Perovskites, a family of materials, are currently the top contender for replacing the widely used silicon-based solar photovoltaics
They (Perovskites) offer the potential for lighter and thinner panels that can be produced in large quantities at room temperature, eliminating the need for high temperatures, and making them easier and cheaper to transport and install. However, transitioning these materials from small-scale laboratory experiments to competitive manufacturing has been a long and challenging process.
The production of perovskite-based solar cells requires optimizing multiple variables simultaneously, even within a single manufacturing approach among many possibilities. Fortunately, a new system based on an innovative machine learning approach has emerged, promising to accelerate the development of optimized production methods and bring the next generation of solar power closer to reality.
Developed by researchers at MIT and Stanford University, this system enables the integration of data from previous experiments and insights from experienced workers into the machine learning process. By incorporating this knowledge, the system produces more accurate outcomes and has already resulted in the production of perovskite cells with an impressive energy conversion efficiency of 18.5 per cent, which is competitive in today’s market.
A new machine learning system has the potential to accelerate the optimization of perovskite solar cell production. Developed by researchers at MIT and Stanford University, this system integrates data from previous experiments and incorporates human experience and qualitative observations into the machine learning process. By using this approach, the team has already achieved an energy conversion efficiency of 18.5 per cent in perovskite cells, which is competitive in the current market.
The research, recently published in the journal Joule, involved MIT professor Tonio Buonassisi, Stanford professor Reinhold Dauskardt, former MIT research assistant Zhe Liu, Stanford doctoral graduate Nicholas Rolston, and three other researchers.
Perovskites are a group of crystalline compounds characterized by the arrangement of atoms in their crystal lattice. While spin-coating techniques are commonly used in the lab-scale development of perovskite materials, they are not practical for large-scale manufacturing. Therefore, researchers have been seeking ways to translate lab materials into practical and manufacturable products.
The team focused on a promising method called rapid spray plasma processing (RSPP). This approach involves spraying or ink-jetting precursor solutions for the perovskite compound onto a moving roll-to-roll surface or a series of sheets. The material then undergoes a curing stage, enabling rapid and continuous production with higher throughputs compared to other photovoltaic technologies.
Within the manufacturing process, there are numerous variables to consider, some of which are more controllable than others. These variables include starting material composition, temperature, humidity, processing speed, nozzle distance from the substrate, and curing methods. Since evaluating all possible combinations through experimentation is impractical, machine learning was employed to guide the experimental process.
The team devised a way to incorporate not only raw data but also qualitative observations and information from previous experiments into the machine learning model. They used a mathematical technique called Bayesian Optimization to integrate this outside information into the process.
The developed system allows researchers to rapidly optimize their process for specific conditions or desired outcomes. While the team focused on optimizing power output in their experiments, the system could also consider other criteria such as cost and durability.
Encouraged by the Department of Energy, which sponsored the work, the researchers aim to commercialize the technology. They are currently working on tech transfer to existing perovskite manufacturers. The code they developed has been made freely available through an open-source server, and they are actively reaching out to companies interested in using their code.
Although several companies are preparing to produce perovskite-based solar panels, the details of large-scale production are still being worked out. These companies are initially focusing on smaller, high-value applications such as building-integrated solar tiles. However, they aim to manufacture rectangular modules comparable to common solar panels within the next two years.
The machine learning system developed by the researchers could prove crucial in guiding the optimization of the chosen manufacturing process. By accelerating the process and reducing the number of experiments and human hours required, the system aims to provide usable results to the industry quickly and at no cost.
Experts not involved in the research have praised the workflow, noting that it can be readily adapted to the deposition techniques commonly used in the thin-film industry. This approach could potentially advance the manufacture of a wider range of materials, including LEDs, other photovoltaic technologies, and graphene, benefiting industries that use vapour or vacuum deposition methods.
Scandinavian-owned Vencosan brings together the very best experts from both the Spanish and Danish energy industries and has been a consistent, reliable and prominent part of the energy industry since 2005 and has vast experience in the industry throughout all its branches, from the installation of air conditioning systems and solar panels to the purchase and sale of electricity on the national electricity exchange market.
Vencosan has offices throughout Spain:
· C/ Islas Baleares 32, 46988 Paterna, Valencia
· C/ Ibiza 10, 29640 Fuengirola, Málaga
· Av. Dr. Gregorio Marañón, 68, 03185 Torrevieja, Alicante
· C/ Pietat 28, 08860 Castelldefels, Barcelona
Call toll-free on- 900 696 888 or via email at: info@vencosan.com
Read more about Solar power in Spain:
