SPE 9341 S P E GLOBAL, A NEW APPROACH “neem TO COMPUTER-PROCESSED LOG INTERPRETATION by ¢. Mayer and A, Sibbit Services Techniques Schlumberger, Paris © Copyright 1980, Amerzon Ths paper wos presented atthe 55th Annvol Fol Technical Conference and Exhibition ofthe Society of Petroleum Engineers of IME, held in alls, Tena, September 2124, 1980 The motel i zbject to correction bythe thor. Perminion Yo copy i retried to.an batoct of mot mare than 200, words. Writer 6200 N. Cental Expwy., Delo, Texas 75208. lof Mining, Metellrgzal, and Patroloum Engineers, Ine. ABSTRACT A new computer processed log interpretation chain, GLOBAL, uses a structure independent of model and logging suite. An error model is defined relating tool ‘measurements to petropbysical parameters such as po- rosity, lithology, and fluid saturation. Then, using a minimization routine, GLOBAL searches for the solu- tion with tbe minimum error. This solution is consid- ered the most probable answer. A quality curve is presented indicating bow well the answers fit the chosen model. The curve belps to determine if the model is inadequate, or if insufficient information is available to solve the interpretation problem. The main advantages shown by the exemples ana- Iyzed are * Simultaneous use is made of all combination of logs, including experimental or recently intro- duced sensors. * The approach works for very complex litholo gies. * Ibis adaptable to all kinds of models and local Painted in USA conditions.SPE 9341 GLOBAL, A NEW APPROACH TO COMPUTER-PROCESSED LOG INTERPRETATION c. By Mayer and A. Sibbit Services Techniques Schlumberger, Paris INTRODUCTION GLOBAL consticuces 2 new methodology for computer. processed log interpretation, transforming 2 set of borehole logging measurements into a set of formation evaluation answers, such as porosity, lithology ad fluid sarurations. In this manner ic resembles other existing CPI programs (SARABAND® ©), CORIBAND® ©), etc.). However, the ‘new concept of GLOBAL offers a versatility of computation, choice of interpretation models and ease of utilizing new Jogging measurements that has been lacking in the pas. HISTORICAL BACKGROUND Uniil recently, most major log evaluation programs were based on crossplot techniques primarily using the Neutron-Density combination. Ia these programs, the com- puter reproduces, step by step, the clasical “manval” inter- pretation process. This approach has the advancage of being easily understood by log interpretation specialists and allows the users to follow the logic of a program in cerms of their owa way of thinking. Such techniques, however, have been outstripped by the present-day evolution of log interpretation in its efore to eal with the increasing complexity of formations in which oil and gas are soughs, and the advent of new sensors and increasing multiplicity of model, developed co understand these formations ‘An increasing number of versions, options, threshold parameters, etc, in log interpretation programs have been developed to cover a large range of special cases, increasing the dificalty of che log analysts’ ask. The existing programs remain valid in their principles, and give good results in ‘most cases; however, they have become complex to coat, and do not always make the bes use of available information, OBJECTIVES OF THE “GLOBAL” APPROACH Te was fele that a new approach to computer-processed log interpretation should be developed, pursuing the follow- ing objectives ‘© 10 use all available information - log measurements geological constrains, local knowledge. © to seek results which make optimum use of this com- plex body of information, * Mark of Schlumberger. ‘* to have a scrong potential for evolution, character- ied by the easy introduction of new sensors and new interpresation models. ‘* to provide a powerful and human-engineered quality coatrol of interpretation results In ceems of computer techniques, the above goals require: ‘use of n-dimensional rather chan 2-dimensional ross- plot techniques. + use of probabilistic concepts to handle the physical daca available, with a view obtaining the most probable solutions. «© the building of a flexible and modular sofeware sy tem, in which tools and models are separate plug: elements STATUS OF THE GLOBAL CHAIN Following the above concepts, a chain of computer pro- ‘grams has been designed, implemented aad tested, and is in tse in Schlumberger field computing centers, The results obtained so far confirm that these goals, for the most part, have been reached, This paper describes the principles and