Researchers Develop i-Melt: A Deep Neural Network That Can Predict Glass Quality Based On Melt Composition

Glass can be found all around us. It’s in our computer screens, next-generation batteries, medical implants, and even volcanoes.

Glass is manufactured by melting something and then swiftly cooling it. The chemical makeup of the molten liquid determines the physical qualities of glass. By alternating these properties, it is possible to control the glass’s strength and clarity, how it responds to chemicals, and whether it conducts electricity by adjusting this composition.

We can create a glass that matches our needs for any specific purpose, such as a new smartphone screen. However, this has traditionally been done through an expensive and time-consuming process of trial and error. Machine learning has the potential to revolutionize glass design.

Researchers from Université de Paris, Australian National University, Carnegie Institution for Science, and Durham University introduce a physics- and chemistry-informed deep neural network called i-Melt, that can effectively predict glass qualities based on melt composition. It is a grey-box model that embeds established physical equations and thermodynamic principles alongside learning correlations. This assures relevant and realistic outcomes. 

The researchers use a database of experimental observations to train i-Melt. The findings suggest that the model that can predict a wide range of properties, including melt viscosity, glass density, optical refractive index, and glass Raman spectra.

i-Melt is built using PyTorch. The framework’s vast flexibility and power made it simple to create models that included physical equations and machine learning, as well as to experiment with and test a variety of approaches. The ability to easily transition between GPU and CPU-based computations allowed the team to take advantage of the new DANTE computing cluster at IPGP — the University of Paris, which is equipped with A100 NVidia GPUs.

The i-Melt prototype focuses on glasses produced from four different ingredients: sodium, potassium, aluminum, and silicon oxides. The viscosity of molten glass varies significantly within this system (from -2 to 14 log Pas). The results show that i-Melt successfully predicts it with a precision of better than 0.4 log units. 

i-Melt can be used for a variety of purposes such as:

  • To assist in developing materials with specific qualities and the optimization of industrial processes.
  • To learn about volcanoes. Some volcanoes have tremendously explosive eruptions. Others are ‘effusive,’ with extensive lava flows but no explosions. The molten rock’s viscosity (or ‘runniness’) appears to be the driving force behind this variance. When the viscosity of the lava is low, it flows smoothly, allowing any trapped gases within the molten rock to escape; when the viscosity is high, the gas bubbles become trapped, eventually exploding. However, minor changes in magma composition significantly impact how the melt behaves. The team’s study suggests that as the content of silicic lava changes, its molecular structure changes abruptly, forming a densely connected network. As a result, the magma becomes more resistive to flow, resulting in a step-change in viscosity. While additional research is needed to completely comprehend the processes at hand, scientists now have a better sense of where to investigate.

The team has so far worked with the ‘aluminosilicates’ class of glass, which has oxides of aluminum and silicon as their main ingredients). They’re expanding i-Melt to a larger spectrum of compositions as a result of their successes, including oxides of magnesium and calcium, which are essential in both volcanology and glass research. In addition to this, they aim to integrate more physical properties into the predictions. They also plan to use PyTorch’s versatility to investigate other machine learning methodologies, such as Gaussian Processes, by combining GPyTorch and Pyro’s capabilities.

Reference: https://medium.com/pytorch/from-windows-to-volcanoes-how-pytorch-is-helping-us-understand-glass-8720d480f4f2

Paper:https://www.sciencedirect.com/science/article/pii/S0016703721005007?via%3Dihub