You are on page 1of 17
Crown-pillar risk assessment—planning aid for cost-effective mine closure remediation T. G. Carter and R. |. Miller ‘Synops Tn many countries mines are now required to file closure plans and post bonds for decommissioning. Establishing whether the remediation of specific sur- face hazards arising from mining activity is necessary in a given situation is not always straightforward. In consequence, particularly because of the high cost of closure requirements, considerable interest is being focused on improving methods of assessing the stability of surface crown pillars. Even with such improvements, however, itis not always a simple matter to determine the most appropriate and cost-effective remediation ‘method for a particular surface crown pillar. ‘Some simple probabilistic approaches to the charac~ terization of a crown-pillar rock mass are suggested (0 assist in determining the risk of crown-pillar failure by reference to previous experience. An outline is then given of methods for rationalizing the selection of optimum remediation solutions based on the use of risk-based decision matrices for the quantification and comparison of costs. Establishing whether remediation is necessary to ensure the long-term stability of surface crown pillars over sil active oF abandoned near-surface stopes in hard rock mines presents a particular challenge for mine closure planning. Decision- ‘making is frequently complicated simply by the difficulty of assessing the stability of the crowns as they are usually weathered and are also often covered by overburden, ‘An attempt to assign a single factor of safety forthe unique set of dimensions and properties of a particular crown pillar is generally not valid, However, some basis is required for ‘practical decision-making so that a rational selection of optimum remediation measures can be made and costed and options for closure evaluated. Appropriate stability analyses obviously form e Key initial component of the overall decision-making process. Legislative requirements for long- term security also constitute one of the more important governing enteria that must be considered in such decision- making.! In fact, the regulations themselves often impose such strie controls on the remediation measures that are acceptable for ‘walk-away’ closure approval that they limit the applicability of many otherwise workable solutions. ‘The present contribution therefore provides some ‘examples of ways in which the application of uncertainty- besed decision-making procedures can assist in the selection of cost-effective remediation measures for submission as part ‘of required closure planning documentation, Paper presented at ce conference Risk atessment in the extractive industries held at che University of Exeter on 22-24 March, 1994 revised: manusenpt received bY the Insiuson of Mining and ‘Metallurgy on 13 October, 1994, Paper published in Trans Ino Mix. Mac. (Sec. Ae Min. indus), 104, Janwary~Apel 1995, ‘© The Inston of Mining and Metallurgy 1995, Assessing crown stability Since in almost al situations where a crown pillar has been used in mining there are significant differences berween the overall rock mass characteristics of the ore zone and those of| the hanging-wall and footwal, it is clear that attempts to deserbe the stability of such crowns by means of a single- value factor of safery will present difficulties, Nevertheless, some assessment must be made of the overall likelihood of failure so that conclusions can be reached on whether or not 4 particular crown zone pases a long-term hazard, Available methods of analy Recent back-analysis and detailed assessment of the ch teristics of past failures of surface crovn pillars have enabled the development of a useful guideline approach for evalu- ating the influence on stability of variability in the rock ‘mass? Examination of documented crown-pillar failures in biocky rock masses suggests that failure most frequently develops were several adversely orientated discontinuities Intersect or where a particular suite of major joints provides 2 release mechanism for gravity collapse. Similarly, in faihures of significant areal extent the geometry is often controlled by the orientations of major individual discontinuities. Where crown pillars are characterized by unfavourable geometry, small block sizes and generally poor-quality rock conditions, failure tends to occur by ravelling and breakdown of the blocks that form the crown zone (Fig. 1(b)), whereas in good-quality rock where jointing is not adverse the inter- locking of blocks may create a stress arch that, by providing some measure of horizontal clamping, promotes stability Fe. 1a, In cases where ubiquitous fabrics are the cause of failure uncontrollable slabbing and sloughing have often been res- ponsible for inducing progressive or chimney-type crown failures, the later being particularly evident where very weak crown material is present between relatively strong foot- and hhanging-wall rocks.> For most other types of failure due to ubiquitous fabrics not only is the structural character of the crown important but so also is the weakness that such = fabric creates on the hanging-wall side of the stope. In fact, the presence of such fabrics can initiate a form of progressive failure that leads, first, to widening of the stope by de- lamination of material from the hanging-wall and ultimately ends in complete destabilization and collapse ofthe crown.