Prevent Thermal Runaway in EV using Machine Learning
Electric vehicle battery packs generate complex thermal and electrical patterns during operation, with internal cell temperatures ranging from 20-45°C under normal conditions and voltage variations of 2.5-4.2V per cell. When these parameters deviate from expected ranges, early intervention becomes critical—thermal runaway can raise cell temperatures above 150°C within minutes, releasing gases and triggering chain reactions across neighboring cells.
The fundamental challenge lies in processing vast streams of sensor data to identify precursor patterns of thermal instability while maintaining an acceptable false-positive rate that won't unnecessarily disable cells or impact vehicle operation.
This page brings together solutions from recent research—including graph neural networks for cell-to-cell interaction modeling, convolutional networks for temporal pattern recognition in charging data, and adaptive fault detection systems that learn from field data. These and other approaches demonstrate how machine learning can enable real-time prediction and prevention of thermal events in production vehicles.
1. Multioutput Convolutional Neural Network for Improved Parameter Extraction in Time-Resolved Electrostatic Force Microscopy Data
madeleine breshears, rajiv giridharagopal, david s ginger - American Chemical Society, 2025
Time-resolved scanning probe microscopy methods, like time-resolved electrostatic force (trEFM), enable imaging of dynamic processes ranging from ion motion in batteries to electronic dynamics microstructured thin film semiconductors for solar cells. Reconstructing the underlying physical these techniques can be challenging due interplay cantilever physics with actual transient kinetics interest resulting signal. Previously, quantitative trEFM used empirical calibration or feed-forward neural networks trained on simulated data extract interest. Both approaches are limited by interpreting signal as a single exponential function, which serves an approximation but does not adequately reflect many realistic systems. Here, we present multibranched, multioutput convolutional network (CNN) that uses addition parameters input. The CNN accurately extracts describing both single-exponential and biexponential functions more reconstructs real experimental presence noise. This work demonstrates application physics-informed machine learning complex processing tasks, enabling efficient accurate ana... Read More
2. Battery Module Cell Monitoring with Statistical Analysis of Capacity Deviation Over Time
SAMSUNG SDI CO LTD, 2025
Early detection of abnormally deteriorated cells in battery modules to prevent safety issues like thermal runaway or ignition. The method involves monitoring capacity deviations of multiple cells over time. It calculates statistical values representing the capacity variations at two different time points. By comparing these values, it can detect cells that have significantly deteriorated compared to the others, indicating an abnormality.
3. Experimental and Simulation-Based Study on Thermal Runaway Characteristics of 18650 Lithium-Ion Batteries and Thermal Propagation Patterns in Battery Packs
yao yao, peng xu, lei gao - Multidisciplinary Digital Publishing Institute, 2025
The thermal runaway of lithium-ion batteries is a critical factor influencing their safety. Investigating the characteristics essential for battery safety design. In this study, 18650 under different SOCs were systematically analyzed by experiment and simulation. It was found that at high SOC (100%), highly lithium state accelerated lattice oxygen release, promoted formation LiNiO intensified electrolytic liquid oxygenation combustion, while low (20%), reduction environment dominated, metal Ni residual graphite significantly enriched. Gas analysis shows CO2 H2 account more than 80%, proportion regulated SOC. Temperature pressure monitoring showed increase in increased peak temperature (100% up to 508.4 C) (0.531 MPa).The simulation results show when pack out control, ejection fire explosion wave are concentrated middle upper region (overpressure 0.8 MPa). This study reveals mechanism which affects path product gas generation regulating oxidation/reduction balance, lays theoretical foundation safe design quantitative evaluation runaway.
4. Long Short-Term Memory Networks for State of Charge and Average Temperature State Estimation of SPMeT Lithium–Ion Battery Model
b chevalier, junyao xie, stevan dubljevic - Multidisciplinary Digital Publishing Institute, 2025
Lithiumion batteries are the dominant battery type for emerging technologies in efforts to slow climate change. Accurate and quick estimations of state charge (SOC) internal cell temperature vital battery-management systems enable effective operation portable electronics electric vehicles. Therefore, a long short-term memory (LSTM) recurrent-neural network is proposed which completes estimation SOC average (Tavg) lithiumion under varying current loads. The trained evaluated using data compiled from newly developed extended single-particle model coupled with thermal dynamic model. Results promising, root mean square values typically 2% 1.2 K Tavg, while maintaining training testing times. In addition, we examined comparison single-feature versus multi-feature network, as well two different approaches partitioning.
5. Numerical Studying on the Influence of State of Charge and Internal Heat Generation on Lithium Battery Thermal Runaway
fangchao kang, wendi zhang, xiaoqing lu - IOP Publishing, 2025
Abstract The widespread adoption of lithium batteries has raised concerns regarding safety due to the risk thermal runaway. mechanisms by which State Charge (SOC) and internal heat generation influence runaway in these remain unclear. In this study, we utilize a three-dimensional electrochemical-thermal coupled model investigate evolution with increasing temperature examining growth rate. Our findings indicate that: (1) both rate during initially increase before eventually decreasing; (2) an SOC leads reduction intercalation concentration at negative electrode, resulting decreased lower peak temperatures runaway; (3) significantly impacts critical time runaway, higher associated increased risk. These results provide theoretical support for optimizing battery preventing fires explosions.
6. Proactive Multisensory Solution for Mitigating Thermal Runaway Risks in Li-Ion Batteries
uladzimir fiadosenka, linxi dong, chenxi yue - Belarusian State University of Informatics and Radioelectronics, 2025
The paper presents the concept and modeling results of a multisensor system designed to prevent thermal runaway in lithium-ion batteries. This is especially true for LCO, NMC NCO integrates three types sensors: capacitive pressure sensor, gas sensor based on metal oxide semiconductor, platinum temperature sensor. Moreover, all sensors are located single chip, which ensures increased reliability safety, minimizing risks fire, explosion, or damage Three battery operating modes proposed: normal, hazardous, critical. In normal mode, concentration remain at safe levels, while hazardous they begin increase, indicating possible onset destructive reactions. critical reaches can lead damage, explosion. was modeled using COMSOL Multiphysics 6.1 package finite element method. approach helps improve safety batteries by solving problems monitoring their condition. scalability makes it suitable applications both portable electronics electric vehicles.
7. 13D Modeling of Thermal Runaway in Pouch Cells: Effects of Surface Heating and Heating Rates
deivanayagam hariharan, santhosh gundlapally, 2025
<div class="section abstract"><div class="htmlview paragraph">Thermal runaway in battery cells presents a critical safety concern, emphasizing the need for thorough understanding of thermal behavior to enhance and performance. This study introduces newly developed AutoLion 3D model, which builds on earlier 1D framework offers significantly faster computational performance compared traditional CFD models. The model is validated through simulations heat-wait-search mode Accelerating Rate Calorimeter (ARC), accurately predicting by matching experimental temperature profiles from peer-reviewed studies. Once validated, employed investigate LFPO under controlled heating conditions, applying heat one or more surfaces at time while modeling transfer non-heated surfaces. primary objective understand how these localized patterns impact profiles, including average core temperatures surface each heated surface, examine evolution distributions over time. Additionally, explores effects varying rates using ARC assess different influence process. comprehensive approach aims provide valua... Read More
8. Experimental and Simulation-Based Characterization of Thermal Runaway in Lithium-Ion Batteries Using Altair SimLab®
luca giuliano, luigi scrimieri, simone reitano, 2025
<div class="section abstract"><div class="htmlview paragraph">Thermal runaway is a critical phenomenon in lithium batteries, characterized by self-sustaining process due to internal chemical reactions, that triggered once certain temperature reached within the cell. This event often caused overheating charge and discharge cycles can lead fires or explosions, posing significant safety threat. The aim of this study induce thermal on single cells different ways characterize validate simulation models present Altair SimLab.</div><div paragraph">The work was conducted several key phases. Initially, an experimental test performed calorimeter (EV ARC HWS test) collect data Molicel 21700 P45B cell during under adiabatic conditions. These were used for cell, allowing detailed comparison with results. Subsequently, operational conditions, overheated using heat pad powered at constant power (non-adiabatic conditions), monitoring through three thermocouples placed For scenario as well, replicating effect source surface.</div><div paragraph">This work, carrie... Read More
9. Battery Thermal Runaway Propagation: A CFD Approach to Cell and Module Analysis
anil wakale, ma shihu, xiao hu, 2025
