classification as a tool in rock engineering

Tunnelling and Underground Space Technology 18(2003) 331–345

0886-7798/03/$ - see front matter� 2002 Elsevier Science Ltd. All rights reserved.doi:10.1016/S0886-7798(02)00106-2

Classification as a tool in rock engineering

Hakan Stille *, Arild Palmstroma, b¨˚

Department of Soil and Rock Mechanics, KTH, Brinellvagen 34, S-100 44 Stockholm, Swedena ¨Norconsult as, Vestfjordgaten 4, N-1338 Sandvika, Norwayb

Received 27 September 2002; received in revised form 19 December 2002; accepted 19 December 2002

Abstract

The role of classification in rock engineering and design is discussed. It is important to distinguish between characterization,classification and empirical design method. The classification systems used today should, strictly speaking, either be described asrock mass characterization systems or empirical design methods, as long as the outcome is not organized into classes. The mainrequirements for a true classification system capable of solving rock-engineering problems are as follows.(1) The reliability ofthe classes to assess the given rock engineering problem must be estimated.(2) The classes must be exhaustive(every objectbelongs to a class) and mutually exclusive,(no object belongs to more than one class). (3) The principles of division(rules)governing assignment into the classes must be based on suitable indicators(ground parameters, etc.) and must include thepossibility of being updated during construction using the experience gained.(4) These rules must also be so flexible thatadditional indicators can be incorporated.(5) The uncertainties, or the quality, of the indicators must be established so that theprobability of mis-classification can be estimated.(6) The useful system should be practical and robust, and give an economicand safe design. In the author’s opinion, none of the main classification systems in use today fulfills these requirements. Theymay, however, serve as supervised systems as a basis in the development of local systems adapted to the actual site conditions.� 2002 Elsevier Science Ltd. All rights reserved.

Keywords: Classification system; Characterization; Rock engineering

1. Introduction

‘Most classification systems are continuously misused becausethe premises for and assumptions made in developing them havenot been carefully studied by users, and because they have beengiven a validity for ‘quantification’ of rock mass behaviour that isfar more general than was intended by their authors.’ Brekke andHoward(1972)

As is evident from the quotation above, classificationsystems are used and misused in numerous connections.This was a main topic in the GeoEng2000 workshop inMelbourne on ‘Reliability of classification systems’,from which the following was concluded:

– The concept of rock mass classification is unclear. Itis an ongoing debate regarding the application of rockmass classification as a design tool. New data areneeded from the field(case histories), linking char-acterization with design(classification).

*Corresponding author. Tel.:q468-790-7912; fax:q468-790-7928.

E-mail address: [email protected](H. Stille).

– Practitioners need to be aware of the limits of thevarious classification databases and the input sensitiv-ity of whichever rock mass classification system theyuse.

– It is important to separate characterization fromclassification.

In this paper, we want to discuss the role of classifi-cation in rock engineering and design. Our aim is tooutline useful applications of classification for designpurposes, and to pinpoint some misuses of both termi-nology and application in today’s design works. Firstly,we will theoretically discuss the requirements of trueclassification systems, and then the rock engineeringprocedures required by their use. Finally, we will exam-ine some of the existing classifications systems in orderto evaluate how they fulfil these requirements and theirpractical applicability.In a subsequent paper, we will discuss rock mechanics

and geological aspects of the art of rock engineeringclassification, and related requirements.

332 H. Stille, A. Palmstrom / Tunnelling and Underground Space Technology 18 (2003) 331–345¨

Fig. 1. Observation, measurement and characterization applied in rock engineering(from GeoEng2000 workshop on classification systems).

2. Definitions

Firstly, it is important to define the terms ‘classifica-tion’, ‘characterization‘ and ‘classification system’.The term classification can be used in various ways.

This has led to confusion when the rules and roles ofclassification are discussed. The word ‘classification’comes from Latin, from the root words ‘classficatio’,´which means ‘class’, and ‘facio’ which means ‘to do’.´Thus, classification is the result of putting objects intodifferent classes. The purpose of such classification isto get a better overview of a phenomenon or set of data,to try to gain an improved understanding of them.By contrast, characterization is the procedure of

describing the condition of, for example, a substance ormaterial, and defining or giving value to the variousfeatures it displays. In practical rock engineering, thetask is to:

1. Identify the features or parameters of importance orrelevance to a project and the assessments to beperformed.

2. Measure andyor describe the properties of theseparameters, giving them values or ratings accordingto their structure, composition and properties.

Thus, the process of rock mass characterization con-sists of describing and quantifying the parameters thatgovern or influence the rock mass behaviour. These canbe expressed as intact rock characteristics, discontinuity( joint) characteristics, and the density and pattern ofdiscontinuities, as shown in Fig. 1. The characterizationcan be simplified by putting the different properties intoclasses—in other words, by classifying them. There aremany examples of this process in engineering geology,

such as describing the strength of the rock material orthe joint spacing and density, and placing them into pre-defined and general accepted categories.Generally, the result of the characterization process

will be used to assess the rock mass quality, accordingto some pre-defined system. This procedure is normallygiven the name rock mass classification. It is importantto point out that, strictly speaking, this is truly a processof classification only if the outcome is a recognizableclass description, such as ‘poor rock’ for example, andnot if the procedure leads to a single number or value,which has been evaluated from a rating system.In projects involving rock construction, a particular

group of empirical design tools(based on adoptingexperience gained previously in circumstances that canbe characterized as similar), are known as classificationsystems. Strictly speaking, it would be better if suchtools always were identified as part of an overall groupof empirical methods used in rock engineering design.Frequently, such empirical design is used in conjunctionwith engineering assessment and other design approach-es (see Fig. 2). Rock mass classifications form thebackbone of the empirical design methods. In fact, onmany projects, the classification system serves as a mainpractical basis for the design of complex undergroundstructures. Most of the tunnels constructed at presentmake use of some classification system. The ratings orvalues of the input parameters(indicator) are oftenfound from a sort of characterization of the differentrelevant rock mass properties.The much-used term ‘classification system’ is correct

and justified only if the design tool divides the assess-ment into certain classes or categories. In general, theterm ‘rock engineering classification’ or more correctly

333H. Stille, A. Palmstrom / Tunnelling and Underground Space Technology 18 (2003) 331–345¨

Fig. 2. The main principles of the design process for undergroundconstructions in rock including the use of classification systems as anempirical design method.

