36 Oilfield Review

Real-Time Openhole Evaluation

Tom Barber Sugar Land, Texas, USA

Laurent Jammes Jan Wouter Smits Clamart, France

Werner Klopf Milan, Italy

Anchala Ramasamy BP Amoco ExplorationAberdeen, Scotland

Laurence Reynolds Aberdeen, Scotland

Alan Sibbit Houston, Texas

Robert Terry BP Amoco ExplorationHouston, Texas

For help in preparation of this article, thanks to David Allen,Schlumberger-Doll Research, Ridgefield, Connecticut, USA; Gordon Ballard, Kevin Eyl and Alison Goligher,Schlumberger Wireline & Testing, Montrouge, France;Vincent Belougne, Geco-Prakla, Gatwick, England; KeesCastelijns, Schlumberger Wireline & Testing, New Orleans,Louisiana, USA; Ollivier Faivre and Pascal Rothnemer,Schlumberger-Riboud Product Center, Clamart, France;Pierre Roulle, Schlumberger Wireline & Testing, Pau,France; Jay Russell, Schlumberger Wireline & Testing,Livingston, Scotland; Bob Mitchell, Schlumberger Wireline & Testing, Sugar Land, Texas, USA; Jim White,Schlumberger Wireline & Testing, Aberdeen, Scotland; and Technical editing Services (TeS), Chester, England.AIT (Array Induction Imager Tool), DLL (Dual LaterologResistivity), EPT (Electromagnetic Propagation Tool), FMI(Fullbore Formation MicroImager), GeoFrame, HRLA (High-Resolution Laterolog Array), InterACT, Platform Express,SlimAccess, UBI (Ultrasonic Borehole Imager) and Xtremeare marks of Schlumberger. HDLL (High-Definition LateralLog) is a mark of Baker Hughes, Inc.

Operators are gaining an accurate, at-the-wellsite first look at pay zones, thanks

to new technology that incorporates logging measurements environmentally

corrected in real time, fast forward modeling and inversion techniques. At

the heart of this technology are good science and innovative engineering.

To maximize asset value, oil and gas operatorscontinually strive to characterize the location andextent of recoverable reserves. Traditionally, open-hole logs generated with ”triple-combo“ toolstrings have provided key information: porosityfrom neutron-density measurements and satura-tion from resistivity measurements. The introduc-tion of an advanced logging system in 1995,based on a platform of integrated sensors, hasresulted in a quantum change in data acquisitioncapability, reliability and efficiency. The shorter,lighter, operator-friendly tool string is capable offaster rig-up at the wellsite and better access todeviated holes. Flexible mechanical design andshort tool length enable the operator to drill shal-lower wellbores while retaining the ability to logimportant pay zones at the bottom of the well.Array-resistivity measurements, microresistivityand three-detector density tools are improvingaccuracy in difficult environments without sacri-ficing logging speed.

But there is more to the story than more flex-ible operations, reduced rig time and improvedaccuracy of standard measurements. Using thelatest technology, new-generation tools providemore complete reservoir characterization right at

the wellsite—opening up a wealth of opportuni-ties for locating and tapping additional reserves.Improved tool response in thin beds, better padapplication in poor holes, greater accuracy inhigh-weight muds and real-time corrected forma-tion evaluation—all presented in clear, easy-to-understand formats—aid decision-making. Newtools, calibration methods and processing tech-niques, combined with a comprehensive log qual-ity-control (LQC) system, allow engineers tomonitor tool measurements and environmentalconditions—validating data acquisition andensuring that high-quality analysis can be per-formed over the entire logged zone. In manywells, this saves valuable time by eliminating theneed for repeat passes for log verification. Fasterwellsite calibrations, real-time environmentalcorrections, quality control, depth matching and acomplete wellsite quick-look contribute to well-site efficiency and put formation evaluation datainto the operator’s hands more quickly.

In this article, we look at three aspects ofnew platform logging technology and illustratethe simultaneous improvement in operationalefficiency that can be achieved while accuratelydetermining formation characteristics underincreasingly difficult environmental conditions.

1. Goligher A, Scanlan B, Standen E and Wylie AS: “A First Look at Platform Express Measurements,”Oilfield Review 8, no. 2 (Summer 1996): 4-15.

Summer 1999 37

First, we discuss the foundations supportingreal-time environmental corrections, includingspeed and depth, as a vital first step toward theuse of forward models and inversion techniquesfor making real-time environmental correctionsto basic logging measurements. Model-basedinversions—along with new measurements,such as those from mud resistivity sensors,microresistivity measurements, array tools withmultiple-depth measurements and densitybackscatter detectors—contribute to a clearerpicture of the borehole and formation.

Second, the latest developments in openholelogging are highlighted, including the new HRLAHigh-Resolution Laterolog Array tool with multipledepths of investigation, a tool capable of resolvingthe effects of shoulder beds, invasion and dippingformations, thereby providing better resistivityevaluation in complex environments with salinemud. Post-log processing techniques, to interpretlogs in extreme environments, are discussed alongwith a new maximum-entropy-based inversiontechnique developed to quantitatively interpretinduction logs from highly deviated wells or withlarge shoulder-bed contrasts.

Finally, we look at two important benefits ofreal-time environmentally corrected loggingdata—complete wellsite log interpretation andlog quality control. We illustrate how LQC isenhanced by real-time environmental information.

Real-Time CorrectionsEvery logging tool suffersfrom environmental effects ofone sort or another. Real-time cor-rections are essential to get accurate logginginformation into the hands of the operator effi-ciently. As a first step, every tool raw sensormeasurement requires a speed-derived depthcorrection. Next, forward models are used to pre-dict each measurement response for a given setof borehole and formation properties. Finally, bycomparing predicted sensor responses with log-ging measurements—a process called inver-sion—the environmentally corrected formationproperties are determined. This sequence ofsteps is performed during data acquisition andapplied to openhole logging measurements madewith the Platform Express tool system firstreviewed in the Oilfield Review three years ago.1

It may seem surprising that real-time depthcorrection is such an important issue. After all,most log analysts can shift logs at the computingcenter simply by ”eye“ or with automatic mathe-matical correlation algorithms. The depth of thelogging tool is traditionally determined from thelength of unwound cable with an approximateadjustment for cable stretch. However, one of thegreatest uncertainties in wireline logging hasbeen the assignment of petrophysical data to thecorrect depth of the subsurface.

As the tool is pulled up the wellbore, chang-ing wellbore conditions such as caves or fric-tional drag will cause the tool speed to changeerratically, and even to stick and slip, while thecable speed measured at surface remains con-stant. Thus each sensor’s motion across the for-mation may not correlate with the motion of thecable at the surface. The high tool speed—up tofive times the normal logging speed—occurringafter a stuck zone often results in lost data.Speed-derived depth corrections are essential attwo basic stages.

First, at the measurement stage, some loggingmeasurements depend on integrating raw datafrom multiple sensors located at different loca-tions along the tool string. Irregular or nonuniformtool motion during the measurement cycle willinvalidate the assumption that all the data comefrom the same volume of the formation. Log pro-cessing will produce an incorrect or unstableresult, especially visible at layer boundaries.

Some tool designs, such as neutron and den-sity tools, have multiple asymmetrically posi-tioned sources and detectors and unequalsource-detector spacings, and their measure-ments—based on comparing the count rates ineach detector—can be affected by nonuniformtool motion. An example from the North Seaillustrates the effects of speed-corrected densitymeasurements (above).

Likewise, the AIT-H Array Induction Imagertool filter-based processing algorithm assumesthat the data from the eight asymmetrical arraysare regularly sampled every 3 in. Irregular toolmotion can give rise to artifacts on the logs.Caliper measurements and auxiliary mud resistiv-ity measurements must be correctly depthaligned with the AIT array measurements toderive AIT borehole corrections.

At the second stage, depth corrections areequally important when integrating logging datafrom different tools to perform petrophysical inter-pretations. For example, log analysts frequentlylook for gas by comparing density and neutronporosity logs. For the characteristic crossover tobe meaningful, the spectral count rates of eachgamma ray detector in the density tool mustchange in phase with the count rates in each neu-tron detector in the neutron porosity tool as eachpasses the gas-saturated bed in the formation.

In addition to speed-based depth corrections,resolution matching is equally important. Sincethe neutron measurement samples a slightlythicker region in the formation than does the den-sity measurement, these measurements must bematched volumetrically. This resolution-matchingprocess is important in all high-resolution inter-pretations, because it ensures that the sensors ofeach tool used in a combined measurement seeexactly the same formation thickness.

Finally, measurements with high vertical res-olution can have errors amplified by irregulartool motion because the acquisition systemobtains data at sampling rates that vary as afunction of the required bed resolution. Whentool speed differs substantially from the recom-mended, then over- or under-sampling results.Rapid acceleration following a stuck-toolepisode can result in lost data. Tool-speed-based depth corrections are a prerequisite togood high-resolution measurements.

Obtaining properly depth-matched high-reso-lution measurements is critical for OceanEnergy’s efforts to evaluate thinly bedded reser-voirs in the Gulf of Mexico (next page). In onewell, high-resolution invaded-zone resistivity,Rxo, measurements from the MicroCylindricallyFocused Log (MCFL) tool in combination withdensity logs clearly show the many thin beds—some less than 1-ft thick [0.3 m]—throughout thereservoir. Comparison with the FMI FullboreFormation MicroImager images confirms thepresence of thin beds, and the real-time porosityderived from the high-resolution density logenables accurate reserve calculations.

Crucial depth corrections are implemented inthe Platform Express system during acquisitionusing a built in tool-axis accelerometer. Thisdevice measures instantaneous tool accelerationto determine tool velocity and the true depth atwhich all the other tool measurements wererecorded. Stability of the depth-correction algo-rithm is maintained by the use of a Kalman filter-based optimization.2 This optimization minimizes

38 Oilfield Review

2. Belougne V, Faivre O, Jammes L and Whittaker S: “Real-Time Speed Correction of Logging Data,” Transactions ofthe SPWLA 37th Annual Logging Symposium, NewOrleans, Louisiana, USA, June 16-19, 1996, paper F.

Gamma ray, speed corrected MD1:200

ft

X150

X169

X183

X200

Standoff

Standoff(density)

in.

