Machine learning for historical PVC analysis

Researchers from the Heritage Science Lab Ljubljana have successfully applied machine learning algorithms to the development of computational models for identification and quantification of plasticizers in PVC objects. This pioneering study on real objects demonstrates that non-destructive analysis of plasticizers in historic collections of degraded PVC objects is possible. Plasticizers play a key role in the degradation of PVC, so the study provides valuable data for collection managers.

The research was published in Nature’s journal Scientific Reports (

The researchers analysed a large collection of more than 100 historical and modern PVC objects using IR spectroscopy and chromatography. The objects differed in terms of plasticizer type and content, thickness, fillers, stabilizers and other additives, degradation stage and storage history, so they are considered representative of objects in heritage collections. The results of this work point at four important findings:

  • A good classification model was established to identify four plasticizers: DEHP, DOTP, DINP, DIDP, a mixture of DINP with DIDP, and unplasticized PVC.
  • Successful regression models were built for DEHP and DOTP, the most common plasticizers found in our collection of modern and historical PVC objects.
  • Overall, machine learning classification and regression models constructed using ATR FTIR spectra were more accurate and robust than those constructed using NIR spectra.
  • The effect of numerical differentiation of the spectra on classification accuracy was investigated.
Fig. 1. Schematic diagram of the research methodology.

You can read more in the paper »Machine learning-assisted non-destructive plasticizer identification and quantification in historical PVC objects based on IR spectroscopy« authored by Tjaša Rijavec, David Ribar, Jernej Markelj, Matija Strlič and Irena Kralj Cigić.

Leave a Comment

Your email address will not be published. Required fields are marked *