THE COMMERCIAL CHARACTERIZATION OF DIMENSION STONES THROUGH THE USE OF MULTIVARIATE STATISTICS: THE CASE STUDY OF THE -BUDDUSO’ GRANITES (, )

CAPPELLI A. * , VIOLO M. *, ZAPPATORE V. ** (*) University of Rome “La Sapienza”, Faculty of Engineering, I.C.M.M.P.M. Department, Via Eudossiana 18 - 00184 Rome, Italy. (**) Engineer, independent professional

ABSTRACT The present work concerns the commercial characterization of granites of the Bitti- Buddusò zone (north-central Sardinia). It investigates the existing correlation between the various commercial types of granite extracted from the above-mentioned mining basin and the physical, chemical and petrographical qualities of the rock so as to set a method for the technical and scientific characterization of the commercial value of the investigated materials. Chemical and petrographical analyses were made for each granite variety. With the aid of image analysis slabs were esthetically characterized by pointing out the index color, the dimensional homogeneity of crystals and the frequency of eventual macroscopic defects. All the measured variables were analyzed with the aid of multivariate statistical techniques to identify which variables were more significant in the rock’s esthetical characterization. By projecting the system’s variables onto the factorial planes of the statistical analysis (MCA) the factorial variable F1 appears to be correlated to the mineralogical composition of the granites, to the color and to the petrographical types of the slabs; variable F2 shows a correlation with some commercial classes and with the texture. Factor F3 seems to clearly discriminate among the various commercial granite classes, thanks to its strong correlation to the slabs’ homogeneity and the presence of visible inclusions.

INTRODUCTION A strong mining activity for ornamental granites has developed in Sardinia since the sixties. The high quality of the product allowed local firms to obtain important contracts and to play a major role in this sector on the national and international level. The production of Sardinian granite is now estimated at about a million tons per year of raw blocks which is over 90% of the entire national production. These quantities mean that every year around two million tons of material are classified as waste of the extractive process or of other working phases thus creating a significant environmental impact (De Carlo et al., 1998). Even if waste is an unavoidable outcome of any productive activity, it is clear that, by paying greater attention to the characterization of the deposits, particularly through a study of the existing links among the jointing status of the massif, the chemical-physical characteristics of the rock and its commercial quality, production can be planned and the environmental impact due to exploitation activities and other working phases can be minimized. The present study derives from this idea and its aim is to verify the most significant parameters for the definition of the quality of the rock.

1. THE BITTI-BUDDUSÒ EXTRACTIVE BASIN Today the productive sector of Sardinian granite counts 1500 operators and represents a business of about 150 million dollars. Almost all the yield comes from around 180 quarries concentrated in a specific “natural industrial district” (De Carlo et al, 1998) which is formed by several exploitation sites concentrated in four main mining basins: Tempio-Calangianus, Arzachena-Luogosanto, Buddusò-Ala dei Sardi and Ovodda. The processing factories are situated next to the quarries or along the main traffic arteries. Local transformation activities represent 25% of all the extracted material and consist of 25 firms that work a volume of material equal to three million square meters per year seen in 2 cm thick slabs; there is a collateral production of two hundred thousand tons of wastes per year (De Carlo et al, 1998). The Bitti-Buddusò extractive basin extends for hundreds of square km into the Figure 1 – The Bitti-Buddusò extractive basin, Isle of Sardinia, north-central part of Sardinia (figure 1) Italy and it is characterized by the presence of 34 active quarries with a production that in 1997 reached eighty thousand cubic meters/year. Ninety per cent of the area is made up of granitoid rocks, the remaining 10% is metamorphic. Plutonic rocks are largely formed of monzogranitic granodiorites, monzogranites and, in lesser part, of heterogranular granodiorites. Among the monzogranitic granodiorites there is a difference of facies based essentially on the textural character; on the geopetrographical chart (figure 2) the following facies are recognizable (Calia et al., 1992):

equigranular monzogranitic granodiorites of gray color and constant medium texture; are present in both the north-western and south-eastern sector, in contact with the metamorphic rocks. The matrix is homogeneous ad the femic microgranular inclusions are the lowest in the entire zone,

eterogranular monzogranitic granodiorites of gray color; are found in the central and north-eastern sector. The texture variations are evident.

