Robotics Join Machine Learning for an Electric Future

Author(s): Salvatore Raieli

Originally published on Towards AI.

How robotics and AI can speed energy transition and reduce emissions

One of the most important steps toward the electrification of transportation and aircraft is the development of high-energy and effective battery technology. However, it may take years for battery advancements to materialize. Electrolyte optimization is time-consuming and difficult in the case of non-aqueous battery electrolyte solutions because of the numerous design factors involved in choosing various solvents, salts, and their relative ratios. Recently, Carnegie Mellon researchers showed how coupling robotics and machine learning it is possible to develop better batteries.

Why do we need better batteries?

This summer has seen an increase in extreme events. News reports showed that this year we witnessed extreme droughts in Europe, temperatures in England reached forty degrees for the first time, and Pakistan suffered devastating flooding. It would be reductive to describe 2022 as an annus horribilis because, in fact, this is instead the result of a trend over the past decades.

In fact, researchers have warned that rising global temperatures make extreme events increasingly likely. Higher temperatures increase the risk of prolonged droughts, famines, wildfires, hurricanes, and flooding. Moreover, it is not only the frequency that increases but also the strength of these events.

Global warming is caused by anthropogenic activities. As explained by researchers, the main culprit is carbon dioxide produced by human activities. In fact, carbon dioxide, although not the most potent greenhouse gas, is produced by virtually all industrial activities, and its increase is directly related to the increase in global temperature.

The transportation industry is responsible for 37 percent of CO2 emissions. In fact, even today, transportation has the highest reliance on fossil fuels of any sector. During the Covid-19 pandemic, the sector was heavily impacted by the various lockdowns. In the last two years, however, goods have started moving again, and people have started traveling again.

Although there are some trucks with electric motors, most of those operating still use fossil fuels. In order to imagine a future with electric trucks and planes, high-performance batteries must be developed. Several researchers have focused on what is in requirements for the batteries of the future, briefly:

• fast charging: to avoid delays and queues at charging stations
• safer batteries: Due to the accumulation of gases brought on by frequent recharging, overcharging, or short circuits, batteries may rupture or explode.
• Long range: the majority of electric cars have 200 miles range, which is not enough for logistic transport
• less expensive: the battery is generally granted for 8–10 years and around 100,000 miles; then is generally more convenient to buy a new vehicle than replace the battery.

In general, developing new electrolytes is crucial to being able to create new batteries with the desired characteristics. However, this is not an easy challenge since an almost infinite number of combinations can be tested. In fact, the traditional method is trial-and-error, but it is difficult and time-consuming.

Out of the materials present in a battery, liquid electrolytes are a particular challenge to optimize. There are many choices for solvent or salt, each potentially yielding vastly different performance; optimized electrolyte solutions often contain more than three or four components. — Original article

Coupling machine learning and robotics

Recently machine learning has been used in several applications to identify potential new materials ( photovoltaic cells, solid-state materials, catalysis systems, and so on). This is only the first part of the story, though; after all, once potential materials are identified, they must be tested in the laboratory and analyzed. Even this is still a difficult and laborious task.

“Most battery labs design electrolytes with legions of graduate students making and testing various electrolytes,” said Venkat Viswanathan, an associate professor of mechanical engineering. “We’re just a team of three who’ve built a robot to do most of this work for us.” — Carnegie Mellon article about the project

As the authors of the article written, there is great hope in the “close loop” approach:

There is a great deal of recent research on coupling automated experiments to these machine-learning methods — Original article

the idea is the automatic execution of experiments coupled with the experiment schedule. Automation would reduce time, also allow for better standardization and reduce costs.

Carnegie Mellon researchers aimed to solve the problem using two components. The first component is Clio, an automated system of pumps and valves that mixes various solvents, salts, and other chemicals together. After that, the system measures how the solution performs on critical battery benchmarks.

Clio enables high-throughput experiments characterizing transport properties over a range of solvents and salts. — Original article

Next, they used a system called Dragonfly. a machine learning algorithm that uses the results of Clio to propose new possible combinations that could produce better results.

Dragonfly is an open-source Bayesian optimization package designed for black-box optimization. The library is open-source on GitHub and enables Bayesian optimization in a scalable way and several tools that can be used in the process (high dimensionality, parallelization, multi-objective, etc…).

In this paper, the authors focalized on trying to find a new combination of electrolytes that would allow a battery to charge faster. As mentioned in the article, such a workflow can be applied to several to the optimization of other battery components. The result is that they achieved a 13% improvement over the top-performing baseline battery cell.

This demonstrates the potential of closed-loop experiments to discover optimal material designs within well-explored and unexplored design spaces. — Original articlen

The authors suggested that this model could be applied in a number of other cases. in fact, conjugating high-throughput experiments and machine learning in closed loops could enable cost and time reductions in a number of industries.

Parting thoughts

The climate emergency is increasingly urgent and we need practical solutions as soon as possible.

Researching new materials for batteries and solar panels are laborious and expensive work. The development of both algorithms and automating systems is crucial.

The merit of this paper is to demonstrate a system that links a machine learning model and a robotics application. Better batteries are critical to being able to reduce emissions from transportation (the main source of emissions). In addition, this approach can be used for various applications that can be used to reduce emissions and help the energy transition (new materials, making processes more efficient, and so on).

If you have found it interesting:

You can look for my other articles, you can also subscribe to get notified when I publish articles, and you can also connect or reach me on LinkedIn. Thanks for your support!

Here is the link to my GitHub repository, where I am planning to collect code and many resources related to machine learning, artificial intelligence, and more.

Or feel free to check out some of my other articles on Medium:

Join thousands of data leaders on the AI newsletter. Join over 80,000 subscribers and keep up to date with the latest developments in AI. From research to projects and ideas. If you are building an AI startup, an AI-related product, or a service, we invite you to consider becoming a sponsor.

Published via Towards AI