* Approaches to assessment Current approaches to assessment of the stability of a surface crown pillar encompass a spectrum of techniques from empirical rule-of-thumb methods through to the application ‘of sophisticated numerical modelling using available ‘computer codes, such as UDEC, FLAC, UNWEDGE or PHASES. However, whenever sophisticated analytical pro- cedures are used the lack of a detailed understanding ef rock mass characteristics and variations imposes severe limitations fon the validity of the results obtained. When crown ‘conditions ere complex probabilistic evaluation techniques AM o Fig. 1 UDEC block-model geometry for (a) good and (fir sock conditions showing distinetelement prediction of stes-arch effect nd inition of hanging all slabbing fo diferent crown geometis have merit over modeling sophistication in that more explicit ‘expression can be piven to uncertainty in the assessment of| the heterogeneity of real rock masses. ‘Rock mass clasiication methods and quantitative, semi- empirical stabiliy-evaluation procedures. are particularly amenable to such probabilistic treatment es they essentially allow a descriptive characterization of all controlling para= meters, Further, as such methods are generally aleady built around some form of ranking assessment of each governing parameter, they provide @ convenient framework that is Suitable for staustical treatment. ‘The approach that is proposed here utilizes these classi- fication systems as 2 means of evaluating the likelihood ot probability that a crown pillar contains an area of poor- quality rock chat is suficiently large or sufficiently weak to result in collapse ofthe pilla. Accordingly, the approach has two themes: identification of the characteristics of an unstable crovsn-pillar rock mast; and assessment of the Statistical variation of rock mass properties within x given crovin to enable calculation of the likelihood that such = poor-quality zone is present Rock mass classification Although various classification methods are available thet have been specifically developed for mining applications— Aa some tailored to the design of deep stopes (the Mathews method"), others o caving design (the MRMS method”) and one to the analysis of surface crown pillas®!°—all rly for base input parameters on the descriptive codes that were first put forward for classifications for use in ranneling (ve. the O system! and/or the RMR system!2", ‘Although information on discontinuities is usually impli- ily included in the data that have been compiled for development of the classification indices, che influence of structure often cannot be adequately evaluated through the use of global classification indices alone. It ie therefore recommended that, before application of the approaches that fare set out in the present contribution, contoured stereoncts should be produced so that the primary joint trends ean be identified and quantified for further analysis and a check can bbe made for the presence of adverse joints—which may not always be members ofthe commonest set—as these may be a critical control on stability. As such joints are not represented by the mean, nor are all joint combinations represented by intersections of the mean planes, careful checking with the use of extremevalue statistics may often be appropriate ‘where weak joints are identified that could be eriical to the stability of a crown! imei dette et nos ate weeny 00) 3.) J o-Ps) eee ect pee a wero eco se tm na en an a than with the Q approach, it does tend co limit the sensitivity of the method when an attempt is made to classify subtle, but important, jint-controlled effects, Rather than being a limitation on their application, these ininsic differences between the classifications are a signi= ficant advantage in chat when both systems are applied and the deduced RMR and Q values are compared some confirmation of the reliability of the data from field obser- vations is obtained. Not only can the inherent vatiabiliry in the rock mass characteristics be stsested but a better appreciation of uncertainties inthe definitions of parameters can also be gained. Used together, therefore, in conjunction With other specific information on the orientation, persis- tence and surface conditions of discontinuities, such classi- fications provide a good basis for the «ype of probabilistic analysis that is the subject of subsequent sections of the present paper. Crown-pillar characterization In addition to the need to cope with the vagaries of the various classifications, considerable and real difficulties arise in attempting to define single classification values for typical ‘rown-pillar rock masses that are both representative and unique In principle, the same basic techniques of geostatistical analysis that are routinely applied to the evaluation of variability in ore grade can also be applied to assessment of| the variation in rock quality across a crown-pillar zone, for, just as ore grade varies across site, so tg0 does rock mass duality. In practice, however, variations in the latter sze generally much more difficult to quantify as an evaluation of| rock quality at a particular point requires 2 relatively large amount of information about the rock mass, whereas ore grade can usually be assessed by measuring a single parameter. There is seldom, if ever, sufficient relevant information to define completely the spatial variation of rock ‘quality across a given crown in a fully quantitative manner. Furthermore, most crown pillars are characterized by the presence of different zones of competent rock (usually the ‘margins to the ore) and weak, faulted ground (often the ore ‘zone). In recognition ofthis, the approach that is developed in the present contribution s based on weating the quality of| the weakest zone of rock in a particular crown pillar 35 a random variable and evaluating the likelihood (probability) that this zone ofthe crown is sufficiently poor or extensive to result in a collapse Variation in rock mass characteristics across the crown snd abutments is one aspect that must be carefully examined, but the dificulties of classification are further compounded when areas close to the surface are assessed Here the rock is not only likely to be variable asa result of| ‘weathering effects but is offen so completely covered by joverburden that it can only be crudely characterized trom borehole core. In these situations classification of the rock ‘mass is by no means straightforward. Estimating represen tative discontinuity characteristics is tricky and defining RQD is particularly difficult. In such circumstances there is merit in determining RQD from core (with specific cor- rections for core size effec!) and comparing the calculated values with estimates based on underground -measurements of fracture frequency and surface observations of scan-ines. Where core is not available for areas ofthe rock mass for which a classification is required representative ROD values for different sections of the crown can be estimated ffom maps of fracture frequency using the exponential decay relationship proposed by Priest and ® as correlated to other published trends (including the range of RQD data compiled by Deere!