‘rock engineering classification system’ is recommendedonly for the practical use of classification to solvevarious rock engineering aspects.

3. General aspects of classification

3.1. Basic types of classification

Bieniawski (1989) defined classification as ‘thearrangement of objects into groups on the basis of theirrelationship’. The role of classification is generally toget a better overview of a phenomenon or set of data inorder to understand them or to take different actionsconcerning them. It is possible to distinguish betweentwo main types of classification, as defined by Hand(1997):

Unsupervised classification refers to the process ofdefining classes of objects. This is sometimes calledcluster analysis. That is, we are presented with a collec-tion of objects and the aim is to formulate a classstructure. In an unsupervised classification, we have todecide how many classes to use, and to link the objects

in the collection to the appropriate classes. The devel-opment of many of the classification systems used todayis an example of this type of cluster analysis.In a supervised classification, the class structure isknown a priori and the principles(rules) of division areformulated, allowing one to allocate objects to theirappropriate classes. This is sometimes called supervisedpattern recognition. Examples from existing classifica-tions include the ISRM classification of rock strength,or the geotechnical classification of soils.

Characterisation, on the other hand, is not a prioriclassification, since it is just an exercise to describe anobject in terms of words like color, strength, grain sizedistribution, and so on. Characterization may be asupervised classification—if the aim is to place theobject in predefined classes or groups.

3.2. Requirements of and usefulness of a classification

The logical requirement of a classification is that itshould be both exhaustive and mutually exclusive. Thismeans that every object in the area of interest has tobelong to a class and no object can belong to more thanone class.The usefulness of a classification depends on the

principle of division (rule) applied in separating theobjects into different classes. To be able to talk about acommon principle of division, the criteria for the deci-sion of putting an object in a certain class should havesome relation to each other. Thus, the objects in acertain class may have some similar property of interest.The benefit of the classification is also related to the

purpose it is meant to serve. Here, two main objectivesof the classification can be distinguished:

1. Communication between different users will be easierif the classification system is well defined and relatedto common understanding and language. From thename given to a sample of rock, a geologist willusually understand how the rock has been formedand whether it has been subjected to different geolog-ical processes. However, the name will not give himdefinitive information on the mechanical propertiesof the rock. The information is only qualitative, andthe aim is to facilitate communication between theparties involved, to get a mutual understanding of theobject.

2. Decision-making. The central problem here is notonly to allocate an object into pre-defined classes,but also to obtain information on the quality of theclassification and the possibility that it may havebeen mis-classified. This is perhaps the most interest-ing use of classification in rock engineering. Thespecial conditions related to most rock engineeringproblems imply that the supervised classification willbe the most suitable.


Fig. 3. The process of engineering classification for design.

3.3. Linking indicatorsyobjects to design or constructionclasses

The archetypal supervised classification problem maybe described as follows: Each object is described interms of a vector of features or indicators. Takentogether, the indicators span a multivariate space termedthe indicator space. For a particular object, the vectorof measurement corresponds to a particular point in theindicator space(see Fig. 3). Often, the vector cancontain numerical values, but it is possible also toinclude general features. The aim is to decide to whichclass the object and its measurement vector should beassigned. In order to be able to do this, rules(principlesof division) must have been constructed in advance,which can be used to predict the classes of new objectsbased solely on their measurement vector.An example of this process is as follows: In a

fractured rock mass, observation of the joint sets andtheir individual properties will be adequate for describ-ing the vector of measurements. To define supportclasses from these data, the rule may be to calculate theindicator(Q- or RMR-value) based on the measurementvector.In more complex conditions, like squeezing ground,

a combination of measurements like rock mass charac-terization, description of the minerals, or deformationmeasurements should be used to define the supportclasses or support actions to be taken. In this case, theprinciple of division may not be uniquely definable. Itwill normally be build up by ‘and’, ‘or’ qualificationsand ‘if’ statements. It is obvious that, in such a case, asingle numerical value cannot fully describe the complexsituation.Since the principles of division may be based on one

of the existing empirical methods normally called clas-sification systems,(e.g. the Q or RMR systems), therequirements regarding classification will be those putforward by Einstein et al.(1979) for empirical methods.

Ideally, the principles of division(rules) should beselected to match the actual problem. However, this isnot the case in many situations, which can give differentproblems:

● One problem with classification is that, typically, thetrue class of the objects is unknown. This problemarises when it is not possible to describe or measurethe objects with exactly the same features or indica-tors as were used to define the true classes. Thissituation is very common in rock mechanics. Forexample, rock support can be decided by numericalcalculations based on input data like the strength anddeformation properties of the rock mass, but suchproperties cannot be observed directly. Instead, otherindicators such as general geological observations ofthe rock mass quality can be used. This will introduceadditional uncertainties and therefore a risk for mis-classification. The measurement vector may indicatea certain class in circumstances where the true classshould be another. Once the object has been mis-classified, any decision based on the classificationwill imply a probability of unexpected behaviour.One strategy to avoid this problem may be to collectmore information. However, in this case, the rule(principle of division) must be constructed in suchway that it allows the user of the classification systemto wait until more information has been collectedbefore the decision is taken. This makes it clear thatthe possibility that additional information can be usedin the design decision is an important option. Thetheoretical background of the observational methodis rooted in this problem(Peck, 1969; Stille, 2000).

● Another problem with classification will arise if thechosen principles of division do not define a spacethat is exhaustive when compared with the actualconditions. If the measurement vector falls outsidethe indicator space defined by the design set, ameaningful classification cannot be carried out. One


Fig. 4. Examples of classification into support classes(left) and excavation classes(right).

strategy to avoid this is to assess the typicality of thenew object to each class, and classify it as ‘other’ ifit is atypical of them all. In this way, the logicaldemand for the classification to be exhaustive maybe fulfilled—but the action required for the ‘other’class normally will be different. For example, it mayinvolve carrying out further investigations or askingfor advice from a panel of experts. An example ofthis is the observation of slaking ground in a fracturedrock mass in a water transfer tunnel. The presence ofsuch ground will require the use of special supportmeasures in order to guarantee the tunnel stability.The vector describing the support to stabilize thefractured ground will not normally contain a mineralanalysis of the rock or an observation of slaking sothat, in this case, a special ‘other’ class has to bedefined.

