Machine learning can prevent thermal runaway in EV batteries

Thermal runaway, a self-sustaining chemical reaction with high heat generation, is a critical issue in electric vehicle (EV) batteries. Comprised of multiple interconnected cells, each operating as part of a larger energy system, the entire EV battery array can be compromised when a single cell degrades. This degradation generates excess heat, which can propagate to neighbouring cells, potentially leading to outgassing and the build-up of combustible vapours. These vapours can ignite and cause a spontaneous explosion, followed by an uncontrollable fire, capable of destroying a vehicle within minutes.

To safeguard against thermal runaway, nations are legislating a minimum five-minute warning before catastrophic failure of a battery pack, allowing passengers to evacuate safely even if thermal runaway occurs. This regulation compels manufacturers to implement cell venting detection in battery management systems (BMS).

Early prevention and detection with machine learning

Efforts to mitigate thermal runaway focus on three main areas: prevention, early detection, and effective containment. During battery production, manufacturers use vision systems and a variety of automated quality inspection and test instrumentation with machine learning (ML) to inspect quality in battery manufacturing. ML-driven, automated high-speed detection of manufacturing faults that could lead to cell failures enables timely corrective actions that human inspectors might miss.

Advanced BMS equipped with sensors and ML algorithms can predict potential failures by analysing data from multiple sensor sources. These systems incorporate models of anticipated battery degradation with charge/discharge cycles and external factors, such as temperature and work cycles, and then compare the real-world sensor data to predictive models and similar data from the aggregated population of similar battery packs. ML helps identify early trends that deviate from the normal population and can also detect early signs of abnormal cell behaviour that might lead to thermal runaway, such as temperature fluctuations and pressure changes, providing crucial warnings before catastrophic events occur.

Rare but severe incidents

EV battery fires, though dramatic, are relatively rare compared to fires in internal combustion engine vehicles. Nonetheless, the high temperatures, hazardous gases, and the self-sustaining nature of lithium fires pose significant challenges. For instance, a catastrophic event involving an EV battery in a metal container can lead to an explosion and ferocious fire, potentially causing extensive damage beyond the immediate area. As battery packs are also difficult to access, this can lead to long duration events that include re-ignition of the damaged cells with indeterminate latency, where re-ignition can occur within minutes, hours, or days after the initial damage event.

The dangerous outcome of thermal runaway events, often triggered spontaneously by undetected faults, underscores the need for robust detection systems. Data from 2019 shows that most EV battery fires start spontaneously, making it imperative to detect fault conditions early.

Monitoring and machine learning

While detecting external factors like faulty charging equipment or physical damage is challenging, monitoring for signs of failure remains essential. Symptomatic indicators such as temperature and pressure changes, though sometimes ambiguous or short-lived, can provide early warnings.

ML models, trained to examine data from multiple sensors, offer a solution by analysing precursor trends that could lead to catastrophic failures with greater accuracy. These models can recognise patterns in anomalies that might be overlooked in isolation, enhancing the reliability of early detection systems. Using ML in BMS enhances early prevention and detection of thermal runaway in EV batteries. By continuously monitoring and analysing sensor data, these systems provide critical warnings that can prevent catastrophic events, ensuring the safety of both vehicles and their passengers.

Designing BMS with ML for thermal runaway detection

Designing BMS to detect early warning signals of thermal runaway presents several challenges. ML can add complexity but improve outcomes by evaluating sensor data for brief, sporadic events that do not always indicate imminent failure but show deviation from a normal trend.

Typically, the BMS operates when the vehicle is in use or being charged, requiring systems to run in low-power mode when the vehicle is parked to conserve battery. This poses a challenge, as low-power modes may result in missed critical sensor events. An ideal solution involves systems that are always active, regardless of the vehicle’s state, which includes parked or stored vehicles. Standalone systems, similar to smoke detectors, could be deployed in garages or storage areas to enhance monitoring.

ML can mitigate low sampling rates by interpreting residual sensor data patterns to infer potential events. This approach enhances early detection of thermal runaway using data like temperature, pressure and vent gases, including concentrations of hydrogen, carbon monoxide, or carbon dioxide that indicate a cell in the array may be undergoing active material decomposition and venting.

With the rise in EV adoption, addressing prevention, early detection, and containment of thermal runaway becomes increasingly vital. The introduction of ML models in BMS is a promising area of development, with numerous patent applications emerging from automotive manufacturers and Tier 1 suppliers.

The opinions expressed here are those of the author and do not necessarily reflect the positions of Automotive World Ltd.

Written by Jason Skoczen, Sales Director, Lightspeed and Transportation, Avnet.

The AutomotiveWorld.com Comment column is open to automotive industry decision makers and influencers. If you would like to contribute a Comment article, please contact editorial@automotiveworld.com