CHEMOMETRIC METHOD FOR THE AUTOMATED IDENTIFICATION OF CUCUTENI CERAMICS BASED ON ATR-FTIR SPECTRA

Transcription

1 European Journal of Science and Theology, April 213, Vol.9, No.2, CHEMOMETRIC METHOD FOR THE AUTOMATED IDENTIFICATION OF CUCUTENI CERAMICS BASED ON ATR-FTIR SPECTRA Mirela Praisler 1, Daniela Domnisoru 1 and Leonard Domnisoru 2 1 Dunarea de Jos University of Galati, Faculty of Sciences and Environment, Department of Chemistry, Physics and Environment, Domneasca Street 47, RO 88, Galati, Romania 2 Dunarea de Jos University of Galati, Faculty of Naval Architecture, Domneasca Street 47, RO 88, Galati, Romania Abstract (Received 1 May 212) A chemometric software application was developed in order to obtain an automatic identification of original Cucuteni ceramic artifacts and distinguish them from fake ceramic samples similar to those found on the black market. The software application is based on Principal Component Analysis (PCA), a method of artificial intelligence that allowed us to build multivariate models that can be used for the efficient spectral discrimination of genuine Cucuteni ceramic samples. The spectra obtained by Fourier Transform Infrared Spectroscopy with ATR-Attenuated Total Reflection (FTIR-ATR) for a set of Cucuteni ceramic samples discovered in archaeological sites from Moldova - Romania were used as input database for the detection system. Keywords: Cucuteni ceramics, FTIR-ATR technique, Principal Components Analysis 1. Introduction The Cucuteni-Trypillian culture, also known as Cucuteni culture (from Romanian), Trypillian culture (from Ukrainian) or Tripolie culture (from Russian), is a late Neolithic archaeological culture which flourished between ca. 5 B.C. and 27 B.C., from the Carpathian Mountains to the Dniester and Dnieper regions in modern-day Romania, Moldova, and Ukraine regions. Ancient ceramic objects are usually considered the most specific indicators of each civilization because they are very resistant and thus maintain their aesthetic characteristics for long periods of time. In the case of the Cucuteni culture, the ceramic samples are the most important archaeological artefacts, as this culture became famous particularly due to the surprisingly beautiful pottery its people produced. On the other hand, the genuine and well-developed aesthetic sense of artistry used in decorating the clayware makes the Cucuteni ceramic artefacts so wanted on the black market of archaeological objects. In addition, due tot the Mirela.Praisler@ugal.ro

2 Praisler et al/european Journal of Science and Theology 9 (213), 2, high interest and demand for Cucuteni ceramic artefacts, fake ceramic samples imitating genuine Cucuteni ceramic artefacts are often encountered (Figure 1). The aim of this study was to determine objective criteria to be used for the characterization and identification of Cucuteni ceramic artefacts for forensic and security reasons. The research is focused on the description and selection of physico-chemical characteristics of Cucuteni ancient ceramics that can be used in order to classify ceramics objects in authentic and fake samples [1]. The positive identification of such ceramic samples was performed based on a nondestructive, sensitive, selective and fast analytical technique, i.e. Fourier Transform Infrared Spectroscopy with Attenuated Total Reflection [2], [Bruker, Bruker Optics FTIR Spectrometer System Tensor 27 IR, ATR Golden Gate, Opus Software, 29, available from: The spectral data was organized in a FTIR database that was then processed by using a flexible artificial intelligence technique, i.e. Principal Component Analysis (PCA) [3, 4]. Modern archaeometry includes the use of artificial intelligence methods, which allow us to process simultaneously databases including a very large number of physico-chemical or spectral variables and derive conclusions that are otherwise so time consuming to obtain [5]. 2. Experimental procedure FTIR spectroscopy is an important tool for the analysis of ceramic materials. The absorptions are associated with the vibrations of atom molecules excited by irradiation with infrared light. Each molecule or chemical group absorbs infrared light at specific wavenumbers, the spectrum being a fingerprint of the absorbing molecular structure [2]. (a) (b) (c) (d) (e) (f) Figure 1. (a-c) Cucuteni ceramic objects discovered in Iasi County, Romania; (d-f) imitations of Cucuteni ceramics. The spectra of 7 ceramic samples were recorded with a Bruker TENSOR 27 IR FTIR-ATR Spectrometer. A number of samples are authentic Cucuteni ceramic artefacts and 2 samples are imitations obtained by experimental 2