The problem of deciding which design decisions areappropriate, i.e. the accuracy of the actions taken, is ageneral rock mechanical problem and is related to thereliability of the design. For the example of designingsupports, and given a certain class, what is the proba-bility that the proposed support will fail? To answer thisquestion, several important issues have to be addressed,such as the model uncertainty, uncertainties in the database arising from natural variations, and the investiga-tions performed.

3.4. Practical requirements

It is very important that the principles of division andthe corresponding indicators frame a practical systemwith easily measurable parameters that result in aneconomic and safe design of the underground opening(Einstein et al., 1979).

The support measures and construction proceduresdefined by the classification should not be overly con-servative, nor should they fail. The method of classifi-cation should be relatively insensitive to normalvariation, as well as robust and repeatable. The para-meters should be easily obtained from outcrops andboreholes, as well as easily observed or measured in thetunnel.

3.5. Summing-up

The requirements of a system for classification forrock engineering purposes can be summarized asfollows:

● Use a supervised classification system;● The classes must be exhaustive and mutually

exclusive;● The indicators must be defined;● The principles of division(rules) must be established;● It must be possible to incorporate additional indicators

in the system;● It must be possible to update the rules(principles of

division);● The uncertainties and the probability of mis-classifi-

cation must be estimated;● The system should be practical and robust and give

an economic and safe design.

4. Rock engineering and classification systems

4.1. General

In all civil construction and building activities, differ-ent decisions have to be taken. A decision can only betaken based on a choice of different available alterna-


tives. In principle, there can be an infinite number ofalternatives but, normally, they are divided into classesor groups in order to facilitate the choice. In manycases, we have to choose between two classes. Forexample, we can accept a result of a calculation or rejectit. We can also make estimations with different assump-tions and get different results. The choice will then bebased on some kind of evaluation of the uncertaintiesinvolved, and the consequences of an improper decision.Normally, this is called a decision analysis based onrisk analysis or risk assessment. This is described inSection 4.3.

4.2. The use of classification in rock engineering

In comparison to many other civil engineering situa-tions, the uncertainties in underground rock engineeringare high. The design and different construction actionshave to be based on:

a. the geological model and assumed ground conditionsfrom various types of investigations during the plan-ning stage; and

b. the actual rock conditions encountered in the tunnelor underground opening during construction.

Pre-defined actions based on the use of classificationsystems have been shown to be an economic option inmany cases. This is a very common situation in rockengineering and will be discussed further here. Someexamples are:

– A common situation during the tunneling process isto take the decision whether or not to use forepolingor spiling. This is a typical choice between twoclasses(alternatives).

– Another example is when to reduce the length of theblast holes drilled to advance a tunnel. It is not verypractical to use a continuous reduction; instead, clas-ses are used like full round length, half round lengthor a quarter round length.

– Even for rock support, it is very common and con-venient to use classes with a stepwise increase inlevel of support measures. Often, choosing betweenpre-defined classes has been found to speed up thetunnel works. One reason for this is that it has oftenbeen found practical to use multiples of the thicknessof a single shotcrete layer instead of a continuousvariation. The inaccuracy in the site characterizationis in many cases so high that it is not meaningful todiscuss the difference in support between, for exam-ple, 2.1 m or 2.2 m rock bolt spacing. In such cases,stepwise-defined classes will be adequate.

Fig. 4 gives examples of two common situationswhere classification is used in rock engineering.What is also special for many rock-engineering prob-

lems is that the decision has to be taken during the

ongoing work and it is therefore under time constraints.Examples of such activities are:

– Decisions on rock support at the tunnel excavationface;

– Decisions on the need for grouting before blasting;– The evaluation of excavation and support proceduresin complex ground conditions.

It is quite obvious that the time and the cost neededto obtain better information must be compared withwhat can be saved by refining the design.In principle, there are two ways that can be used in

order to establish the classes:

– One way is to use classes based on some existingclassification system. Better or more correctly pre-defined actions are selected, based on the existingempirical design methods. However, as pointed outin many recent publications, such systems are notperfect and can sometimes lead to the selection ofinadequate ground support or an inappropriate design.This will be further discussed in Section 5.

– The other way is to develop a specific system tailoredfor the site in question, based on adequate siteinformation. It is the authors’ experiences that, inmany cases, this approach will give the optimumdesign(Nilsen et al., 1999; Brantmark et al., 1998).The reason for this is logical. Every tunnel project isunique. A tailor-made system can take into accountthe actual conditions and local construction experi-ence in a more accurate way. This approach has morechance of creating the best solutions than the appli-cation of a general system developed to meet everycondition worldwide. It also implies that a combina-tion of empirical design rules and refined rockmechanics models can be used, and both can serveas input to the prescribed support classes.

Rock engineering problems may be solved by differ-ent means(existing empirical methods, analytical meth-ods, numerical modeling, or observational methods)depending on the situation, see Figs. 2 and 5. All thesemethods are associated with uncertainties related to theproblems of ground conditions and project related fea-tures. In some cases, experience alone may be adequate.In other cases, when the possible consequences areserious, all the available tools have to be used. This isillustrated in Fig. 6.The design of an underground project is a result of a

long and complex process involving different steps likecharacterization, description of the project related fea-tures, and the processing of the acquired informationwith different design tools. Many authors have describedthis, maybe in different ways, with more emphases onthe complexity of the process than the fact that thereexist fundamental different opinions on the design pro-cedure, as illustrated in Fig. 7.


Fig. 5. Main principles in the process of ground characterization androck engineering.

Fig. 6. Examples of procedure for block instability(left) and squeezing(right).

As discussed above, more than one design tool willnormally be used for complex underground structures.This implies that it is reasonable to assume that aclassification system should be structured according tothe information from the different design methods usedin the engineering process.The key question is to use a classification system that

has an acceptable level of uncertainties. This is the basicquestion for all designers. In one way or another, everydesign must evaluate the safety level, and reliability ofthe design and describe its factor of safety, or probabilityof failure, or some related parameter. The accuracy ofthe classification system and the risk for mis-classifica-tion must always be evaluated.As they are empirical methods, it is essential to

understand how the use of classification systems isstructured and what requirements have to be satisfied.

