The researchers at the ARC Centre of Excellence in Exciton Science have created a new type of machine learning model that could significantly speed up designing more efficient solar cells. Its importance in reducing carbon emissions is more significant than ever. This program helps to predict the power-conversion efficiency (PCE) of substances for application in next-generation organic solar cells, including virtual compounds that don’t exist yet. This method is fast and convenient to use, unlike other time-consuming and complex models. The code for this approach is also available for all scientists and engineers.
More manageable and chemically interpretable signature descriptors replace the analyzed molecules’ complicated and costly computational parameters, requiring quantum mechanical calculations. This makes the model more efficient and user-friendly. They generate information that can be applied to design advanced materials by providing essential data about various vital chemical fragments in materials that affect PCE.
Earlier professionals were highly dependent on Silicon, which is comparatively expensive and lacks flexibility. Organic photovoltaic (OPV) solar cells have gained increasing attention as they are more affordable to develop using printing technologies. They are also versatile and more convenient to dispose.
A vast number of chemical compounds are suitable for use in OPVs that scientists can synthesize. Therefore, the foremost concern is sorting through these vast numbers of compounds. Many earlier machine learning models that were developed to address this issue were time-consuming, required significant computer processing power, and were challenging to replicate. They also didn’t provide adequate direction for scientists investigating to develop unique solar devices.
Many of these challenges are now successfully addressed by the research led by Dr. Nastaran Meftahi and Professor Salvy Russo of RMIT University, in association with Professor Udo Bach’s team at Monash University.
Dr. Nastaran says that most other models use electronic descriptors that are complicated, computationally expensive, and not chemically interpretable. The experimental scientist can’t get approaches from those models to design and synthesize materials in the lab. Her team’s newly developed model is simple as they used chemically interpretable descriptors, supporting them to see the crucial fragments. Her work was greatly assisted by her co-author Professor Dave Winkler who co-created the BioModeller program, which provided the basis for the new, open-source model.
Using it, the researchers have produced robust and predictive results. It has also generated quantitative relationships between the molecular signatures under research and future OPV devices’ efficiency. The researchers now aim to extend their work scope to include more significant and accurate computed and empirical datasets.
Related Codes: https://github.com/Nas796/Machine-learning-for-photovoltaic-material-property-prediction