® and ‘hat compiled by Bieniawski (see Kendorski et af)) RQD = 100e**(72.+ 1) @ where Lis the fracture frequency (fractures/m) and t= 0.1 m defines the theoretical limst-cate, upper-bound RQD range, Values computed from this approach can also be checked using the method proposed by Palmstrom:?! ROD= 115-331, @ where J, is a measure of the number of joints erossing a 1-m? block of rock Although correct characterization of the weakest part of the rock mass inthe crown zone isthe key to appreciation of | the inherent strength of the pilla, aecurate information on the geomeuy of the underground stope excavation is also essential to a proper assessment of stability. Obtaining sufficiently accurate profile is sometimes no tival matter as all too often no reliable data ate available on the final ge0- ‘metry of the excavation, Furthermore, in many cases the sope crown and hanging-wall may have caved and ravelled to such an extent thatthe shape created by mining may have changed completely. Resolution of the considerable un- certainties that can sometimes arise in defining the geometry of a particular stope crown requites careful site inves: tigation—often with the use of faiky sophisticated techniques, such as those discussed by Cole et al? Carter ral and Steed and co-workers. Empirical assessment of crown stability Although various rule-of-thumb methods for the design of| surface crown pillars have been applied in mining practice for well over a century, recent research has documented ‘numerous failures that have occurred where the rules were simply inappropriate*% Attempts have therefore been ‘made to improve the existing rules by undertaking detailed checks of available data to establish rock mass characteristics and pre-ulure geomerry for as many failed and non-feiled surface crown pillars as possible” Early evaluation led to the development of an improved relationship of the form shown in Fig. 2 whereby the thick ress to span ratio that was employed in the old rule-of- ‘thumb approach, rather than being defined as single value, ‘was replaced by an expression related to rock mass quality: ssquse a where Tis thickness and $ i span. Initially, it was considered that this method of evaluation would provide simple, guideline relationship that would be suitable for checking crowns in geological serings similar to those for which assessments were held in the database. It was, however, ‘quickly realized that since the relationship was not scale independent, its use without calibration could very easly lead to significant errors. Accordingly, efforts were made 10 develop a relationship that would describe crown geometry ‘more accurately. Sealed crown span Some success was achieved by examining different beam- bending analogies by methods of dimensional analysis, This allowed the definition of a sealing expression that could be effectively applied to provide a unique characterization ofthe a3 Noi-a lo 19 © 0 400 000 ° i: t S | store span OPEN STOPE \ | Sone 2ove o rcuas ctses | o.0l T > ap a KO GEOMECHANICS RMR Fig. 2 Summary of erown-plar ease records plotted as thickness to span ratios versus rock mass quality of weakest zone within crown feomcuy, NGI-O, Norwegian Geotechnical Insitute tunseling ‘ual index, three-dimensional geometry ofa given surface crown pila!” In addition, it was subsequently found that ths ‘scaled crown span’ concept enables fuirly reliable comparisons to be made of the stability of different pillars that have been excavated in Giflerent rock masses of different overall quality. The approach was based on a simple sealing expression of the form Cy = 5x Ky where a geometric scaling factor, Ky is used co modify the actual span, S, t0 take into account differences in three-dimensional pillar geometry. The scaling telationship that was developed to consider all the critical dimensions of crown thickness and the dip and geometry of the rock forming the stope walls and surface crown pillar ig. 3) can be expressed as - x O° S FG e MH Oaeosd] @ where Ge is scaled erown span, m; S is actual crown-pilla span, mis specific gravity ofthe rock mass; Tis thickness of the crown pillar, m; 8 is dip of the orebodyloliation, degrees; and Sy i8 the span ratio SIL (span of crown pillar Aivided by its strike length). Derailed discussion of the fll development of this expression is beyond the scope of the present work, but i should be noted that in the overall expression the foliation dip angle, 8, reflects also the dip of the controlling hanging- \wall ofthe stope. Thus, for a tabular orebody (Fig. 3(@)), a8 the dip of the foliation and, hence, of the stope sidewalls shallows from 90 to less than 45, the effective span of the stope should no longer be considered as the width of the ore zone but rather as the dip length ofthe hanging-wall, Although by no means perfect, this simple scaling approach represents @ considerable’ improvement on the ‘iraight thickness to span ratio as a means of defining pillar stability a8 some account can be taken of each of the most {important variations in dree-dimensional geometry. With time and the accumulation of more case records the approach is likely to undergo further development and refinement, At this stage, however, it can be considered a useful index approach that, like the Mathews method for ddeep-stope design or the Q and RMR systems for tunnel evaluation, provides a ready means for quick assessment of the likely stability of a given pillar geometry. (For 8 fuller ata Fig. 3 Terminology fOr definion of crowm-pllar sealing relationship. (a) Schematic view of stope showing span, S; lena, L fand dip, 0. (5) Stope croseseotion showing S, 6 and crown, thickness, T description of the development and application of the approach the interested reader is referred 10 Carter! oto the original (1990) report by Golder Astocistes to CANMET.), “Taking the scaling expression given as equation 5 and recasting the database records plotted in Fig. 2 allow generation of the graph presented as Fig. 4. Again, all the case records in the Golder-CANMET database have been plotted against the rock quality of the weakest zone in the crown. As in Fig. 2, for convenience of reference the