Autonomous Battery Optimization With Machine Learning, Robotics

Lithium-ion rechargeable batteries have become indispensable in today's portable electronics and electric vehicles. They power up smartphones, laptops, cars, and homes. Therefore, high energy and efficient battery design is a crucial aspect of advancing the electrification of transportation.

However, the battery design process is lengthy and can take years to deliver an optimum battery for targeted specifications. Most batteries used currently consist of non-aqueous electrolytes. For a superior battery design, optimizing the ion conductivity of the electrolyte is crucial. There are many design variables in selecting multiple solvents, salts, and proportions.

To speed up the process, researchers at Carnegie Mellon University have developed an experimental design that combines an automated robot, Clio, to make electrolytes for optimized battery designs and machine-learning algorithms that suggest improvements based on the data from Clio and other sensors in the electrolyte.

 

Image used courtesy of Adobe Stock

 

Aspects of Battery Development

A cell consists of four main parts: an anode electrode, cathode, separator, and electrolyte. The electrodes are made up of two different materials, one is positively charged, and the other is negatively charged. The separator isolates the electrodes, and the electrolyte creates the reduction and reduction process. It also provides a medium for the ions to flow between the two electrodes. Each component has unique properties that affect how a battery performs and various parameters like voltage, lifecycle, storage capacity, safety, and operating temperature.

Out of these components, electrolytes are challenging to optimize. There are many salts and solvents to choose from, each leading to a different performance. Both salts and solvents and their relative proportions matter while designing a battery, as they can create an efficient or low-performance battery.

Moreover, the market demands continue to push more energy in the smallest volume possible, aggravating the safety issue. Therefore, the electrical and mechanical design of the battery becomes crucial, not only for high performance but also for a safe battery.

 

AI and Robotics for Battery Development

The researchers at Carnegie Mellon University aimed to develop fast-charging battery electrolytes that can transport lithium ions at high current rates of 5 - 10 mA/cm2.

For developing an optimum electrolyte, they created a robotic platform called Clio that can perform closed-loop optimization (a system that finds optimum parameters capable of minimizing the value of a cost function) of non-aqueous lithium-ion electrolyte solutions. The researchers report that the closed-loop experiments discover an optimum result faster and with fewer trials and can find an optimal design within a given design specification.

 

Illustration of the closed-loop optimization system. Image used courtesy of Carnegie Mellon University

 

Clio allows for high-throughput experiments and autonomously optimizes conductivity over solvent mass fraction and salt in a design space. The optimal electrolytes pass through a sequence of fast-charging electrochemical tests conducted in graphite pouch cells.

 

Schematic of Clio. Image used courtesy of Carnegie Mellon University

 

The researchers combine this robotic platform with a computer running deep learning artificial intelligence (AI) called Dragonfly that accepts data from Clio and sensors in electrolytes to suggest improvements. Clio accepts these improvements and makes new samples. The researchers found this feedback-based system can gradually improve the electrolyte samples. Their best battery design was 13% better than the top-performing batteries.

The researchers will continue improving their system to allow for more specifications and make it run faster.