Applying multivariate analysis to X-ray diffraction computed tomography: the study of medieval applied brocades

Abstract

X-ray diffraction tomography is a well-developed technique to study the structure of heterogeneous materials which makes it a tool of choice for the non-destructive investigation of cultural heritage microsamples. Characterizing such complex materials with a high enough spatial resolution requires acquiring large amounts of diffraction images, followed by a complex sequence of data management to localize the crystalline phases in the sample. Here we propose to use multivariate analysis in order to automatically decompose the data in a small set of components, each representing the diffraction pattern of one or a small number of phases. This makes phase identification and quantification of each component much more efficient and leads to a quantitative knowledge of the phase content in each voxel of the tomographic reconstruction. We show that non-negative matrix factorization is very efficient for this purpose, with a computing time well compatible with in-line data analysis in order to assess the quality of measurements during experiments at synchrotron beamlines. Here, we apply the method to the investigation of microsamples from medieval applied brocade decors selected from a wooden statue, and the results are validated a posteriori by comparison with ex situ destructive techniques.