<div class="section abstract"><div class="htmlview paragraph">This Paper will focus on simulating thermal runaway propagation within a battery cell and module. The model parameters are derived from accelerating rate calorimeter (ARC). simulation involves that converts the stored energy of materials into energy, thereby runaway. initiation is modelled through nail penetration event, represented by heat profile in region. resulting temperature rise this area triggers short model, leading to spread runaway.</div><div paragraph">For single-cell simulation, 1-equation used, focusing direct conversion cell. In contrast, module more complex scenario. Here, an initial near region activates which subsequently 4-equation abuse model. higher activation energies required initiate cascading effect, driving process throughout As increases further, internal intensifies, turn reactivates associated with compared those create effect accelerates process.</div><div paragraph">The results obtained both levels be validated against ARC data. A detailed discussion these ... Read More
10. A Fault Prediction Model for Electric Vehicle Charging Equipment Based on Adaptive Dynamic Thresholds
hao wang, ning wang, yuan li, 2025
<div class="section abstract"><div class="htmlview paragraph">The surge in electric vehicle usage has expanded the number of charging stations, intensifying demands on their operation and maintenance. Public often exposed to harsh weather unpredictable human factors, frequently encounter malfunctions requiring prompt attention. Current methods primarily employ data-driven approaches or rely empirical expertise establish warning thresholds for fault prediction. While these are generally effective, artificially fixed they prediction limit adaptability fall short sensitivity special scenarios, timings, locations, types faults, as well overall intelligence. This paper presents a novel model equipment that utilizes adaptive dynamic enhance diagnostic accuracy reliability. By integrating quantifying Environmental Influence Factors (EF), Scenario (SF), Fault Severity (FF), Charging Equipment Status (CF) into cohesive predictive framework, our dynamically adjusts based comprehensive analysis factors. Using dataset 560,000 records from Hangzhou, employs batch offline reinforcement... Read More
11. Numerical Model of the Heat-Wait and Seek and Heating Ramp Protocol for the Prediction of Thermal Runaway in Lithium-Ion Batteries
antonio gil, javier monsalveserrano, javier marcogimeno, 2025
<div class="section abstract"><div class="htmlview paragraph">Interest in Battery-Driven Electric Vehicles (EVs) has significantly grown recent years due to the decline of traditional Internal Combustion Engines (ICEs). However, malfunctions Lithium-Ion Batteries (LIBs) can lead catastrophic results such as Thermal Runaway (TR), posing serious safety concerns their high energy release and emission flammable gases. Understanding this phenomenon is essential for reducing risks mitigating its effects. In study, a digital twin an Accelerated Rate Calorimeter (ARC) under Heat-Wait-and-Seek (HWS) procedure developed using Computational Fluid Dynamics (CFD) framework. The CFD model simulates heating cell during HWS procedure, pressure build-up within LIB, gas venting phenomena, exothermic processes LIB degradation internal components. validated against experimental NCA 18650 similar conditions, focusing on temperature domain pressure. effectively captures heat released by undergoing TR through convection radiation surrounding air while providing temporal spatial resolution compo... Read More
12. Acoustic Sensor-Based Detection System for Thermal Runaway in Electric Vehicle Batteries
VOLVO TRUCK CORPORATION, 2025
Early warning system to detect thermal runaway in electric vehicle batteries using acoustic sensors. The system monitors sound waves emitted from battery cells to predict thermal runaway before it escalates. Acoustic sensors detect low frequency infrasound generated by gas bubbles forming during early stages of thermal runaway. An algorithm analyzes the acoustic data to predict thermal runaway. This allows earlier intervention to prevent escalation compared to temperature sensors.
13. Battery Internal Temperature Estimation via Impedance Measurements Across Specific Frequency Bands Using a Learning Model
BATTLAB INC., 2025
Estimating the internal temperature of a battery using impedance measurements at specific frequencies. The method involves applying a signal of a set frequency band to the battery, measuring the impedance, and using that impedance to estimate the internal temperature using a learning model. The set frequency band is determined based on a correlation between impedance and state of charge. By estimating internal temperature this way, it enables predicting battery fires before they occur.