API0 75Density, speed corrected

g/cm31.95 2.95

Density, no speed correctionCaliperin.4 14

Gamma ray, no speed correction

API0 75 1 0

A

A

>Speed-corrected density logs. This example, from a 97º-deviated borehole in the North Sea, comparesthe speed-corrected Platform Express density log (red) with the uncorrected log (black) in track 2. The green band on the left of track 1 is the accelerometer LQC flag indicating—by turning black—excessive stick and slip experienced in Zone A. The gamma ray in track 1 also shows shifts whenspeed-corrected. The logs highlight the benefits of speed corrections in poor hole conditions. The twohigh-porosity thin beds seen at X169 ft and X183 ft are incorrectly identified in the uncorrected log.Potential hydrocarbons in these zones could be missed if uncorrected depths were used to guide theperforation interval.

Rxo (8 in.)

Rxo (1 in.)0 API

Gamma ray150

Rxo

0.2 ohm-mAIT 90-in. resistivityDepth

ft

X760

X770

X780

X790

X800

X810

X820

X830

20

0 FMI image

North

Rxo

ohm-m

360

0.5 8

g/cm3

Densityg/cm3

Density (1 in.)

1.65 g/cm3

Density (8 in.)2.65

60 p.u.Neutron porosity

0

>Detecting thin beds in the Gulf of Mexico. The FMI Fullbore Formation MicroImager tool image shown in track 1 confirms thepresence of many beds less than 1-ft thick detected by the high-resolution Micro-Cylindrically Focused Log (MCFL) Rxo log shownin track 2. The high-resolution (black) and very high-resolution (blue) density logs are shown in track 3. The high-resolution densitylogs are quantitative in beds over 8-in. thick. The high-resolution, deep induction log shown in track 2 detects many of the thinbeds and can be used quantitatively in beds 1 ft or thicker. Track 4 compares all the density logs with the neutron porosity log.

Summer 1999 39

the overall error in the depth correction by solv-ing a system of simultaneous equations linkingthe exact moment each sensor measurementwas made with the true tool position—derivedfrom the instantaneous accelerometer measure-ment. Cable depth is used as a constraint to helpstabilize the solution. All raw sensor measure-ments and optimization solutions are performedin the time domain. This allows them to be easilyconverted to true depth or cable depth,whichever is required. Time-domain processingalso helps to overcome limitations encounteredin high-frequency depth sampling during high-resolution logging operations.

Forward Modeling for Environmental CorrectionsIn the language of log analysts, the phrase for-ward modeling refers to computing a loggingsensor response in the presence of the environ-ment surrounding the logging tool. Almost anytool response can be linked to formation proper-ties through the physics of the sensor measure-ment and its interaction with the materials of theformation and borehole environment. Comptonscattering and photoelectric absorption governthe interaction of low-energy gamma rays usedto measure formation density. The physical prin-ciples embodied in Maxwell’s equations are well

understood, and with enough knowledge of aresistivity tool design and its environment, thevoltages and currents that make up the toolresponses are predictable.

Forward modeling is important because itallows prediction of tool response under anygiven conditions. These predictions can then becompared with observed measurements, in a pro-cess known as inversion—described later in thisarticle—to understand the real conditions underwhich the measurements were made. In this arti-cle, unless otherwise specified, tool or sensorresponse means the raw measurement, such ascount rate in a nuclear detector, or voltage andcurrent measured on an electrode or inductiontool antenna coil.

Mudcake FormationMud

b

c

a

b Short-spacingdetector

spectrumW2

W3

W4

W1

c Backscatterdetector

spectrumW2

W3

W1

a Long-spacingdetector

spectrum

W1

Window count rates (W)

W2

W3W4

Density

Formation

Mudcake

Mud

Distance

Coun

t rat

eCo

unt r

ate

Coun

t rat

e

Energy

A forward model is used in Platform Expressreal-time density analysis to calculate eachwindow count rate as a function of formation and mudcake properties. The formation modelgeometry consists of a homogeneous forma-tion and mudcake corresponding to a one-dimensional (1D) radial step profile varying indensity and photoelectric properties. Within thisframework, the different detector count-rateresponses depend on only five environmental

40 Oilfield Review

3. Allioli F, Faivre O, Jammes L and Evans M: “A NewApproach to Computing Formation Density and Pe(Photoelectric Factor) Free of Mudcake Effects,”Transactions of the SPWLA 38th Annual LoggingSymposium, Houston, Texas, USA, June 15-18, 1997,paper K.

4. Ellis D: Well Logging for Earth Scientists. New York, NewYork, USA: Elsevier Science Publishing Co., Inc, 1987.

5. Anderson B, Druskin V, Habashy T, Lee P, Luling M,Barber T, Grove G, Lovell J, Rosthal R, Tabanou J,Kennedy D and Shen L: “New Dimensions in ModelingResistivity,” Oilfield Review 9, no. 1 (Spring 1997): 40-56.

6. In this article, we shall confine our discussion to thedimensionality of the formation model—the number ofindependent coordinates needed to describe the way inwhich resistivity varies. See “The Vocabulary ofResistivity Modeling,” Anderson et al, reference 5.

>Three-detector density logging sonde. Multiple Compton scattering and photoelectric absorptionlead to a spectrum of gamma ray photons entering the detector windows from the borehole and forma-tion environment (right). The forward model represents a homogeneous formation behind a thick layerof mudcake (bottom). Each detector spectrum is partitioned into broad count-rate windows used toestimate properties of the gamma ray scattering environment (left).

>Density calibration database. The density for-ward-model database was recorded in the Envi-ronmental Effects Calibration Facility in Houston,Texas, USA, which was built for the characteriza-tion of nuclear logging tools. This database cov-ers the range of environments to which the toolwill be exposed. The facility manager, John Spal-lone, is shown lowering the Platform Expressdensity tool into one of the calibration blocks.Since the Platform Express tool first becameavailable, a continuous effort has been underway to enhance the density measurement capa-bility in heavy mud environments.

degraded photons from multiple Compton scatter-ing and photoelectric absorption contribute to anoverall continuous gamma ray spectrum seen inthe formation by each of the three detectors.

Typically, increasing density causes anincrease in the gamma ray flux near the sourcebecause there are more scattering targets in ahigher density material. This increases theobserved backscatter detector count rate. On theother hand, increasing density tends to cause adecrease in the observed count rates in the twodetectors spaced farther from the source becauseof the long attenuating path to these detectors.Also, changes in formation lithology can bedetected by variations in the low-energy windowcount rates due to photoelectric absorption.These count rates are also strongly affected bybarite in the mudcake, and its presence can makethe photoelectric effect measurement intractable.

Density forward models—The three-detectordensity tool in the Platform Express tool stringuses a gamma ray source that emits 662-keVphotons from a source capsule located in thelogging pad (above).3 Although density measure-ments are sensitive to a relatively small volume ofthe environment between the source and detector,the increasing source-to-detector spacing of eachdetector enables each to see progressively deeperinto the mudcake and formation.

Gamma rays from the source enter the mud-cake and formation and typically scatter severaltimes before being detected by each detector inthe logging pad. Each Compton scatteringencounter causes the incident gamma ray to loseenergy and change direction, eventually bendingmany gamma rays back towards the detectoraperture in the tool. The comined effect of energy-

Summer 1999 41

parameters—formation density and photoelec-tric factor; mudcake density and photoelectricfactor; and mudcake thickness.

An ideal implementation of the gamma rayphysics in this formation forward model would begiven by an exact solution to the Boltzmannequation for gamma ray transport. Unfortunately,the Boltzmann equation has no simple analyticform for this environment.4 Instead, a proxy forthe exact sensor response physics parameterizesthe detector window count rates as exponentialfunctions in terms of formation and mudcakeproperties. Each window count-rate responsefunction contains empirical coefficients, whichaccount for source strength, detector collimation,average gamma ray track length for each partic-ular source-detector spacing, and energy-depen-dent Compton scattering and photoelectricabsorption cross sections. These coefficientsform the calibration for this nonlinear parametricforward model, and are determined by aweighted least-squares fit to a database of labo-ratory measurements.

The database measurements were obtained in a laboratory with a calibrated density tool in known formations and mudcake conditions(previous page, right). Today, over 1130 calibra-tions in boreholes with barite mudcake and 420calibrations with nonbarite mudcake have beenmade, and recent calibrations for the density for-ward model have extended the operating rangeof the density tool to mud weights of up to 17 lbm/gal [2.04 g/cm3].

Resistivity forward models—In contrast tonuclear measurements with their small volume ofinvestigation, all deep-resistivity responses,whether from induction or laterolog tools, areinfluenced by the resistivity distribution in a largevolume surrounding the logging instrument.Correct interpretation requires corrections for bore-hole, invasion and other large-scale environmentalor geometrical effects such as shoulder effects.

The response of a laterolog, such as the High-Resolution Azimuthal Laterolog Sonde (HALS)measurement, is computed by solving Laplace’sequation for the electrostatic potential in theborehole and formation environment. Laplace’sequation follows from Maxwell’s equations inthe low-frequency limit and dictates conserva-tion of current everywhere in the environmentaldomain. Knowing the voltage and currents every-where along the tool allows prediction of theapparent resistivity reading. Unfortunately, ana-lytical solutions exist for only a limited number ofproblems with simple geometries. In practice, the

solution requires the use of numerical methods.The two methods most frequently employed arethe Finite Element Method (FEM) and the FiniteDifference Method (FDM). Development of for-ward modeling techniques using FEM and FDMfor resistivity measurements was discussed twoyears ago in the Oilfield Review.5

Both FDM and FEM divide the environmentaldomain into grid cells and solve for the potentialin each cell. The interactions between neighbor-ing cells are controlled by Laplace’s equation.Combining all the cell interactions together withthe boundary conditions yields a large system oflinear equations, which is solved by the computerto find the electric potential at the vertices of each cell. Depending on the complexity and dimension of the formation model, eithertwo-dimensional (2D) or three-dimensional (3D) solutions of the electromagnetic field areneeded (right).6

Multidimensional forward modeling tech-niques have been successful in rapidly and accu-rately predicting the tool response of resistivitytools. The models are used to optimize tooldesigns, characterize their response and providethe basis for inversion procedures designed tofind formation characteristics. Although modelingspeed has increased considerably in recent years,it is frequently too slow to allow real-time inver-sion of measurements while logging, especiallywhen a 3D model is needed. When the number offormation parameters is limited, it is often morepractical to create a database of the toolresponse to variations in parameters and inter-polate from this database to build the numericalforward model used during the inversion.

