Machine learning used to identify quality graphene

Wednesday, 02 September, 2020

Machine learning used to identify quality graphene

Engineers at Monash University have developed technology that should help industry identify and export high-quality graphene cheaper, faster and more accurately than current methods, helping graphene and graphene oxide manufacturers boost the reliability of their supply. Their work has been described in the journal Advanced Science.

Discovered in 2004, graphene is touted as a wonder material for its outstanding lightweight, thin and ultraflexible properties. It is produced through the exfoliation of graphite — a crystalline form of carbon with atoms arranged hexagonally, which itself comprises many layers of graphene.

“Graphene possesses extraordinary capacity for electric and thermal conductivity,” said study leader Professor Mainak Majumder, from Monash University’s Department of Mechanical and Aerospace Engineering and the ARC Research Hub for Graphene Enabled Industry Transformation. “It is widely used in the production of membranes for water purification, energy storage and in smart technology, such as weight loading sensors on traffic bridges.”

However, the translation of this potential to real-life and usable products has been slow. One of the reasons is the lack of reliability and consistency of what is commercially often available as graphene.

“Graphene is rather expensive when it comes to usage in bulk quantities,” Prof Majumder said. “One gram of high-quality graphene could cost as much as $1000; a large percentage of it is due to the costly quality control process. Therefore, manufacturers need to be assured that they’re sourcing the highest quality graphene on the market.”

The most widely used method of producing graphene and graphene oxide sheets is liquid phase exfoliation (LPE). In this process, single-layer sheets are stripped from their 3D counterpart such as graphite, graphite oxide film or expanded graphite by shear forces. But this can only be imaged using a dry sample (ie, once the graphene has been coated on a glass slide). Manufacturers can therefore only detect the quality and properties of graphene used in a product after it has been manufactured.

“Although there has been a strong emphasis on standardisation guidelines of graphene materials, there is virtually no way to monitor the fundamental unit process of exfoliation; product quality varies from laboratory to laboratory and from one manufacturer to other,” said Dr Mahdokht Shaibani, a co-author on the study.

“As a result, discrepancies are often observed in the reported property-performance characteristics, even though the material is claimed to be graphene.”

Using a quantitative polarised optical microscope, the researchers identified a technique for detecting, classifying and quantifying exfoliated graphene in its natural form of dispersion. To maximise the information generated from hundreds of images and large numbers of samples in a fast and efficient manner, the researchers developed an unsupervised machine-learning algorithm to identify data clusters of similar nature, and then use image analysis to quantify the proportions of each cluster.

The team applied the algorithm to an assortment of 18 graphene samples — eight acquired from commercial sources and the rest produced in a laboratory under controlled processing conditions. It was found to successfully characterise graphene properties and quality, without bias, within 15 min.

“Our technology can detect the properties of graphene in under 14 minutes for a single dataset of 1936 x 1216 resolution,” Prof Majumder said. “This will save manufacturers vital time and money, and establish a competitive advantage in a growing marketplace.”

The algorithm, which has the potential to be rolled out globally, thus means graphene producers can be assured of quality product and remove the time-intensive and costly process of a series of characterisation techniques to identify graphene properties, such as the thickness and size of the atomic layers.

“The capability of our approach to classify stacking at sub-nanometre to micrometre scale and measure the size, thickness and concentration of exfoliation in generic dispersions of graphene/graphene oxide is exciting and holds exceptional promise for the development of energy and thermally advanced products,” said study co-author Md Joynul Abedin. He added that the method has the potential to be used for the classification and quantification of other two-dimensional materials.

Professor Dusan Losic, Director of the ARC Research Hub for Graphene Enabled Industry Transformation, concluded, “These outstanding outcomes from our ARC Research Hub will make significant impact on the emerging multibillion-dollar graphene industry, giving graphene manufacturers and end users a simple quality control tool to define the quality of their produced graphene materials, which is currently missing.”

Image credit: © Mark

Please follow us and share on Twitter and Facebook. You can also subscribe for FREE to our weekly newsletters and bimonthly magazine.

Related Articles

Leveraging AI in clinical trials critical to the advancement of pharmaceuticals

Trials leveraging AI technology and big data in the planning phases can optimise trial design and...

Microscopy deep learning can identify viral infections

Using fluorescence images from live cells, researchers have trained an artificial neural network...

One step closer to a 100% digital laboratory

When Bumrungrad International Hospital first went live with the InterSystems TrakCare electronic...

  • All content Copyright © 2021 Westwick-Farrow Pty Ltd