applications of the GLOBAL mechod, and outlines its main advantages with field examples. GENERAL PRINCIPLES To illuscrate the principles of the method, let us con- sider a dual mineral formation. The inpues of the GLOBAL program are for each sample, or level, ofthe well: « all available environmentally corrected logs, which may be written in array notation: a) = (Pr, by, Raw Ry GR, SP), 7) and ‘© asecof zoned” parameters: Ry, clay parameters, mud characteristics, ete. ‘The outputs, or unknowns, in such a formation are: 55 (, Ves Se Soo Pine) @ TOOL RESPONSE EQUATIONS “The relationships between inputs and outputs can be expressed by a set of rool response equations, one equation ( throughout the paper, arrays are underlined,SPE 9341 for each tool. For example, the density relationship may be expressed Po= Seo Pa +6 (1-See) Pat Ve Pa + (1-O-Ver) Pras @) Using the array notations introduced in (1) and (2), the tool response equations may be writeen: a =h() f(x) te “ where & oF (a1, 2s, .), is the set of inputs at a parcicular level, and f; isthe tool response function of the i" tool type. Ie should be noted thar the functions f, may depend on variables other than x1, x, ete. These variables, such at Pe, Pav etc, have traditionally been called "zoned parameters” and are assumed to be constant within a given zone. Deter mining these parameters must be done prior to the applica sion of GLOBAL Equations (4) must be solved in order to find an inter- pretation solution, In addition, a solution must comply with certain constraints delineating the likely domain of che results ‘The system of equations (4) may be either under- determined (less equations than unknowns) or overdeter- ‘mined. In the first case, an infinite number of solutions may be consistent with the system; thus if chere ace no valid rea- sons to select any one of them, no interpretation can be per- formed. Therefore, throughoue chis paper we shall assume that che syscem is overdetermined, or at least balanced. Since there is no exact solution ro an overdetermined system, we shall confine ourselves to looking for an approximate one, This is quite consistent wich the fact chat the rool response equations are only approximations to che physical reality, and thar the logs are subject ro dispersion, errors and sta- ‘APPROXIMATE SOLUTION AND THE INCOHERENCE FUNCTION ‘An approximate solution, x of system (4) is one where the residuals, ¢, defined for each tool by =a f(s) ) are as small as possible. Combining all rool response func- sions and constrains, we can define an error function, A (2, x), to bea weighted sum ofthe residuals, iacluding ad- itional penalty terms corresponding to non-satisfied con- straints: Sau = 3 (SREY + consrane penies (6) ‘This function is called the incoherence function, since it express the lack of coherence between logs, eslts, re sponse equations and constraints. In equation (6), ui is the standard deviation of the error distribution arising from the 48 rool response equation and results from dispersion on che C. Maver AND A. Sipart 3 log as well as from the response function itself. The incoher- cence function is explained in more derail in the nex« section. GLOBAL METHOD ‘The basic principle of the method may be stated a fol lows: for « given set of logs a, the GLOBAL method deter- mines che unknown vecwor x which yields the minimum ‘value of the incoherence function A(a,x). Under reasonable assumptions, it can be shown that the results obtained in this manner are the most probable ones. The GLOBAL method is thus a maximum likelihood method, finding for a set of log, sesponses the most probable log interpretation using all logs and theit responses. GLOBAL CHAIN ‘The practical implementation of the GLOBAL method requires a sequence of operations, performed under the con- trol ofthe log analyst; some of these operations are manual, others are partly or fully automacized. The main functions performed are: + eaviconmental cozrections of logs. + computation of error parameters, ‘* selection of the unknown answer array, x = (1 xa ), according to che formation interpresation model being use. + selection of tool response equations, fi + decermination of zoned parameters ‘application of the GLOBAL method to compute the ‘most likely values ofthe interpretation answer, x, at each level ‘QUALITY CONTROL ‘The GLOBAL chain provides the log analyst with a set cf powerful quality contzol displays. These help him judge the results and diagnose remedial action in the case of in- consistency ot error. DETAILS OF THE GLOBAL METHOD In general, che