1D-radial

1D-layer

2D

3D (2D plus dip)

3D

Formation model dimensions. In HALS laterologresistivity modeling, a 1D radial formation modelis concerned only with the influences of theradial invasion profile (A). It assumes that thereading is being taken in an infinitely thickformation bed. At other times, a 1D layered for-mation is used to model the effects of thick andthin shoulder beds with resistivity contrasts (B). A 2D model assumes that the formation is com-posed of homogeneous layers perpendicular tothe borehole (C). In this model, both the invasionof each layer and the shoulder-bed effect ofadjacent layers are taken into account, resultingin a more accurate Rt calculation in beds wheresignificant shoulder-bed effects exist. A 2D modelis defined by the values of Rxo, Rt, the invasionradii and thicknesses of the formation layers.The 2D model can be taken one step further byincorporating the formation dip relative to theaxis of the borehole. This results in a 3D-formationmodel (D). This scenario could be due to struc-tural formation dip, deviated borehole or both.More complicated 3D formation models can beconstructed in various ways. One way is to parti-tion the layers into azimuthal sectors (E). Thismodel takes into account azimuthal anisotropy,shoulder-bed effects and invasion as well asvariations in layer thickness with distance fromthe wellbore.

A

B

C

D

E

>

Boreholeradius AIT tool

Boreholemud resistivity

Standoff

Formationresistivity

rRm

Rf

The HALS tool uses two such FEM-deriveddatabases for processing its measurements.7 Thefirst database is used to correct apparent resis-tivities for borehole effects. It models the toolresponse in an uninvaded formation as a functionof formation-to-mud-resistivity contrast, bore-hole size and tool eccentricity. Because of tooleccentering, a 3D model was needed to constructthis database. The second database is used toinvert apparent resistivities for Rt. It describesthe tool response as a function of invasion radiusand contrast between formation resistivity, Rt,and invaded zone resistivity Rxo. Even though theresponse was originally determined using a 3D-based computation, the forward model iscalled 1D because it models resistivity variationsonly in the radial direction.

The AIT Array Induction Imager tool bases itsborehole corrections on a forward model thatincludes a 2D plus eccentricity effects (above).8

The borehole forward model is based on solutionsto Maxwell’s equations in a cylindrical boreholewith resistivity Rm surrounded by a homogeneousformation of resistivity Rt. The tool can be locatedwith a standoff anywhere in the borehole, but is assumed to be parallel to the borehole axis. Inthis model, the signal in any given AIT array ispredicted as a function of four environmentalparameters—Rm, Rt, borehole size and tool stand-off. This model is used to develop a database oftool responses that provides calibration for thefitted forward model used in real time to predicttool responses in any borehole environment.

From Forward Models to InversionsGiven a forward model with a system of equa-tions governing tool responses, one simply entersthe formation and borehole parameters into themodel and, after computation, the desired toolmeasurements are predicted. The prediction is aset of tool responses that would be observed if aphysical experiment were performed in the givenformation and borehole environment. When thegoverning equations are linear, the process isreversible—or invertible in one step. Iteration isnot needed. Given the tool response, the modelparameters can be estimated by multiplying thevector of observed tool responses by the gener-alized inverse of the same matrix used in thecorresponding forward problem.

Unfortunately, nature does not always poselinear problems. Density and resistivity toolresponses are not linear with respect to forma-tion properties. For these responses, a moreversatile technique is to find a solution by itera-tively solving the forward problem. Inversion isthe process of creating a model, mathematicallymodeling the physical response to that model,and then varying the parameters in the modeluntil the modeled response matches the oneseen by the logging tool. There are two ways toperform this task, manually and automatically. Alog analyst often uses manual inversion, adjust-ing model parameters based on previous experi-ence or knowledge, with advanced post-loggingprocessing programs at a computing center.

Automatic inversion is performed by an algo-rithm that computes the response to a model, thenfollows certain rules to modify the model to con-verge to a solution. Common algorithms minimizethe difference between the observed responseand the modeled synthetic response by adjustingthe model parameters to reduce the differences ateach step. Real-time environmental borehole cor-rections in the Platform Express system all dependon automatic inversion techniques.

Density inversion—Density inversion isbased on a maximum likelihood method that usesthe nonlinear parametric forward model dis-cussed above to link the depth-correctedobserved count rates in each of the detectorcount-rate windows to the formation parameters.At every depth, the inversion predicts the countrates in each detector window and comparesthem with those measured by the tool.Minimizing a cost function containing threeterms optimizes the solution.9

The first term in the cost function is based onthe best fit of all the observed window countrates. This is done by minimizing the ”recon-struction error,“ which is a term proportional tothe average squared difference between all themeasured and modeled count rates in eachdetector energy window—each weighted by theexpected error based on counting statistics andmodel uncertainty.

The second term in the cost function mea-sures the difference between the current model-predicted environmental parameters and those atthe previous logging depth. This term, called thesmoothness condition, helps ensure compatibil-ity between sampling rate and measurement ofvertical resolution.

The final cost term helps control the stabilityof the solution when formation and mudcakeparameter estimates are far from the databaserange, which can happen when large standoffsoccur. This term vanishes when the solution iswithin the limits of the database.

A powerful example of density inversionrobustness can be seen when comparing densitymeasurements derived during extreme condi-tions (down logging) with normal measurementstaken while logging uphole (next page).Occasionally, for precautionary reasons, logging

42 Oilfield Review

7. Smits J, Benimeli D, Dubourg I, Faivre O, Hoyle D,Tourillon V and Trouiller J-C: “High Resolution From a New Laterolog with Azimuthal Imaging,” paper SPE 30584, presented at the SPE Annual TechnicalConference and Exhibition, Dallas, Texas, USA, October 22-25, 1995.

8. Anderson BI and Barber TD: Induction Logging. Sugar Land, Texas, USA: Schlumberger Wireline &Testing, 1997.

9. The concept of a cost function comes from the mathe-matics of operations research and making decisionswith multiple objectives. A cost function is an equationthat contains a measure of the error in the decision-making or optimization problem. The function cancontain many terms, one of which usually describes the overall reconstruction (or fit) of the forward modelpredictions to the observed sensor responses. Themagnitude of the cost function decreases as the opti-mization improves.

10. Statistical (count rate) uncertainties and forward modelerrors are propagated through the density inversion pro-cessing by the measurement covariance calculation.

< AIT borehole correction model.A tool can be located anywherein the borehole with any valuestandoff. The borehole corrections are based on an iterative inversionprocessing.

Summer 1999 43

measurements are taken while going into theborehole. The caliper arm and pad are closed toprevent getting stuck against the borehole wallwhile going down the borehole. This results in alarge standoff with excessive mud and mudcakebetween the density pad and formation. Undersuch large standoff conditions, two-detector den-sity measurements based on a graphical spine

and ribs algorithm cannot account for theseextreme standoff conditions. The different radialsensitivity of each of the three density detectorshelps the inversion algorithm provide a final envi-ronmentally corrected density log that compen-sates for the extra standoff. This corrected logagrees well with the density log obtained undernormal logging conditions.

The use of a parametric forward-model-basedinversion for density measurements has severaladvantages. It makes the most efficient use of allthe sensor measurements while simultaneouslyobtaining formation and mudcake properties. Forexample, all density, photoelectric and standoffinformation contained in the entire spectrumfrom each gamma ray detector is automaticallytaken into account in the inversion algorithm,providing a strong degree of redundancy in theinformation available from the density tool. Bothstatistical uncertainties and forward modelerrors are considered in the inversion calculation,and provide realistic output uncertainties.10

These take into account count-rate statistics, cal-ibration errors and model errors, and establishreliable confidence limits for LQC and subsequentpetrophysical analysis.

Resistivity inversion—Inversion is central tothe borehole correction algorithm in the process-ing chain for AIT logs. It uses the AIT polynomialforward model to correct the eight depth-corrected raw array measurements for tooleffects in a nonstandard borehole environment.The parameters for the inversion are the fourcomponents of the database—borehole radius,tool standoff, mud resistivity and formation resis-tivity. The inversion is an optimization that findsthe set of borehole parameters that best repro-duces the four shortest arrays (6-, 9- 12- and 15-in. measurements). However, since thesemeasurements overlap considerably in theirinvestigation range, their information content isnot sufficient to solve for all borehole parameterssimultaneously.

In practice, the inversion process reliablydetermines only two of the four parameters. Theother two parameters are always measured orfixed. Since formation resistivity is always anunknown, there is only one additional freeparameter for the inversion to determine.Accordingly, there are three modes of boreholecorrection—depending on which parameter issought. If an accurate hole diameter from thedensity caliper and mud resistivity from the aux-iliary mud resistivity, Rm, sensor are used, thenthe borehole correction inversion determines toolstandoff. Mud resistivity needs to be measuredwithin 5% of its true value, which can be metwith the AIT mud sensor. By solving for formationresistivity and standoff, the borehole correctionproblem can be solved with no intervention fromthe engineer. Likewise, hole diameter can becomputed with an accurate standoff and mud-resistivity measurement, or else the mud resistiv-ity can be computed from accurate hole size andstandoff information.

Gamma ray

Depthft

Standoffupin.

API0 150Backscatter density (down)

g/cm31.65 2.65

Short-spacing density (down)

Inversion density logging downg/cm31.65 2.65

Inversion density logging up

Long-spacing density (down)

Backscatter-spacing density (up)

Short-spacing density (up)

Long-spacing density (up)

Caliperin.4 14

1 0

Standoffdown

in.1 0

X575

X600

>Using inversion to obtain the correct answer under extreme conditions. When a density tool is low-ered into the borehole, the caliper is kept closed to prevent getting stuck in the hole. Under theseconditions, the density pad is not pushed firmly against the borehole wall, resulting in a large standoff(black dashed) derived from the inversion algorithm shown in the depth track. All three detectors (bluecurves) see low density in track 2 because of extra mud and mudcake encountered in this configura-tion. However, the inversion algorithm is still capable of accounting for the extra mud and standoff inthis environment. It produces a density curve (blue) in track 3 that agrees with the density reading(red) obtained under normal conditions—logging up the borehole with the caliper open and padpushed against the formation.

invasion profiles. Typically, radial inversion isanother 1D four-parameter optimization that usesa monotonic-conductivity invasion profile modelto produce logs of Rxo, Rt, and the limits of a tran-sition zone (left).