RMR and O seales have both been used for the plotting, their relative positioning being based on Bieniawski's 1976 cor relation, RMR =9l0g,0¢ 4.2 Although it must be recognized that for certain rock conditions significant divergences can occur in the cor- Felation between the two indices, this widely published relationship has been utilized as, for most rock masses, it adequately reflects the relationship between the two classfcatons. However, in cases where comparative classi- fications suggest wide divergences either these must be treated explicly or a specific site andlor rock mass calibration relationship must be developed. Critical span tine In addition to the plot of the database records, Fig. 4 includes three power-law relationships forthe assessment of | maximum spans in different rock conditions—two of which plot as linear traces and one as a curved envelope. Although ‘each vas originally formulated for the definition of span only, they provide # useful framework for checking scaled spans on the premise thatthe scaling coefficient, Ky, incorporates all appropriate three-dimensional factors to ensure that Cy is suitably sealed. However, examination of the slopes of the linear expressions shows that neither adequately reflects the trend of the boundary between failed and stable cases. The line proposed by Barton in 19762 to define the maximum. span of generally unsupported civil engineering openings the non-linear trend to increasing stability atthe very good- ‘quality end of the O-RMR scale as indicated in Barton’s original Q data!!#® and as suggested by some of the case records in the Golde-CANMET crown-pillar database It should also be appreciated that this curve, rather than defining a theoretical limit, represents a transition line between stable and failed cases as defined at a specific moment in time. Tt is, thus, a curve that, although it represents some low probability of filure, isolates most cases for which failure is likely. However, exceptions may well ‘ecur: some ofthe pillars for which records are held filed eee eo | ee [Ke neff] em SE ane “a Lento C= p= Fig, 4 Summary of crown pillar case rconds in Golder-CANMET database plore as scaled crown Spans versus rock mass quality using Q or RMR claseation methods. (References: Barton (1974)! Barton (1976) Carter (1989) Golder Associates (1990)? (critical span, S = 20°) tends to the conservative side for poor rockquality conditions, whereas the power-law ex- pression for average critical span proposed by Carter” to fit the mean trend 10 the Various mining engineering class fications (critical span, S = 4.49%?) tends, by contrast (mainly because it essentially addresses only short-term mining requirements), to underestimate the time-dependent influences on failure that are seen in some of the older case records at the poor-quality end of the scale (see also the discussion on stand-up time by Franklin and Palas), However, if the shape of the original empirical ‘unsupported span’ curve oulined by Barton and co-workers in 19743" is plotted together with a non-linear tall to encompass Barton’s various data points?42® at the very good-quality end of the Seale (these are not shown in Fig. 4), the resulting curve tends not only to separate the case records in the crown-pillar database better but also tracks other available data for good fand very good rock conditions more accurately, The following expression, termed the ‘critical span line’, has therefore been developed to match the shape of Barton and co-workers’ original curve:!! Se 30° x (sinh216Q) ® where Soy m, provides a megsure of the maximum scaled span fora given pillar beyond which failure may occur. It should be appreciated, however, that the hyperbolic sin term in equation 6 has been introduced simply to account for soon after excavation whereas others stood for a hundred ‘years before failure, and some currendy stable pillars could ‘et fail owing to unfavourably located joints or other factors. ‘Nevertheless, in its present form Fig. 4 provides a basic means of rapid crown-pillar stability assessment by allowing the scaling of different crown geometries for diferent rock ‘mass conditions. Further, because it is based on a fairly fextensive database of more than 200 surface crown-pillar ccse records, it does provide a ready basis for sensitivicy ‘evaluation when used in conjunction with more sophisticated techniques of numerical analysis. Application to existing crowns Although Fig. 4 can be used for basic deterministic analysis (oy simply calculating the scaled span of a given pillar and then, with the controlling rock mass quality defined, plotting its postion on the chart to determine its stability state), the raph can also form the basis for more extensive probabilistic assessment of variations in stability Ie is important to realize, however, chat when the origina database records were plotted crowns were only known either 10 be standing or to have failed. Actual factors of safety for both ‘stable’ and failed crowns were not known with any certainty at the time, although now it would be possible to assexs how close to marginal some of the crowns were from the final plotted position for each case. This same limitation ‘exists when the chart is used deterministically for crown design. If, for example, the geometry of a particular crown Ass | Table t ‘Moss? Priest and Brown, Barton, Pine and Cole*!) (Comparative significance of crovn pillar failure (modified from McCracken and Jones)? with data from Kirsten and Class Probability Minimum Reliability Design guidelines fr pillar acceptability offailure, factor of indices ‘Serviceable ite ofcrown pillar Public. Regulatory Operating % safety access positionon surveillance 8 ESR Expectancy ‘Years Elosure required A sono SOLOS“ Bifevely zero <05 Forbidden Totally Ieletve ‘unacceptable ee eee Continsous (cemporay mining prevented acceptable sophinicated purposes eal ronitorng ‘unacceptable isk of aur fr cemporary cv ‘tunnel poral) © 10-20 1206 Veryshortterm (quasi: 245 Actively igh velo Consinaous semporary stope crowns: prevented concer monitoring Undesirable risk of fire wie for temporary civil works) instruments Dy oan 1S 1214 Short-term Gem 5-10 Prevented Moderate vel Continuous temporary crowns, eg fconcern simple ‘under non-ensive