3 Chemometric method for the automated identification of Cucuteni ceramics archaeology in the laboratory. In order to record a FTIR spectrum, a small amount of powder (1-3 grams) was collected from the surface of the ceramic sample, and then the powder was placed on to the diamond radiation area of the ATR Golden Gate device. The spectrometer collects the spectral data and turns the interferogram in numeric format, then it performs the Fourier Transformation (FT) of the analytical signal and yields the FTIR spectrum file [Bruker, Bruker Optics FTIR Spectrometer System Tensor 27 IR, ATR Golden Gate, Opus Software, 29, available from: The procedure is very fast, lasting 1-2 seconds, with fully automatic spectrum recording. Each spectrum was recorded twice, once on each side of the ceramic samples. A number of 24 scans were recorded for each sample in the 4 to 5 cm -1 spectral range, using the reflection mode with a resolution of 4 cm Results and discussion Representative FTIR-ATR spectra of authentic and fake ceramic samples are presented in Figure 2, in absorbance vs. wavenumber (5 4 cm -1 ) format. We can see that the selectivity of the FTIR-ATR spectroscopy is extremely helpful in discriminating authentic Cucuteni ceramic samples from their imitations, based on the intensity of the main peaks. The authentic samples are characterized by a strong (partially overlapped) pair of peaks found around 24 cm -1, characterized by an absorbance of.15, while the same pair of peaks have an absorbance of only.1-.8 in the spectra of the imitating ceramic samples. On the other hand, the large peaks showing around 9 cm -1 have a medium to low intensity (.8.4) in the spectra of Cucuteni ceramic objects, and a very strong intensity (.14.22) in the spectra of the fake ceramic samples. In other words, the high selectivity of the FTIR-ATR technique recommends it as a powerful tool that can be used to distinguish between Cucuteni ceramic objects and imitations without any further spectra processing. However, the exquisite sensitivity of the FTIR-ATR spectrum shape to very small variations in the structure and composition of the sample becomes in this case a disadvantage. The variation in peak intensity encountered in the spectra of authentic Cucuteni samples becomes a challenge in recognizing a Cucuteni sample as a true one, i.e. in classifying a Cucuteni sample as such, based on its FTIR-ATR spectrum. These intensity variations appear due to the fact that the manufacturing process used at the time of the Cucuteni culture (5-27 BC) was rudimentary. As a result, variations in manufacturing parameters, such as temperature, yielded variations in the structure and composition of the pottery [6]. The same variations are observed in the spectra of the ceramic samples obtained by experimental archaeology (see Figure 2 c, d), in which the clayware manufacturing techniques are kept as close to the most probable original ones as possible. In conclusion, the sensitivity of the FTIR- ATR technique becomes in our application a challenge for the capacity of FTIR- ATR spectra to recognize a true Cucuteni sample as such (true positive), or a fake sample imitating Cucuteni ceramics (true negative). 1

4 Praisler et al/european Journal of Science and Theology 9 (213), 2, A [u.rel.] absorbanta.16 ATR-FTIR Spectru : Proba A12INT / A_SV A [u.rel.] absorbanta.16 ATR-FTIR Spectru : Proba A28EXT / A_SV [cm -1 ] 4 (a) [cm -1 ] 4 (b) A [u.rel.] absorbanta ATR-FTIR Spectru : Proba FA54INT / FA_SV [cm -1 ] 4 (c) A [u.rel.] absorbanta ATR-FTIR Spectru : Proba FA63EXT / FA_SV [cm -1 ] 4 (d) Figure 2. FTIR-ATR spectra of two: (a, b) authentic Cucuteni ceramic samples; (c, d) fake ceramic samples. In order to diminish as much as possible the rate of classifying Cucuteni artefacts as false negatives or of classifying imitating ceramic objects as false positives, artificial intelligence was used to process the data obtained by ATR- FTIR spectroscopy. A successful multivariate method that has been applied for the identification of ceramics belonging to various cultures or ages is Principal Component Analysis (PCA). This data processing method is based on the eigen values and vectors of the multivariable covariance matrix. It allows the extraction of the relevant information present in large experimental databases, through the decomposition on principal components (PC) resulting eigen vectors, according to the significance of the eigen values (principal components variability) [7, 8]. In order to characterize the authentic ceramic samples with PCA, we have build a spectral database consisting of 12 ATR-FTIR spectra, representing 1789 variables, i.e. the absorptions measured in the range 5 4 cm -1 at 1.93 cm -1 apart. PCA has been carried out with the XLSTAT software [XLSTAT Addinsoft Program User Guide (educational licence), 29, available from accessed: 21/4/211], developed as a Visual Basic application under Microsoft-Excel environment. A number of 99 eigenvalues were used for the decomposition on significant principal components in the case of the authentic ceramic samples and 42 eigenvalues for the fake ceramic samples. 2

