Due to the adverse image created by battery-originated fires in EVs and their potential consequences, OEMs are face the challenging task of regaining consumer trust. Several tech companies are working overtime to resolve the battery issue and UK-based Eatron Technologies is one of them. While the company acknowledges the rarity of EV battery fires, but it also recognises the need to minimise catastrophic battery failures. Eatron Technologies says that by leveraging AI technologies in these areas, EV manufacturers can enhance the safety, reliability, and performance of their battery systems.
Vehicle fires are nothing new, regardless of the powertrain type (ICE/EV), but the recent emphasis has been on highlighting those involving electrified vehicles. “The reality is that EV battery fires are incredibly rare, but even one is one too many,” explains Dr Umut Genc, CEO at Eatron Technologies. “As an industry, we need to ensure the number of catastrophic battery failures reaches zero, and then stays there. Our intelligent, connected and safe automotive-grade battery management software has demonstrated that AI holds the key to achieving this.”
The causes of battery failure are complex, and often involve a combination of factors. One of the most common causes of these mishaps is lithium plating, wherein metallic lithium deposits form around the anode. This is most likely during fast charging at low temperatures and over time, these deposits erode the performance of the battery. Left unchecked, this can lead to the growth of dendrites, needle-like structures that can pierce through the separator between the anode and the cathode, causing a short circuit within the cell. This in turn leads to a rapid self-discharge that can initiate thermal runaway, a self-sustaining chain reaction that is difficult to extinguish. Detecting lithium plating without opening the battery cell and examining the electrodes, which is largely impossible once mounted in a vehicle, is a challenge that has been the subject of intense research. While various techniques have been developed over the years, each has their own limitations, particularly when it comes to distinguishing lithium degradation.
AI to help resolve battery issues
Eatron Technologies has co-developed an AI diagnostics kit that can predict cell failures with great accuracy and zero false positives. Early detection allows for more effective and convenient management of potential failures, such as altering battery management or scheduling a service visit. “Using a technique called feature extraction, we transform the raw health data coming from the battery into a format that makes anomalies easier to identify. By combining this with our proprietary AI pipeline that accurately captures battery behaviour, our AI diagnostics can predict cell failures before they occur, with up to 90% accuracy and zero false positives,” says Dr Genc.
Detecting a failure before it happens opens the door to dealing with it far more effectively and conveniently. That could mean altering the way the battery is managed to minimise any further damage in the short term, and ultimately to schedule a service visit for rectification at the driver’s convenience.
Machine learning solutions
Eatron has partnered with Infineon Technologies to integrate advanced machine learning solutions and algorithms into a micro controller unit (MCU). The partnership aims to address challenges in EV adoption, including range anxiety, charging speed, and battery health. Infineon’s PPU (Parallel Processing Unit) is an on-chip digital signal processor (DSP) designed for efficient processing of SIMD (Single Instruction, Multiple Data) vector operations. One of the key advantages of the PPU is its ability to dramatically reduce computation time compared to traditional CPUs when executing vectorised operations. This efficiency stems from its architecture, which is optimised for parallel processing of data elements.
To make the adoption and utilisation of the PPU easier for customers, Infineon provides an automated toolchain within its ecosystem. One notable aspect of this toolchain is its ability to automatically convert existing models or algorithms into vectorised code that can be efficiently executed on the PPU.