A.I.T.R.E.S.™: An Advanced Concept Mathematical Predictive Model for Real-Time Thermal Runaway Risk Prediction in Electric Vehicle Lithium-Ion Battery Systems
By Mohd Taufiq Bin Abd Sattar
Abstract
Electric vehicle adoption is rapidly transforming modern transportation, but it also introduces new challenges for fire safety, emergency response, battery diagnostics, and predictive risk management. One of the most critical hazards in electric vehicle lithium-ion battery systems is thermal runaway, a complex failure process involving heat generation, electrical instability, gas release, mechanical damage, and propagation between cells or modules.
The A.I.T.R.E.S.™ (Advanced Integrated Thermal Runaway Evaluation and Scoring System) model is proposed as a multi-parameter predictive framework designed to transform battery telemetry, environmental conditions, and sensor fusion data into three practical outputs:
Thermal Runaway Risk Score (TRRS)
Confidence-Adjusted Risk Output (CARO)
Estimated Time-to-Failure (ETTF)
The framework integrates six primary risk domains:
Thermal Instability
Electrical Anomalies
Gas Release
Mechanical Damage
Contextual Operating Conditions
Propagation Likelihood
Drawing upon lithium-ion battery thermal runaway research, heat transfer theory, electrochemical degradation mechanisms, and probabilistic risk modelling, A.I.T.R.E.S.™ provides a structured pathway for early detection and prediction of battery failure. The framework is intended for future application in EV battery management systems, charging infrastructure safety, battery testing laboratories, fire and rescue operations, and AI-assisted emergency response platforms.
This article introduces the conceptual and mathematical foundation of A.I.T.R.E.S.™, its operational value for electric vehicle safety, and its potential contribution to fire and rescue decision-making, battery monitoring systems, research validation, and future AI-assisted emergency response platforms.
1. Introduction
Lithium-ion batteries have become the dominant energy storage technology for electric vehicles due to their high energy density, long cycle life, and favourable power characteristics. Modern EV battery packs commonly operate between 300 V and 800 V and may contain hundreds or even thousands of individual cells storing energy levels exceeding 50 - 150 kWh.
Despite these advantages, lithium-ion batteries remain vulnerable to thermal runaway, a self-accelerating exothermic process that can result in:
Rapid temperature escalation, sometimes exceeding 800°C in localized regions
Toxic and flammable gas release
Jet flames and fireballs
Cell-to-cell thermal propagation
Reignition after suppression
Structural battery pack failure
Research by Feng et al. (2018), Finegan et al. (2017), Larsson et al. (2017), and various NREL battery safety studies demonstrates that thermal runaway is rarely triggered by a single parameter. Instead, it develops through the interaction of thermal, electrical, chemical, and mechanical degradation processes.
In practical terms, a battery may appear stable externally while internal decomposition has already begun. Temperature increases may only become obvious after significant internal damage has occurred. In other cases, gas generation, voltage abnormalities, or mechanical damage may provide earlier warning signs than temperature alone.
For this reason, predictive safety systems must move beyond simple threshold alarms and adopt a multi-dimensional risk assessment methodology. A.I.T.R.E.S.™ was developed to address this requirement.
2. Scientific Basis of Thermal Runaway
Thermal runaway generally progresses through several distinct operational and chemical stages:
| Normal Operation |
20 – 60°C |
Stable electrochemical operation |
| Early Degradation |
60 – 90°C |
SEI layer decomposition begins |
| Electrolyte Instability |
90 – 120°C |
Gas generation increases |
| Separator Failure |
120 – 150°C |
Internal short-circuit risk rises |
| Thermal Runaway Initiation |
150 – 250°C |
Exothermic reactions dominate |
| Full Thermal Runaway |
>250°C |
Self-sustaining heat generation |
Experimental studies indicate the following extreme environmental conditions during a failure event:
Internal temperatures may exceed 800 – 1000°C.
Cell pressure may exceed 1 – 2 MPa before venting.
Heat release rates may exceed 5 – 20 kW per cell depending on chemistry.
Thermal propagation between adjacent cells may occur within seconds to minutes.