Real-time wellsite processing for laterologsalso involves borehole corrections followed by a1D inversion for Rt. For example, HALS boreholecorrections adjust for the presence of the bore-hole, taking into account the borehole size andthe ratio of apparent resistivity to mud resistivityRa/Rm. It also includes an eccentricity correctionthat allows for the eccentered position of the toolin the borehole. The mud resistivity needed in the

44 Oilfield Review

Annulus profile

Rt

RannRxo

r1

r2

Rxo

Rt

Form

atio

n re

sist

ivity

pro

file

Slope profile

r1

r2

Distance from wellbore

Invasionmidpoint

Rxo

Ramp profile

Rt

ri

Step profile

Rxo

Rtri

>Resistivity invasion models. The simple piston-invasion, or step profile, and the ramp profilerequire three parameters; the slope profile is a four-parameter model; the annulus profile is a five-parameter model. A step profile is usedfor real-time HALS and HRLA radial inverse inva-sion modeling, and the slope profile is used forreal-time AIT radial invasion inversion.

Depthm

X150

X250

Gamma ray0 150API

Caliper6 11in.

HRLA array resistivity 1

ohm-m

HRLA array resistivity 2

HRLA array resistivity 3

HRLA array resistivity 4

HRLA array resistivity 50.1 10

DLL shallow resistivity

DLL deep resistivity0.1 10ohm-m

>Eliminating the Groningen effect. The DLL Dual Laterolog Resistivity deep-resistivity (red)and shallow-resistivity (blue dashed) curves in track 3 show a large separation (yellowshaded) caused by the Groningen effect. The HRLA curves in track 2 are not affected bythe Groningen effect because all the currents return to the tool string itself, rather than thesurface. The fact that all five HRLA curves are reading the same resistivity over this intervalalso establishes that the formation is uninvaded.

After all the raw array measurements havebeen corrected for nonstandard borehole effects,they are processed by the usual AIT log-formingtechniques. This involves generating the stan-dard depth (10-, 20-, 30-, 60- and 90-in.) andsimultaneously resolution-matched (1-, 2- and 4-ft) logs by convolving the corrected raw arraymeasurements with the Born-approximation-based weighting functions.11 Subsequently, real-time interpretations are based on inverting theseprocessed measurements radially to obtain anestimate of Rt.12 This radial processing also givesa quantitative estimate of invasion geometry aswell as an accurate estimate of Rt in complex

correction can be derived from the tool itself orfrom an external mud-resistivity measurement.Following borehole correction, an inversionbased on a 1D three-parameter step-profile inva-sion model is used to determine the formationparameters—Rt and the invaded-zone radius—that best describe the borehole-corrected deepand shallow measurements. Since only twoformation parameters can be determined fromthe two measurements (shallow and deep resis-tivity), the value of Rxo must be supplied to theinversion. It is obtained from the MCFL tool,which gives a resolution-matched microresis-tivity measurement.

Summer 1999 45

A similar Rt inversion technique is used for thenew HRLA High-Resolution Laterolog Array tooldiscussed below.13 For this tool, there is enoughinformation in the five array measurements toallow an accurate estimate of Rt, and determinethe invasion profile—independent of an externalRxo under most conditions. However, if an addi-tional Rxo measurement is used, it helps constrainthe inversion processing and improve the derivedRt. In the HRLA inversion, weights are assigned toeach measurement based on the magnitude of itsborehole correction. The measurement with thesmallest borehole correction is given the highestweight in the inversion algorithm.

New High-Resolution Resistivity TechnologyAlthough the concept of an array laterolog toolhas existed since the 1950s, Shell was first torecognize that an array-resistivity tool couldimprove thin-bed saturation evaluation, and pro-posed a point-electrode array tool called theMulti-Electrode Resistivity Tool (MERT).14 Theirtool design is similar to a multiple-spaced”normal“ measurement to which lateral andsecond-difference voltage measurements areadded.15 Recently, Baker Atlas commercializedtheir HDLL High-Definition Lateral Log service,which is an implementation of the MERTarchitecture.16

In response to this need, Schlumbergerdeveloped a new HRLA High-Resolution LaterologArray tool that can be used with the PlatformExpress system. This tool achieves multipledepths of investigation through a segmented array of six simultaneous, symmetrical andactively focused laterolog measurements. Thisdesign gives a coherent set of high-resolutionresistivity measurements that can be inverted to correct the deepest measurements for theenvironmental influences of invasion and shoulderbeds. By having all the laterolog currents return to the tool body, the HRLA tool minimizes the two most unwanted laterolog parasitic distor-tions—reference effects and shoulder-bedeffects. In addition, the surface current return and insulated bridle are no longer needed—reducing cost and risk.

Reference effects—Traditionally, laterologtools operate in a deep mode with currentsreturning to a reference electrode at the surface.This requires isolating the logging cable from thetool by use of a long insulating bridle. A system-atic shift in the resistivity measurement, calledthe Groningen effect, arises when high-resistivityformation layers above the tool force returningcurrents—following the path of least resis-tance—into the borehole. This leads to a poten-tial drop along the cable, and subsequently thevoltage reference of the tool can no longer beconsidered to be infinitely far away. As a result,formation resistivity computed using this refer-ence reads artificially high (previous page, right).Long tool strings and drillpipe have a similareffect—artificially increasing the measured for-mation resistivity.

Shoulder-bed effects—Shoulder beds withlarge resistivity contrasts have a strong influenceon most laterolog measurements. The measure-ment and focusing currents from the laterologtool tend to flow along zones of least resistance(left). A distortion in the focusing current distribu-tion allows the measurement current to flow dif-fusely across intervals with significant resistivityvariations. This defocusing introduces a couplingbetween the vertical and radial response charac-teristics of the resistivity measurement.

11. Generalized (finite conductivity) 1D tool geometricalresponse functions are derived using a forward modelsolution to similar to the single scattering Bornapproximation formalism traditionally used in quantummechanics. See Gianzero S and Anderson B: “A NewLook at Skin Effect,” The Log Analyst 23, no. 1 (January-February, 1982): 20-34.Barber TA and Rosthal R: “Using Multiarray InductionTool to Achieve High-Resolution Logs with MinimumEnvironmental Effects,” paper SPE 22725, presented atthe 66th SPE Annual Technical Conference andExhibition, Dallas, Texas, USA, October 6-9, 1991.

12. Howard AQ: “A New Invasion Model for Resistivity LogInterpretation,” The Log Analyst 33, no. 2 (March-April,1992): 96-110.

13. Griffiths R, Smits J, Faivre O, Dubourg I, Legendre E andDoduy J: “Better Saturation from a New ArrayLaterolog,” Transactions of the SPWLA 40th AnnualLogging Symposium, Oslo, Norway, May 31-June 3, 1999,paper DDD.

14. Vallinga PM, Harris JR and Yuratich MA:”A Multi-Electrode Tool, Allowing More Flexibility in ResistivityLogging,” Transactions of the SPWLA 14th EuropeanFormation Evaluation Symposium, London, England,December 9-11, 1991, paper E. Vallinga PM and Yuratich MA: “Accurate Assessment of Hydrocarbon Saturation in Complex Reservoirs From Multi-Electrode Resistivity Measurements,”Transactions of the SPWLA/CWLS 14th FormationEvaluation Symposium, Calgary, Alberta, Canada, June13-16, 1993, paper E.

15. The use of computed focusing makes it possible, in prin-ciple, to obtain a Laterolog-7-style measurement fromthe MERT tool.

16. Itskovich GB, Mezzatesta AG, Strack KM and TabarovskyLA: “High-Definition Lateral Log-Resistivity Device: BasicPhysics and Resolution,” Transactions of the SPWLA39th Annual Logging Symposium, Keystone, Colorado,USA, May 26-29, 1999, paper V.

SqueezeAntisqueeze

>Squeeze and antisqueeze. Squeeze (top) occurswhen adjacent outer beds have significantlyhigher resistivity than the middle bed. Laterologfocusing currents tend to migrate toward the bedof interest and squeeze into this bed of compara-tively low resistivity. This results in a deeper mea-surement than in a homogeneous formation. Thesqueeze can result in resistivity curve separationthat imitates an invasion profile even in theabsence of invasion. When both invasion andsqueeze are present, the resistivity curve separa-tion will indicate deeper invasion than is actuallypresent. The reverse situation, antisqueeze (bot-tom), results when adjacent outer beds have lowerresistivity than the middle bed. In this case, mea-surement currents tend to flow into beds of lowerresistivity rather than flow through higher resistiv-ity beds. The resulting defocusing causes the deepmeasurements to have a reduced depth of investi-gation and thus read a lower formation resistivity ifthe invaded-zone resistivity is lower than Rt. Thiseffect can lead to underestimation of reserves.

The HRLA tool addresses these problemswith multiple modes of tightly focused arraymeasurements.17 Multifrequency operation of thesegmented electrodes enables the simultaneousresistivity measurement modes to be distin-guished (left). Software focusing by linear super-position of each mode is used to provide activefocusing. The shallowest mode is the most sen-sitive to the borehole and is used to estimatemud resistivity, Rm. The deepest mode has aresponse comparable to the deep-resistivitymeasurement of the HALS tool. The spacing ofthe other arrays is such that they have responsecharacteristics that optimize the information con-tent of their measurements with respect to theinvasion profile.

The addition of shallower curves improvesthe radial sensitivity to resistivity change, whichresults in greater log curve separation in thepresence of invasion (below). This is especiallyhelpful in thin beds, where deeper measure-ments tend to lose both depth of investigationand vertical resolution because of antisqueezeeffects. In addition, with reference effects goneand shoulder-bed effects reduced, the separationbetween the deep and shallow measurementscaused by these effects is also reduced. Theimproved invasion discrimination in thinly bed-ded formations leads to better vertical resolutionand more accurate inversion processing forformation resistivity.