method may be applied to any set of logs and any set of unknowns, providing an appropriate GLOBAL model is defined, For the rest of this paper, we shall call: #1 GLOBAL program, any program based on the inimization, level-by-level, of an incoherence func- 5L0BAL PROGRAM ‘toeat INCONERENCE ‘MaoeL FUNCTION Fig. 1 — Structure of the GLOBAL Model.4 GLOpAL, A New APPROACH TO COMPUTER-PROCESSED LOG INTERPRETATION #1 GLOBAL model, the corsesponding set of un knowns, tool response equations, ew. being inpur into the program. We now describe in detail the various constituents of a GLOBAL model INPUTS As before, a = (2, a5...) isthe set of log readings to be used by the GLOBAL program. In genera, 2 is a set of environmentally corrected logs, but additional corrections may be necessary, for example shoulder bed and invasion Examples of inputs are: 2 = (Rete, Rute Rina, Ruger) a8 used in the Ry GLOBAL program (RTGLOB) 1 = (Po dss & Re, Room GR, SP), a5 used in che Reservoir Interpretation GLOBAL program (RIG) L0G DISPERSION ‘The acquisition process leading to a introduces many sources of error. These include: ® dispersion in the measurement apparatus (sensor, eleetronics,...) «dispersion in the raw data corrections (surface ac- quisition) «© dispersion in the environmental corrections (pre- interpretation sofeware) For each input a, we write o for the standard deviation of the error distribution of the acquisition process. If the errors are asymmetric (positive and negative ersors having differ- ent amplitudes), we use two dispersion coefficients: 0,~ for negative errors (log reading too low) and oy! for positive ‘errors (log reading too high). This often occurs with pad tools, where pad application problems yield errors mainly in one disection. UNKNOWNS & is the array of unknowns to be determined by the GLOBAL program. Examples are: = (Ry Reodi) in RTGLOB = (f,Vew Sur Sws Pine) inthe dual mineral model X= ((,VouSm Sus Pu) in shaly sands, Xi, « «+ Xn Should correspond to the variable petrophysical characteristics of the formation. In order to maintain the overdetermination of the system, x should have n0 more components than a, TOOL RESPONSE DISPERSION Like the logs, the tool response equations are subject to dispersion. The error in an equation a, = f(x), is the differ. cence berween the two parts, assuming that both the log a, SPE 9341 and the results x are correct. Le the standard deviation of such an error be Ty, T; may be caused by a combination of the following factors: «© the formulae f, are simplifications of complex phys- ical phenomena (eg. Archie, Wyllie equations, et. ‘* some zoned parameters may be erroneously selected, ‘© there exise hidden parameters, which represent vati- able characteristics of the formation effecting the Togs, bue which are not taken into account in the ‘yeetor x These would include such things as textural parameters which affect the acoustic and the elec- ‘tical properties of the formation; or special miner- als, assumed co be absent, Since the errors on the response equations may vary with the formation encountered, T, must be a function of x. For ex- ample, results from the sonic equation become less accurate in gas-bearing and unconsolidated formations. TOTAL ERROR ‘The total error of the system (logs and response func- tions) was defined as us, Assuming the acquisition and tool response errors are independent, the standard deviation of the total error is given by: a CONSTRAINTS ‘The general type of constraint we consider is of the form: g(x) 20, (8) where j is the constraint aumber, and g(x) is a function involving the unknowns (xy, x2, -..), 8 well as some log analyse chosen constants, bt no log measurements. As with the tool response functions, consteaints may also have dis- persion, T,, Four types of constraines are presently used: (a) Mathematical Constraints ‘These constraints are rigid and allow 0 dispersion; they result from the very definition of the unknowns. Exam- plesare: (b) Geological Constraints ‘These express approximate relacions between the un- knowns, caused by geological or physical laws, These expres sions are subject co dispersion. For example, SeSSue < Pues (IVa) (€) Local Constraints These express a priori knoviledge of the likely results, and are set by the log analyst. Thus we may impose a maxiSPE 9341 8, (Po dus Ts PeRO) Outputs: (ty X39 Xe), (formation description) 8 (San Sun Ves Vi Tool Responses: Funetions =f, (x) Constraints: y (x) > 0 Uncertainties: 6; on logging measure, ‘Tyon log response function f, Ton constraint a, INCOHERENCE FUNCTION