These findings form the scientific foundation of the A.I.T.R.E.S.™ framework and reinforce the need for predictive monitoring rather than reactive detection.
3. Project Overview
A.I.T.R.E.S.™ stands as an advanced predictive model for electric vehicle lithium-ion battery safety. Its purpose is to support real-time thermal runaway risk prediction through mathematical modelling, sensor fusion, confidence adjustment, and dynamic escalation analysis.
The project is designed to support several future applications:
Electric vehicle battery safety research
Fire and rescue operational decision support
Battery management system enhancement
EV charging station safety monitoring
AI-assisted emergency response systems
PhD and journal-level research development
Thermal runaway experimental validation
Prototype algorithm and simulation development
The main objective of A.I.T.R.E.S.™ is to convert raw battery and environmental data into three practical outputs: Thermal Runaway Risk Score, Confidence-Adjusted Risk Output, and Estimated Time-to-Failure. These outputs allow responders, engineers, researchers, and system designers to understand not only whether a battery is at risk, but also how reliable the prediction is and how quickly the situation may escalate.
4. Problem Statement
Thermal runaway prediction remains difficult because lithium-ion battery failure is multi-dimensional. A battery may appear stable externally while internal degradation is already occurring. In some cases, temperature increase becomes obvious only after internal decomposition has already advanced. In other cases, electrical anomalies, gas release, or mechanical damage may occur before a visible fire or smoke condition is detected.
This creates several operational problems:
Firefighters may arrive after the battery has already entered an unstable phase.
EV users may receive insufficient early warning before a serious battery event.
Charging station operators may lack a proper risk score to determine whether to isolate power or evacuate the area.
Battery monitoring systems may generate alerts without explaining confidence level or escalation speed.
Researchers may have sensor data but lack a structured method to convert it into a predictive risk output.
A.I.T.R.E.S.™ attempts to solve this gap by creating a structured mathematical pathway from multi-sensor data to real-time risk interpretation.
5. Modelling Philosophy of A.I.T.R.E.S.™
The A.I.T.R.E.S.™ model is based on five core principles:
5.1 Multi-Parameter Fusion
Thermal runaway is not determined by one variable alone. Temperature, voltage, gas emission, impact history, state of charge, cooling performance, and propagation potential must be evaluated together.
5.2 Normalized Scoring
Different sensor values use different units. Temperature may be measured in degrees Celsius, voltage in volts, current in amperes, gas concentration in ppm, and deformation in millimetres. A.I.T.R.E.S.™ converts these raw values into normalized scores so they can be compared and combined seamlessly.
5.3 Weighted Contribution
Not all indicators contribute equally. Thermal instability may carry higher predictive importance than some contextual indicators, but contextual stress may still influence the final risk score. The model therefore allows different indicators to be weighted according to experimental evidence and calibration.
5.4 Confidence-Aware Prediction
A prediction is only useful if the system understands how reliable the data is. A.I.T.R.E.S.™ includes a confidence factor based on sensor reliability, data completeness, model agreement, and similarity to known training cases.
5.5 Dynamic Forecasting
Battery risk is not static. A battery may move from normal to advisory, warning, high risk, and critical condition within a short period. The model therefore includes an escalation momentum factor to reflect how fast the hazard is developing.
6. The Six Main Risk Indicator Groups
A.I.T.R.E.S.™ uses six major indicator groups to evaluate the risk of thermal runaway.
6.1 Thermal Instability Index (TII)
The Thermal Instability Index evaluates heat-related battery behaviour. It includes temperature rise rate, temperature acceleration, hotspot intensity, cell-to-cell temperature deviation, and module-to-module thermal deviation.
