piece of evidence
A closer look at the chips, labeled according to the colors blue, yellow and red that once belonged to art painted on the Berlin Wall, reveals handwriting, multiple layers and the pigments used .
Credit: Adapted from F. Armetta et al., 2024
Nondestructive techniques like Raman spectroscopy are often used to determine the molecular signatures of pigments, dyes, and other compounds, but this typically requires taking the sample to a laboratory. Handheld Raman instruments are used when analysis must be performed in the field, but they are much less accurate than full-sized laboratory instruments. Almetta et al. therefore decided to employ a machine learning approach to improve the accuracy and sensitivity of the spectral data collected by these handheld devices.
The research team collected 15 painting fragments in five different colors from the Berlin Wall paintings. They used handheld Raman spectroscopy on paint chips, compared the spectral data to a commercially available pigment spectral library, and confirmed their findings with X-ray fluorescence and fiber optic reflectance spectroscopy.
Most of the debris had two top layers painted with a brush rather than spray paint. In some cases, handwriting was clearly visible under the microscope. The third layer below, bordering the masonry, is white and was probably used to prepare the surface for painting.
The most abundant elements in all samples were calcium and titanium. The green sample contained chromium and lead, which the authors believe was mixed with another color to obtain a specific shade. The blue and green samples also contained trace amounts of copper.
Armetta et al. We also created our own mock-up samples by mixing commercially available German acrylic paints (commonly used since the 1800s) in various proportions to determine the color and color of the fragment, which is important information for restoration. I tried to match the shades. This is where their machine learning algorithm (called SAPNet) proved useful. They trained it on Raman spectral data from Berlin Wall samples and used it to determine the proportion of the dye. The model concluded that paint chips on the Berlin Wall contained titanium white and as much as 75% pigment.
“Identification of most of the components of the debris was only possible through a comprehensive evaluation of the results obtained by all techniques.” [combined]“While SAPNet is specifically tailored for pigment mixture analysis, its robust framework demonstrates the transformative potential of deep learning methods in Raman spectral analysis across a variety of scientific and industrial applications.” The authors concluded, further enhanced by the development of SAPNet.
DOI: Journal of the American Chemical Society, 2024. 10.1021/jacs.4c12611 (About DOI).
