GORs for reservoir engineering purposes. When production separator conditions are different from the laboratory or test separator conditions, a tuned equation of state is often used to model fluid properties for either compositional simulation or PVT table generation. Understanding what data are appropriate for tuning purposes is a critical step in equation of state fluid characterization.

Plotting reported properties as a function of pressure is a simple first step in assessing oil data quality. Properties for undersaturated oils should exhibit smooth monotonic trends with pressure. When plotting properties for undesaturated oils, compressibility should decrease with increasing pressure, while density and viscosity should increase with increasing pressure.

These might seem like relatively simple checks, but they can be very revealing. Figure 9 represents viscosity measurements for a highly underaturated oil that were recently obtained. The original viscosity measurements show a subtle change in slope at about 12000 PSIA. Because viscosity is such an important parameter for understanding performance of undersaturated oils, this issue needed to be resolved. Sometimes, phenomena such as asphaltene precipitation can cause similar behavior, but this did not appear to be the case in this instance. Calibrations were subsequently checked and viscosities remeasured using multiple techniques. Once all of the data were collected, the original viscosities were updated to the values shown in the figure.

The Y-function is commonly used for smoothing CCE data and assisting in precisely determining bubble points for oils. Ahmed(8) describes the Y-function and its usage for smoothing and interpreting bubble point data. Bubble point and relative volume data reported in hydrocarbon PVT reports for oils have typically been interpreted and smoothed in this fashion.

Because the data have already been smoothed, consistency problems with relative volume data are usually hidden. One indication that there might be issues with relative volume data or the bubble point measurement is when there are few points reported near the saturation pressure. Often the laboratory does not report all of the measurements so checking with the laboratory to verify the measurement is a good idea if the data is difficult to match during equation of state tuning.

Another very simple qualitative test for oil data consistency is to compare the residual oil API gravity from the differential liberation with the stock tank API gravity from a staged flash or separator test. Typically, the oil remaining from a differential liberation will be denser and thus have a lower API gravity than the stock tank oil from a staged flash. This is because the differential liberation is usually performed at a much higher temperature. When there is little difference between the differential liberation temperature and the flash temperature, this might not be true, so this test is not rigorous. While this test is not completely rigorous, it can be very useful for identifying inconsistent data. Possibly the most useful technique for determining what differential liberation data should be matched during equation of state tuning is to convert solution gas-oil-ratio (GOR) to cumulative GOR referenced to the bubble point volume and convert formation volume factors to liquid shrinkages. These changes are important because the reported solution GOR and formation volume factor data will include any error that occurs during the experiment in all of the data points. Differential liberation formation volume factors will include any error in liquid measurement during the experiment because the values are all referenced to the residual liquid from the experminet. Differential liberation solution GOR will include any errors in gas measurement as well as any error in liquid measurement. In addition to the challenges associated with measurement of the final depletion point, the theoretical simulation of the last step might not replicate the laboratory process. The most common points when errors are introduced result from depressuring to atmospheric pressure. This can make understanding the data and tuning an equation of state very difficult. The conversions to eliminate the propagation of these errors are described as follows:

Rcbp = (Rsbp – Rs)/Bobp

V/Vsat = Bo/Bobp

Where:

Rcbp = Cumulative GOR referenced to bubble point volume

Rsbp = Solution GOR from differential liberation at bubble point

Rs = Solution GOR from differential liberation

V/Vsat = Liquid shrinkage

Bo = Formation volume factor from differential liberation

Bobp = Formation volume factor from differential liberation at bubble point

If any errors were introduced during the laboratory procedures, these conversions will remove the errors from any points that are at pressures higher than where the errors were introduced. This technique makes it much easier to identify when errors might have occurred and enables the engineer to utilize the data measured before the errors were introduced.