46 Oilfield Review

17. Smits et al, reference 7.

150

100

50

0

-50

-100

-150

Mode-0 Mode-1 Mode-2

150

100

50

0

-50

-100

-150

-100 -50 0Mode-3

50 100 -100 -50 0Mode-4

50 100 -100 -50 0Mode-5

50 100

0 5 10 15 20 25 30 35 40 45 50

Invasion radius, in.

1.0

10

Appa

rent

resi

stiv

ity, o

hm-m

HRLA mode 1

HALS deep

HRLA mode 5

HRLA mode 4

HRLA mode 3

HRLA mode 2

HALS shallow

Radial response of resistivity tools. The bore-hole-corrected HALS deep-resistivity radialresponse compares well with the Mode-5response from the HRLA array measurement,while the HALS shallow-resistivity response isintermediate between the Mode-2 and Mode-3HRLA responses. The additional HRLA resistivitymeasurements provide improved definition ofradial resistivity changes, which helps evaluateinvasion profiles.

>Current distributions for HRLA focusing modes. By increasing the number of centralelectrodes that are kept at the same potential, the tool current return in the formation ismoved farther away, and the depth of investigation is increased. Six modes with increas-ing depth of investigation are used. In Mode-0, current flows directly from the centralelectrode to the nearest array electrodes. This mode is sensitive to the mud column envi-ronment and is used to estimate mud resistivity and borehole diameters. The deepestmode, Mode-5, sends current out from all but the outermost electrodes. The spacing ofthe array has been designed to optimize the information content of the measurement datawith respect to the invasion profile.

>

Summer 1999 47

An inversion for the HRLA tool based on a 2D-formation model with a piston-invasionprofile is currently being developed for theGeoFrame log interpretation system. The inver-sion processing proceeds by detecting bedboundaries and segmenting the log into discretebeds with average thickness of 1 to 2 ft [0.3 to0.6 m]. The inversion will iteratively refine for-mation parameters until the forward modelaccurately reproduces the input logs. Becausethe formation model includes the boreholeshape and mud resistivity, the input logs do notneed to be borehole-corrected.

More accurate representations of theformation environment lead to more accurateestimates of Rt, especially in thinly beddedformations. When the reservoir and water layerthickness are on the order of 2 to 5 ft [0.6 to 1.5 m], it is not unusual to see differences of 50 to 100% between the 1D- and 2D-inversionresistivity estimates (right). Usually the 1D-inversion estimates are too pessimistic. An obvi-ous extension of these models includes dippinglayers with 3D models.

Getting More Pay fromResistivity Logs

HRLA array resistivity 3

Rt from 2D inversion

Rxo from 2D inversion

Rt from 1D inversion

MCFL microresistivity

ohm-mohm-m

HRLA array resistivity 2

HRLA array resistivity 1

HRLA array resistivity 4

Rxo from 2D inversion

HRLA array resistivity 5

Rt from 2D inversionAttenuation

Propagation time Differential

EPT

1 30 1 30100 0 20700dB/m

XX00

XX10

XX20

XX30

XX40

XX50

XX60

XX70

XX80

XX90

X100

27 7ns/m 0 3in.

Crossover Washout

2D invasion

Invasion radius

Bit radius

Hole radius

in.

Caliper

Rxo > Rt

Rxo < Rt

Rxo > Rt

Rxo < Rt

Improved Rt estimate with 2D inversion. The 2Dinverted formation resistivity Rt (wide red) andinvasion resistivity Rxo (green) are shown in track 3along with the raw HRLA curves. The shadingbetween Rxo and Rt indicates where the invasion isnormal (Rxo < Rt) or reversed (Rxo > Rt). In track 4,the 2D inverted resistivities Rt (red) and Rxo

(green) are compared with the 1D inverted forma-tion resistivity Rt (magenta) and the Rxo (black)from the MCFL tool. The 2D inversion shows a sig-nificant increase in Rt obtained in thin beds—suchas those between XX30 and XX70 ft—over the 1D-inversion results. A good match between the 2Dinversion-derived Rxo and the one independentlyobtained from the MCFL measurement—addsconfidence to the inversion results. The EPT Elec-tromagnetic Propagation Tool dielectric attenuationand propagation time curves, confirming the pres-ence of thin beds in track 1, were used to constrainthe inversion for the uninvaded formation model inthe shales.

>

48 Oilfield Review

Optimal array focusing is enhanced by thesymmetric tool design, ensuring that all the sig-nals are measured at exactly the same time andat the same logging tool position. This helpsavoid horns and oscillations produced by irregu-lar tool motion, and ensures that the measure-ments are exactly depth aligned. The coherentnature of the focused, depth-aligned, resolution-matched measurements from the HRLA toolproduces a more intuitive LQC, as the curves sep-arate following the invasion resistivity profile.(below). At the wellsite, operational safety andefficiency are improved by the elimination of thebridle and surface current system.

Tackling Difficult EnvironmentsModel-based inversion processing is a delicatebalancing act. Two conflicting factors need to beconsidered. On one hand, the accuracy of theresult depends on how much additional informa-tion can be built into the model. On the other, thespeed with which the result is delivereddecreases with the complexity of the model. Fast1D-inversion models are needed for real-timeenvironmental corrections to help the operatorprocess, interpret and evaluate logs quickly atthe wellsite. However, sometimes the wellsiteanswer isn’t enough. For example, various para-sitic effects on resistivity measurements—

shoulder beds, spiral boreholes and dipping for-mation beds with invasion—can continue tocause errors when computing Rt. For these, post-logging processing techniques available at thecomputing center can help.

Shoulder beds—The presence of shoulderbeds can lead to overestimating Rt in the”squeeze“ case and underestimating it in the”antisqueeze“ case. Consequently, in both cases,water saturation, Sw, estimation will be affected.For example, a log analyst will estimate thewater-filled resistivity, Ro, from a water-saturatedbed (squeeze case), and estimate Rt in the payzone (antisqueeze) using Archie’s equation. Errorsin both resistivities contribute to overestimatingwater saturation in the pay zone, which can leadto overlooked hydrocarbons.

How to tackle this problem? The 1D radialinversions used in real time are for simpler casesand do not address the fact that nearby adjacenthigh-contrast beds may influence the resistivityreading. Improved methods involve the combina-tion of 1D shoulder-bed corrections followed by a1D radial inversion, but they do not address thefact that the radial and vertical responses arecoupled, leading to significant errors in resistivitydetermination. It has been long recognized thatresistivity estimation can be improved by the useof inversion techniques that take into account true2D or 3D formation structure (see ”Getting MorePay from Resistivity Logs,“ previous page).Application of 2D formation models to the inver-sion technique can double the calculated reserves,particularly in thinly bedded formations.18

Spiral boreholes—Some drilling practicesproduce a borehole with a 3D shape that has aspiral groove on top of the bit-sized hole. Suchboreholes produce a quasi-periodic character inthe logs, and are variously referred to as”corkscrew“ or ”threaded hole.“ These havebeen associated with downhole drill motors andhigh-angle wells. Displaced stabilizers can pro-duce the same effect in vertical wells. The effectis seen as a periodic oscillation on the caliperlog. The effect on other logging tools depends on the physics of the measurement. The impactof spiral boreholes on induction measurementshas been extensively studied by BP AmocoExploration.19 Recently, filter techniques havebeen developed to reduce the effects of spiralborehole rugosity on logging tool measurements(see ”Dealing with Spiral Boreholes,“ next page).For density tools, the effect of the spiral-groovedhole is to produce a cyclic mudcake effect.

XX20

XX00

1D Rt Increase

HALS invasion

Depthft

0

Invasiondiameter

HALS

in. 60

0

Bit size

in. 60

1

1D Rt from HRLA

ohm-m 100 1

1D Rt from HALS

ohm-m 1000

Invasiondiameter

HRLA

in. 60

MCFL microresistivity MCFL microresistivity

1

1D Rt from HRLA

A

ohm-m 100

HRLA array resistivity 5 HALS shallow resistivity 1D Rt from HALS

HRLA array resistivity 4 HALS deep resistivity

HRLA array resistivity 3

HRLA array resistivity 2

HRLA array resistivity 1

>Reducing shoulder-bed effects on laterolog measurements. In this example, HALS resistivities,shown in track 3, are out of sequence because of shoulder-bed effects. Therefore, the HALS real-timeprocessing forces invasion diameter to bit size—thus computing zero invasion. As a result, the HALS-derived formation resistivity, Rt, defaults to the deep-reading resistivity value because the data areinconsistent with the 1D formation forward model. Over the same interval, the HRLA tool is much lesssensitive to shoulder-bed effects, and the additional data from the array measurements, shown intrack 2, provide a realistic estimation of the invaded zone. The invasion corrections are applied to allHRLA array resistivity measurements for an accurate Rt evaluation. Results show that the maximumresistivity from HRLA real-time inversion is 45% higher than that estimated by the HALS measurement,leading to a 16% increase in reserves estimated in Zone A.

18. Smits et al, reference 7.19. Webster M: “AIT Performance in Prudhoe Bay,” BP

Exploration Memorandum, October 1997.

Summer 1999 49

For the AIT tool, saline mud combined withspiral or corkscrew boreholes can produce astrong signal and completely smear an arrayinduction tool log (below). The origin of theinduction tool response distortion is likely dueto changes in standoff.1 Practical experienceshows that the effect is worst in 6-in. holeswhere the standoff distance is limited. As thetool moves along the borehole, the standoff ribson the tool tend to fall into the grooves, allowingthe induction sonde to approach the boreholewall—periodically reducing tool standoff. Using the fact that the spiral borehole introduces a periodic effect or distortion to the log, several

signal-processing techniques effectively cancelthe unwanted periodic signal.