This index is one of the strongest predictors because abnormal heat generation is closely linked to internal failure, separator breakdown, electrolyte decomposition, and propagation risk.
\[TII\ = \ 0.35T\_ r\ + \ 0.25T\_ a\ + \ 0.20H\_ s\ + \ 0.10\backslash Delta\ T\_ c\ + \ 0.10\backslash Delta\ T\_ m\]
Where:
\(T\_ r\) = temperature rise rate
\(T\_ a\) = temperature acceleration
\(H\_ s\) = hotspot severity
\(\backslash Delta\ T\_ c\) = cell temperature deviation
\(\backslash Delta\ T\_ m\) = module temperature deviation
Typical Interpretation:
Normal rise rate: \(< 1{^\circ}C/min\}\)
Elevated concern: \(> 3{^\circ}C/min\}\)
Critical escalation: \(> 10{^\circ}C/min\}\)
Example Case: If temperature increases from 45°C to 75°C within 5 minutes:
\[\frac{75\, - \, 45}{5}\, = \, 6\text{°C}\text{/}\text{min}\]
This significantly exceeds normal operating behavior and indicates severe abnormal internal heat generation.
6.2 Electrical Anomaly Index (EAI)
The Electrical Anomaly Index evaluates abnormal electrical behaviour inside the battery system. It includes voltage imbalance, voltage decay rate, current oscillation, internal resistance growth, and overcharge stress. Electrical anomalies are important because internal short circuits, overcharging, and resistance growth may occur before obvious external signs of fire break out.
\[EAI = 0.30V_{i} + 0.25V_{d} + 0.20I_{o} + 0.15R_{g} + 0.10OC_{s}\]
Where:
\(V_{i}\) = voltage imbalance
\(V_{d}\) = voltage decay
\(I_{o}\) = current oscillation
\(R_{g}\) = resistance growth
\(OC_{s}\) = overcharge stress
Typical Warning Indicators:
Cell imbalance \(> 50mV\)
Resistance increase >20\%
Unexpected sudden voltage collapse
Abnormal high-voltage charging behavior
### 6.3 Gas Release Index (GRI)
The Gas Release Index evaluates the presence and trend of gas emissions associated with battery degradation and decomposition. This includes hydrogen, carbon monoxide, volatile organic compounds, and gas release rate. Gas release is a critical early-warning indicator because it signals that internal chemical reactions are occurring before visible flaming begins.
\[GRI = 0.40CO + 0.25H_{2} + 0.20VOC + 0.15G_{r}\]
Where:
| \[\text{H}_{\mathbf{2}}\] |
Electrolyte decomposition |
| \[\text{CO}\] |
Cathode degradation reactions |
| \[\text{CO}_{\mathbf{2}}\] |
Solvent breakdown |
| \[\text{CH}_{\mathbf{4}}\] |
Hydrocarbon structural decomposition |
| \[\text{VOCs}\] |
Electrolyte off-gassing and venting |
Strategic Value: Diagnostic studies suggest detectable off-gas venting events may occur between 5 and 30 minutes prior to full thermal runaway propagation, making gas sensing an invaluable early-warning layer.
6.4 Mechanical Damage Index (MDI)
The Mechanical Damage Index evaluates physical degradation conditions that increase the structural probability of battery failure. These include impact severity, vibration anomaly, enclosure deformation, and delayed post-impact anomaly behaviour.
\[MDI = 0.35I_{s} + 0.25D_{f} + 0.20V_{a} + 0.20P_{a}\]
Where:
\(I_{s}\) = impact severity
\(D_{f}\) = deformation factor
\(V_{a}\) = vibration anomaly
\(P_{a}\) = post-impact anomaly
This metric is highly relevant for emergency responders attending automobile accidents, severe underbody strikes, battery enclosure intrusions, and post-crash structural battery assessments.
6.5 Contextual Operating Index (COI)
The Contextual Operating Index evaluates environmental and macro-operational factors that elevate baseline risk profiles. A battery operating at high state of charge, under rapid fast-charging parameters, exposed to extreme ambient temperatures, or running with compromised cooling lines exhibits a significantly increased vulnerability to terminal runaway.
\[COI = 0.35\text{SOC} + 0.25A_{T} + 0.20C_{R} + 0.20C_{D}\]
Where:
\(\text{SOC}\) = State of Charge
\(A_{T}\) = Ambient Temperature
\(C_{R}\) = Charging Rate
\(C_{D}\) = Cooling Degradation
Empirical Vulnerability Constraints:
\(\text{SOC} > 90\backslash\%\) exponentially increases structural runaway thermal severity.