The utility of these conversions is best illustrated with an example. A reservoir engineer had acquired some PVT data for an oil and started equation of state tuning. He soon realized that extreme modifications of critical parameters were required to match the differential liberation data and that he could not simultaneously match densities and multi-stage flash data along with the differential liberation data. At this point, the reservoir engineer was advised to make the above conversions and compare those predictions. Figure 10 shows a comparison of laboratory measured and equation of state predicted solution GOR and cumulative GOR referenced to the bubble point volume. The comparisons of solution GOR cause the equation of state to appear biased, while the comparisons of cumulative GOR show the equation of state to make accurate predictions of the oil behavior down to the 700 PSIA depletion point. Likewise, comparisons of formation volume factors and liquid shrinkage in Figure 11 show the equation of state to make accurate predictions of the liquid volumetric behavior, while the formation-volume-factor comparison alone would lead you to believe that the equation of state did not accurately predict the liquid volume behavior. Once the reservoir engineer tuned the equation of state to match the cumulative GOR referenced to the bubble point volume and liquid shrinkage data instead of the reported solution GOR and formation volume factors from the differential liberation experiment, predictions for densities and separator GOR were easily matched as well.

Comparisons with Equation of State Predictions

With the widespread availability of equation of state software, comparisons of laboratory data with equation of state predictions may very well be the most commonly used checks for laboratory data consistency. Equation of state comparisons can be a powerful tool for evaluating laboratory data, but caution should be exercised to ensure that bad data are not matched and that good data are not ignored.

The following example illustrates some of the pitfalls that can occur from equation of state predictions. PVT data were acquired on a relatively high-shrinkage oil and a fluid characterization was performed using the volume translated Peng-Robinson(9) equation of state.

The fluid characterization required minimal adjustment of critical properties or interaction parameters and most of the data such as saturation pressures, GOR’s, and densities were matched very closely. Liquid shrinkages and high-pressure gas gravities were two exceptions to the very close matches. The liquid shrinkage matches as shown in Figure 12 were not considered bad for a high-shrinkage oil, especially considering that the measurements were probably more challenging than for a typical oil. More confidence was initially placed in the equation of state predictions for the gas gravities than for the laboratory reported gas gravities for the first couple of depletion points. As additional data from the field were collected, a discrepancy between the equation of state predicted gas pressure gradient and the gradient observed in the field became apparent.

Yield data from the gas cap also indicated that the equation of state should be re-examined. Initial attempts to retune the Peng-Robinson characterization met with little success.

At this point, the question of whether the data should be matched was given serious consideration and a test with another equation of state, the BWRS(10) was performed to try to gain a better understanding. Surprisingly, the BWRS required little tuning and matched the liquid shrinkage and gas gravity data very closely. Figure 12 aslo shows the results of the BWRS equation of state predictions for liquid shrinkage. The BWRS also predicted gas gradients consistent with field observations. In this case, data that had been questioned based on the match with one equation of state were now considered very good based on the match with another equation of state. The BWRS equation of state, being derived from the Virial equation, also provided a good contrast to the cubic Peng-Robinson for testing the reasonableness of the data.

Because compositional reservoir simulation was being performed for this fluid, a Peng-Robinson characterization was still required. Now armed with additional data and confidence in the data, the Peng-Robinson was aggressively retuned (while still maintaining reasonable parameter adjustments) to improve the match to the gas gravities, expected yields, and liquid shrinkages. While the final Peng-Robinson match was not as good as the match with the BWRS, the results were considered suitable for future reservoir simulation and engineering.

Conclusions

Major conclusions from this work include:

• While most laboratory PVT data are of high quality, poor data quality can be identified using a variety of quality assessment and consistency evaluation techniques. These techniques are straightforward and easy to apply. These techniques have been demonstrated through a series of examples and illustrations.

• While many of these PVT techniques are available in the open literature or easily understood, they are often not being used. This has been observed through the continued identification of poor quality PVT data.

• These techniques can be applied to assess data quality and consistency to identify high quality data. The high quality data then provide improved bases for reservoir engineering calculations, including equation of state fluid characterization.