mine smnitoing infrastructure) E155 18 11.3 Mediumeretm (emi «15-20 Discouraged Low to Conscious permanent crowns, moderate level superficial possibly under structures) ofconceen monitoring Fo osas 2 eee 50-100 Allowed Oftimited —_tacidemal Dermanent exons, cil once superficial portals, nearsueace monitoring Sewer tele) G <5 m2 308 Very longterm 3100. Free Ofnoconcem No (permanent crowns over monitoring el tinsel) required ESR, Excavation suppor rao point plots in the potential eaving zone—defined as the area above the eritical-span line~the likelihood of failure is high unless the crown is sufficiently supported or fill isto be used in mining, Establishing the exact factor of safety of the designed crown, however, is not a simple matte. Probabilistic assessment of crown stability Although the empirical croven-pillar stability relationships outlined above were essentially formulated on the basis of deterministic evaluation of the database records, the approach lends itself to use as a means to quantify probabilsticaly how likely a given pillar is to fail, This can, readily be accomplished for a single crown geometry with reference 10 defined variations in rock mass quality or, where pillar geometry also varies (for example, along the strike of long stope section), it can be done for both geometry and rock mass quality. In fact, there is considerable technical ‘mevit in using a sensitivitybased approach that permits assessment ofthe full range of likely rock qualities. Such an approach not only allows better understanding to be gained during evaluation of the sablity state ofa given pillar but it also takes into account more effectively the variations in rock mass characteristics that occur in real situations. Aside from the technical perspective, the use of probabi- liste approaches for assessing the stability of crown pillars hhas some other significant advantages, particularly where a6 such pillars underlie areas accessible to the public, Better Public awareness and understanding of the risk of failure are ‘usually achieved when the results of critical analyses are ‘expressed in terms of the probability of a factor of safety smaller than 1, or the probability of failure (Pz). Thi, in fact hhas been the thrust of much recent work connected with ‘mine hazard assessment studies. 102227.9031 Jt has alo beet atthe centre of thinking directed to the quantification of risk snd uncertainty (See Table 1, as modified and extended from tabulations compiled principally by McCracken and Jones,™? Kirsten and Moss, Priest and Brown, Barton, Pine?® and Cole), Table 2 Average geometrical and rock mass quality conditions for crown zone sbove first-level stope east of highway in Fig. 5 (Grown span, S,m 145 ‘Thickness, 185 Sure length, m 50 Foliaton ip, degrees 70 Rock quality Mean RQD, 2 Mean Q 93 ‘Mean RMR. 6 Sad. (©), RMR units " R INFLUENCE ZONE Posse Ee? ‘ or f Nee | ON Sr a STOPE, / see B\ ouTLNe oe Abst LEVEL era XR | pI STOPE ey hee) | Hat even AS, Fig. 5 Brample geometry of two areas of near-surface stoping for which vsk assessment and emelton solutions are required to sbepuand hiphs Graphical assessment Let us suppose that we wish to analyse the stability of the crown pillar over the eastern stope shown in Fig 5 as there is some concem that its failure could lead to collapse of the highway. Fig. 6 shows the basic approach that can be used for graphical probability assessment in this situation and illustrates how readily an estimate of the current stability state can be derived for any given crown, The average ‘geometrical and rock mass quality conditions for the crown zone considered as an example are given in Table 2. By substituting into the scaling relationship and assuming. that the specifi gravity, of the rock mass is 2.7 the equi= valent scaled crown span, Cy, can be calculated as follows: 2a veal (.(i83 PE] 5-145. 52m ® ‘This value can now be plotted against rock mass quality on the chart shown in Fig. 4 using the Q scale along the top axis or the RMR scale along the bottom axis, To enable @ rapid determination of whether the crown is generally sable or is liable to cave the mean values for Qand RMR can be plotted either with appropriate ranges oF as single-vale estimates. In this ease the crown plots in the stable section ofthe graph— bat how stable isi? This ean be faisly readily assessed ifnow, rather than plot a rock-qualty range or mean value only, a probability density function is plotted to reflect the likely variability in rock mass quality within the crown. For this level of preliminary assessment with the simple graphical approach # normal distribution curve has been used can be assumed that rock mass quality values will be normally yor exos-sectonal geometry so Fig 9) distributed provided that the coefficient of variation (CoV) ‘remains reasonably small Tn assessing the spread of such a distribution it is Important, however, to take care not to mix the charac teristics of diferent parts of the crown rock mass by taking the simple option of choosing a distribution with a large standard deviation. In most crowns one part—usselly the one zone of the immediate margins of the ore zone—is less competent than the remainder ofthe pillar. ris important in such situations that the distribution selected match as best as possible the spread of actual conditions. In this regard it should be noted that the weakest area is not always che ore zone. Because the majority of ores originate by injection of| ‘mineralized fluids into a fault zone, sometimes, when such injection is predominantly siliceous, the ore zone—being (quarz-rich—ean be much more competent than either the hhanging-wall or footwall rock masses, In such cases the distribution should be chosen to reflect the characteristics of| the controlling (weakest) zone of the rock mass. For the ‘example shown in Fig. 5 the characteristics of the footwall ideration. In this case, in inary stability assessment the mean Q of 9.3 and the mean and standard deviation in RMR units of 63, and 11 for the foorwall rocks would be utilized as these ‘Table 3 Summary rock mass characteristics for highway stope example @ RMR (o) Hanging-vall 25 mas (aie volcanics) Ore200e 8 Bo (enetconglomerste) Foot 93 ean (lary meragreywacke) Aq RITICAL QUALITY o=29 ance i CAVING