5 Chemometric method for the automated identification of Cucuteni ceramics Table 1. Eigenvalues of the covariance matrix for the first principal components in the case of the Cucuteni ceramic samples. PCA PC1 PC2 PC3 PC4 PC5 Eigenvalue Variability (%) Cumulative % PCA PC6 PC7 PC8 PC9 PC1 Eigenvalue Variability (%) Cumulative % Table 1 presents the first ten eigenvalues of the covariance matrix, PCs and the corresponding variability, for true ceramic samples. The diagram of first five PCs obtained for the 12 spectra characterizing the Cucuteni ceramic samples is presented in Figure 5. We can see that the cumulated explained variance increases significantly up to the 3 rd PC (PC1 51.6%, PC2 32.4%, PC3 7.4%), for higher order PCs the explained variance of all individual PCs dropping down under 4%, practically fading out in significance for PCA data decomposition. In conclusion, only PC1, PC2 and PC3, which are cumulating % of the information, were kept for further investigation. Figures 3b, c and d present the projections of the spectra of the Cucuteni ceramic sample onto the main three PCs. In order to characterize the fake ceramic samples PCA, we have build up a database formed with 42 ATR-FTIR spectra by using the same variable distribution, i.e spectral variables (absorptions measured at wavenumbers in the range 5 4 cm -1, 1.93 cm -1 apart). Table 2 presents the first ten eigenvalues of the covariance matrix and the corresponding variability for each PC in the case of fake ceramic samples. Figure 4a presents the amount of spectral information imbedded in the first five principal components (PC). In the case of fake ceramic samples, the first three PCs are also explaining most of the information contained by the FTIR-ATR spectra (PC1 4.4%, PC2 37.7%, PC3 9.%), the rest of the PCs bringing little valuable information to the system. Still we should notice that the variability in the spectra of fake samples is larger than in the case of true Cucuteni samples. The first three PCs account for a cumulative explained variance of only 87.14% in the case of fake samples, i.e. less than in the case of Cucuteni ceramic samples. The projections of spectra of fake ceramic sample onto the main three principal components are presented in Figures 4b, c, d. Comparing Figures 3b and 4b we can notice that the scores of the Cucuteni ceramic artifacts are grouped around PC1, in the up-left and down-right quadrants, while the scores of the fake ceramic samples are scattered in all PC1-PC2 projection area. 3