From the most interesting paper of Einstein et al.(1979)on this issue, the following requirements of classifica-tions systems are summarized:

● They should promote economic, yet safe designs.● They must be correctly calibrated against test cases,

and those test cases must be representative of thefield of application.

● They should be complete in that all relevant factorsare included, yet they must be practical.

● They should have general applicability and robustnessto the varieties of use.

In this connection, it is important to recall thatBieniawski(1988) pointed out:

‘Rock mass classifications were never intended as the ultimatesolution to design problems, but only as a means towards this end.Nor were they intended to replace analytical considerations, fieldobservations and measurements, or engineering judgement. Rockmass classifications were developed to create some order out ofchaos in site investigation procedures and to provide desperatelyneeded design aids.’ ‘Nevertheless, these new ‘tools’ were sopowerful and successful that soon a tendency developed to ignoreeverything else and use rock mass classifications as the ultimateanswer. If this did not work, then rock mass classifications wereblamed!’

4.3. Decision analysis

In recent years, a new philosophy has been developedfor the use of applied probability in decision-making.The new theory, Bayesian statistical decision theory,provides a mathematical model for making engineeringdecision in the face of uncertainty(Benjamin andCornell, 1970). The basic principle is that the conse-quences of a decision depend on some factor, which isnot known with certainty. This factor is called a ‘stateof nature’ which has a certain probability to be the truestate. The decision-making problem will be formulated


Fig. 7. The process of rock engineering based on the principles in Fig. 1.

so as to choose an action between a set of availablealternatives. To every combination of states and actions,the engineer will be able to understand the consequence.It is also possible to incorporate new information in

the decision-making process by a process known asterminal analysis. This type of analysis will give theengineer the possibility of evaluating the benefit ofsearching for new information(e.g. by further investi-gations) before the search has been carried out.Strictly, when based on decision theory, the require-

ments to solve rock-engineering problems are thereforethe following:

● The possible states of nature have to be defined.● The different actions that may be undertaken have to

be defined.● Provided that the true state of nature is known, the

accuracy of any action to solving the given problem,must be estimated.

● The probability for any state to be the true state ofnature has to be estimated.

● The consequence of different combinations of actionsand states of natures has to be estimated.

● It must be possible to take the necessary decisionduring ongoing work without exceeding timeconstraints.

The different states of nature correspond to the dif-ferent possible rock conditions. The different actionscorrespond to classes of measures to be taken. Theaccuracy for the design and the probability to be thetrue state describe the uncertainties, and the correspond-ing probability of unexpected behaviour or failure.

5. The systems of today—how do they fulfill therequirements?

‘After over a decade of extensive use, rock mass classificationscan indeed serve as useful design aids in tunneling. However, thereare a number of pitfalls in using rock mass classifications and thesemust be understood by potential users.’ Bieniawski(1988)

5.1. General

The classification of rock masses continues to be asubject of discussion, as shown by the great number ofnew proposals that are being made in the literature. Aspointed out in many recent publications and governmentmanuals,(e.g. USACE, 1997), the classification systemsof today are not perfect, and can sometimes lead to theselection of inadequate ground support.Rock mass classification systems(or more correctly,

rock engineering classification systems) have beendeveloped over the years to describe the rock mass orground and to formalize an empirical approach to tunneldesign. Most of the classification systems were devel-oped from civil engineering case histories. The differentclassification systems place different emphasis on vari-ous engineering geological parameters.Many systems are developed and used for different

purposes. The same rock mass classification system canbe used both to describe and to characterize the rockmass and to estimate different design measures byempirical design rules. They are also used to giveindicators to rock engineering classifications. Neverthe-


less, it is important to distinguish their fields of appli-cation, either as a part of the process of characterization,or as an empirical method of design(Russo et al., 1998)Many classification systems have evolved as engi-

neers have attempted to apply their experience of rockmass behavior to a wider range of engineering problems.In recent years, classification systems have often beenused in tandem with analytical and numerical tools.Therefore, there has been a proliferation of work linkingclassification indexes to material properties, such asmodulus of elasticity, rock mass strength,m and s forthe Hoek and Brown failure criterion, etc. The valuesare then used as input parameters for the numericalmodels. Consequently, the importance of rock massclassification systems has increased over time(Milne etal., 1998).As summarized by Riedmuller and Schubert(1999)¨

and discussed in USACE(1997), the major shortcom-ings of the rock mass classification systems used fordesign of underground structures include:

● Classification parameters are not well defined orsufficient to select adequate design parameters androck support;

● Complex properties of a rock mass cannot be satis-factorily described by a single number;

● The same rating can be achieved by various combi-nations of classification parameters, even though therock mass behavior could be different;

● The user is led directly from the geological charac-terization of the rock mass to a recommended groundsupport without the consideration of possible failuremodes. It is necessary to examine the available rockmass information to determine if there are any appli-cable failure modes not addressed by the empiricalsystems. A number of potential modes of failure arenot covered by some or all of the empirical methods,and must be considered independently;

● The understanding of the geological setting and fea-tures of importance for the underground constructionis not seriously evaluated;

● Normally, the use of skilled people experienced inthe collection and assessment of data is not specifi-cally required.

The most common classifications systems used world-wide today are the RMR system published by Bieni-awski in 1973 and the Q system first described in 1974by Barton et al. More recently developed systems arethe RMi system, developed by Palmstrom in 1995.¨These classification systems have a quantitative estima-tion of the rock mass quality linked with an empiricaldesign rule to estimate adequate rock support measures.Quantitative rock mass classification systems(such

as the RMR, Q, or RMi systems) are most usefullyapplied during the early phases of design. These methodsprovide a means to compare quantitatively different

cavern layouts or tunnel alignments when only limitedrock mass data are available. They also provide a meansto communication and to develop construction costparameters, either for comparative purposes or to devel-op a construction cost budget(Hoek, 2002).The support charts or tables used by the various

classification systems to determine rock support arebased on experience from numerous underground pro-jects. Being statistically based, a support chart can neverreplace or accurately represent the ground conditions atsite. A main reason for this is, for example, that all theactual geometrical features of discontinuities cannot beincluded in a support chart. During tunnel construction,application of these rock mass classification systemscan, however, be very useful as one way of documentingactual conditions encountered by tunnel and cavernsconstruction.Summaries of these systems are presented below,

together with a discussion of their merits for character-izing the rock mass or being used in an empirical designmethod, or as an indicator in a true classification system.The systems and their use have been described innumerous papers and reports and it has not been feasibleto study all this material. The opinion presented belowis, therefore, based on a subjective selection, and on ourpersonal experiences from tunnel projects in Scandinaviaand around the world.

