Peer-reviewed research: predicting ABS properties from fewer measurements

Milad Golkaram, co-founder and CSO of polySCOUT, is the corresponding author of a peer-reviewed paper in the Journal of Chemical Information and Modeling (volume 64, issue 19, 2024). He wrote it with Jonah Poort, Pieter Janssen and Jan Harm Urbanus. The paper is open access.

What the paper shows

Characterising a plastic grade means measuring many properties, which costs time and money. The authors use compressed sensing, a data reconstruction method from signal processing, to find a small set of properties worth measuring and to reconstruct the remaining ones from them. On a dataset of 42 ABS samples with 18 measured properties, six well chosen measurements were enough to reconstruct the other properties with an average error below 5 percent. With nine measurements the average error dropped below 3 percent.

Why it matters

Data on polymers is scarce, fragmented and expensive to produce. Methods that get more out of a few measurements are useful for materials such as ABS. The authors also note that the approach could be applied to recycled materials, whose properties are harder to predict.

Context

The work was carried out at TNO and Maastricht University. Trinseo provided the samples and their characterisation, and the project was funded by the Horizon Europe project ABSolEU on ABS recycling. It is separate from the polySCOUT platform, but it reflects the same focus on reliable property data for polymers.

Read the paper

Prediction of Acrylonitrile-Butadiene-Styrene Mechanical Properties through Compressed-Sensing Techniques, Journal of Chemical Information and Modeling, 2024, 64(19), 7257 to 7272. Open access under a CC BY-NC-ND 4.0 licence.