There are three main steps: automaticallydetecting the primary signal and its harmonics,estimation of their frequency, and subsequentremoval of all unwanted components. For AITmeasurements, the method involves frequencyspectrum estimation and peak identificationover short logging segments, in which each arraydata segment is replaced by a parametricautoregressive model. The advantage of thisapproach is automatic detection and frequencyestimation of the dominant periodic componentsin each segment. Typically, up to three sinu-

soidal harmonics are detected in AIT logs. Oncethe harmonic components are determined, thetask is to filter them out. For each array datasegment, a notch filter is designed with theappropriate transfer and phase characteristicsto remove—without phase distortion—all har-monics from the spiral borehole (bottom). Thefilter is applied to the raw signal just after bore-hole correction. Similar methods have beenproposed for handling the effects of corkscrewrugosity on nuclear logs.2

Dealing with Spiral Boreholes

1. Barber et al, reference 21, main text.2. Betts P, Blount C, Broman B, Clark B, Hibbard L,

Louis A and Oostoek P: “Acquiring and Interpreting Logs in Horizontal Wells,” Oilfield Review 2, no. 3 (July1990): 34-51.

Spec

tral p

ower

, dB

50

40

30

20

10

0

-10

-200 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Frequency

0 0.1 0.2 0.3 0.4Normalized frequency, Nyquist = 1

0.5 0.6 0.7 0.8 0.9 1

Phas

e sh

ift, d

egre

es

100

50

0

-50

-100

Atte

nuat

ion

resp

onse

, dB

0-5

-10-15-20-25-30-35

LQCValid flag1-ft2-ft4-ftORMag mud flagNonMag

LQCValid flag1-ft2-ft4-ftORMag mud flagNonMag

Hole sizein.

Array resistivity10 in.20 in.30 in.

50

100

150

60 in.90 in.

4 14

Rmohm-m

1.0CBM 10.0Resistivity, ohm-m Resistivity, ohm-m

Before filtering After filtering

100.0 1000.0

0 10

Hole sizein.

Array resistivity10 in.20 in.30 in.

50

100

150

60 in.90 in.

4 14

Rmohm-m

1.0CBM 10.0 100.0 1000.0

0 10

Induction logs in a spiral borehole. Onthe left is a Platform Express inductionlog from a Canadian well with a severespiral borehole (left). Filtering the datafrom the spiral borehole well hasremoved unwanted harmonic boreholenoise, and now the environmental flagsindicate that the 2-ft resolution logs arevalid (right). Chart-based C, bad hole Band magnetic mud M log quality-control(LQC) flags are shown to the left ofeach log. The color of the chart-basedand bad hole flag shows the recom-mended resistivity bed resolution. Thecolor of the magnetic mud flag indicateseither magnetic mud (red) or nonmag-netic mud (yellow).

Frequency spectrum and notch filter.A power spectrum of the logging datashows several peaks (arrows) causedby cyclic standoff variations in toolstandoff. A frequency response showsthe attenuation (top right) and phase-shift (bottom right) characteristics ofa three-frequency notch filter designedto remove periodic noise found in log-ging data.

>>

Handling dip—The large volume of investi-gation of induction tools complicates the inter-pretation of their logs. Modern inductiondevices such as the AIT tool are designed foruse in vertical wells, and are carefully focusedto limit their response to a relatively thin forma-tion layer perpendicular to the borehole.However, in wells at high relative dip, theresponse cuts across several beds, and themeasurement is no longer focused in an iso-lated single layer. The effect of high relative dip

50 Oilfield Review

Window 1Backscatter

Processing flag statistical analysisPEF flags upDensity flags up

BS average reconstruction errorSS average reconstruction errorLS average reconstruction error

W20.50%0.03%2.29%

W31.29%0.41%

W4

-0.58%

Window 2

Window 3

%-10 10 % %

Window 4

-10

XX300 XX300

XX400 XX400

10

Window 3

Window 2

Short spacing, Window 1

Density

2

Window 4

-20 20

Window 3

Window 2

Cost function

0 200

Long spacing, Window 1g/cm3 3

Form factor-0.5 % 0.5 6 in.

Caliper16

Crystal resolution5 % 25

Total count rate0 cps 1M

High voltage1600 V 1700

Backscatter detectorLow-energy window CR

0 cps 10K

High voltage1600 V 1700

Form factor-0.5 % 0.5

Crystal resolution5 % 25

Total count rate0 cps 500K

Low-energy window CRShort-spacing

0 cps 5K

High voltage1600 V 1700

Form factor-0.5 % 0.5

Crystal resolution5 % 25

Total count rate0 cps 50K

Low-energy window CRLong-spacing

0 cps 1K

Detector reconstructionerrors Black areas show that the corresponding error flag is set

Offset errorTau loop errorStabilization loop or crystal resolution error

0.7%0.0%Window 1

1.93%-0.53% 3.32% -0.29%

0.70%

B

A

>Density LQC logs. In a well drilled with nonbarite mud in the North Sea, the density inversion processing reconstruction errors (left) for all detectors are seento be nearly zero—closely tracking the center of each track, as expected. The density curve (yellow) from this interval is superimposed to highlight the largechange in densities computed across this interval. The global cost function log is low throughout the entire interval indicating good reconstruction and highconfidence in the inversion results. Statistical analysis of the reconstruction errors shows that every energy window is below its maximum bias level. Thehardware LQC logs (right) from this well show stable tool operation. As expected, the backscatter total count rates (black) in track 1 anticorrelate in Zones Aand B with those of the from the short-spacing detector (black) in track 2 and long-spacing detector (black) in track 3. No detector hardware error flags weredisplayed in the green columns shown at the left side of each track.

is to blur the log response and to introducehorns at the bed boundaries.

Traditional dip-correction algorithms forinduction logs are limited in practice to anglesless than 50º because of an increasing nonlinearresponse to dip and adjacent bed-boundary con-trasts. As a result, interpreting resistivity frominduction logs at apparent dip angles over 50ºhas been limited to iterative inversion using 1Dforward models. The presence of invasion hasadded additional complexity to the geometry,requiring processing based on 3D inversioncodes. Even with fast computers, the processing

has been excessively time-consuming for longlog sections or when many thin beds areencountered. Recently, a new algorithm basedon a maximum-entropy inversion of raw, bore-hole-corrected array data through a fast 1Dforward model has dramatically improved thespeed and ability to interpret multiarrayinduction logs in invaded formations at high rel-ative dip angles (see ”Interpreting InductionLogs in High Dip Angle Formations withInvasion,“ page 54).

Summer 1999 51

Window 1Backscatter

Processing flag statistical analysisPEF flags upDensity flags up

BS average reconstruction errorSS average reconstruction errorLS average reconstruction error

Window 2

Window 3% % %

Window 4-10-10 1010

Window 3

Window 2

Short spacing, Window 1

Detector reconstructionerrors

Window 4-20 20

Window 3

Window 2

Long spacing, Window 1

Cost function0 200

XX800XX800

XX900XX900

Form factor-0.5 % 0.5

2.5 in.

Densitystandoff

Standoff

0

Crystal resolution5 % 25

Total count rate0 cps 1M

BackscatterTotal count rate

1650 V 1750

Black areas show that the corresponding error flag is setOffset errorTau loop errorStabilization loop or crystal resolution error

High voltage

1600 V 1700

Form factor-0.5 % 0.5

Crystal resolution

5 % 25

Short-spacingTotal count rate

Short-spacing Long-spacingBackscatter

0 cps 500K

High voltage

1620 V 1720

Form factor-0.5 % 0.5

Crystal resolution

5 % 25

Long-spacingTotal count rate

0 cps 50K

W21.44%1.44%

W32.91%0.87%

W4

-2.48%

4.9%1.4%Window 1

-0.53%-0.72% -12.59% -2.13%

0.48%-1.04%

B

A

>Density LQC logs. In another 52º deviated well drilled with heavy barite-weighted mud for a different operator in the North Sea, a section of the density pro-cessing LQC logs tells a different story. The long-spacing and backscatter detector reconstruction errors (left) shown in tracks 1 and 3 are large because ofthe presence of heavy barite mud (14 lbm/gal) [1.67 g/cm3] and a highly fractured borehole environment through a known coal bed. In other parts of thewell—with better borehole conditions—reconstruction errors were lower. The density algorithm automatically detects the presence of barite and changesthe weighting of the window count rates from each of the detectors to obtain the most accurate answer in this difficult environment. The hardware LQC logs(right) from a higher section in this well show an unusual phenomenon. The total count rates (black) in the lower coal bed in Zone A, from the backscatterdetector (track 1) anticorrelate with those from the short-spacing detector (track 2) and long-spacing detector (track 3) as before, and as expected. However,in Zone B, they all decrease uncharacteristically. This is apparently due to increased attenuation in the backscatter gamma ray flux due to thick barite mud-cake in this zone. The density standoff curve (red) in the depth track confirms the increased thickness of mudcake over this zone.

indicates a problem in calibration, excessive padwear or tool standoff. High intermittent values inthe reconstruction errors indicate an abnormalnoise level on the measurement (hardware prob-lems), or instabilities in the inversion processthat may be associated with bad borehole condi-tions. A log of the cost function used in the inver-sion modeling helps to evaluate confidence in theestimations. Hardware LQC logs along withdetector count-rate logs are used to help confirmtool response, calibrations and stabilization inunusual environments such as in thick mudcakewith heavy muds (above).

Active Log Quality ControlMany questions arise when logs appear strange.Raw data may be fine, but the computed forma-tion parameter, such as density, may look abnor-mal. This leads to questions: Is the tool in anunusual formation? Is the software correct? Is thecalibration correct? Is the tool working properly?Good log quality control (LQC) resolves theseissues, and with the addition of real-time envi-ronmental corrections provides insight duringacquisition into both the effects of the loggingenvironment on every tool measurement and theway these measurements are being processed.

Numerous LQC analyses, logs and flags areavailable in the Platform Express system toensure quality measurements and processing.

These literally form a log quality-control hierar-chy from the top level of environmental process-ing down to the bottom level of tool-specificsensor performance and calibration. Followingare some examples showing how LQC activelyworks to provide better logging answers both atthe wellsite and afterwards.20

Density LQC—Window count-rate reconstruc-tion errors from the density inversion algorithmare a measure of the ”health“ of the inversionprocess (previous page). They pinpoint significantdifferences between modeled window countrates and those measured for each detector. Alarge systematic bias in the observed reconstruc-tion error log for more than one energy window

20. For more on LQC and tool specifics see: PlatformExpress User’s Guide. Houston, Texas, USA:Schlumberger Wireline & Testing, 1999.