6 PC3 (7.4 %) PC3 (7.4 %) Eigenvalue Cumulative variability (%) PC2 (32.47 %) Praisler et al/european Journal of Science and Theology 9 (213), 2, ATR-FTIR PCA ANALYSIS - TRUE CERAMIC SAMPLES 1 ATR-FTIR PCA ANALYSIS - TRUE CERAMIC SAMPLES Observations (axes PC1 and PC2: 84.4 %) A23ext A15ext 9 A39pigm 8 8 A43int A32int axis PC1 PC2 PC3 PC4 PC5 (a) A6ext A3ext A29ext A31int A12ext A31ext A12int Aint A33ext A13int A35int A36int A61int A28ext A6int A34ext A24ext A37ext A24int A27ext Aext A27int A26ext A35ext A44int A37int A61ext A13ext A4ext A4int A21int A34int A19int A22int A26int A39ext A43ext A6ext A45ext A33int A15int A19ext A1pigm A3int A21ext A8ext A1ext A36ext A4ext A2int A23int A16int A28int A8int A29int A16ext A22ext A58ext A59ext A38int A62ext A62int A44ext A45int A58int A6int A59int A2ext A41int A39int A1int A4int A1ext A3ext A41ext A38ext A3int A42ext A42int A14int A14ext A1int A17ext A9int A18ext A7ext A9ext A7int A18int A17int A5ext A32ext A2ext A2int A5int A11int A11ext - - PC1 (51.58 %) (b) ATR-FTIR PCA ANALYSIS-TRUE CERAMIC SAMPLES Observations (axes PC1 and PC3: %) ATR-FTIR PCA ANALYSIS-TRUE CERAMIC SAMPLES Observations (axes PC2 and PC3: %) - A6ext A22ext A11ext A18int A11int A17int A22int A21ext A7int A4ext A4int A15int A3ext A8ext A43ext A24int A27ext A27int A21int A13int A24ext A12ext A16int A5ext A7ext A2int A8int A13ext A17ext A18ext A34int A6extA6int A19int A35ext A44int A37int A61ext A23int A33ext A34ext A35int A36int A61int A28ext A37ext Aext A29ext Aint A29int A19ext A28int A43int A2ext A16ext A2ext A2int A26int A38int A39pigm A31ext A58ext A31int A33int A45int A45ext A58int A39ext A26ext A62ext A62int A12int A6int A3int A59ext A41ext A59int A38ext A15ext A44ext A4int A1pigm A39int A41int A4ext A23ext A1ext A1int A3int A5int A1int A1ext A14ext A3ext A42ext A36ext A14int A32ext A9ext A9int A42int - A11ext A11int A18int A17int A22ext A7int A21ext A22int A5ext A7ext A16int A8ext A15int A4ext A4int A18ext A17ext A21int A24int A27ext A24ext A3ext A12ext A2ext A2int A2ext A29int A43ext A8int A2int A13ext A27int A23int A16ext A28int A19ext A34int A19int A35ext A44int A61ext A37int Aext A13int A28ext Aint A29ext A38int A6ext A34ext A37ext A61int A33ext A35int A36int A6int A26int A31ext A31int A33int A38ext A58int A6int A3int A26ext A12int A41ext A45int A59int A58ext A62int A62ext A39ext A4int A39int A41int A44ext A59ext A45ext A4ext A1pigm A1ext A5int A14ext A1int A3int A6ext A1int A3ext A1ext A42ext A36ext A32ext A14int A9ext A9int A42int A43int A39pigm A15ext A23ext A32int A32int PC1 (51.58 %) (c) (d) Figure 3. (a) The diagram of first significant five PC for 12 spectra of Cucuteni ceramic samples; (b) Score plot PC2 vs. PC1 characterizing the spectra of Cucuteni ceramic samples; (c) Score plot PC3 vs. PC1 characterizing the spectra of Cucuteni ceramic samples; (d) Score plot PC3 vs. PC2 characterizing the spectra of Cucuteni ceramic samples PC2 (32.47 %) Table 2. Eigenvalues of the covariance matrix for the first principal components in the case of the fake ceramic samples. PCA PC1 PC2 PC3 PC4 PC5 Eigenvalue Variability (%) Cumulative % PCA PC6 PC7 PC8 PC9 PC1 Eigenvalue Variability (%) Cumulative %

