Engineers apply physics-based machine learning to solar cell production

0


PICTURE: Despite recent advances in the power conversion efficiency of organic solar cells, knowledge about the thermomechanical stability induced by processing of bulk heterojunction active layers is helping to advance the field …. see After

Credit: Department of Mechanical and Mechanical Engineering / Lehigh University

Today, solar power provides 2% of America’s electricity. However, by 2050 renewables are expected to be the most widely used source of energy (overtaking oil and other liquids, natural gas and coal) and solar will overtake wind as the main source of energy. ‘renewable energy. To achieve this and make solar energy more affordable, solar technologies still require a number of breakthroughs. One is the ability to more efficiently convert photons from sunlight into usable energy.

Organic PV achieves a maximum efficiency of 15-20%, which is substantial, but limits the potential of solar energy. Lehigh University engineer Ganesh Balasubramanian, like many others, wondered if there were ways to improve the design of solar cells to make them more efficient?

Balasubramanian, associate professor of mechanical engineering and mechanics, studies the basic physics of materials at the heart of solar energy conversion – organic polymers passing electrons from one molecule to another so that they can be stored and operated – as well as the manufacturing processes that produce commercial solar cells.

Architecture of the OPV bulk-heterojunction structure and design scope. [Credit: Ganesh Balasubramanian, Joydeep Munshi, Lehigh University]

Using the Frontera supercomputer at the Texas Advanced Computing Center (TACC) – one of the most powerful on the planet – Balasubramanian and his graduate student Joydeep Munshi executed molecular models of organic solar cell production processes and designed a framework for determining optimal engineering choices. They described the computational effort and the associated results in the May issue of IEEE Computing in Science and Engineering.

“When engineers make solar cells, they mix two organic molecules in a solvent and evaporate the solvent to create a mixture that helps convert excitons and transport electrons,” Balasubramanian said. “We mimicked the way these cells are created, specifically the massive heterojunction – the absorption layer of a solar cell. Basically we are trying to understand how structural changes correlate with the efficiency of the cell. solar conversion? “

Balasubramanian uses what he calls “physics-based machine learning”. His research combines coarse-grained simulation – using approximate molecular models that represent organic materials – and machine learning. Balasubramanian believes this combination helps prevent artificial intelligence from coming up with unrealistic solutions.

“A lot of research uses machine learning on raw data,” Balasubramanian said. “But more and more there is an interest in using machine learning trained in physics. This is where I think the most benefits lie. Machine learning in itself is just math. there’s not a lot of real physics involved in this. “

Write in Computational Materials Science In February 2021, Balasubramanian and Munshi along with Wei Chen (Northwestern University) and TeYu Chien (University of Wyoming) described the results of a set of virtual experiments on Frontera testing the effects of various design changes. These included changing the proportion of donor and receptor molecules in bulk heterojunctions, as well as the temperature and time spent on annealing – a cooling and hardening process that contributes to product stability.

They exploited the data to train a class of machine learning algorithms called support vector machines to identify the material and production process parameters that would generate the most energy conversion efficiency, while also maintaining structural strength and stability. By coupling these methods together, the Balasubramanian team was able to reduce the time required to achieve an optimal process by 40%.

“Ultimately, molecular dynamics is the physical engine. It’s what captures fundamental physics,” he said. “Machine learning examines numbers and patterns, and evolutionary algorithms facilitate simulations.”

Tradeoffs and limitations

Like many industrial processes, compromises are necessary to fine-tune any facet of the manufacturing process. Faster cooling can help increase energy efficiency, but it can make the material brittle and prone to shattering, for example. Balasubramanian and his team used a multi-objective optimization algorithm that balances the pros and cons of each change to derive the overall optimal manufacturing process.

Flowchart describing the steps of a typical coupled Cuckoo Search-Coarse Grained Molecular Dynamics (CS-CGMD) algorithm. The dotted box represents the augmented machine learned exploration of regions of interest to supplement poorly executed nests with newer alternatives during each generation of CS optimization. [Credit: Ganesh Balasubramanian, Joydeep Munshi, Lehigh University]

“When you try to optimize a particular variable, you look at the problem in a linear fashion,” he said. “But most of these efforts have multi-faceted challenges that you are trying to solve simultaneously. There are trade-offs you have to make and synergistic roles that you have to grasp, to get to the right design.”

The Balasubramanian simulations corresponded to the experimental results. They determined that the composition of the heterojunction and the temperature / time of annealing have the greatest effects on overall efficiency. They also found what proportion of the materials in the heterojunction is best for efficiency.

“There are certain conditions identified in the literature which people believe are the best conditions for efficacy for these selected molecules and processing behavior,” he said. “Our simulation was able to validate these and show that other possible criteria would not give you the same performance. We were able to realize the truth, but from the virtual world.”

With more time allocation on Frontera in 2021-2022, Balasubramanian will add additional layers to the machine learning system to make it more robust. It plans to add experimental data, as well as other modalities of computer models, such as electronic structure calculations.

“The heterogeneity of the data will improve the results,” he said. “We plan to do material principle simulations first, and then feed that data into the machine learning model, along with coarse-grained simulation data.”

Balasubramanian believes that today’s organic photovoltaic may be reaching the limits of its efficiency. “There is a hard wall to penetrate and that is the material,” he said. “These molecules that we have used cannot go any further. The next thing to try is to use our framework with other advanced molecules and materials.”

His team tapped into the literature to understand the characteristics that increase solar efficiency, then trained a machine learning model to identify potential new molecules with ideal charge transport behaviors. They published their research in the Journal of Chemical Information and Modeling. Future work on Frontera will use the Balasubramanian framework to explore and computer test these alternative materials, assuming they can be produced.

“Once established, we can take realistic molecules made in the lab and put them into the framework that we created,” he said. “If we find new materials that work well, it will lower the cost of solar power generation devices and help Mother Earth.”

Balasubramanian’s research exploits the two things for which computer simulations are essential, he says. “One is to understand the science that we can’t study with the tools we have in the real world. And the other is to speed up science – streamline what we really need to do, which reduces our cost. and our time to make things and physically test them. “

###



Source link

Leave A Reply

Your email address will not be published.