Breaking up agglomerates with sound

Engineered nanoparticles (ENPs) have been used in commercial products, such as foodstuff, cosmetics, and personal care products for over two decades. Typical materials are zinc oxide (found in sunscreen), titania (toothpaste), ceria, and silica. These particles are usually produced, sold in bulk, and stored in a dry powder form. When stored this way, particles can form agglomerates - single particles clumping together. These powders must be transferred to a liquid by a process called dispersion for many applications. Ultrasonication is a widely adopted and standardized method to disperse powders in liquids and homogenize nanoparticle dispersions. It uses sound waves to break apart clumps of nanoparticles without changing their properties. However, the outcome is often unpredictable. To address this issue, Adolphe Merkle Institute researchers from the BioNanomaterials group investigated the use of machine learning to analyze the effects of ultrasonication and improve experimental design and reproducibility.

In a first step, four types of nanoparticles were considered, and supervised machine learning and dynamic light scattering were used to analyze the size of the nanoparticle aggregates after ultrasonication. The BioNanomaterials researchers relied on an experimental design that allowed them to extract maximum information from a small number of experiments. They varied the sonication parameters and particle properties to understand their effects on the degree of particle dispersion. The particles were also characterized using dynamic light scattering, which provided information about their size and polydispersity.

To establish a quantitative relationship between the input parameters and the output labels (particle size and polydispersity), the researchers trained a machine learning model to predict the outcome of ultrasonication based on various parameters, such as particle concentration, dispersion volume, sonicator type, duration of sonication, and particle properties like size and surface coating. The model was validated and tested using known data. A decision tree algorithm allowed the researchers to rank the importance of different parameters and gain insights into the underlying physical processes involved in ultrasonication.

In a second step, the researchers then performed a meta-analysis of published data on ultrasonication of different particle systems. They focused on nanoparticles commonly found in consumer products like zinc oxide, silicon dioxide, cerium dioxide, and titanium dioxide. This allowed them to compare the experimental machine-learning study results with a baseline of previous measurements.

“We observed a strong alignment between our predicted and simulated data,” explains PhD student Christina Glaubitz. “This outcome instills confidence in the predictive capabilities of our model.”

Overall, the study demonstrates the potential of machine learning to improve the reproducibility of nanoparticle dispersion. By understanding the complex relationships between input parameters and outcomes, researchers can optimize the ultrasonication process and avoid undesired changes in the nanoparticles' properties.

The research has the potential to be expanded further. “Our future research endeavors involve constructing a comprehensive database to develop a more inclusive model,” says Glaubitz. “This model will predict the behavior of a wider range of material types and more intricate dispersants.”

Reference: Glaubitz, C.; Rothen-Rutishauser, B.; Lattuada, M.; Balog, S.; Petri-Fink, A. Designing the Ultrasonic Treatment of Nanoparticle-Dispersions via Machine Learning. Nanoscale 2022, 14 (35), 12940–12950.