afd? yy BC % +t? + tT? ne term per tool Max ‘One term per constraint Fig. 2— Summary of GLOBAL Model and Inco- erence Function. ‘mum clay concent, a minimum water saruration, a porosity range, etc. (d) Continuity Constraints Continuity constraints express the fact that the results cannot have a better vertical resolution chan the logs them- selves. Their dispersions are sec accordingly to prevent too much discontinuity in the computed results MINIMIZING THE INCOHERENCE FUNCTION Fig. 2 summarizes che inputs and outputs of the GLOBAL ‘method and incoherence function. A GLOBAL program es- sentially minimizes the incoherence function, A(a, x), re: sulting from che associated model, with respect to x A trial-and-error approach is used, An initial value, x, is com- pured using « heuristic or quick-look interpretation; then after each step an incermediare result, x is computed. The next result, x» 1, is derived from xq by a “steepest descent’ technique, (see Fletcher'®) and Powell). The successive iterations are stopped when a coavergence criterion is sats fied, The flowchart of the GLOBAL method is shown in Fig. 3. THEORETICAL LOGS In the incoherence function above, we may write a" instead of f(x). ay° may be thought of asthe theoretical log corresponding to unknowns x and fonction f, Ia the coatzol displays provided by the GLOBAL programs, a, and a,° are displayed for comparison and quality control C Maver anp A. Sissir 5 EXAMPLES OF GLOBAL MODELS Ia this section, we briefly describe three of the GLOBAL ‘Models that have been implemented. —RIGIOB (Ry GLOBAL) Model —RIG- Malkimineral Model RIG - Dual Water/Wexman-Smits Models RTGLOB MODEL In this model the unknowns are given by Ruan) 0 This is a step cesistivity profile assuming no annulus or sgadual evolution of the invasion depth, All available resis- tivity logs are used including any micro-devices, Laterologs, Induction logs, Sphericaly Focussed logs. Tool response ‘equations employ the classical radial geomecrical factor con- cept. For example: R=JRe+(1—J) ReforLaterologs (10) C= GC + (1G) C for Induction logs, ay where G and J are, respectively, the radial geometrical and ‘pseudo-geomerrical factors of the tools, themselves functions Fig. 3— Flowchart of the GLOBAL method.6 G1onat, A New APPRoAcH 10 COMPUTER-PROCESSED LOG INTERPRETATION of Re, Ryo and di'®*?, Very few constraints are used, mainly ‘mathematical ones. The geological constraint Rus < Ror R SR is used when Ry is known, (MULTIMINERAL MODEL ‘The unknowns in this model are: ',S:o)Sqp VesVion-sVone) (12) where Vas. Vans af€ the bulk proportions of up to 6 ‘minerals. These minerals may be standaed ones (sands, car- bonates, anhydrite, silt...) or more unusual (siderite, glau- conite, tufites ...). The log analyst chooses the minerals required for the interpre All the presently available logs can be used as inpats: © Rand Ry, from RTGLOB ‘© Porosity Logs (Neutron, Formation Density, Sonic) ver nn ‘or TOON orem Lovet ‘eo ao rane in cae) cn or aio ranges wot os auuyst SRECTIN oF Lacan “0 Mauer fron scr é z Fa eS — Z & soeuan aaron ws Fig. 4 —The chain of computation for a GLOBAL open-hole interpretation. SPE 9341 © Natural Radioactivity Logs (Gamma Ray, Nesural Gamma Ray Spectrometry) © Thermal Decay Time Log ‘* Electromagnetic Propagation Tool (Transit ime, ‘Accenuation) « Litho-Density Tool (Density, Litholog) + Spontaneous Potential For each tool, a response equation has been established, using cither standard formulae in the ease of eaditional logging tools, or new relationships obtained from recent studies for the newer tools. AS special cases of the multi- mineral model, we have: ‘¢ Dual Mineral Model (Using only Vaui and Vau2) + Shaly Sand Model (Using miner 1 for snd and ‘mineral 2 for sie) DUAL WATER” AND WAXMAN-SMITS® MODELS For the Dual Water model, the unknowns are given by: = (br ScotsSety Seis Viat--sVaas) (13) ‘where Sw Sw te the water saturations pertinent to “total” porosity, and Suis che bound-water fraction of the total po- rosty. Minerals 1 20 6 may include standard and special minerals, sit or various dry clay minerals Again, any com- bination of these can be selected by the log analyst The Wax sman-Smits model uses a similar set of unknowns. The same logs are used asin the Multimineral Model GLOBAL CHAIN DESCRIPTION ‘The minimization method, described above, constitutes the “heart” of che GLOBAL Method, but in practice, a com- plete processing sequence is necessary 0 implement the method, This sequence, partly manual, partly computerized, includes the following functions: © Preincerprecation (depth matching, environmental corrections), « Selection of the GLOBAL model to be used in each ‘¢ Computation of the unknowns by minimizing the in coherence function, ‘© Quality control and diagnostics. ‘The flowchart of Fig. 4 shows these functions. On the right of each box is the name of existing programs written «0 perform some ofthese functions Several of the operations remain, however, under the personal control of 2 log analyst. ‘The main components ofthe present chain are LODET (Log Quality Determination), RTGLOB (Ry GLOBAL) and RIG (Reservoir Interpretation by GLOBAL). Let us briefly re- view the elements of the flowchartSPE 9341 C Maver ano A, Sisorr 1 rrr evRONENTAL TOOL DESCRIPTION his cep is standard in any computer processed iater- InocaTORS sweets pretation chain and isnot affeced by the GLOBAL method. fe musosiry sous oF tom cans PRESS SH axe 1 e40H oveATOR his program performs the usual envisonmental correc- tions which remain unaffected by GLOBAL. Because corrections are considered as a pare of the log acquisition process, errors made during chis phase combine with the measurement errors £0 contribute to roel error. Acquisition uncertainties (1) are consequently computed at this time 10G QUALITY DETERMINATION (LQDET) For each logging measurement, ersor may result from many different factors, such as a caved hole, rugosity, the preseace of mudcake, noise in the tool's electronics, auclear statistics, cycle skipping, et. If we assume these factors eo be ‘mutually independene, we can write eat = Eo «ay where 0, are the individual standard deviations resulting from each factor. In practice LQDET computes at each level, ‘certain number of environmental indicators, such as rugos iy and mud cake thickness, which may affect the log read- ings. Then, foreach too, formulae describing the sensitivity of the measurement eo these indicators permit a computation of each 04, giving finally, via equation (14), gaa The pro- cess of LQDET can be summarized by Fig. 5. RIGLOB RTGLOB is the first GLOBAL program of the chaia, Here environmentally corrected values of resistivity logs are used ro calculae the most consistent values of resist the invaded and virgin formations SELECTION OF UNKNOWNS FOR RIG Depending on the interpretation model desired (mult ‘mineral, dual-waer, et.) and the lithology, set of unknowns is chosen. These can vaty in different zones and are selected by the fog analyst from a set of options. In making his choice, the principle of overdetermination muse be respected. SELECTION OF RESPONSE EQUATIONS ‘AND PARAMETERS ‘A umber of equations has been established for each tool, related to each model option, The log analyst must select the response equation he desires, and choose the parameters in the tool response equations which are considered to be known, Traditional crossplots and related techniques are commonly used for this purpose, as they have been tradi- tionally. CHOICE OF LOCAL AND GEOLOGICAL CONSTRAINTS. ‘These constraines are preset for each model but may be modified by the log analyse. unceRramy ue To Je saount oF Pes ‘ as 1 mOIcATOR IS ey CORRECTION WADE Tora Fig. 5 —A schematic of log quality determination (LQDET). IDENTIFICATION OF SPECIAL MINERALS (MINSPE) ‘This program fags the intervals where special non- permeable minezals are encountered, eg: salt, cron, anky- Arie, aypsom, potash and coal. The special minerals are detected when a set of conditions is satisfied by the logs. ‘When RIG reads che corresponding flag, 20 minimization is performed and a special coding is displayed. RIG (Level-by-level formation evaluation) ‘Ac his stage, all the ingredients of the RIG model have been gathered, except for the dispersions - T,, Each Ty de- pends on the formation conditions, and is therefore a func- tioa of the unknown asray x. In RIG, for simplicity, Ts (x) is computed from the initial estimate xo, and redevermined from time co time if the minimization process carries the fray, Xp £00 far from che inital estimate. RIG outputs a standard presentation of the results as well as a aumber of quality control displays and statistics. First, the results are presented with standard coding, with distinc pavteras used to represent che various minerals in the ‘multimineral option. Second, a “control display” comparing the original and theoretical logs, a and a", is shown. Four curves are played back for each input log. ‘the log itself, a, as read from the input file ‘© two curves representing a confidence interval around the log, a1 = uy us is the standard deviation of the