The granitic rocks in this zone are characterized by the following paragenesis: Figure 2 – Geopetrographical map of the Bitti-Buddusò basin: quartz, orthoclase, microcline, plagioclase, 1) old and recent alluvials; 2) quartzitic, aplitic and pegmatitic dykes; 3) microgranites; 4) equigranular monzogranitic biotite and muscovite. Quartz is abundant and granodiorites; 5) eterogranular monzogranitic granodiorites; can be seen under the microscope as large 6) eterogranular granodiorites; 7) metamorphic rocks; 8) certain and presumed faults (from Calia et al., 1992). crystals and as small irregularly shaped individual spots; often it is included in bigger crystals of feldspar. Orthoclase and microcline are present in seemingly large crystals; plagioclase is more abundant than K-feldspar and it is present in idiomorphic crystals (Calia et al., 1992).

2. COMMERCIAL CHARACTERIZATION OF THE BITTI-BUDDUSÒ GRANITE The work of characterization was carried out on a great number of granite slabs chosen as representative of the various commercial types of granite present in the zone using the empirical classification criteria common to quarry workers. All the slabs were classified as first or second choice and divided into four commercial types: Malaga Gray, Pearl Gray, Brown Gray and Mount Limbara Gray. The latter does not belong to the examined production basin but to the one in Tempio; it has however been included in the analysis in order to compare the procedure’s results and test its credibility. Every sample was chemically and petrographically analyzed and the aspect of the slabs noted through image analysis techniques (Cappelli et al., 1998).

2.1 Chemical and petrographical analyses All samples were analyzed for their content in iron, sodium, silica, calcium, magnesium, potassium and aluminum oxides. On the basis of the composition of oxides each slab was petrographically classified (chart 1). All the samples belonging to each commercial type showed a clear petrographical homogeneity with the Brown Gray as the only exception. To sum up:

Malaga Gray (1st e 2nd choice) monzogranitic granodiorites Monte Limbara Gray (1st e 2nd choice) monzogranites Pearl Gray (1st e 2nd choice), Brown Gray (1st choice) monzo-leucogranites Brown Gray (2nd choice) leucogranites

2.2 Analysis of the external aspect of slabs; image analysis All the slabs were scanned with a 300 dpi scanner and then processed with the Adobe Photoshop 5.0 image analysis software. The single mineralogical components of the rock were shown through the study of the textural and chromatic differences calculating average color and frequency.

For all the slabs a Gabbros Qz-diorites Tonalites Granodiorites Monzogranites Leucogranites color index was SiO2 48.24 53.96 57.38 61.45 64.28 67.40 70.10 73.34 75.06 77.70 defined, from the TiO2 1.49 0.67 0.94 0.85 0.63 0.52 0.43 0.25 0.12 0.05 weighed average of Al2O3 15.37 20.81 16.43 17.06 15.53 15.42 14.53 13.92 13.22 12.42 the three color Fe2O3 3.94 2.21 1.38 1.38 1.37 0.91 0.96 0.50 0.28 0.11 components (RGB). FeO 9.11 3.95 5.43 4.70 3.24 2.83 1.95 1.65 1.08 0.26 In the slabs that had MgO 5.77 3.02 4.58 2.35 2.41 1.75 1.08 0.30 0.23 0.12 “inclusions” the color CaO 10.02 10.38 6.59 4.81 4.10 3.38 2.69 1.54 1.19 0.47 Na O 2.39 2.38 3.34 3.09 3.02 2.94 3.23 3.02 3.40 2.64 index was calculated 2 K O 0.85 1.11 2.24 2.90 3.63 3.37 4.00 4.79 4.65 5.30 by “cutting off” the 2 anomalous zone and Chart 1 – Mineralogical composition (in %) of different petrographical types of the Sardinian temporarily excluding batholith (from Bralia et al., 1993). it from the analysis. The inclusions were then individually analyzed noting their color and their extensions. Most of the anomalous zones were essentially made up of femic minerals; in many samples however, the inclusions were made of quartz, feldspar and mica crystals, which presented a different texture from the rest of the slab. Then, for each slab, crystal dimensional homogeneity and grain size were assessed using an empirical method.

