AI streamlines search for recyclable food packaging polymers, study finds

Researchers from Cornell University in the US have streamlined the process of identifying chemically recyclable polymers using AI and machine learning. Suitable polymers can act as a “drop-in replacement” for existing food packaging monomers. 

The study, “AI-assisted design of chemically recyclable polymers for food packaging,” created an optimized workflow that identifies single and multilayer replacements for polymer-based packaging materials. 

It aims to find a replacement for conventional polymers used in food packaging that are difficult to recycle due to their complex chemical structures, such as PP, PE, and EVOH. 

“Food preservation is a contemporary challenge that requires materials providing protection and sustainability. Appropriately optimized polymeric materials may serve as effective alternatives to conventional solutions,” say the researchers. 

“Present-day packaging plastics often rely on a multilayer architecture. While effective, these materials are notorious for their persistence in landfills, fragmenting into microplastics that contribute to long-term environmental pollution.”

Moreover, the researchers highlight that multilayer polymer composition can pose a “significant obstacle” to recycling since chemically distinct layers must be separated — “a time and resource intensive process.”

Identifying sustainable polymers

As part of recyclable polymer identification, machine learning models predicted eight key properties that a recyclable polymer needs to meet. 

The eight performance properties identified by the researchers include: tensile strength, flexibility, enthalpy of polymerization, oxygen and water vapor permeability, and degradation, melting, and glass transition temperature.

Enthalpy of polymerization is the amount of energy released or absorbed when monomers chemically bond together to form a polymer. It is a “critical metric for chemical recyclability.”

The machine learning model found 7.4 million feasible polymers that met all eight requirements. The researchers then screened these potential polymers using AI predictions, based mostly on the enthalpy of polymerization. 

After reducing the number of potentially suitable polymers, the researchers further experimented on poly-p-dioxanone (poly-PDO), an existing polymer that has not yet been considered as a food packaging material. 

Recyclable alternatives

The study reveals that experimental validation of poly-PDO showed promising results as a chemically recyclable alternative to conventional food packaging monomers, such as PP, PE, and EVOH. 

The results showed that poly-PDO has water vapor barriers that meet packaging targets, as well as thermal properties that closely meet the AI performance indicators. 

Moreover, it shows “reasonable” mechanical performance, though experimental values were lower than predicted, highlighting the need for further optimization.

Most importantly, poly-PDO exhibits high chemical recyclability, achieving a monomer recovery rate of over 95% within six hours. 

The researchers conclude: “The strong agreement between our experimental measurements and machine learning predictions for poly-PDO affirms the robustness of our predictive models.”

The mechanical properties showed discrepancies between predicted and measured values. This is said to highlight the need for further optimization of polymer samples and of the machine learning models.

“Addressing such shortcomings will be critical to extend both experimental optimization and predictive reliability, in a quest to design truly useful and practical polymers for a sustainable world.”