tozal etror of the system, acquisition and tool re sponse, This confidence incerval is lightly shaded, «the theoretical log, a;" = f(x), where x isthe result found by the program. If a," is found inside the shaded confidence interval, the corresponding log response equation is within tolerance ‘Third, an indicaror, called the Reduced Incoberence is derived, level by level, from the incoherence function, The8 G1onat, A New APPROACH 10 COMPUTER-PROCESSED LOG INTERPRETATION seduced incoherence should read everywhere less than I (ex- cept for isolated levels), indicating a reliable minimization. Fourth, statistics are calculated to summarize the above quality controls, zone by zone, forthe use of the fog analyst. QUALITY CONTROL Quality Control is facilitated by the control displays and statistics provided by RTGLOB and RIG. After GLOBAL SPE 9341 processing, che log analyst can compare retical logs, and monitor the reduced incoherence. If the reduced incoherence is oo high (> 1), then the ‘model has noe been satisfied, Causes ofthis may come from a poor measurement, en incorrect function or constrsing, 2 poor and theo- ‘choice of a zone parameter, or an incorrec selection of x. If the reduced incoherence is good (< 1), the tool response CONTROL DISPLAYS INVASION PROFILES RESISTIVITY PROFILE (CORRECTED) CALIFER 6" iy . Nua potas ft oi 02 200 le" ze] [02 2000] T 7 REDUCED r T ' INCOHERENCE Soo. 2000) L 9900 T 0000} Fig. 6 — Example of R, GLOBAL determination using four resistivity logs — ULd, Ld, LLs, MSFL to evaluate Ry, Rio and di In the "control displays” dashed curves are the original logs; solid curves are the theoretical logs after mini- ‘mization of the incoherence junction. Uncertainties om input logs are shown stippled.SPE 9341 equations, 2s well a the constraint, have all been satisfied by the results within acceptable tolerances. Te may happen that incorrect results satisfy log response ‘equations and constraints. This occasionally happens when ‘here are an insufficient aumber of active logs and constrains, and the system is therefore somewhat undetermined. When this situation is diagnosed, more information muse be made available, by either adding logs and constraints, or reducing errors FIELD EXAMPLES, EXAMPLE OF GLOBAL R, Fig 6 shows a resule of processing using the RTGLOB program. The headings "Invasion profile” and “Resistivity profile” are the standard presentation. “Resistivity profile” dlisplays che logs after environmencal cosrections together with the results, Ry and Ry. "lavasion profile” displays graphically the amount of filtrate invasion (area between invasion diameter and caliper) and the quality curve - the reduced incoherence, Immediately we can see that the re- duced incoherence is less chan 1 chroughour, implying chat the results are consistent with the interpretation model. In the section, “control displays", log analyse controls are shown for the deep Laterolog (LLd) and for the deep Induction (ILA) inputs. For both logs the dashed curve represents the original input log; the solid curve represents the theoretical log, calculated after minimizing the incoher- ence function; and the stippled area indicare the uncertainty oon the log inputs Ifthe model fi is good, the solid theoretical carve should remain within the stippling. We can see that this is indeed che case. In this example, Ry = .021 Qm and Rye = 052 Om at tocal depth, so in water-bearing zones we ‘would expect the Induction ro bea better measure of Ry. The reservoir is a complex sequence of layers separated by resis- tive thin streaks. The thickness of the layer is such thatthe deep Laterolog is severely affected by shoulder beds, account- ing forthe large uncertainty on the deep Laterolog in zones {9995-10070 ft and 9900-9920 ft. In the higher resistivity areas, the Induction, being a conductivity measurement, has the higher uncertainyy. RTGLOB weighs all these factors and computes the Ry and Ry» curves shown. This combination of deep ivasion, high and low resis- tivities and layering shows the need for both the Dual Latero- Jog and Induction for Re decermination, EXAMPLE OF THE GLOBAL CHAIN - RIGLOB AND RIG Fig, 7 shows another example of R, GLOBAL process ing - 2 preliminary stage in a complete GLOBAL chain of open-hole interpretation. Fig, 8 shows the results of the reservoir interpretation made by GLOBAL (RIG). The presentation is traditional, but with the addition of the reduced incoherence curve, and the possibilty of a more detailed lithological description. ‘The lithology in this example is san, sileand