AI helps ID paint chemistry of Berlin Wall murals
Shards that reveal secrets
Close examination of the fragments that were once part of the Berlin Wall artwork, labeled according to their blue, yellow or red color, reveals the brushstrokes, multiple layers and pigments used.
Image source: Adapted from F. Armetta et al., 2024
Non-destructive techniques such as Raman spectroscopy are often used to identify the molecular signatures of pigments, dyes and other compounds, but this often requires taking the sample to a laboratory. Handheld Raman equipment is used when analysis must be performed in the field, but is far less accurate than full-scale laboratory equipment. So Almeita et al. It was decided to use machine learning methods to improve the accuracy and sensitivity of the spectral data collected by these handheld devices.
The team collected 15 image fragments in five different colors from Berlin Wall paintings. They used a handheld Raman spectrometer on paint chips and compared the spectral data to a commercial pigment spectral library, confirming the findings through X-ray fluorescence and fiber optic reflectance spectroscopy.
Most pieces have two top layers and are painted with brushes rather than spray paint; in some cases the brushstrokes can be seen clearly under a microscope. The third layer below, in contact with the masonry, is white and was probably used to prepare the surface for painting.
Calcium and titanium are the most abundant elements in all samples. The green sample contained chromium and lead, which the author believes was mixed with another color to obtain a specific hue. Trace amounts of copper are also present in the blue and green samples.
Almeita et al. They also created samples of their own models by mixing commercial German acrylic paints (commonly used since the 1800s) in different proportions to try and match the color and tone of the fragments – a key message for the restoration. This is where their machine learning algorithm, called SAPNet, proves useful. They trained it on Raman spectroscopy data from Berlin Wall samples and used it to determine the percentage of pigments. The model concluded that Berlin Wall paint chips contained titanium white and up to 75% pigment.
“Only through a comprehensive evaluation of the results provided by all techniques can the majority of the fragment’s components be identified [combined]”, the authors conclude, the development of SAPNet further enhances this capability. “While SAPNet is specifically tailored for pigment mixture analysis, its powerful framework demonstrates the utility of deep learning methods in Raman spectroscopy analysis across different scientific and industrial The transformative potential of applications.
DOI: Journal of the American Chemical Society, 2024. 10.1021/jacs.4c12611 (About DOI).
2024-12-12 17:38:05