Studies On EDXRF Analysis Of Ancient Ceramics And Data Processing Methods

Chiese cultural relics could be divided into 3 parts: inorganic nonmetal cultural relics, metal cultural relics and organic cultural relics according to their material properties. The inorganic nonmetal cultural relics include ancient porcelain. About 10000 years' history has witnessed the manufacturing of porcelain in China. If wanting to understand them essentially, we must searching for the provenance and dynasty information based on chemical compositions, structures and properties of silicate materials and regard it as the basis for identifying cultural relics. So it is a very important approach for scientists to research ancient porcelains that examining and analyzing the chemical compositions of bodies and glazes, which comprises 2 aspects: (1) acquiring exact and creditable data; (2) processing these data scientiflcly.With the development of the element analysis technologies, the Energy-dispersive X-ray Fluorescence (EDXRF) technique is now well established as a powerful tool for qualitative and quantitative elements analyses when studying on the cultural relic materials. Being characteristic of quick and non-destructive analysis, involving high-precision, a wide concentration range, easy sample preparation, good reproducibility and synchronously analyzing the elements from Na(Zl 1) to U(Z92), it is appropriate to analyze precious ancient ceramics.In the study, we examined the chemical compositions of bodies and glazes of 92 ancient porcelain pieces from Guan Kiln, Yue Kiln, Longquan Kiln and Royal Palace site in Hangzhou in the Southern Song Dynasty. Then, 4 kinds of multivariate statistical analysis methods were used to analyze the major and minor element compositions of 78 ancient porcelain pieces, and the classification characteristic and discriminant functions were acquired. Then the method of SOM was introduced into the clustering analysis based on the major and minor element compositions of the bodies, the results manifested that 48 samples could be perfectly distributed into 3 provenances: Hangzhou, Cixi and Longquan. Because the major and minor element compositions of two Royal Kilns were similar to each other, the classification accuracy over them was merely 76.92%. In view of this, authors have made a SOM clustering analysis again based on the trace element compositions of the bodies, the classification accuracy rose to 84.61%. These results indicated that discrepancies in the trace element compositions of the bodies of the ancient ceramics excavated in two Royal Kiln sites were more distinct than those in the major and minor element compositions, which was in accordance with the fact. We argued that SOM could be employed in the clustering analysis of ancient ceramics. Then, 38 porcelain pieces from two Guan kilns in the Southern Song Dynasty were classified by the method of LS-SVM according to the chemical composition discrepancies of the major, minor and trace element in bodies and glazes. The classification effect was validated by the method of Leave-One-Out and compared with the SVM and SOM methods.The results show that the methods of SVM and LS-SVM are preferable to SOM in solving the classification problems on "small sample learning". Generally, the classification accuracy of SVM is higher than that of LS-SVM. However, the calculation of LS-SVM is quicker than that of SVM when running in MATLAB 7. For two Guan kilns, the chemical composition discrepancies in glazes are larger than those in bodies, and discrepancies of the trace element compositions are larger than those of the major and minor element compositions both in bodies and glazes.