- Home
- Institut Néel
- Research teams
- Technical Groups & Services
- Work at the Institut
- Partnerships
The defence will be in French.
Abstract: The questions of when, where, how and by whom an object was made, as well as its evolution over time, have always been central to the work of art historians, curators and archaeologists. For several decades now, they have developed close relations with researchers in materials science and related disciplines, in order to expand their knowledge of the ancient skills of artists and craftsmen. However, there are two main obstacles to physical or chemical analysis: on the one hand, some artefacts cannot be moved because of their size or subject to potential damage during transport; on the other, the usual method of micro-sampling suffers from invasiveness and a lack of representativeness. Against this background, the main aim of this thesis is to design and develop methods for the analysis of works of cultural heritage. Our approach has been to combine, along a measurement chain, X-ray fluorescence spectroscopy (XRF) and X-ray powder diffraction (XRPD), whose complementarity ensures elemental and structural characterization of materials. Two bodies of work, the Grande Chartreuse manuscripts and the “applied brocades” of the former Duchy of Savoy, were used as case studies, enabling the implementation and development of these methodological aspects. A non-invasive approach using a mobile instrument, designed and built in-house then calibrated and optimized during this thesis, provided a preliminary insight into the materials aspects of the artworks and guided the collection of micro-samples to ensure their relevance. The latter, whose stratigraphy is crucial to understanding, were studied using high-resolution synchrotron imaging techniques (XRPD and XRF in tomography and mapping mode), which generate massive amounts of data. A huge part of the work in this thesis therefore concerned the development and testing of an original method for quantitative analysis of this data, using deep learning algorithms. We developed a method for generating training data and optimized a simple neural network to obtain the proportions of the crystalline phases at each point of the sample. The software resulting from this work is open-source and available to the community, and can naturally be used in other materials science contexts.