14. Battery Thermal Runaway Early Warning System with Multi-Model Fusion and Real-Time Monitoring
Taiyuan University of Technology, TAIYUAN UNIVERSITY OF SCIENCE AND TECHNOLOGY, 2024
Multi-model fusion power battery thermal runaway early warning system that uses multiple warning models to accurately predict and warn of power battery thermal runaway failures. The system uses voltage, temperature, and heat transfer models to comprehensively monitor battery modules. It involves real-time acquisition of single cell voltages, historical voltage data, and internal temperature prediction using neural networks. The models consider voltage anomalies, physical heat generation, and cell-to-cell heat transfer. A decision model combines warnings for overall thermal runaway prediction. The multi-model fusion provides accurate early warnings and informed driver response to mitigate battery failure hazards.
15. Multi-Stage Lithium Battery Thermal Runaway Prediction Using BiGRU Neural Network with Electrothermal Coupling
SHANGHAI UNIV OF ELECTRIC POWER, SHANGHAI UNIVERSITY OF ELECTRIC POWER, 2024
Lithium battery thermal runaway early warning method to improve accuracy and reduce calculation time compared to traditional methods. The method uses a deep learning model that integrates an offline electrothermal coupling model with an online model for surface temperature and voltage. A BiGRU neural network is trained using COA optimization to predict battery internal temperature. The method divides overcharge into three stages based on heat sources. This multi-stage prediction improves accuracy compared to single-stage direct prediction.
16. Battery Temperature Management Method with LSTM-Based Predictive Modeling
GAC EON NEW ENERGY AUTOMOBILE CO LTD, 2024
Power battery temperature management method using a machine learning model to predict battery temperature and determine safe operating conditions. The method involves acquiring battery temperature and charge/discharge power data, extracting temperature correlation features, feeding them into a pre-trained LSTM model to predict future battery temperatures, and comparing the predictions to a threshold to determine temperature safety. This improves response speed and accuracy compared to conventional PID control.
17. Battery Thermal Runaway Early Warning System Utilizing Electrochemical Mechanism Model with Real-Time Data Integration
SHANGHAI INST OF SPACE POWER SOURCES, SHANGHAI INSTITUTE OF SPACE POWER-SOURCES, 2024
Battery thermal runaway early warning system using an electrochemical mechanism model to improve accuracy and reduce false alarms compared to traditional methods. The system predicts and warns of battery thermal runaway based on analyzing the battery's heat generation and iterating a model using real-time temperature, heating rate, and voltage data. It proposes a situation-based early warning scheme based on thermal trigger mechanisms rather than just monitoring temperature.
18. Battery Thermal Runaway Detection System with Strain, Temperature, and Growth Rate Monitoring Using CNN-LSTM Neural Network
SHANDONG UNIVERSITY, UNIV SHANDONG, 2024
An advanced warning system for battery thermal runaway using strain, temperature, and growth rate monitoring. The system uses a CNNLSTM neural network to detect when a battery is entering thermal runaway based on filtered strain, temperature, and growth rate data. It also defines a thermal runaway state using envelopes of the growth rates to quantify the severity of the runaway. The system provides early warning of thermal runaway by accurately predicting when it will occur and quantifying the runaway state in real-time.
19. Energy Storage Battery System with Real-Time Monitoring and Simulation for Thermal Runaway Risk Detection
ELECTRIC POWER RES INST STATE GRID HENAN ELECTRIC POWER CO, ELECTRIC POWER RESEARCH INSTITUTE OF STATE GRID HENAN ELECTRIC POWER CO, ZHENGZHOU INST EMERGING IND TECH, 2024
Early warning system for energy storage battery systems to prevent thermal runaway and improve safety. It uses real-time battery monitoring and simulation to detect and mitigate thermal runaway risks. The system obtains parameters like cell voltage, temperature, and gas concentration. It simulates cell behavior based on these inputs. By combining parameters like smoke concentration and module temperature, it determines thermal runaway risk levels. The system then takes appropriate actions based on the level to prevent escalation and protect the battery system.
20. Neural Network-Based Thermal Runaway Detection System for Energy Storage Assemblies
Shanghai Makesens Energy Storage Technology Co., Ltd., 2024
Early warning system to detect thermal runaway in energy storage systems like battery packs in electric vehicles or grid storage applications. The system uses a neural network model trained on historical temperature data to predict future temperatures. If the predicted temperature deviates significantly from actual readings or the reconstruction error is high, it indicates potential thermal runaway. This provides a more reliable and proactive warning compared to just monitoring individual sensor readings.
Get Full Report
Access our comprehensive collection of 68 documents related to this technology
Identify Key Areas of Innovation in 2025
