From integrating intermittent energy with demand, to releasing carbon-free electricity for the transportation sector through electric vehicles, trains, and ships, to a range of advanced electronic and robotic applications, batteries are becoming increasingly important in a variety of applications.
However, a key challenge is that batteries degrade rapidly with changing operating conditions. Currently, it is difficult to assess battery health without interrupting battery operation or undergoing lengthy charge-discharge procedures using specialized equipment.
In a recent study published in the journal *Nature Machine Intelligence*, researchers from the Intelligent Systems Group at Heriot-Watt University in Edinburgh, in collaboration with researchers from the CALCE Group at the University of Maryland, developed a new method to assess battery health without considering battery operating conditions, battery design, or chemistry by inputting raw battery voltage and current operating data into an artificial intelligence (AI) algorithm.
Darius Roman, a PhD student designing the AI framework, stated that progress in data-driven models of battery degradation to date has relied on the development of algorithms that can perform inference faster. While researchers often spend a significant amount of time on model or algorithm development, few dedicate time to understanding the engineering environment in which these algorithms are applied. In contrast, our work started from scratch. We first learned about the battery degradation problem through a collaboration with the CALCE group at the University of Maryland, which conducted internal battery degradation testing. We then focused on the data, designed features to capture battery degradation, selected the most important features, and only then deployed AI technology to assess battery health.
Furthermore, the researchers found that current data-driven models for battery health assessment do not consider model confidence levels. However, this is often crucial for understanding how AI models arrive at certain conclusions and for decision-making regarding the reliability of those models. In their work, they propose an AI model that quantifies uncertainty in predictions, thereby better supporting operational decisions.
The framework for development has been expanded to include new chemical substances, upcoming novel solid-state batteries, battery design and operating conditions, and has the potential to unlock new strategies for how and how batteries should be used.
Valentin Robu of SmartSystems Group stated that batteries are becoming increasingly important for a wide range of applications, from robotics to renewable energy integration. A key challenge in these areas is making accurate and reliable estimates of battery health. For example, ensuring the health of the batteries deployed on robotic equipment operating in remote environments (such as deep-sea underwater monitoring) is a critical task. Similarly, for energy applications, accurately estimating the remaining battery life is often essential to the economic viability of the project.