CONDITIONS: UR RANGE 21570, CRTICAL SPAN LN ESSENTIALLY STABLE GRAPHICAL STABILITY ASSESSMENT ROCK QUALITY ASSESSMENT RMRaens | 69 Puce | 6a irrenence | Standard Deviation, 2 (AMA unis) | 11 Prosasiuty oF FaLure | 20% (hated on offeet,o duiions wom Mean im AMA-untel | (0.82) p= 330" x ah (OH (neve WITH RESPECT TO RMR s DeViaTION 1a, (Gow ia epan cow plese rg) whore = Stns Onan Rock Quy AA ts Fig. 6 Example application of sealed crown-span concept t evaluate probability of alte for surface crown pla of known geometry would reflect conditions in the weakest part of the crown appropriate ‘At this stage it should be appreciated that, when per- forming probabilistic assessments of rock mass quality i is ‘much more convenient to express all varabiities in RMR. units than to use the Q scale. This is because the later is a logarithmic index, which makes ie dificult to refer to con- sistent coefficients of variation or standard deviations except in logQ terms. Thus, irrespective of the classifcation approach that has been used to assess the rock quality, for the purposes of probabilistic evaluation it becomes eficient to adopt the linear RMR scale simply because it aids ‘comprehension in ploting and analysis. Whichever classification is the preferred choice, the approach that should always be followed in identifying dis- tributions for the parameters used in a probabilistic analysis is to try to reflect closely the best possible understanding of the geology of the erown zone. To achieve this the rock mass forming the crown and sidewalls of the stope should be Aas examined with the utmost care, For the example shown in Fig. 5, where itis considered that the footwall rocks, being the weakest and most intensely sheared, will dominate crown, ‘behaviour, a normal probability distribution range at least one standard deviation either sie of the mean (Fig. ) would bbe selected as representative, A further conservative sssumption is then made tha failure will occur whenever the RMR js locally less than the cntical minimum rock mass ualityrating required for stability. (The assumption is con- servative because, a5 shown in Fig. 4, limited number of ‘crowns were stable even when the RMR was less than the tical value—i.e, above the bold curve.) This «ritical ‘minimum rock quality can readily be approximated for the crown considered in the example, for which the equivalent scaled span is 5.2 m, The value can be read off the chart or calculated for the linear portion of the erticel-span curve for values of Q <50 or RMR <80 using either of the following fegression relationships (which are bated on the straight-line Portion ofthe curve) Que = 0.06252 ® RMR, eae = 2108S +19 ® For the example the critical rock mass quality, as read off the graph or calculared from the above linear expressions, would be RMR a= 54 (0F Ong = 2.9) a0) As shown by the calculations in the top right-hand comer of ig. 6, this critical quality value falls nine RMR units (Le. 0.82 standard deviations) below the mean rock mass rating ‘of 63; thus, from the table inthe lower right-hand comer of | the figure the probability of failure can be estimated as about 20%. This suggests that, although the crown is essentially sae, there is about one chance in five chat some combination fof the poorer parts of the rock mass will preva right across fone section of the crown such that a failure could occur. @RISK approach Although the simple graphical approach described above is ‘quite adequate for rapid assessments of crown stability (for instance, at a preliminary level when it is necessary 10 develop hazard rankings), ii also possible co apply the same basic procedures somewhat more rigorously by using quan titative probabilistic methods. Assessment can then be carried out not only of variations in rock quality in terms of| RMR or O but also of variations within the geometry of the ‘crown. This type of more rigorous evaluation can stil be fairy readily accomplished by frst deseribing the ‘of all Key parameters in terms of appropriate probability distributions and then using a statistical analysis package, such as the @RISK" or Crystal Ball add-ins for Lotus 1-2-3, and Microsoft Excel, to evaluate probabilities of failure. Essentially, the approach requires that rock mass quality variation and geometric characteristics be randomly chosen from within defined probability density functions that have ‘been selected on the basis oftheir appropriateness for each of| the key parameters. Distributions must be defined for each of the controlling dimensions of the crovin geometry end for ‘each of the basic parameter descriptor codes that aze used in formulation of the rock-quslty indices for both the Q and RMR values For stochastic analysis RQD and rock strengths can typically be described by lognormal or truncated normal distributions, such truncations being implemented toy nt the random selection of meaningless values (for exauple, negative RQD or strength). The remaining parameters that are used to calculate the rack mass classifications can then be described using discrete values from the standard ranges siven in the and RMR classification tables. This allows, for ‘each specific parameter, the shape of the distribution to be ‘matched to observed vatiations in the rock mass, Examples fae provided in the form of the histogram plots on the right side of the @RISK spreadsheet illustrated in Fig. 7. In these plots discrete function variables have been used for each of | the Jy, Jge Condition, etc. terms to match the division codes set by Barton’ or by Bieniawskil for each classification des- criptor group. Each description used in the @RISK spread sheet has therefore been assigned a value, x, thatis based on the corresponding rock mass classification weighting and a "Palisade Corporation. @RISK Risk analysis and simulation, Addin to Locus 1-2-3, version 1.35. Ue guide, 1000. ‘weight or likelinood, p, that specifies the probability of occur- rence of the outcome on the basis of field observation and ata collection information. One of the significant advantages of discrete functions of the type that is used ro describe some of the parameters in Fig. 7 is that they allow geomechanics variables that are not normally distributed, such as joint spacing and persistence, to be readily defined. Most of the other common variables ‘that are used in the calculations have approximately normal distributions and, although most natural variables tend t0 ‘exhibit coefficients of variation (CoV = fy) that are somewhat more constant than their standard devistions,?