Using AI to predict how bioplastics behave under heat

A recent study by ANIPH researchers at the Agricultural University of Athens has demonstrated the potential of machine learning models to predict how PHB and PHBV bioplastics behave under heat. These bio-based, biodegradable plastics, produced through bacterial fermentation, can be challenging to process due to their brittle nature and chemical breakdown at temperatures close to their melting points.

The study, published in the journal Polymers, focused on predicting three key thermal properties of PHB/PHBV formulations: melting temperature, glass transition temperature, and crystallization temperature. To build their models, the team compiled a dataset of 572 data points, sourced from 109 scientific studies and 14 in-house experiments. This dataset included information on each sample's composition, molecular weight, and any additives used, such as plasticizers, fillers, or stabilizers.

Two machine learning methods were employed and compared, with the best models explaining a significant portion of the variation in the thermal properties between different formulations. The results showed that formulations with a higher proportion of the hydroxyvalerate building block tended to have lower melting and crystallization temperatures, as it disrupts the orderly packing of polymer chains.

Conversely, higher molecular weight tended to increase these temperatures, as longer chains become more entangled and harder to melt. The type and amount of additives also played a crucial role, consistent with their intended functions.

The study emphasizes the importance of carefully curated datasets and well-defined formulation ranges for the predictions to be reliable. The team has made the curated dataset and modeling code openly available, following ANIPH's commitment to open-data principles, enabling other researchers and manufacturers to leverage this work in designing bioplastic formulations with the desired processing characteristics.