52 Oilfield Review

XX850

XX900

XX950

X1000

2 g/cm3

Density

32 g/cm3

Backscatter density

ADRP

Standoff

Depthft

Gamma ray

3 0 API 150

Standoffdensity

0.5 in. 0

2 g/cm3

Short-spacing density

LQC flags

PEF computation

Density computation

Density detector

Accelerometer 3

2

UBI amplitudeLow High

UBI transit timeLow High

g/cm3

Long-spacing density

3

A

>Density profile caused by pipe groove. In this 97º-deviated wellbore, pipe groove filled withmudcake produces a mudcake density profile, shown in the depth track. The backscatter (dotted),short-spacing (short dashed) and long-spacing (long dash) detector density curves show anincreasing density profile as they each see deeper into the formation. LQC flags, shown on theleft of track 1, verify that the tool readings in this environment are normal. The UBI UltrasonicBorehole Imager tool images—amplitude (track 1) and transit time (track 2)—confirm that thepipe groove contains mudcake. In other sections of the wellbore that have been scraped cleanof mudcake, each detector reads the same density. The characteristic parallel lines caused bythe drillpipe scraping the borehole wall shown in Zone A confirm this. The standard resolutionformation density (solid blue) derived from inverse modeling is shown in track 2.

A typical example of significantly high recon-struction errors occurs when the logging engi-neer selects the wrong mud-type-processingoption—Barite or No-Barite. These two modescorrespond to different tool-response models aswell as specific inversion schemes, and thechoice has to be made according to the impor-tance of photoelectric absorption in the mud,which is linked to the barite content. If the engi-neer does not have exact knowledge of the mudcomposition, the wrong option might be selected,resulting in poor estimation of density and Pe.However, the raw count rates are alwaysrecorded, and since processing is applied on theraw counts, it is always possible to recomputecorrect results regardless of the acquisition mode.

Recently, a new ”switchless“ mud-densityalgorithm was designed to eliminate problemswith incorrect mud-algorithm selection. Currentlybeing field-tested, the algorithm uses a uniquetool-response model, valid for all mud properties,and a generalized inversion scheme based on abarite indicator. The barite indicator provides anestimation of the amount of barite in the mud ateach depth, and is based on low- and high-energy window count-rate ratios in the back-scatter detector.

Another example of how LQC provides activecontrol of the log acquisition process is theautomatic offset recalibration for high-resolutiondensity logging measurements. The robust low-resolution window count-rate reconstructionerror helps identify slowly varying count-rate off-set errors on the backscatter and short-spacingdetector measurements—used primarily for thehigh-resolution density logs. Any offset count-rate error detected in these measurements is cor-rected before being used in the inversionalgorithm to estimate the high-resolution forma-tion parameters. Whenever LQC detects extremerugosity or bad borehole, the more robust low-resolution density logs are recommended.

Summer 1999 53

An example from the North Sea shows howLQC helps build confidence in the tool measure-ments when the unexpected happens in unusualborehole conditions (previous page). Each of thethree detectors in the Platform Express densitytool sees progressively farther into the formation.Like the invasion profile produced by an array-resistivity tool, array density measurements pro-duce density profiles. These profiles depend onthe mud weight used. In wells drilled with lightnonbarite mud, the density profile tends toincrease from low to higher density, as the detec-tors look deeper into the formation. Typically, inwells drilled with high-density barite mud, thenormal density profile will be from high to low aseach detector looks farther into the formation.

When first seen in boreholes, unexpectedapparent density profiles were thought to be dueto hardware problems. However, LQC quickly ver-ified that the tool was functioning correctly.Further examination revealed the answer: Indeviated wells, where pipe grooves frequentlyoccur, the borehole is scraped clean of mudcake,giving an unpredicted density response. Nowthat the phenomenon is well understood, theinversion algorithms are designed to accommo-date this effect.

Environmental LQC for resistivity—The oper-ational limits of the AIT induction measurementhave been incorporated into a ”fuzzy-logic“algorithm that uses real-time inputs of caliperand mud-resistivity measurements to determinethe best resolution logging output consistent

with the environment (above).21 The output of the logic has four possible states—1-ft valid,2-ft valid, 4-ft valid or “Out of Range.” The laststate is flagged when environmental parametersare completely outside the range of the leastrestrictive induction tool measurement. Thismeans laterolog tools would provide better resis-tivity measurements.

However, the chart-based induction tool LQCalgorithm is based on smooth boreholes, anddoes not always detect when the environment isunfavorable for induction measurements. Inwells where the borehole is very rough or whenstandoff is inadequate, spurious spikes and otheranomalies might render high-resolution logsunusable. Research has shown that high-fre-quency induction array signals come from nearthe borehole, confirmed by the extremely sharpspikes near the tool axis seen on the shortestarray Born-response tool sensitivity function. Inthese cases, a rugosity-detection algorithm com-bines high-pass-filtered, short-array data withmud-resistivity information to make certain thatrugosity detection is dominant. The default well-site presentation is the most appropriate com-posite-resolution log, which varies smoothlybetween the 1-ft to 4-ft resolution log—basedon the combination of the chart and hole-rugositylogic. In all cases, the three basic-resolution andcomposite-resolution logs are always recorded.

Remote witnessing—Recently, BP AmocoExploration initiated a program of remote wit-nessing on their wells in the Andrew field in theNorth Sea by combining the capabilities ofPlatform Express real-time LQC and interpreta-

tion with the InterACT communications system.The InterACT system provides the capability forreal-time transmission of log data and wellsitegraphics to distant locations. This allows directand immediate communication and interactionbetween the offshore wellsite and consultants inAberdeen and London during log acquisition forbetter and more timely decision-making.

For example, in one well, a wireline loggingtool was unable to reach target depth due towellbore deviation. The situation was confirmedwhile logging. An immediate decision was madeto pull out of hole and go straight to a drillpipe-conveyed logging option. In other cases, irregu-larities in borehole dimensions shown on thecaliper were witnessed during logging and aquick decision was taken to pull out of the holeand rig down—eliminating the possibility of astuck tool. In all cases, real-time LQC providedconfidence in the tool measurements during log-ging, enabling operators to make appropriatedecisions based on environmental constraints,and not on limitations in the tool performance.

Remote witnessing has also decreased costsand improved safety by reducing personnel andtransportation requirements at the wellsite.Logs and evaluations are immediately availableto the experts who need them, and real-timeLQC ensures that logging measurements arevalid and can be trusted. If problems occur,expert opinions are available to help with con-tingency plans and decisions.

Limit of 4-ft logs

Possible large errors on shallow logs – 2-ft limit

Limit of 1-ft logs

AIT-family toolsrecommended operating range

using computed standoff method

Possible largeerrors

on all logs

(Rt/Rm)(hole diameter/8)2 (1.5/standoff)

R t, o

hm-m

1000

100

10

0.01 0.1 1 10 100 1,000 10,000

1

Possible large errors on all logs

Possible large errors on 2-ft logs

Possible large errors on 1-ft logs

Recommended range

R t, o

hm-m

1000

100

10

0.01 0.1 1 10 100 1,000 10,000

1

(Rt/Rm)(Bit size/8)2

>Operating range for AIT induction measurements. Borehole resistivity, Rm, hole diameter and standoff limit the range of acceptableAIT induction measurements (left). The chart shows that the effect of the borehole environment is most significant on the high-resolu-tion logs. A web-based job planner can be used to determine what resolution is usable for a range of expected formation resistivity andborehole parameters: Rt, Rm and bit size (right). The scatter of data represents the range of uncertainty on the input parameters.

21. Barber T, Sijercic Z, Darling H and Wu X: “InterpretingMultiarray Induction Logs in Difficult Environments,”Transactions of the SPWLA 40th Annual LoggingSymposium, Oslo, Norway, May 31-June 3, 1999, paper YY.

A new algorithm based on maximum-entropyinversion of borehole-corrected multiarrayinduction data through a fast 1D forward modelhas been developed and tailored for highly devi-ated wells.1 This algorithm provides the sameinterpretation for invasion that has previouslybeen available only for vertical wells. The key to maximum-entropy inversion is a fast forwardmodel. For this model, an analytical solution is used to compute the response of the AIT toolin a layered formation with dip. The response of each array is computed by finding exactsolutions to Maxwell’s equations for the beds at a given dip angle. In implementation, it is desirable to have the layer thickness lessthan the resolution of the sensors—typicallylayers 6- to 12-in. [15- to 30-cm] thick.

The inversion is formulated on finding theunknown formation conductivity that minimizesa cost function. Like the other inversions usedin the Platform Express system, the first term in this cost function is a measure of how wellthe forward model predicts each array raw

measurement—weighted by the expected erroror noise in each measurement. For example, if the model predicts array voltages that agreewell with those observed on each AIT receivercoil, then its contribution to the cost function is low. The second term is one proportional to the total entropy in the resistivity log. Thisterm adds stability to the solution. Finally, there is an empirical smoothing term includedin the cost function. The smoothing term also helps add stability to the solution, but is used sparingly because it tends to decreasevertical resolution.

The concept of entropy as applied to log datais not intuitive. In physics, entropy is a measureof the degree of disorder in a system, and thesecond law of thermodynamics states that thetotal entropy of a system can never decreaseduring a change. Applied to log data, entropy is a measure of the departure of log values fromlocally averaged values. For example, by settingup a simple least-squares inversion of the thinlylayered model—done by including only the first

term in the cost function—the results can havemany high-frequency “wiggles” (below).Including the entropy term in the cost functionhelps smooth out all the extraneous high-frequency information content—meaninglessinformation below the resolution capability of the tool.

The implementation of maximum-entropyinversion processing is more robust when pairsof arrays are inverted at the same time to obtainformation resistivity. As a result, rational sets of array pairs are inverted together, with thesurprising property that the depth of investiga-tion of each pair is the same as that of the deeper-reading array. This means that radialresponse functions for the arrays can be used to define the radial response of the invertedlogs. Furthermore, by weighting the results of inverting the array pairs, the inverted logscan be focused radially to give logs with stan-dard AIT depths of investigation.

