Micro-SORS and data analysis routines for the non-invasive study of subsurface pigments degradation

In fifteen words or less, explain the significance of this contribution (Novel Aspect).: We explore the potential of data anlysis and machine learning routines for cultural heritage

Abstract Text:

Degradation processes in layered samples are among the greatest problematics in many scientific fields, and they are often complicated not only to prevent but also to study and mitigate. The extent of the decay is troublesome to identify in both quantity and extent, and it usually requires destructive analyses  to evaluate. Thanks to its high versatility and chemical specificity, Spatially Offset Raman Spectroscopy (SORS)^1,2 has proved effective for the non-invasive analysis of the subsurface of materials both in-situ and in laboratory environments, demonstrating its capability of retrieving signal from the inner parts of the analyte. Moreover, micro-SORS³ is particularly suited when dealing with micrometre scale decay processes, as it often happens in heritage science. A typical complication of these samples is provided by the compositional heterogeneity both in depth and on the surface; also, multiple unknown degradation substances can be found during the investigation, which could create complex signals that greatly differ from literature references. In this study, we will present the potential of micro-SORS and data analysis routines when dealing with pigment degradation processes. This research exploits micro-SORS mapping of artificially aged paint mock-ups consisting of multiple layers (aged and unaged) of different chemical composition; the maps are reconstructed via Python and allow the immediate visualization of the distribution of degradation products on and below the surface, in order to evaluate its extent. Since in case of cultural heritage samples visualization is key in order to interpret and understand the obtained results, micro-SORS maps are a simple but efficient tool to have a preliminary indication of the presence and distribution of known substances. Additionally, we will present the impact of machine learning routines for analysis and interpretation of micro-SORS data, which render the analyses more accurate and able to deal with large datasets. Most of all, they allow the “blind” investigation of unknown samples, thanks to its capability of retrieving the spectral differences between the unaffected substrate and the degraded surface of the sample. These outcomes are valid also for other scientific fields, such as forensic or biomedical, where data visualization and pattern identification can be relevant.