Decoding paintings through their materials : enhancing Macro X-ray Powder Diffraction for artworks
Arthur Gestels
2025
Scientific research in cultural heritage has increasingly shifted toward non-invasive imaging methods that can reveal material information across entire artworks. Because paintings often exhibit complex stratigraphies and heterogeneous compositions, traditional point-based techniques and small-scale sampling frequently fail to provide representative insight into the full material structure of an artwork. In this context, hyperspectral imaging techniques have become essential tools for conservators, researchers, and art historians.
Macroscopic X-ray powder diffraction (MAXRPD) is one such hyperspectral technique. It enables direct identification of crystalline compounds, making it highly valuable for the study of pigments and degradation products in painted artworks. Despite its high chemical specificity, MAXRPD is limited by low spatial resolution and long acquisition times, restricting its applicability for large-scale investigations.
To address these limitations, this thesis investigates the combination of MAXRPD with faster hyperspectral imaging modalities, including reflectance imaging spectroscopy (RIS) and macroscopic X-ray fluorescence (MAXRF), using machine learning approaches to reduce acquisition time and improve efficiency. The research begins with a case study on Rembrandt’s The Night Watch, where MAXRPD was used to assess the material impact of an acid vandalism attack. The results revealed the formation of anglesite as a crystalline degradation product in the upper paint layers, demonstrating the value of compound-specific imaging for conservation assessment and treatment planning. Next, a proof-of-concept study on corroded metal plates showed that quantitative MAXRPD data could be combined with RIS data to train machine learning models capable of predicting material distributions with reasonable accuracy.
This demonstrated the feasibility of extrapolating material-specific information from faster imaging techniques. The methodology was then adapted to cultural heritage objects, enabling the generation of high-resolution, material-specific images of artworks. Case studies on an illuminated manuscript and oil paintings showed that fusing MAXRPD with RIS or MAXRF data produces reliable predictions.
Finally, the approach was applied to more complex historical paintings to assess its generalizability, demonstrating that accurate compound mapping is possible even in unscanned areas when representative training data are available. Together, these results establish a scalable framework for accelerating MA-XRPD-based imaging while preserving its analytical power.