5.2. The RMR system

Bieniawski (1973, 1974) published the details of arock mass classification called the Geomechanics Clas-sification or the Rock Mass Rating(RMR) system.Significant changes have been made over the years withrevisions in 1974, 1975, 1976 and 1989; our discussionis based upon the 1989 version of the classificationsystem.The following six parameters are used to classify a

rock mass in the

1. RMR system:2. Uniaxial compressive strength of rock material;3. RQD value;4. Spacing of discontinuities;5. Condition of discontinuities;6. Ground water conditions;7. Orientation of discontinuities.

The rating of each of these parameters are summarizedto give a value of RMR. The rating is an outcome of asupervised classification of each parameter. The calcu-lated RMR value may be used to find which of fivepre-defined rock mass classes the rock mass belongs to,(going from very good rock to very poor rock). In thisrespect, the system can be described as supervisedclassification of rock mass quality. All parameters are


measurable in the field and some of them may also beobtained from borehole data.In applying this classification system, the rock masses

are divided into a number of structural regions. Theboundaries of the structural regions usually coincidewith major structural features(Bieniawski, 1984, 1989).However, from the practical point of view, the rating isalso related to length of the blasting round or the recentlyexcavated tunnel section.Bieniawski (1989) published a set of guidelines for

estimating the stand-up time(Lauffer, 1958), and forselecting rock support in tunnels, based on the RMRvalue. Other authors have modified the system andgiven guidelines for design especially for mining engi-neering and for slope stability. However, Bieniawskistrongly emphasizes that a great deal of judgment isneeded in the application of rock mass classification tosupport design.The RMR value has also been used to estimate rock

mass properties. Bieniawski(1984, 1989) and Serafimand Pereira(1983) have given a relationship betweenthe RMR and the rock mass deformation modulus. TheRMR value is also used as one way to estimate themand s factors in the Hoek–Brown failure criterion(Wood, 1991; Hoek, 1994; Hoek and Brown, 1998) aswell as the GSI value to evaluate the rock mass strength.These give, however, only empirical relations and havenothing to do with rock engineering classification in itstrue sense.The experiences of the authors from using the system

indicates that it works well to classify the rock massquality, since it is relatively well defined and the ratingfor each parameter can be estimated with acceptableprecision. The relatively small database makes the sys-tem less applicable to be used as an empirical designmethod for rock support.The RMR system has been used in many tunnel

projects as one of the indicators to define the supportor excavation classes. However, RMR cannot be usedas the only indicator, especially when rock stresses ortime dependent rock properties are of importance forthe rock engineering issue.

5.3. The Q system

On the basis of an evaluation of a large number ofcase histories of tunnel projects, Barton et al.(1974) ofthe Norwegian Geotechnical Institute(NGI) proposed aTunneling Quality Index(Q) as a classification systemfor estimating rock support in tunnels. It is a quantitativeclassification system based on a numerical assessmentof the rock mass quality. Later, Barton et al. havepublished several papers on the Q system aiming atextending its applications. Some of these use additionaladjustments of the Q system.

The numerical value of the index Q is defined by sixparameters and the following equation:

QsRQDyJn=JryJa=JwySRF

where

– RQD is the rock quality designation;– Jn is the joint set number;– Jr is the joint roughness number;– Ja is the joint alteration number;– Jw is the joint water reduction factor;– SRF is the stress reduction factor.

In explaining the system and the use of the parametersto determine the value of Q, Barton et al. have giventhe following explanation:

● The first quotient(RQDyJn) represents roughly theblock size of the rock mass.

● The second quotient(JryJa) describes the frictionalcharacteristics of the rock mass.

● The third quotient(JwySRF) represents the activestress situation. This third quotient is the most com-plicated empirical factor and has been debated inseveral papers and workshops. It should be givenspecial attention, as it represents four groups of rockmasses: stress influence in brittle blocky and massiveground, stress influence in deformable(ductile) rockmasses, weakness zones, and swelling rock.

The Q system can be used as supervised classificationof rock mass quality. Nine different rock mass qualityclasses are defined, ranging from ‘exceptionally poor’to ‘exceptionally good’.The Q-system is normally used as an empirical design

method for rock support. Together with the ratio betweenthe span or height of the opening and an excavationsupport ratio (ESR), the Q value defines the rocksupport. The accuracy of the estimation of rock supportis very difficult to evaluate. It is the authors’ experiencefrom using the system that, especially in the poorer rockclass (Q)1) the system may give erroneous design.The true nature of the rock mass that is essential for thedetermination of the support measures(e.g. swelling,squeezing or popping ground) is not explicitly consid-ered in the Q system. Nor are such issues as the timingfor installation and the need for an invert strut. For theconditions in faults and weakness zones, the supportsshould be checked or designed by complimentary engi-neering methods.In fractured ground, the orientation of the joints is an

essential parameter. In such cases, it is very importantto follow the guideline given by Barton et al.(1974)that the parameters Jr and Ja should be related to thejoint surface most likely to allow failure to initiate.From the rock mechanics point of view, it is obviousthat even such a simple load case as block instability is


Fig. 8. The input parameters to RMi(from Palmstrom, 1996).¨

Fig. 9. Possible applications of RMi(from Palmstrom, 1996).¨

much more complicated than can be given by a singlenumber like a Q value.Of course, the Q system can be used as an indicator

for rock support or other types of rock engineeringclassification. The value is, however, normally used asthe only indicator to define the classes in question. Theauthors strongly argue against such use of an engineeringclassification system, since it may be too rigid and willnot allow other types of observations to be taken intoaccount.Grimstad and Barton(1993) have also presented an

equation to use the Q value to estimate the rock massdeformation modulus(for values of Q)1). The Q valueis also used as one way to estimate them and s factorsin the Hoek–Brown failure criterion(Hoek, 1983; Hoekand Brown, 1988). In this respect, it is only an empirical

relationship and has nothing to do with engineeringclassification.

