A Machine Learning-Based Approach to Discover Nanocomposite Films for Biodegradable Plastic Alternatives

The accumulation of plastic waste in the natural environment is of great concern, as it is contributing to the destruction of ecosystems and harming aquatic life. In recent years, materials scientists have thus been trying to identify all-natural alternatives to plastics that can be used to package or manufacture products.

Researchers at the University of Maryland, College Park recently devised a new method to explore alternatives to biodegradable plastics. Their proposed method is described in a paper published in Nature Nanotechnologycombines the latest machine learning techniques with molecular science.

“My inspiration for this research came from a 2019 visit to Palau in the western Pacific,” paper co-author Prof. Po Yin Chen told TechExplorer. “The impact of plastic pollution on the marine life there – plastic films tricking fish and sea turtles into mistaking plastic waste for food – was disturbing. It made me apply my expertise to this environmental issue. inspired me to come up with the car and focused my attention on finding a solution when setting up my research lab at UMD.”

Conventional and previously used methods for finding sustainable alternatives to plastics are time-consuming and inefficient. In many cases, they also give poor results, for example, identifying materials that are biodegradable but do not have the desired properties of plastics.

The innovative method for identifying plastic alternatives introduced in this recent paper relies on a machine learning model developed by Chen.

In addition to being faster than traditional methods of material discovery, this approach can be more useful in discovering materials that can realistically work in manufacturing and industry settings. Chen applied his machine learning techniques to the discovery of all plastic alternatives in close collaboration with his colleagues Teng Li and Liang Binghu.

“By combining automated robotics, machine learning, and molecular dynamics simulations, we accelerated the development of environmentally friendly, all-natural plastic alternatives that meet the required performance standards,” Chen explained. “Our integrated methodology combines automated robotics, machine learning, and active learning loops to accelerate the development of biodegradable plastic alternatives.”

First, Chen and colleagues compiled a comprehensive library of nanocomposite films derived from various natural sources. This was done using an autonomous pipetting robot, which can independently prepare laboratory samples.

The researchers then used this sample library to train Chen’s machine learning-based model. During training, the model gradually became more adept at predicting material properties based on their structure, through a process known as iterative active learning.

“The synergy of robotics and machine learning not only accelerates the discovery of natural plastic alternatives, but also allows the targeted design of plastic alternatives with specific properties,” said Chen. “Our approach significantly reduces the time and resources required compared to traditional trial-and-error research methods.”