Applying machine learning to soft material design

May 23, 2023

(nanowork spotlight) Soft materials such as polymers, rubbers and hydrogels play an important role in our daily life. From car tires to the stretchy material in your favorite loungewear, from cutting-edge flexible electronics to wearable technology like e-tattoos, these versatile materials are everywhere. However, to effectively use these materials in a wide range of applications, their mechanical properties must be precisely tuned.

For example, wearable sensors such as electronic skins and tattoos need to be soft enough (scientifically speaking, they should have a relatively low Young’s modulus) to conform to the natural strains of the skin. It also needs to have the right amount of stretch to move with your body without causing discomfort. When used in soft robotics, these materials require different properties to mimic biocompatibility and biological design. Furthermore, when used in artificial skin, the material should be strong, durable and robust.

The key to achieving these unique properties lies in careful manipulation of aspects such as the polymer chain, monomer composition and intermolecular hydrogen bonding. However, coordinating these materials is not an easy task. It requires a deep understanding of material chemistry and multiple experimental trials. This process can present significant hurdles for end-users who require materials with specific mechanical properties for their application. A new approach to this design process is therefore important. Demonstration of soft material design system and design process. a) The process of designing soft materials using design systems. The system takes the desired mechanical properties as input and returns the experimental conditions in the form (X1, X2, and X3). b) Examples of an elastomer (b-1) with a high Young’s modulus and an elastomer (b-2) with a high strain at break (ie, high stretchability). (Reprinted with permission from Wiley-VCH Verlag)

The Rise of Machine Learning in Material Design

This is where the power of machine learning and materials informatics comes into play. These advances have greatly accelerated the materials discovery process. Machine learning algorithms can detect subtle patterns in datasets that are difficult to discern using human intuition alone. This feature enables reverse design of materials. This means using a set of desirable material properties to determine experimental parameters, which greatly accelerates the design process.

However, using machine learning models in experimental research comes with challenges. Collecting the large amount of high-quality experimental data required to train a model can be time-consuming and laborious. Fortunately, there are innovative strategies for collecting high-quality data with less effort, such as using archived lab notebooks and applying design of experiments techniques such as Design of Experiments (DoE). Appeared.

A data-driven approach to soft material design

One of the interesting developments in this field is the application of data-driven approaches for tuning the mechanical properties of soft materials. In recent studies, Advanced functional materialsIn (“A Data-Driven Approach to Tuning the Mechanical Properties of Soft Materials”), researchers at Stanford University demonstrated this approach using a common type of soft material, polyurethane (PU) elastomers.

The research team tuned the mechanical properties of the PU elastomer by changing the mixing ratio of the components. They collected data on the material’s mechanical properties such as Young’s modulus, strain at break, ultimate strength, and toughness. We used this data to train a machine learning model that predicts these traits based on the mixing ratio. Five samples were prepared for experimental validation. Desired and measured mechanical properties labeled “predicted” in the graph are displayed: Young’s modulus c), strain at break d), ultimate strength e), toughness f). Error bars for ‘measured’ data points indicate 1-standard deviation calculated from three measurements. (Reprinted with permission from Wiley-VCH Verlag)

The advantage of this method is that you can do a “reverse design”. You enter the desired mechanical properties and the model spit out a synthetic recipe to achieve those properties. The researchers tested this by making elastomer samples using these recipes and found that the mechanical properties obtained closely matched the input properties.

The researchers conclude that this data-driven approach to soft material design powered by machine learning can accurately predict and tune the mechanical properties of these materials with remarkably small data sets. This approach can provide soft materials with close to desirable properties by focusing on macroscopic structural information controlled by the synthesis recipe.

The success of this study may inspire further discussion between the materials research and artificial intelligence research communities. It may also facilitate the development of new algorithms specifically designed for small datasets, a common challenge in the field. Using a data-driven approach and machine learning allows us to explore different soft material systems and design processes more efficiently, bringing us one step closer to lab automation.

To

Michael
Burger

– Michael is the author of three books by the Royal Society of Chemistry: Nano-Society: Pushing the Boundaries of Technology, Nanotechnology: The Future is Tiny, Nanoengineering: The Skills and Tools Making Technology Invisible Copyright ©




Nanowork LLC