clay. C. Maven aNp A. Simprt 9 aa a Tess PRE oo 5 ws m . ty ts ‘ tam sr ws te XY Fig.7 — Another example of the R; GLOBAL program. Inputs are resistivity logs — Deep Induction and Literolog, MSEL*; and out- puts are Ry, Roy and diameter of invasion Figs. 9, 10 and 11 are che control displays for the input logs. In each display the four curves discussed previously are presented. The shaded areas represent confidence limits for the measurements. As can be seen from the Ry, and Py con: tol displays, these limits can be asymmetric. In this example good agreement can be seen between all logs and their eheorercal counterparts indicating that the results are consistent with the interpretation model. This is reflected in the reduced incoherence which is nearly every ‘where less than one. This condition is no satised, however, in zones A and B, indicating chat we should look more closely at the solution in these areas. The problem arises be: cause the LDT, EPT* (Fig. 11) and the R.. output from RTGLOB (Fig. 9) with their excellenr vertical resolutions see fine laminations in the two zones. The other measure- Mark of Schlumberger.10 Giosat, A New APPROACH To CoMPUTER-PROCESSED LOG INTERPRETATION SPE 9341 F.uiO ANAL.| ANAL. | LITHOLOGY THEORETICAL LOGS bs sal wo. Win ou MAT reoucen [100 O {50 o INCOHERENCE) DIFF. CAL. 5 0 4°03" hoo p----- cTITy ORIGINAL LOGS VA nent Ni Ly Mf hal iat 4 eee ily aah a Fig. 8 — Standard presentation of the results from 4 GLOBAL RIG program. In most respects, this is identical to the traditional present tion. The added "Reduced Incoberence” curve indicates the success of the model fit and minimization process. Fig. 9—Control displays for Ray and Ru, which are outputs from the GLOBAL R, program ‘and inputs $0 the GLOBAL interpretation phase (RIG).SPE 9341 C. Maver AND A. SimpiT THEORETICAL LOGS THEORETICAL LOGS |___ORIGINAL [ORIGINAL Logs UNCERTAINTY UNCERTAINTY ama | | Ue Th RAY on Po 1 o AM y00j60 Yo Fatt bamsiee | asim @B/m 0 s]o "at Fig, 10— Control displays for Gamma Ray, Den- Fig. 11 —Control displays for Litbodensity and sity and Neutron logs, after RIG process Electromagnetic Propagation tools. With ing. GLOBAL's flexibility, shese new tools were easily incorporated into the interpretation presented in Fig. 8.2 G1osat, A New APPROACH TO COMPUTER-PROCESSED LOG INTERPRETATION ‘ments do not have the same vertical definition and indicare 4 porous formation, GLOBAL makes a good compromise indicating the presence of clay and chin laminations. The constancy of the residual water volume bear out this inrer- pretation, ADVANTAGES OF THE GLOBAL METHOD ‘The main advantage of the GLOBAL approach is that, by making full use ofall available measurements, it leads t0 ‘more reliable results than conventional programs. Its ability to deal with different models and rools allows a far more flexible approach, particularly in the more difficult interpre- tation problems. ‘Other advantages, though les immediately obvious, are important when we cosider the problem of interpretation in ts human and economic context. These concern the better control of processing and results, and the potential evolution ‘of the GLOBAL CPI chain with advances in geology, geo- physics and sensor des Let us review briefly these three categories of advan- sages. RELIABILITY Better Behavior in Bad Holes Due to the smooth operation of the confidence factors, the coos lease affected by bad hole are used optimally in bad hhole zones; no useful information is los. Use of all available tools ‘The GLOBAL structure permits essy processing of all available logs, withoue sacrificing the information given by any of them. Ie makes a synthesis rather than a choice. The ‘most recent Schlumberger sensors have been introduced into GLOBAL, including the Natural Gamma Ray Spectrometry log, the Litho-Density log and the Electromagnetic Propaga- tion log. Standard sensors not conventionally used in open- hole formation evaluation have also been incorporated, e.g the Thermal Decay Time log. Use of the Most Adapted Model ‘The option of varying up to six minerals at a time com- bined with the use of more aumerous tools, allows us to SPE 9341 process very complex or special lithologies, encountered ‘more and more frequently ia the search for oil and gas. ‘The flexibility of the approach permits the selection of the model which is best adapted to the type of formation analyzed. Local knowledge can be naturally introduced by ‘way of local constraints. QUALITY CONTROL ‘The proper use of the incoherence and control displays, comparing real and theoretical