5.4. The RMi system

The rock mass index, RMi, is a volumetric parameterindicating the approximate uniaxial compressive strengthof a rock mass. The system was first presented byPalmstrom(1995) and has been further developed and¨presented in several different papers. It makes use ofthe uniaxial compressive strength of intact rock(s )cand the reducing effect of the joints penetrating the rock(JP) given as:

RMiss =JP for jointed rock massesc


RMiss =f for massive rock having block sizec s

3larger than approximately 5 m(where f -JP)s

The jointing parameter(JP) is by empirical relationsrelated to the joint condition factor, jC, and the blockvolume, Vb. The joint condition, jC, can be estimatedby:

– the joint roughness, jR;– the joint alteration, jA; and– the joint size, jL.

The massivity parameter, f , represents the scale effects

of the uniaxial compressive strength(which for intactrock samples or massive rock has a value of approxi-mately 0.5).The RMi system has some features similar to those

of the Q-system. Thus, jR and jA are almost the sameas Jr and Ja in the Q-system. The connection betweenthe different input parameters applied in the RMi isshown in Fig. 8.The different input parameters can be determined by

commonly used measurements and mapping and fromempirical relationships presented by Palmstrom in his¨work. It requires more calculation than the RMR andthe Q system, but spreadsheets can be used from whichRMi values can be found directly.Based on a characterization of the rock mass by RMi

combined with the geometrical features of the openingand ground factors like rock stresses, different rockengineering issues such as relevant rock support can beestimated using support charts(Palmstrom, 1996). The¨charts have been developed from experience of morethan 25 different projects and locations as well aspersonal experience from numerous underground con-structions in hard rock.As shown in Fig. 9, the RMi value can be applied as

input to other rock engineering methods, such as numer-ical modeling, the Hoek–Brown failure criterion forrock masses, and to estimate the deformation modulusfor rock masses(Palmstrom and Singh, 2001).¨The RMi system can be characterized as a typical

empirical design method and is not a classification inits true sense. However, Palmstrom has given five¨different strength classes of rock masses from very lowto very high and, in this respect, it can be used as asupervised classification for rock mass strength.The system applies best to massive and jointed rock

masses where the joints in the various sets have similarproperties. It may also be used as a first check forsupport in faults and weakness zones, but its limitationshere are pointed out by Palmstrom(1995). For special¨ground conditions like swelling, squeezing, ravelingground, and weakness zones(fault zones, etc.) the rocksupport should be evaluated separately for each andevery case. Other features to be separately assessed are

connected to project specific requirements such as thelife-time required and safety.Like all the other empirical design methods, it is not

possible to evaluate the accuracy of the system. Thefactor of safety or the probability of failure for a givenset of indicators cannot be evaluated.

5.5. The GSI system

The geological strength index, GSI, introduced byHoek (1994) and Hoek et al.(1998) provides a systemfor estimating the reduction in rock mass strength fordifferent geological conditions as identified by fieldobservations. The rock mass characterization is straight-forward and based on the visual impression of the rockstructure, in terms of blockiness, and the surface condi-tion of the discontinuities indicated by joint roughnessand alteration. The combination of these two parametersprovides a practical basis for describing a wide range ofrock mass types. Note that there is no input for thestrength of the rock material in the GSI.Visual determination of GSI parameters represents the

return to quality descriptions instead of advancing quan-titative input data as in RMR, Q and RMi systems. GSIwas found mainly useful for weaker rock masses withRMR-20.As GSI is used for estimating input parameters

(strength), it is only an empirical relation and hasnothing to do with rock engineering classification.

5.6. The NATM

The new Austrian tunneling method was developedby Rabcewicz (1964y1965) and Pacher(1975). Inpractice, the NATM involves the whole sequence ofrock tunneling aspects from investigation during design,engineering and contracting, to construction and moni-toring as described by Brown(1981). It is important tonotice that the NATM has been developed for tunnelingin weak or squeezing ground. The NATM has beenapplied successfully in a large number of tunnels inmany parts of the world, some of which were construct-ed in poor and difficult ground conditions. Considerablecost savings have often been gained when compared totraditional tunneling, as well as reduced constructiontime. The NATM has, however, also experienced manyunpleasant rock falls and some tunnel collapses.In Austrian tunneling practice, the ground is described

behaviorally and allocated a ground class in the field,based on field observations. Construction and supportcan be estimated from this classification. The qualitativeground description used is associated, rather inconsis-tently, with excavation techniques, together with princi-ples and timing of standard support requirements.Therefore, the NATM is not a rock engineering classi-fication system, but a construction strategy(in German


‘bauweise’) containing several methods for assessingthe amount and timing of rock support, constructionsteps etc.(Jodl, 1995).

6. Conclusions and recommendations

Several papers have been published on the use andmisuse of classification systems. Some of these areEinstein et al.(1979), Bieniawski (1988), Milne et al.(1998), Hoek (2002) and Riedmuller and Schubert¨(1999) as discussed above.The primary object of all rock mass classification

systems is to quantify different engineering propertiesof, or related to, the rock mass, based on past experience.One important use of the classification system today hastherefore been to serve as a kind of checklist.Three different types of output can be distinguished

from the rock mass classification systems discussed inthis paper.

1. Characterization of the rock mass expressed as overallrock mass quality, incorporating the combined effectsof different geological parameters and their relativeimportance for the overall condition of a rock mass.This enables the comparison of rock mass conditionsthroughout the site and delineation of regions of therock mass from ‘very good’ to ‘very poor’, thusproviding a map of rock mass quality boundaries.

2. Empirical design with guidelines for tunnel supportcompatible with rock mass quality and the method ofexcavation. Traditionally, this is often seen as themajor benefit from the use of rock mass classificationsystems.

3. Estimates of rock mass properties. Rock mass char-acterization expressed as an overall rock mass qualityhas been found useful for estimating the in situmodulus of rock mass deformability and the rockmass strength to be used in different types of designcalculations.