2.3 Multiple Correspondence Analysis (MCA) The search for some correlation among the numerous variables involved in the analysis of the granite samples, imposed the use of a method of factorial analysis capable of representing a complex system of several variables into a more simple and variable name class specifications SiO2 S1 SiO ?70% comprehensible one. The Multiple 2 S2 70%

Correspondence Analysis (MCA) uses S3 SiO2>75% discrete input variables and transforms Al2O3 A1 Al2O3?13% A2 ? the variables of a multivariate system 13% 14,5% Fe O F1 ? variables each of which is a linear 2 3 Fe2O3 2% F2 2 % 3% 1986). GRAIN D1 Homogeneity The new system is called factorial UNHOMOGENEITY D2 Unhomogeneity and the new variables are the factors; and/or INCLUSIONS D3 Inclusions GRAIN SIZE G1 Fine the first factor indicates the maximum G2 Medium possible quantity of the total variance G3 Coarse of the initial multivariate system, the COLOR L1 L?151 second the maximum residual variance L2 151< L ?167 and so on. In this way the initial L3 L > 167 COMMERCIAL GM1 Brown Gray 1st choice. multivariate system is transformed into CLASS GM2 Brown Gray 2nd choice. a simplified equivalent one in which GP1 Pearl Gray 1st choice. few variables (factors) express most of GP2 Pearl Gray 2nd choice. the information contained in the whole ML1 Mt. Limbara Gray 1st choice. ML2 Mt. Limbara Gray 2nd choice. set of data (Saporta, 1990). All this MA1 Malaga Gray 1st choice. can be graphically visualized by MA2 Malaga Gray 2nd choice. projecting the system’s initial PETROGRAPHICAL CP1 Monzogranitic granodiorite variables onto the factorial planes: an CLASS CP2 Monzogranite interpretation of the significance of the CP3 monzo-leucogranite CP4 leucogranite factors can be given according to the positions that the variables occupy on Chart 2 – List of the variables used in the MCA: all continuous variables have these planes (Cappelli et al., 1998, been discretized into three intervals. 1999).. In our case the variables are those shown in chart 2. It is clear how the continuous variables of the initial system (percentage in SiO2, Al2O3 and Fe2O3, color index) have been discretized to allow the application of a multiple correspondence analysis.

2.3.1 Interpretation of MCA results The factorial plane F1F2 (figure 3a) shows how some categories are strongly correlated to the positive values of factor F1; particularly evident are the clustering of the first and second class of Malaga Gray (MA1, MA2), the petrographical classes of the monzogranitic granodiorites (CP1), the unhomogeneity (D2), the low content of silica oxide (S1), the high aluminum and iron oxides content (R3,A3) and the dark color of the slabs (L1). The correlation of the last four categories is particularly significant considering how the complementary categories are correlated to the negative values of F1, while the mean values of the variables are correlated to nil or slightly negative values of this factor. From this a first consideration can be made: F1 can be interpreted as a color index and as an index of the mineralogical composition of the rock. This shows the strong correlation between F1 and the petrographical qualities: the factor seems to gradually discriminate between CP1 and CP4, the more basic petrographical groups (positive F1 values) to the more acidic ones (negative F1 values). The correlation between the chemical composition and color variables shows a strong dependency relation: the presence of Silica tends to increase the slabs’ brightness inversely to

Figure 3a Figure 3b

Figure 3 - Projection of the variables onto factorial planes: a) F1F2; b) F2F3; c) F1F3

