Machine Learning Identifies Effective Green Chemicals for De-Icers

Researchers have used a machine learning system to develop a new de-icer that outperforms commercial products while being more environmentally friendly. 

Researchers from Osaka Metropolitan University in Japan used machine learning to analyze and identify possible chemical agents that could be used in de-icing products.

The machine learning system identified a mixture that would melt ice on roads and runways while minimizing environmental impact.

Traditional de-icers contain chemicals that are harmful to the environment, including sodium and calcium chloride. When applied to ice, the melting runoff can contaminate soil and water bodies, damaging plants and aquatic life.

Chemicals found in de-icers can also damage infrastructure, including roads and bridges.

The researchers tasked the machine learning system with finding an organic, chloride-free solvent for use in a de-icer.

The AI was directed to consider factors such as density, viscosity, melting point and topological polar surface area when analyzing potential chemical compounds.

Using machine learning techniques like XGBoost and SHAP (SHapley Additive exPlanations), the AI identified key ice-melting mechanisms. Their SHAP analysis, for example, suggested different ice-melting factors and mechanisms for salts and organic solvents.

Related:MIT Uses AI to Find New Antibiotics to Kill Superbugs

The machine learning eventually settled on two: Propylene Glycol, an organic solvent that already sees use as a de-icer for airplanes and sodium formate, an organic salt that the AI identified as having a “low toxicity and minimal corrosive effect.”

The system ultimately suggested combining the two, allowing a de-icer to apply multiple ice-melting mechanisms in one product

The chosen chemicals were more environmentally friendly and when mixed and demonstrated higher ice penetration capacity than six commercial de-icers.

Publishing their findings in the journal Scientific Reports, the researchers wrote that machine learning “presents an effective strategy for creating powerful yet eco-friendly deicers.”

“The development of such a deicer with large ice penetration capacity allows for the reduction of de-icing agent usage, thereby decreasing environmental impact,” the paper reads. “Moreover, mixing salt solutions and organic solvents include can lower the concentrations of both components in the de-icer, further reducing environmental impact.”