In nature, lotus leaves, moth eyes, and butterfly wings have repellent characteristics that keep them free of fog, dirt, and water.
Recreating such valuable omniphobic qualities in glass, however, has always been a challenge — that is, until a team from the University of Pittsburgh turned to machine learning.
Using an automated probability theory known as Bayesian optimization, the Pitt researchers created an anti-fogging, liquid-resisting, and self-cleaning glass that they hope will soon support technologies like displays, laptops, tablets, and solar cells.
"We used machine learning to suggest variables to change, and it took us fewer tries to create this material as a result,” said Paul Leu, PhD, associate professor of industrial engineering, whose lab conducted the research.
The fabrication process produced randomized nanostructures on the top surface of the glass — an arrangement much like the glasswing butterfly, whose transparent wing features are smaller than the wavelengths of visible light.
Then, with the help of machine-learning models from San Francisco-based software company SigOpt, the team determined the optimal characteristics for the material. (Learn more about the process parameters in this SigOpt blog post .)
The surface demonstrated static water and ethylene glycol contact angles of 162.1 ± 2.0° and 155.2 ± 2.2°, respectively. The glass also exhibited resistance to condensation, an antifogging efficiency more than 90%, and a departure of water droplets smaller than 2 μm.
The nanostructured creation has self-healing properties as well. While abrading the surface with a rough sponge damages the coating, heating restores the glass to its original function.
The Pitt team envisions the glass being used in a variety of optical applications where self-cleaning, antifouling, and antifogging functionalities are essential. High-transparency reduces a display’s brightness and power demands, extending battery life, for example, and lower haze results in clearer images and text.
Findings from the study were recently published in Materials Horizons.
Dr. Leu and Sajad Haghanifar, lead author of the paper and doctoral candidate at Pitt, spoke with Tech Briefs about how their already "super" glass can get even better.
Tech Briefs: How is the nanostructured glass like the butterfly? What inspired this choice?
Sajad Haghanifar: The glass-wing butterfly has random sub-wavelength nanostructures on its wing, which provide for its remarkable anti-reflection properties across a wide variety of wavelengths and incidence angles. This transparency makes it difficult for predatory birds to track the butterfly during the flight. Inspired by the properties in these natural surfaces, this research focuses on creating random nanostructures in glass.
Tech Briefs: What role did machine learning play in the design effort? What could be accomplished with machine learning that couldn't be accomplished otherwise?
Haghanifar: The butterfly wings took millions of years to develop such properties. We took the challenge of creating these structures in glass by using machine learning. The fabrication of such random nanostructures involves many fabrication parameters, such as the flow rates of various gases, pressure, time, and electromagnetic field power. If we were to consider all the possibilities, this would involve over 1012 experiments, and each experiment can take 2 to 3 hours to complete. Machine learning accelerates the experimental design process, allowing us to quickly identify promising fabrication parameters and test them. With this active search approach, we were able create the supertransmissive, superclear glass in less than 70 experiments.
Tech Briefs: How would you describe the nanostructures being added? What are they exactly?
Dr. Paul Leu: The nanostructures are tiny glass structures smaller than the wavelength of light. They have somewhat random positions and sizes like those on the glasswing butterfly wing, which give them their superclear, supertransmissive properties. There are zoomed-in images provided on this blog post .
The nanostructures are not “added” on the glass but instead created by etching or removing material.
Tech Briefs: And where are they being placed on the glass? Is this part of the process difficult?
Dr. Leu: The nanostructures are etched on the surface of the glass. They can also be etched on both sides of the glass to provide even better antireflectivity. The process to create them involves a maskless etching method where the nanostructures can be created directly into the glass without the use of a mask.
Masked etching methods are commonly used, where some other material (like a metal or polymer) creates a pattern for the etching. Our maskless method saves both cost and time for fabrication.
Tech Briefs: What are the most exciting features of this nanostructured glass, and in what applications do you see these materials being most useful?
Haghanifar: Not only is the glass very transparent (over 99%) and very clear (with haze less than 0.1%), but it also has good antireflection properties over high incidence angles (less than 5% at 50º). Furthermore, it has additional functionalities such as liquid repellency, anti-fogging, and self-healing. The glass repels a wide variety of liquids such as water, oils, blood, milk, coffee, and ketchup. When the glass is abraded, it loses its liquid repellency, but by heating the glass up, it can self-heal and recover its excellent liquid repellency properties. There could be a wide variety of applications for this glass such as laptops, tablets, watches, displays, e-readers, and solar panels.
Tech Briefs: What's next regarding this research?
Haghanifar: We would like to scale up the fabrication process where we can create such nanostructures in very-large-area glass. We can create these structures currently on 4 x 4-inch areas, but we would like to be able to create them on larger areas that could be used in a large screen TV, windows, or solar panels. Also, we are considering different ways to enhance the mechanical durability of the glass.
What do you think? Have you used machine learning in your designs? Share your questions and comments below.