™>* it is generally accepted practice simply to use tabulated ‘means, y, and standard deviations, 6, co represent variations in the parameters, as has been done for the Q and RMR indices obtained in Fig. 7. ‘Further, in carrying out probabilistic evaluation of Q and, particularly, of RMR values attention must be given to the problem of cross-correlation, Of particular importence in calculating RMR values is the significant cross-correlation bberween spacing and RQD. Although some correlation also can be considered to exist berween J, the number of joint sets, and RQD, detailed checks of this interrelationship and ‘other, possibly important, correlations suggest that these weaker dependencies are of little significance to the overall, calculations. Thus, as only the correlation between spacing and RQD in the RMR classification is significant, this relationship has been specifically dealt with in formulating the @RISK spreadsheet illustrated in Fig. 7, This has been achieved by making the sampling of the spacing distribution conditional on the sampled RQD, where the latter is represented by 2 truncated lognormal distribution with the ‘upperlimit correlation between RQD and spacing specified to conform with equation 2c. to be consistent with the form of the correlation relationship suggested by Priest and Hudson."*1° Although the more obscure and less significant correlation relationships (eg. berween roughness and surface characteristics) have not been specifically treated, both G@RISK and Crystal Ball are capable of incorporating any specified correlations into their random sampling schemes without affecting the individual probability distributions of the input parameters. ‘Once such a spreadsheet has been created its use in practice is quite straightforward. Fist, for each iteration of the @RISK spreadsheet shown in Fig. 7 a pair of Q and RMR values is calculated independently. Next, two specifi estimates of critical span, Sc, that are consistent with these RMR and Q values are computed from equation 6, The wo Sc values are then compared on a cate by cate basis with ‘stochastically assessed values of scaled crown span, Cg that have been independently derived from the probabilistic descriptions of the geometry of the crown zone using ‘equation 5 (See tabulation section at lower left of Fig. 7) If either of the crtical-span values from the Q or RMR calculations is found to be less than the scaled erown-span vlue representing. che actual crown geometry for that iteration step, then for that crown-pillar geometry and rock ‘mass quality the pillar plots within the caving area of Fig. 4 and, thus, a case of ‘allure’ is registered. With at least 500 iterations to ensure an adequate representation of the input dliscibutions two separate estimates of the most likely probability of failure are derived, one bated on the stochastic spread of computed Q values and one based on the RMR. calculations, These values can then simply be averaged 10 provide a best-estimate failure probability forthe pillar under consideration, In this case, as shown in Fig.7, an approximately 18% probability of failure is suggested, which aio Dear (900 Fatnge| UP ete Novotbinsetilashe Cesteamon =) 8S Bette heinoed| S105. I Spang | 30 | csc noc| ne ee tee tom toca | RelatneLitalnoed | 23% 47% 78% e34N O58 ZEON | | | jones ge noel rr ; [SN ETON Oy TSN (SLASSIFIGATION [oRix CORE GURL ap) = VeyPoot Poot Far Good Baer —| J a ‘OTHER CATER ‘Cosma vals 2 ‘spetedLbibood ™ Teetegon RGD) ages = Sai a "BNTROUGHNESS [4 CONDITON lek Smooth Acuph Smoot Rough Noi SufaceCharacrens| Piarar Planar Pura! Wary Way Cont Cetin o] 08 dS Sg Vas mwa] 86'S 8S A FeitveUiminecd| 10% Oh 2H SRI e avenges =x conan Te ae Bh "SORTALTERATON (ay INIAGT STRENGTH bi ining Head Stoned Sand Sk Ciay-SUt, SR Saling) ott Lisihosd| 15% 40% _s0%_s_o& om on Tris! Seength| US >So tb-z Seo “Bne) Wo-35 ,..), thea, by Bayes’ theorem, the updated probability of crown collapse due to the presence of a fault zone would be 65%: 0.18 x0.85 0.2350 = 06511 as) here the updated probability of failure due to a fault zon (current probability of failure due to poor rock—eg. fault zone) x (probability of intersecting poor-quality rock if fat exists) + (Current probability of encountering poor-quality rock). I however, the core drilling proved that rock com- petence was high throughout the pillar, the updated proba- bility oF failure due toa faule would be only about 2%: 018x010 0235 an 0.7650 where the updated probability of failure due 10 a fault (current probability of future due to poor rock—e.g. a faut zone) x (prababilty of not intersecting poor-quality rock as no fault zone exists) « (current probability of not imersecting ‘poor-quality rock), Ifthe mine did dlecide to undertake the core driling and it showed that the rock quality was indeed poor (a 24% chance), what would be the most cost-effective decision? Using the updated probability of failure for collapse of the crown pillar, dhe expected costs for the two options that are being considered would be calculated as follows, Fencing Fencing costs $100 000 to install and has @ (1 ~ 0.6511 = 0.3489) chance of success with @ 25% of 65.11% chance that a collapse could break through and affect the highway, thereby incurring additional expenditure of '$8 000 000. Thus, the total weighted cost is $100 000 + (0.25 x 0.6511 > $8 000 000) = $1 402 200. Backfill Backilling costs $1 000 000 and provides 100% security, giving arora weighted cost of $1 000 000, Obviously, with poor rock conditions proved by the drilling, ‘backfling becomes the more cost-effective approach senso POF ITEROALLNG swat POF Fae T vEsTCaTON PROBABLTY OF FLRE 1,1 ace ouALiTy Arevsen a maak Fig 11 Conceprus) changes in probability distbusion with ‘improved undemanding of rock “quality as consequence of addiuona investigation. PDF, probably density function If, on the other hand, as suggested by the original proba- Dility distribution (shown diagrammatically in Figs. 6 and 11), the drilling did not find poor