A family of materials known as perovskites has shown it can be used to create solar panels with several benefits over current dominant designs.
Panels can be made far thinner, lighter and more flexible than the existing silicon-based photovoltaics.
They are cheaper and easier to install and can react to various wavelengths of light, allowing them to convert more of the sunlight that reaches them into electricity.
Bringing perovskite panels from the lab to production has proven challenging, however. It is not least because the process requires the optimisation of a dozen or more variables, even when considering just one of many possible approaches.
Now researchers at MIT and Stanford University are using novel machine learning techniques to improve the process and speed up the development of optimised means of production.
The AI-powered system allows researchers to feed data from previous experiments and information based on observation by people with experience in the field into the machine learning process.
It makes the process more accurate and has resulted in perovskite cells with a credible energy conversion rate of 18.5%.
Researcher Nicholas Rolston said it could be challenging to take a process from a laboratory environment to a start-up or manufacturing line.
One of the main issues with perovskites is the thousands of possible compounds and numerous ways to make them.
The main one in laboratory settings is a spin-coating technique, but that is not practical at larger scales.
Production technique creates highly complex sets of variables
The researchers looked primarily at a method known as rapid spray plasma processing (RSPP), which involves spraying precursor solutions for the perovskite compound onto a moving series of sheets.
The material then passes onto a curing stage. But numerous variables can affect the result, including temperature, humidity, the composition of the starting materials, and the distance of the nozzle that sprays onto the substrate.
Many of these variables can also interact with each other, creating highly complex combinations that can only really be evaluated using artificial intelligence.
While most machine learning gathers and evaluates raw data, few consider previous experiments or observations from human experts.
In incorporating this qualitative data, the new system allows researchers to guide their process more quickly to optimise it for any given conditions or required outcomes.
And this has the potential to pave the way for perovskite-based panels to become the next-generation standard in solar energy.
Today’s news was brought to you by TD SYNNEX – the UK’s number one solutions distributor.
Read more of our latest Industry Updates stories