However, none of the discussed rock mass classifi-cation systems is a ‘classification’ in the true sense.They are all, as a matter of fact, empirical designmethods based on characterization of rock masses. Theuse of the word classification is therefore misleading.It is interesting to notice the conclusion presented by

Einstein et al.(1979) that the accuracy of the existingempirical design methods is not established. The meth-ods probably overestimate the support requirements andthe relationships to the ground support pressure are oftennot very accurate.In numerous cases, it has been necessary to adapt an

existing classification system to the actual condition andproblem, and calibrate the existing rock mass classifi-cation systems against the experience gained from aspecific project. This means that tailor-made supervisedclassification have been developed, where the index of

the rock mass quality derived from the existing classi-fication system has been an indicator to evaluate thesupport class. In many cases, the index has been usedas the only indicator. This has created contractual prob-lems when unforeseen geological conditions have beenencountered, and where the system has not been appli-cable. Typical conditions that are not covered are swell-ing, squeezing, raveling, or popping ground.None of the rock mass classification systems studied

is able to incorporate other types of information, suchas results from deformation measurements. This is agreat disadvantage as, especially for complex under-ground structures, more than one design tool is normallyused and also will be followed up during construction.Guideline for observational systems with alarm thresh-olds as discussed by Olsson and Stille(2002) may beused in order to form a system for classification thatincorporates deformation measurements or visualinspections.In the early stages of a project, the existing quantita-

tive rock mass classification systems(empirical designmethods) can be applied as a useful tool to establish apreliminary design. At least two systems should beapplied(Bieniawski, 1984, 1989). They are not recom-mended for use in detailed and final design, especiallyfor complex underground openings. For this purpose,they need to be further developed.Classification systems are unreliable for rock support

determinations during construction, as local geometricand geological features may override the rock massquality defined by the classification system. Restrictionson their use here is also pointed out by Bieniawski(1997).The main core of the classification systems is the

assessment of the rock mass quality that, preferably, canbe used as one of the indicators for a supervisedclassification of a rock engineering case.The following requirements can be put forward to

build up such a system to be able to adequately solverock engineering problems:

● Use a supervised classification adapted to the specificproject.

● The reliability of the classes to handle the given rockengineering problem must be estimated.

● The classes must be exhaustive and mutuallyexclusive.

● Establish the principles of the division into classesbased on suitable indicators.

● The indicators should be related to the different toolsused for the design.

● The principles of division into classes must be soflexible that additional indicators can be incorporated.

● The principles of division into classes have to beupdated to take account of experiences gained duringthe construction.


● The uncertainties or quality of the indicators must beestablished so that the probability of mis-classificationcan be estimated.

● The system should be practical and robust, and givean economic and safe design.

Our conclusion is that none of the existing classifi-cation systems fulfills the requirements mentioned abovefor a true classification system for rock engineeringproblems. The classification systems, or better the empir-ical design methods, cannot be used as the principles ofdivision for a true system without further development,since it is not possible to define the accuracy of themethods. They can, however, be used as one of theindicators defined by the principles of division. Theauthors strongly argue against using the existing classi-fication system as the only indicator to define the rocksupport or other rock engineering items.We want also to emphasize that tools like decision

theory can be very useful in order to select the mostsuitable supervised classification for a specific rock-engineering problem and project, and also to determinethe need for further investigation.

7. Other references

The following references have been most useful forus to form our opinion on rock classification, thoughthey have not directly been quoted in the text above:Barton et al., 1980; Barton, 1987; Bhawani et al., 1992;Bieniawski, 1976, 1992; Brosch, 1986; Coates, 1964;Deere, 1963; Deere and Miller, 1966; Einstein, 1991;Kirkaldie, 1988; Krauland et al., 1989; Milne andPotvin, 1992; Patching and Coates, 1968; Riedmuller,¨1997; Singh and Goel, 1999; Terzaghi, 1946.

Acknowledgments

The authors are most thankful to Dr Don Moy forvaluable text suggestions.

References

Barton, N., Lien, R., Lunde, J., 1974. Engineering classification ofrock masses for the design of rock support. Rock Mech. 6, 189–236.

Barton, N., Lien, R., Lunde, J., 1980. Application of Q-system indesign decisions concerning dimensions and appropriate supportfor underground installations. Proceedings of the InternationalConference on Subsurface Space. Pergamon Press, pp. 553–561.

Barton, N., 1987. Predicting the behaviour of underground openingsin rock. 4th Manual Rocha Memorial Lecture, Lisbon(Also inNorwegian Geotechnical Institute, Publ. No. 172, 21 p).

Bhawani, S., Jethwa, J.L., Dube, A.K., Singh, B. 1992. Correlationbetween observed support pressure and rock mass quality. Tunnel.Underground Space Technol. 7(1), 59–74.

Benjamin, J.R., Cornell, C.A., 1970. Probability, statistics and deci-sion for civil engineers. McGraw-Hill, New York.

Bieniawski, Z.T., 1973. Engineering classification of jointed rockmasses. Trans. S. Afr. Instn. Civ. Eng.15(12), 335–344.

Bieniawski, Z.T., 1974. Geomechanics classification of rock massesand its application in tunneling. Proceedings of the Third Interna-tional Congress on Rock Mechanism. ISRM, Denver, pp. 27–32.

Bieniawski, Z.T., 1976. Rock mass classifications in rock engineering.In: Bieniawski, Z.T., Balkema, A.A.(Eds.), Proceedings Sympo-sium on Exploration for Rock Engineering. Rotterdam, pp. 97–106.

Bieniawski, Z.T., 1984. Rock Mechanics Design in Mining andTunneling. A.A. Balkema, Rotterdam.

Bieniawski, Z.T., 1988. Rock mass classification as a design aid intunneling. Tunnels Tunneling, July 1988.

Bieniawski, Z.T., 1989. Engineering rock mass classifications. JohnWiley & Sons, New York, pp. 251.

Bieniawski, Z.T., 1992. Design Methodology in Rock Engineering.A.A. Balkema, 198 Rotterdam.

Bieniawski, Z.T., 1997. Quo vadis rock mass classifications? Felsbau15(3), 177–178.

Brantmark, J.,Taube, A., Stille, H., 1998. Excavation of a sub-searoad tunnel at Hvalfjordur, Iceland. 8th international IAEG congress¨Vancouver. Balkema.

Brekke, T.L., Howard, T.R., 1972. Stability problems caused by seamsand faults. Rapid Excavation Tunnel. Conf. 1972, 25–41.

Brosch, F.J., 1986. Geology and the classification of rock masses-examples from Austrian tunnels. Bull. IAEG 33, 31–37.

Brown, E.T., 1981. Putting the NATM into perspective. TunnelsTunnel. Nov. 13–17.

Coates, D.F., 1964. Classification of rocks for rock mechanics. RockMech. Min. Sci. 1, 421–429.