The combination of maximum-entropyprocessing, resolution-matched inverted forma-tion resistivities and radial focusing is calledMaximum Entropy Resistivity Log INversion(MERLIN). This processing works on AIT data at all dip angles from 0º to approximately80º (next page, top). In principle, it will work up to 90º, but in practice, the parameterizationrequires that the wellbore cut all beds of interest. MERLIN processing replaces theBorn-response function filter-based standardAIT processing. The resulting logs can be inverted for invasion parameters as if theywere in a vertical well, but at any dip angle. In addition, although maximum-entropy inver-sion was developed to remove the effects of high dip, the exact solution forward model at the heart of the method works at any dipangle, including zero.

Interpreting Induction Logs in High Dip Angle Formations with Invasion

Maximum-entropy solution. A simple least-squaresinversion of the thinly layered model results in a logwith a high noise component (left). In the high-entropy solution (right), the dotted curve indicatesthe increased entropy found in the maximum-entropy solution.

54 Oilfield Review

Resistivity, ohm-m Depthft

Highentropy

Lowentropy

0

10

20

30

40

50

60

70

80

90

100

110

120

130

140

150

1.0 10.0 100.0 1.0 10.0 100.0

Resistivity, ohm-m

>

For example, formations with high shoulderconductivity and with shoulder-bed contrastscan be modeled correctly with the exact-solution-based MERLIN processing. In a verticalwell drilled with oil-base mud, anomalies areseen on the bottom and top of the reservoirsection in the real-time AIT logs (right). Thehigh contrast between the reservoir resistivity(100 ohm-m) and that of the conductive clayshoulder beds (less than 1 ohm-m) causes thestandard logs to overshoot in the resistive reser-voir. This happens because standard AIT pro-cessing—based on Born-based responsefunctions—includes contributions from manydepth intervals around the tool. Although thesefunctions are accurate models of the inductionresponse in low-to-moderate-contrast envi-ronments, when the shoulder-bed contrastapproaches 500:1, the Born-based approxima-tion for the tool response is poor. The MERLINlogs are well-behaved in these difficult induction-logging environments.

1. Barber TD, Broussard T, Minerbo G, Sijercic Z andMurgatroyd D: “Interpretation of Multiarray InductionLogs in Invaded Formations at High Relative Dip Angles,”The Log Analyst 40, no. 3 (May-June, 1999): 202-217.

AIT 10 in.

Wellsite (1-ft)resistivity

AIT 20 in.

AIT 30 in.

AIT 60 in.

ohm-m

AIT 90 in.

AIT 10 in.

MERLIN-processedresistivity

AIT 20 in.

AIT 30 in.

AIT 60 in.

ohm-m

AIT 90 in.

AIT 10 in.

Wellsite (2-ft)resistivity

AIT 20 in.

AIT 30 in.

AIT 60 in.

ohm-m0.2 2K 0.2 2K 0.2 2K 0.2 2K

AIT 90 in.

AIT 10 in.

Wellsite (4-ft)resistivity

AIT 20 in.

AIT 30 in.

AIT 60 in.

ohm-m

AIT 90 in.

Bit size

12 in.

Caliper

17

0 API

Gamma rayDepthm

X240

X245

X250

X255

X260

X265

150

Correcting for dip. In a section of a deviated welldrilled with oil-base mud, real-time processed AITinduction wellsite logs show non-normal invasion profiles, resistivity horns and overshoots in four zones,shown in track 2. The first post-log MERLIN processingresults in track 3 assumed that the wellbore was at 75ºrelative to the beds, but better results in track 4 wereobtained assuming a 65º deviation. The final resultshows coherent resistivity separation in the top zoneand a dramatic increase in Rt, indicating more pay in each zone.

AIT 10 in.

Wellsite-processedresistivity

AIT 20 in.

AIT 30 in.

AIT 60 in.

ohm-m

AIT 90 in.

AIT 20 in.

MERLIN-processedresistivity (75 deg)

AIT 30 in.

AIT 60 in.

ohm-m

AIT 90 in.

AIT 20 in.

AIT 10 in. AIT 10 in.

MERLIN-processedresistivity (65 deg)

AIT 30 in.

AIT 60 in.

ohm-m

AIT 90 in.-40 degrees

Azimuth

360

70 degrees

Hole deviation

120

4

Bit size

14Caliper

C2

in.4

C1

14 0 API

Corrected gamma ray

Depthm

150 0.2 2K 0.2 2K 0.2 2K

X830

X840

X850

<Resolving shoulder-bed anomalies. The gamma raycurve (green) in track 1 clearly shows shale bedsabove and below the reservoir section. The wellsiteAIT 1-ft induction logs in track 2 show anomalousbehavior with large resistivity oscillations and a non-coherent separation over a large zone near the bottomand top of the reservoir. The 2-ft and 4-ft logs (tracks4 and 5) display similar behavior. The anomalies arecaused by high shoulder-bed contrast coupled with low-resistivity shoulders which violates the Born-approxi-mation used in the wellsite processing. After MERLINprocessing, the anomalies are eliminated from the logs in track 3, leaving the middle thick section of the reservoir unaffected.

Summer 1999 55

>

Real-Time InterpretationsWellsite interpretation during log acquisition isanother benefit of real-time environmental cor-rections. Platform Express interpretations featureintegrated petrophysical computations andgraphic presentations that help operators maketimely decisions about the reserves in their fieldduring the logging run. These include a completeformation volume and lithology evaluation, fluidsaturation analysis, invaded-zone gas saturationand a special horizontal well presentation.

For example, formation porosity is derivedfrom the traditional crossplot of tool-measuredbulk density, ρB, and the thermal neutron poros-ity. Comparing neutron response with that of thedensity measurement, a lithology-corrected for-mation porosity is determined. With theseresults, porosity-corrected formation grain den-sity, ρmaa, and volumetric matrix photoelectricfactor, Umaa, are computed, to provide inputsneeded for a standard ρmaa-Umaa crossplot-basedmineral interpretation. By its nature, this cross-plot method is independent of porosity. Finally,clay volume, Vcl, derived from gamma ray orspontaneous potential (SP) measurements,provides a third dimension to the standardmineralogy crossplot (below).

56 Oilfield Review

increasesVCI

Quartz

Umaa

ρmaa

Dolomite

Anhydrite

Calcite

> Formation lithology analysis. The ρmaa-Umaacrossplot, derived from neutron and density toolmeasurements, forms the basis for a standardmineralogy interpretation (right). Porosity iseffectively removed from these inputs, resultingin a crossplot that is a function only of the miner-alogy in the formation. End-points for pure anhy-drite, sandstone (quartz), dolomite and limestone(calcite) are shown. The lithology column colorchange depends on clay volume, Vcl, and isderived from the SP or gamma ray measurement.Colors assigned to the plot are derived from acolor cube with the corners of yellow, green,cyan and white on one face, which representszero Vcl, and corners of red, black, blue andmagenta on the opposite face, which represents100% Vcl. The Platform Express lithology columnshows significant lithology changes in the lowersection of the BP Amoco Catoosa test well (above).

70Depth

ft

Standoff(resistivity)

Standoff(density)

ohm-m

Computed Rt120

0.2 ohm-mAIT 90-in. resistivity

2K

0.2 ohm-m

Rxo

2K

1.95 g/cm3Density

2.95

45 p.u.Neutron porosity

-15

0

Zone of interest

Pe

10

90 0

10.0020.0030.0040.0050.0060.0070.0080.0090.00

RadialSw image

90

0 lbf

Tension

5K

0 MV

SP

200

6 in.

Caliper

16

0 API

Gamma ray

150

1600

1500

Summer 1999 57

The clay volume and the mineralogy crossplotdrive the real-time lithology column profile. Thecolor change depends on Vcl and the position ofeach point on the ρmaa-Umaa crossplot. This lithol-ogy image provides a valuable method to observelithology changes in the formation. Although notintended to provide a volumetric interpretation ofthe mineralogy of the formation, it can be usedeffectively for well-to-well correlation.

Next, porosity and resistivity measurementsare coupled with Archie’s equation to determinewater saturation. An optional invaded-zone gas orsteam saturation—computed from the density-neutron crossover separation—is used to com-pute a gas-corrected neutron-density porosity.

Finally, a special wellsite Platform Express pre-sentation has been developed for highly deviatedand horizontal wellbores (above). A real-time truevertical depth (TVD) computation—based on thedeviation derived from the built-in tool-axisaccelerometer—is used to plot a wellbore shapeversus cable depth using the lithology image as anarea fill in the wellbore trajectory curve. The avail-ability of a real-time well deviation and lithologyprofile at the well helps to explain unexpected log-ging results often encountered when crossingbeds and fault zones in highly deviated wells.

The Road AheadMore than two-thirds of the ”triple-combo“logging operations performed by Schlumberger in 1998 were done with the new-generationplatform logging technology. Real-time depth and environmental corrections provide the key togetting accurate formation information into thehands of the operator when and where it doesthe most good—at the wellsite. Real-time,speed-based corrections are being implementedon every tool platform that contains a built-inaccelerometer, such as the Xtreme quad-combotool string designed for high-pressure and high-temperature environments and the SlimAccessquad-combo tool string designed for the slim andcomplex-geometry borehole environment. Newtechnology, such as the HRLA tool, addresses thetraditional problems encountered in conductiveboreholes and promises to bring more accurateresults in a wider range of environments.

Improved 2D and 3D forward models arehelping to clarify increasingly more difficultlogging environments. Breakthroughs in com-puting speed will eventually lead to theirincreased application. Advanced processing, suchas MERLIN and HRLA 2D processing are availablein computing centers and may be availablesomeday for real-time operation. These develop-ments promise an exciting time for geologists,reservoir engineers and petrophysicists who useopenhole-logging measurements. —RH

Well trajectory

Anhydrite

Anhydrite

Sand reservoir

Resistivity PorosityDepth, m

X100

X150

X200

X250

X300

Horizontal well presentation. In this well,the wellbore penetrated salt and anhydritelayers, and entered the sand reservoir. Belowthe 75-m sand reservoir, the lithology log intrack 3 indicates that the well appears tohave entered another anhydrite layer andthen turned back into a second sand interval.However, the well trajectory plot (track 1)clearly shows that the wellbore turnedupwards prior to entering the anhydritelayer—suggesting that the well simply reen-tered the caprock anhydrite and then turneddown—back into the first sand reservoir.Without the well trajectory plot, analysiswould have been delayed until other logs anddeviation information could be correlated tothe log data to explain the lithology changes.

>