Chemical Analysis and AI to Preserve Berlin Wall Murals | What are foundation models in generative ai | Google ai | Generative ai in banking mckinsey | Turtles AI

Italian scientists have developed a neural network to analyze spectral data from portable Raman spectroscopy, studying the chemical composition of paints used in Berlin Wall murals to support the conservation of historic street art.

Key Points:

• Using AI to improve the spectral analysis of complex pigments.
• Studying paint fragments to reconstruct the original materials of murals.
• Applying non-destructive methods to preserve the integrity of samples.
• Innovative potential of machine learning in the conservation of street art.

The conservation of street art, with its ephemeral nature and vulnerability to time and vandalism, represents a unique challenge for science and technology. The murals of the Berlin Wall, symbols of a crucial moment in twentieth-century history, are particularly precious and require innovative approaches to safeguard their integrity. A group of Italian researchers, led by Francesco Armetta of the University of Palermo, combined advanced spectroscopic analysis and machine learning to investigate the chemical composition of the paints used on the fragments of the wall, as reported in an article in the Journal of the American Chemical Society.

Street art paint materials, unlike classical masterpieces, are generally less durable and more complex. Modern acrylic paints are composed of intricate mixtures of pigments, binders, solvents, and chemical additives, often undocumented by the manufacturers or the artists themselves. To address this complexity, the researchers employed portable Raman spectroscopy devices, a non-destructive method ideal for on-site analysis. However, the use of portable devices entails a loss of precision compared to laboratory instruments. To overcome this limitation, the team developed a machine learning algorithm, called SAPNet, to optimize the interpretation of the spectral data.

Fifteen pictorial fragments of five different colors were collected from the Berlin Wall and subjected to Raman spectroscopy. The results were verified by X-ray fluorescence and fiber optic reflectance spectroscopy. The analyses revealed the presence of multiple layers of paint: a white base layer for surface preparation, overlaid by two colored layers applied with a brush. The predominant chemical elements were calcium and titanium, while traces of chromium and lead were identified in a green sample, suggesting the use of specific pigment mixtures to obtain certain shades. Pigments containing copper were found in the blue and green samples, highlighting the heterogeneity of the compositions.

In parallel, the researchers created simulated samples by mixing German acrylic paints to reproduce the observed colors and hues, with the aim of faithfully reconstructing the original techniques and materials. SAPNet was instrumental in analyzing the proportion of pigments in the samples, showing a predominance of titanium white and a pigment concentration of up to 75%. The synergy between spectroscopic methods and machine learning provided a detailed understanding of the chemistry of the materials, laying the foundation for more effective and precise restoration strategies.

The work demonstrates the important role that advanced technologies and AI can play in the preservation of contemporary artistic heritage, offering new perspectives to address the challenges of street art conservation.