Street art takes many forms, and the vibrant murals on the Berlin Wall both before and after its fall are expressions of people’s opinions. But there was often secrecy around the processes for creating the paintings, which makes them hard to preserve. Now, researchers reporting in the Journal of the American Chemical Society have uncovered information about this historic site from paint chips by combining a handheld detector and artificial intelligence (AI) data analysis.

“The research highlights the powerful impact of the synergy between chemistry and deep learning in quantifying matter, exemplified in this case by pigments that make street art so captivating,” says Francesco Armetta, a co-author of the study.

To restore or conserve art, it’s important to collect information on the materials and application techniques. But the painters of the Berlin Wall didn’t document this. In previous studies of other historic artifacts, scientists brought fragments or even whole objects into the lab and, without destroying the samples, identified pigments on them using a technique known as Raman spectroscopy. Although handheld Raman devices are available for on-site investigations, they lack the precision of full-sized laboratory equipment. So, Armetta, Rosina Celeste Ponterio and colleagues wanted to develop an AI algorithm that could analyze the output of portable Raman devices to more accurately identify pigments and dyes. In an initial test of the new approach, they analyzed 15 paint chips from the Berlin Wall.

The researchers first magnified the chips and observed that they all had two or three layers of paint with visible brush strokes. The third layer in contact with the masonry appeared white, which they suggest is from a base coat used to prepare the wall for painting. Next, the researchers used a handheld Raman spectrometer to analyze the chips and compared them to spectra collected from a commercial pigment spectra library. They identified the primary pigments in the samples as: azopigments (yellow- and red-colored chips), phthalocyanins (blue and green chips), lead chromate (green chips) and titanium white (white chips). These results were confirmed with other non-destructive techniques, including X-ray fluorescence and optical fiber reflectance spectroscopy.

Then, the researchers mixed pigments from a commercial acrylic paint brand (used in Germany since the 1800s) with different ratios of titanium white, trying to match colors and the range of tints typical for painters. A knowledge of these ratios could help art conservators prepare the right materials for restoration, say the researchers. Using the mixtures’ handheld Raman spectral data, they trained a machine learning algorithm to determine the percentage of pigment. The approach indicated that the Berlin Wall paint chips contained titanium white and up to 75% of pigment, depending on the piece analyzed and according with the color tone. The researchers say these results indicate that their AI model could provide high-quality information for art conservation, forensics and materials science in settings where it’s hard to bring lab equipment to a site.