7 PC3 (9.3 %) PC3 (9.3 %) Eigenvalue Cumulative variability (%) PC2 (37.66 %) Chemometric method for the automated identification of Cucuteni ceramics On the other hand, by comparing Figures 3c and 4c, we can observe that the scores of the Cucuteni ceramic artefacts are grouped around PC1 in centre of the PC1-PC3 projection area and in the down-right quadrant, while the scores of the fake ceramics are grouped in the up-right and down-left quadrants. Figures 3d and 4d indicate that the scores of the Cucuteni ceramic artefacts are grouped around PC1, in the central and down-left quadrant, while the scores of the fake ceramic samples are scattered in the whole PC2-PC3 projection area. ATR-FTIR PCA ANALYSIS- FALSE CERAMIC SAMPLES 8 1 ATR-FTIR PCA ANALYSIS- FALSE CERAMIC SAMPLES Observations (axes PC1 and PC2: 78.8 %) FA46int FA7int FAext FA51int FA51ext FA64int FA55int FA65int FApigm FAint FA56ext FA7ext FA56int FA52ext FA53ext FA65ext FA54ext FA68ext FA69ext FA57ext FA69int FA54int FA57int FA55ext FA46pigm FA52int FA46ext FA68int FA64ext FA66ext FA48int FA63int FA48ext 2 1 axis PC1 PC2 PC3 PC4 PC5 (a) 2 - FA49int FA63ext FA67ext FA53int FA49ext FA47int - FA66int FA67int - - FA47ext PC1 (4.41 %) (b) ATR-FTIR PCA ANALYSIS-FALSE CERAMIC SAMPLES Observations (axes PC1 and PC3: %) ATR-FTIR PCA ANALYSIS-FALSE CERAMIC SAMPLES Observations (axes PC2 and PC3: %) - FA55int FA47int FA52int FApigm FA64ext FAint FA46ext FA51ext FA7ext FA52ext FA54int FA65ext FA53int FA46pigm FA47ext FA51int FA55ext FAext FA67int FA54ext FA49ext FA68int FA7int FA57ext FA65int FA68ext FA56ext FA57int FA64int FA53ext FA56int FA69ext FA67ext FA69int FA66int FA63ext FA49int FA46int FA63int FA66ext FA48ext FA48int FA47ext FA67int FA49ext FA66int - FA47int FA53int FA52int FA64ext FApigm FAint FA54int FA46ext FA46pigm FA65ext FA52ext FA7ext FA55ext FA51ext FA51int FAext FA54ext FA46int FA68int FA63ext FA49int FA57ext FA57int FA68ext FA67ext FA69ext FA69int FA48ext FA63int FA56ext FA65int FA53ext FA56int FA64int FA66ext FA48int FA55int FA7int PC1 (4.41 %) (c) (d) Figure 4. (a) The diagram of first significant five PC for 42 spectra of fake ceramic samples; (b) Score plot PC2 vs. PC1 characterizing the spectra of fake ceramic samples; (c) Score plot PC3 vs. PC1 characterizing the spectra of fake ceramic samples; (d) Score plot PC3 vs. PC2 characterizing the spectra of fake ceramic samples PC2 (37.66 %) 4. Conclusions FTIR-ATR spectroscopy is a powerful technique for the characterization of the origin of ceramic samples. The analytical differences between Cucuteni ancient ceramic samples (positives) and fake ceramic samples (negatives) obtained by FTIR-ATR spectroscopy allow an efficient discrimination of the two classes of objects by visual inspection. However, in the particular case of ancient ceramic 5

8 Praisler et al/european Journal of Science and Theology 9 (213), 2, artefacts, significant differences in peak intensity appear due to instabilities in the manufacturing process (e.g. temperature control). This study has demonstrated that the problem can be successfully overcame. Further processing the FTIR-ATR spectral information can decrease the rate of false positives (fake ceramic samples identified as Cucuteni ancient ceramics) or false negatives (Cucuteni ancient ceramics identified as fake ceramic samples) significantly. Performing a PCA analysis is the key to obtain archaeological criteria to positively identify an unknown sample and assign its class identity. The criterion of sensitivity will certainly be improved by enlarging the sample collection and associated spectral database with ceramic artefacts of different cultures, ages, geographical origin, manufacturing technique or structure. We should also notice that the proposed combination of methods can be successfully applied for the evaluation of the origin or the quality of industrially present-day manufactured ceramics. Reliable identification criteria based on FTIR-ATR spectroscopy coupled with artificial intelligence techniques such as PCA can be very useful for consumer protection or forensic tests. Acknowledgements This study has been performed within the research project no. 8141/27-21 entitled Archaeometric expert system for the intelligent fight against the trafficking of cultural and historical heritage, financed by the PARTNERSHIP Program of the 2 nd National Plan for Research, Development and Innovation of the Romanian Government. References [1] C.R. Brundle, C.A. Evans and W. Wilson, Encyclopaedia of Materials Characterisation, Butter-worth-Heinemann, Boston, 1992, [2] B. Stuart, Infrared Spectroscopy. Fundamentals and Applications, John Wiley & Sons, New York, 24, [3] G.E. Benedetto, B. Fabbri, S. Gualtieri, L. Sabbatini and P.G. Zambonin, J. Cult. Herit., 6 (). [4] P. Fermo, E. Delnevo, M. Lasagni, S. Polla and M. de Vos, Microchem. J., 88 (28) 1. [5] L.F. Bubuianu and R. Diaconescu, Eur. J. Sci. Theol., 5(2) (29) 37. [6] L. Moraru and F. Szendrei, Eur. J. Sci. Theol., 6(2) (21) 69. [7] S. Gosav, M. Praisler and M.L. Birsa, Int. J. Molec. Sci., 12(1) (211) [8] S. Gosav, M. Praisler, D.O. Dorohoi and G. Popa, J. Molec. Struct., ()