Deere, D.U., 1963. Technical description of rock cores for engineeringpurposes. Felsmechanik und Ingenieurgeologie 1(1), 16–22.

Deere D., Miller R.D., 1966. Engineering classification and indexproperties for intact rock. University of Illinois, Technical Report.No. AFWL-TR-65–116, 1966.

Einstein, H., Steiner, W., Baecher, G.B., 1979. Assessment of empir-ical design methods for tunnels in rock. RETC 1979, 683–705.

Einstein, H.H., 1991. Observation, quantification and judgement:Terzaghi and engineering geology. J. Geotech. Eng. 117(11), 1772–1778.

GeoEng2000 workshop on classification systems. The reliability ofrock mass classification used in underground excavation andsupport design. ISRM News, Vol. 6, No. 3, 2001. 2 p.

Grimstad, E., Barton, N., 1993. Updating the Q-system for NMT.Proceedings of the International Symposium on Sprayed Concrete,Fagernes, Norway, 1993. Norwegian Concrete Association, Oslo,20 pp.

Hand, G.J., 1997. Construction and assessment of classification rules.John Wiley & Sons.

Hoek, E., 1983. Strength of jointed rock massesThe Rankine Lecture1983, Geotechnique 33(3), 187–223.

Hoek, E., Brown, E.T. 1988. The Hoek–Brown failure criterion—a1988 update. Proceedings of the 15th Canadian Rock MechanicsSymposium, pp. 31–38.

Hoek, E., 1994. Strength of rock masses. News Journal of ISRM2(2),4–16.

Hoek, E., Marinos, P., Benissi, M., 1998. Applicability of thegeological strength index(GSI) classification for very weak andsheared rock masses. The case of the Athens schist formation.Bull. Eng. Geol. Env. 57, 151–160.

Hoek, E., Brown, E.T., 1998. Practical estimates of rock massstrength. Int. J. Rock Mech. Min. Sci 34, 1165–1186.

Hoek, E., 2002. Rock mass classification. Hoek’s Corner, www.rocscience.com; (accessed August 2002).

Jodl, H.G., 1995. Construction method NATM. IACES, Bureau ofVienna, Summer course in NATM, University of Technology,Vienna.

Kirkaldie L., 1988. Rock classification systems for engineeringpurposes. STP 984, Am. Soc. Test. Mater., p. 167.


Krauland, N., Soder, P., Agmalm, G., 1989.@ ‘Determination of rock¨mass strength by rock mass classification—Some experiences andquestions from Boliden mines.’ Int. J. Rock Mech. Min. Sci.Geomech. Abstr. 26(1), 115–123.

Lauffer, H., 1958. Classification for tunnel construction(in German).Geologie und Bauwesen 24(1), 46–51.

Milne, D., Potvin, Y., 1992. Measurement of rock mass propertiesfor mine design. Eurock 1992. Thomas Telford, London, pp.245–250.

Milne, D., Hadjigeorgiou, J., Pakalnis, R., 1998. Rock mass charac-terization for underground hard rock mines. Tunnel. UndergroundSpace Technol. 13(4), 383–391.

Nilsen, B., Palmstrom, A., Stille, H., 1999. Quality control of a sub-¨sea tunnel project in complex ground conditions. Proceedings ofthe World Tunnel Congress 1999. Oslo. A.A. Balkema.

Olsson, L., Stille, H., 2002. Observational systems with alarmthresholds and their use in the design of underground openings. Tobe published as SKB report in 2002.

Pacher, F., 1975. The development of the New Austrian TunnelingMethod and the main features in design work and construction.16th Symposium on Rock Mechanics, Minneapolis, pp. 223–232.

Palmstrom, A., 1995. RMi-a rock mass characterization system for¨rock engineering purposes. Ph.D. Thesis University of Oslo, p. 400(also on web site www.rockmass.net).

Palmstrom, A., 1996. Characterization of rock masses by the RMi¨for use in practical rock engineering. Tunnel. Underground SpaceTechnol., Vol. 11, No. 2, pp. 175–186(part 1); Vol. 11, No 3, pp.287–303(part 2). (also on web site www.rockmass.net).

Palmstrom, A., Singh, R., 2001. The deformation modulus of rock¨masses-comparisons between in situ tests and indirect estimates.Tunnel. Underground Space Technol. 16(3), 115–131.

Patching, T.H., Coates, D.F., 1968. A recommended rock classificationfor rock mechanics purposes. CIM Bull. Oct. 1968, 1195–1197.

Peck, R.B., 1969. Advantages and limitations of the observationalmethod in applied soil mechanics. Geotechnique 19, No. 2.

Rabcewicz, L.V., 1964y1965. The new Austrian tunneling method.Water Power, Part 1, November 1964, pp. 511–515, Part 2, January1965, pp. 1195–1197.

Riedmuller, G., Schubert,W., 1999. Critical comments on quantitative¨rock mass classifications. Felsbau 17(3), 164–167.

Riedmuller, G., 1997. Rock characterization for tunneling—Engineer-¨ing geologist’s point of view. Felsbau 15(3), 167–170.

Russo, G., Kalamaras, G.S., Grasso, P., 1998. A discussion on theconcepts of geomechanical classes, behaviour categories, and tech-nical classes for an underground project. Gallerie e grandi operesoterranee 54, 40–51.

Serafim, J.L., Pereira, J.P., 1983. Consideration of the geomechanicsclassification of Bieniawski. Proceedings of the International Sym-posium on Engineering Geology and Underground Constructions,pp. 1133–1144.

Singh, B., Goel, R.K., 1999. Rock mass classification: A practicalapproach in civil engineering. Elsevier Publisher, Amsterdam, pp.267.

Stille, H., 2000. Squeezing behaviour-Observation and monitoring.Italian Geotechnical Journal Anno XXXIV, n1-Gemaio-Marzo2000.

Terzaghi K., 1946. Introduction to tunnel geology. In Rock tunnelingwith steel supports, by Proctor and White, pp. 5–153.

USACE, 1997. Engineering and design; Tunnels and shafts in rock.US. Army Corps of Engineers, Manual no. 1110-2-2901, 236 p.

Wood, D., 1991. Estimating Hoek–Brown rock mass strength para-meters from rock mass classifications. Transp. Res. Rec. 1330,22–29.

classification as a tool in rock engineering

Documents