logs, gives a better under- standing of the problems met daring the interpretation process. This helps the log analyst replace or modify the ‘model used until 2 consistent interpretation is made. Ar the end of the process, the log analyst has nor only a set of results, buc all the quality control information allowing him to est- ‘mate the confidence he can pat on the results, POTENTIAL OF EVOLUTION ‘The key t0 the pocential evolution of GLOBAL. is the use of tool response equations as the central past of the model. These equations are che most natural way to express the knowledge of the functioning of a wireline too, derived 1s they are from the physical principles of the tool, and fur- ther adjusted by laboratory and core studies, CONCLUSIONS ‘The GLOBAL method is a general framework for log interpretation. A variety of ools and formation models can be used at che log analys’s discretion. The method makes the bese use of these selections, by computing a maximum likli- hood solution. Iesexibility enables ito easily grow with the inroduc- tion of new tools and increasing complexity of models, Pro- vided that the logging suite has sufcient resolution, and individual tool responses are knows, GLOBAL should be able to process satisfactorily all rypes of formations. As of row, the GLOBAL chain delivers results comparable to those ‘of CORIBAND or SARABAND jin standard cases, and shows improvement over them in many complex lithology or bad hole examples. In the future, the concept of GLOBAL will be extended 0 new tools, new models and automatic parameter selectionSPE 9341 GR G(R, Randi) TR Raopd) Rua Rusa Rute Ruse, Rae R Ry C. Mayer AND A, Susi NOMENCLATURE Log reading ofthe i* logging measurement ‘Theoretical log reading of the "logging ‘measurement \N-dimensional array of log readings Conductivity Diameter of invasion i cool response function j**conscraine Gamma Ray log Radial geometrical factor function for Induction Radial pseudo geometrical factor function for Lacerolog Resistivity from deep Induction Resistivity from deep Laterolog Resistivity from shallow Lacerolog Resistivity from MSFL* Resistivity of mud fileate Resistivity of virgin formation Resistivity of connate water Resistivity of invaded zone Water sacuration ia invaded zone “Total” porosity water saturation in invaded ‘Water saturation in virgin formation ‘Bound water fraction ofrotal porosity “Total” porosity water saturation in viegin formation * Mark of Schlumberger. A(x) Pr Pa Pra Poe % $ buas be 13 Volume fraction of clay ‘Volume fraction of mineral; Vis eC. defined similarly Spontaneous potential Sonic interval transit time Standard deviation of total error ia i logging measurement ‘Multidimensional array of incespretation unknowns Incoherence function Residual in i tool response equation Density of formation Densiey of clay ‘Density of hydrocarbon Density of matrix Density of mud filtrate ‘Standard deviation error on logging Assymetric negative error on i" logging ‘measurement Assymetric positive error on i" logging Standard deviation error on a logging ‘measurement due ro one factor (see text) Standard deviation of error in a tool response equation or constraint Porosicy Porosity maximum limit Porosity from Neutron log4 Gronat, A New APPROACH To CoMPUTER-PROCESSED LOG INTERPRETATION REFERENCES Poupon, A., Clavier, C, Dumanoir, J, Gaymard, R., and Misk, A. — "Log Anaifsi of Sand-Shale Sequences — A Systematic Approach”, J. Pes. Tech., 22, No. 7, 867-881, SPE paper 2897. Poupoa, A., Hoyle, W. R, and Schmidt, A. W. — “Log Analysis ia Formations vith Complex Lithologies”, J Pes. Tech, 23, No. 8, 95-1005, SPE paper 2925, Fletcher, R. — "Generalized laverse Methods for the Best Least Squares Solution of Systems of Non-Linear Equa- sions", Computer Journal, 10, 392-399, 1968. Powell, M. J.D. — "A Hybrid Method of Non-Linear Equations” ia "Numerical Methods for Non-Linear Al aebraic Equations’, Editor, P. Rabinowitz, Gordon and Brent, London (1970). Schlumberger Log Interpretation Charts — 1978 edition. 6. Schlumberger Log Interpretation Principles, Volume 1 — 1972, Clavier, C, Coates, G, ard Dumanoir, J. — “The The oretcal and Experimental Basis for the ‘Dual Water Model for che Inerprettion of Shaly Sands", 52nd An- ‘nual Fall Meeting of Soc. Pet. Eng. of AIME, Denver, (ce. 9:12, Paper SPE 686). ‘Waxman, M. H. and Thomas, EC — “Hlectrical Con- lucivcies in Shaly Sands — The Relation Berween Hydrocarbon Saturation and Resistivity Index; TI. "The ‘Temperature Coefficient of Electrical Conductivity", J Pes. Tech, 26, No. 2, 213-225, SPE 9341