AI modeling uses raw data to predict battery status

Researchers at Heriot-Watt University in Edinburgh have developed an AI modeling framework that can accurately predict battery status using simple voltage and current data.

Estimating battery status (SOH) without interrupting battery operation or running long charge / discharge cycles is a current challenge. A team at Heriot-Watt’s Smart Systems Group worked with researchers at the CALCE Group at the University of Maryland to devise a solution for building algorithms from the first principles of battery degradation. The algorithm is supplied with raw battery voltage and current operational data and can provide accurate SOH measurements under all conditions, regardless of battery design or chemistry.

Opinion: UK needs to address the looming skill gap in battery technology

“So far, advances in data-driven models of battery degradation have relied on developing algorithms to perform inference faster,” explains Darius Roman, a student at Heriot-Watt University who designed the AI ​​framework. Did. “Researchers often spend a considerable amount of time developing models and algorithms, but few spend time understanding the engineering context to which an algorithm applies.

“In contrast, our work is built from scratch. First, we understood battery degradation through a collaboration with the University of Maryland’s CALCE group, where in-house battery degradation tests were conducted. Focus on the data, design the ability to capture battery degradation, select the most important features, and then deploy AI technology to estimate battery status. “

The AI ​​framework will scale up to new chemicals, such as future advances in solid-state batteries and evolving battery designs and operating conditions. According to researchers, this technology could help unleash new strategies on how and how batteries can be used. The proposed AI model can also quantify forecast uncertainty to better support operational decisions and show how and models reach specific conclusions and whether the model is reliable.

“Batteries are becoming more and more important for applications ranging from robotics to renewable energy integration,” said Valentin Rob of Heriot-Watt’s Smart Systems Group.

“A key challenge in these areas is to estimate battery status in an accurate and reliable way. Consider, for example, a robotic asset that operates in a remote environment, such as deep sea surveillance. Similarly, for energy applications, an accurate estimate of the remaining battery life is important to the economic feasibility of the project. Often. “

Job is Was announced in Nature Machine Intelligence. AI modeling uses raw data to predict battery status

Back to top button