Researchers from Heriot-Watt University and the University of Maryland created an artificial intelligence (AI) algorithm that can predict the health of a battery.
Electric batteries are important for a variety of applications, particularly in electric vehicles. But these batteries degrade quickly and it is difficult to estimate battery health without interrupting operation or using the lengthy process of charge-discharge with specialized equipment. The team used their AI algorithm to estimate electric battery health, without interrupting battery operating conditions or harming the battery.
They found that current data-driven models for battery health do not take model confidence into consideration. This is a critical piece for decision-making to help understand how a model came to its conclusion and if the model can be trusted.
Source: Unsplash
To create data-driven models for battery degradation, the team relied on the development of algorithms that quickly work on inference. The team built their algorithm from the ground up.
Data from in-house battery degradation testing was analyzed and the algorithm's features were engineered to capture degradation mechanisms and rates. Researchers selected the most important features of battery health and used that data to teach the algorithm.
To estimate battery health, the team fed their AI algorithm raw battery voltage data and current operational data. The proposed model can quantify uncertainty in predictions and support operating decisions. The framework could be scaled up with new chemistries, like solid-state batteries, evolving designs and more. It also has the potential to unlock new strategies for how a given battery can and should be used.
A paper on this new AI algorithm was published in Nature Machine Intelligence.