the presence of aluminum and iron. The analysis of factorial plane F1F2 shows furthermore a strong correlation between the negative F1 values and some important categories concerning commercial types and petrographical classes. These categories cluster in variously compact groups: this is the case of GP1, GP2 and GM1 that, with respect to F1, are correlated with the petrographical class of the monzo-leucogranites (CP3). The same is true for the Mount Limbara Gray commercial Figure 3c type (first and second choice ML1 and ML2), for the monzogranite (CP2) and for the fine grain category (G1); this latter association is correlated to strongly positive values of factor F2. On the contrary the association made by the second choice Brown Gray (GM2) and by the petrographical class of the leucogranites (CP4) seems correlated with strongly negative values of factors F1 and F2. The texture seems to be discriminated by both F1 and F2: positive F1 values are related to coarse grain commercial types while negative values are related to medium to fine grain types; with respect to F2 however the correlation is positive for fine grain types while negative values are related to medium and coarse grain types. The unhomogeneity factor seems to be related only to factor F1. From the study of the F2F3 factorial plane (figure 3b) an interesting peculiarity of factor F3 can be noticed: this discriminates among the first and second choices of all the commercial types except for the second choice Brown Gray. The categories ML1, MA1, GP1 and GM1 seem to be related to positive or slightly negative F3 values, while the second choices (GP2, MA2, ML2) are all correlated with F3 values that are from slightly to strongly negative. As above-mentioned, the exception to this behavior is the second choice Brown gray which seems to have a strongly positive and negative relation with F2 and F3 respectively and, above all, seems to be related to the petrographical class to which all the analyzed samples belong. The slabs’ homogeneity shows an evident correlation with positive F3 factor values, while the non-homogeneity seems unrelated to F2 or F3. The presence of inclusions seems strongly related to negative F3 values and this could explain its discriminatory capacities from the quality point of view: in fact, for many of the analyzed slabs the association to first or to second choice is a function of the visible inclusions. For the Brown Gray type however, since the second choice slabs do not show any macroscopic defects, the discrimination is based on petrographical parameters since both choices belong to different petrographical classes. Therefore the differentiation can be based on those factors that mostly discriminate among the petrographical classes: that is F1 and, especially, F2. From the study of the factorial plane F1F3 (figure 3c) it is possible to see how, apart from the content in oxides, both texture and unhomogeneity influence the slabs’ coloring: a coarse grain and non homogeneity mean a less brightness while a medium-fine grain is brighter.

CONCLUSIONS This study distinguished the chemical and physical parameters that largely influence the esthetic qualities of a rock: its mineralogical composition, texture, unhomogeneity, presence of inclusions and color. Through the use of multivariate statistics, and particularly from the Multiple Correspondence Analysis (MCA) it has been possible to define and quantify the entity of the relationships among the parameters that can commercially characterize the slabs. By projecting the variables onto the factorial planes strong correlations between mineralogical composition, color and petrographical class were evidenced. The commercial type showed however a strong correlation with homogeneity, the presence of visible inclusions, color and petrographical class. The factorial variables are strongly representative of these relations: factor F1 can be considered as an index of the rock’s mineralogical composition and as a color index and it appears to be connected to characteristics of texture and non homogeneity of the slabs. F1 discriminates also among the various petrographical classes of the analyzed samples. Factor F2 seems to partially discriminate among the commercial types and the corresponding petrographical classes, while F3, thanks to its clear correlation with the characteristics of homogeneity of the slabs and with the presence of visible inclusions, operates an effective separation among the various commercial classes. REFERENCES  BRALIA A., GHEZZO C., GUASPARRI G., SABATINI G. – Aspetti genetici del batolite sardo-corso, Rend. Soc. It. Mineral. Petrol., 38/2, 701-764, 1983.  CALIA G.S., MARINI C., MASSOLI-NOVELLI R. – I graniti di Bitti (NU): potenzialità estrattive e limitazioni connesse all’uso del territorio, Geologia tecnica e ambientale, 1/92, pp.23-31, 1992.  CAPPELLI A., HOXHA I., VIOLO M. – The use of multivariate statistics for the characterization of the magnesite deposit of Euboea Island (Greece), Proceedings of the 5th annual conference of the International Association for Mathematical Geology, Trondheim (Norway), 6-11 August 1999, Edited by S.J.Lippard, A Naess and R.Sinding- Larsen vol. I, pp. 151-156, 1999.  CAPPELLI A., DI FIORE L. - The use of multivariate statistics in the commercial characterization of ornamental stones: a case study of the Botticino limestone (Brescia, Italy), Proceedings of the 4th annual conference of the International Association for Mathematical Geology, Isola d’Ischia, Italy, 5-9 October 1998, Edited by A.Buccianti, G.Nardi and R.Potenza, vol 1, pp 403-409, 1998.  DAVIS J.C.- “Statistics and data analysis in geology”, 2nd edition, J. Wiley & Sons, New York, (1986).  DE CARLO I., MANCA F., SIRIGU E. – I bacini minerari e i poli estrattivi delle pietre ornamentali della Sardegna, proceedings of the conference “le materie prime minerali sarde: problemi e prospettive”, , 23-24 June 1997, pp. 47-54, 1998.  FRISA MORANDINI A. - La qualificazione tecnica delle rocce ornamentali: normativa vigente e proposta di intervento, proceedings of the International Conference “La cava nel 2000”, Carrara, 29-30 May 1986, pp. 129-133, 1986.  SAPORTA G. -“Probabilités analyse des données et statistique”, Éditions Technip, Paris, 1990.