rock within the pillar (a 76.5% likelinood—P,.4) the updated decision analysis would suggest the following. Fencing Cost, $100 000, with (1 ~ 0.0235 = 98%) chance fof success and only 25% of 2.35% chance of incurring Additional expenditure of $8 000 000. Thus, total weighted ‘ost = $109 000 + (0,25 x 0.0235 x $8 000 000) = $147 200. Backfill Cost, $1 000 000; security, 100%; giving total weighted cost of $1 000 000. Clearly, proof that good rock existed in the crown pillar would considerably reduce the weighted risk-based cost. The rine would therefore be even more justified in opting simply to fence off the area ofthe hazard. Is the driling information worth che expendinure? The previous analysis showed that ifthe drilling were not cazried fut, the optimal decision would be to fence the site at an expected cost of $460 000. If drilling is undertaken, the optimal decision depends on the outcome: ifthe rock quality is poor, the best choice isto backfill if tis good, a fence i the most sensible option, The overall expected cost if the Grilling were to be carried out would be: weighted cost of| Alling = (cost of investigation) ++ (weighted cost of poor ground) + (weighted cost of good ground)—i. '$50 000 + (0.2350 x $1 000 000) + (0.7650 x $147 200) = $397 608 ‘Thus, the drilling program would be a good investment as it ‘would reduce the overall expected cost from $460 000 to $397 608. Assessment of uncertainties ‘The approach presented above, while straightforward in concept, makes assumptions of representativencss that the [parameters truly reflect actual conditions. However, as is ‘evident from the example, a large degree of uncertainty results simply ffom lack of detailed knowledge of crown ‘conditions. Only by gaining improved information from detailed field investigation is it possible to reduce this type fof uncertainty. In such critical situations enough data ‘must be collected that the uncertainty is reduced 10 that arising from the true variability in rock mass conditions Gig. 12), wn SITE INVESTIGATION EXPENDITURE wweemTany Fig. 12. Relationship between uncerninses due to natural vatabilty and data inadequacies as function of data accuracy and cost of site investigations. J, Uncertainty due wo ignorance and eos and insecraces in data; 2, uneerainy due to wue variation in rock sass and geomewe factors ‘As can be seen, formal techniques of decision analysis are ‘merely aid to the making of explicic decisions on the basis of 4 careful analysis of the problem at hand.*! In principle, all ASS such decisions are simple: the real difficulties lie in ensuring that the initial database is valid and in developing a full ‘understanding ofthe ramifications ofthe decision. Once that understanding is reached and the reliability of the database has been verified the decision may be readily apparent. In fac, the general techniques illustrated here merely formalize jn mathematical terms the judgments that go into making any reasonable decision. Bayes" theorem and the principles of decision analysis are the mathematical expressions of ‘common sense and judgment. Decision analysis can be of great value in reducing apparently very complex decisions to more manageable parts and in providing a more balanced perspective on. the advantages and disadvantages of @ decision. However, it rust be appreciated that itis easy to be misled by a decision analysis that embodies false premises, In considering stopes, like those in the example, that were directly adjacent to hhighway one would want to think Jong and hard before committing expenditure to a fencing-only option unless driling had been undertaken that had proved good-quality rock in the pillar area. Where decisions affect lives other prions always need to be explored. In such cases not only should the cos alternatives be verified but 0, 190, should the jeomechanical factors thet form the basis for the decision- ‘making. Any of these issues could, if overlooked, invalidate the analysis—with catastrophic consequences, as actually ‘occurred in the case example, where one particularly weak zone controlled che overall behaviour characteristics, Another complication for effective decision-making— although usually of a more minor narure—is inaccuracy in the estimating. Although the associated uncertainties can usually be ignored in deriving overall cost estimates for pre- liminary decision-making, it should be appreciated that at times wrong decisions can be made as a resule of such un- certainties or because of wide error margins in the costings ‘Typically for the level of costing required for the ‘ype of decision analysis that has been discussed i is rare that there is a nced 0 extend the accuracy beyond @ Class C estimate (ie. 425% accuracy). Costs can therefore usually be based fon previous practice and experience. This issue, however, ‘becomes significant when lite cost differential separates the various alternatives. In these cases itis worth undertaking ‘more detailed estimating since costs developed from base- case information (derived from applicable labour and ‘material rates) usually allow the error band of uncertainty to be reduced significantly, as indicated in Table 4. Table 4 Usual levels of acceptable accuracy for cost. estimating purposes Class Accuracy, % Estimate "Binal construction tender Deign-evel engineering Feasibility or prefeastbity level ome Nevertheless, even with these provisos, for crown-pillar work all estimating of costs should be considered very tentative as actual remediation costs can vary dramatically ‘owing to the unforeseen cor “cations that can atise when the work is in progress.2® In ny eases where itis evident that remediation will be difficult it is wise to consider undertaking detailed contingency costing as well as base-case costing when preparing final decsion-matsix tabulations, 56 Conclusions Establishing the stability and deciding whether remediation is required to ensure the long-term integrity of surface crown, pillars over active or abandoned near-surface mine workings ccan be a complex task. However, the application of risk: bbased decision-making procedures can assist significantly in the rational costing and selection of remediation alternatives s0 that a beer evaluation can be made of options for

You might also like