Fast charging parameters \(\left( > 2\text{C} \right)\) elevate continuous local thermal stress.
Ambient temperatures \(> 40\text{°C}\) accelerate internal chemical degradation pathways.
6.6 Propagation Likelihood Index (PLI)
The Propagation Likelihood Index evaluates whether a localized single-cell failure will spread cascade-style to adjacent target cells or modules.
\[PLI = 0.40N_{H} + 0.30M_{G} + 0.30S_{R}\]
Where:
Experimental studies demonstrate cascade propagation velocities ranging widely from \(0.5\text{ to }10\text{ cm}\text{/}\text{min}\) depending heavily on specific pack chemistry, interstitial cell spacing configurations, cooling architecture, and structural pack materials.
7. Composite Thermal Runaway Risk Score
The six individual indices are combined into a standardized composite base thermal runaway risk score (TRRS), mapped to a robust scalar range from 0 to 100. Each distinct index contributes according to an engineered calibration weight, prioritizing thermal and electrical telemetry over environmental operational inputs.
| 0 – 20 |
Normal |
Standard Monitoring |
| 21 – 40 |
Advisory |
Increase Telemetry Frequency |
| 41 – 60 |
Warning |
Pre-emptive Cooling / System Alert |
| 61 – 80 |
High Risk |
Isolate / Evacuate Area |
| 81 – 100 |
Critical / Imminent Runaway |
Emergency Response Activation |
8. Confidence-Adjusted Risk Output (CARO)
A.I.T.R.E.S.™ evaluates structural data confidence alongside raw risk parameters to prevent sensor drops, noise, or incomplete datasets from skewing system predictions.
The Confidence Factor (CF) is mathematically formulated as:
\[CF\ = \ 0.35SR\ + \ 0.25DC\ + \ 0.20MA\ + \ 0.20TS\]
Where:
\(SR\) = Sensor Reliability
\(DC\) = Data Completeness
\(MA\) = Model Agreement
\(TS\) = Training Similarity
The final Confidence-Adjusted Risk Output (CARO) is evaluated as:
\[CARO\ = \ TRRS\ \times CF\]
Operational Safety Guardrail: A raw risk score of 82 paired with a degraded confidence factor of 0.60 yields a CARO value of 49.2. This adjustment indicates high data uncertainty rather than a safe battery condition. In safety-critical automation, an alert must be triggered if either TRRS or CARO crosses their respective safety thresholds, ensuring that compromised data confidence never dangerously masks an imminent thermal crisis.
9. Escalation Momentum Factor (EMF)
Battery hazard progression is dynamic. A cell displaying a moderate risk score paired with an aggressive rate of temperature and gas escalation presents a more critical hazard than a high-score cell that has stabilized. The Escalation Momentum Factor (EMF) addresses this by quantifying the rate of failure acceleration.
\[EMF = 0.40\frac{dT}{dt} + 0.25\frac{d^{2}T}{dt^{2}} + 0.20\frac{dV}{dt} + 0.15\frac{dG}{dt}\]
Where:
\(\frac{dT}{dt}\) = temperature rise rate
\(\frac{d^{2}T}{dt^{2}}\) = thermal acceleration
\(\frac{dV}{dt}\) = voltage decay rate
\(\frac{dG}{dt}\) = gas generation rate
Mathematical Scaling Note: Because raw telemetry derivatives utilize radically different physical units (e.g., °C/min versus V/min or ppm/min), all raw input gradients are passed through a localized, non-linear min-max scaling function prior to index calculation. This normalizes each derivative into a dimensionless scalar range \((0 \leq x \leq 100)\), ensuring that no single parameter disproportionately dominates the EMF weight matrix.
10. Estimated Time-to-Failure (ETTF)
A primary contribution of the A.I.T.R.E.S.™ architecture is computing a clear remaining time window before terminal structural failure occurs, providing vital operational data to emergency crews.
A simplified hazard-based model for ETTF estimation is expressed as:
\[ETTF = \frac{K}{TRRS \times EMF}\]
Where:
System Gating and Boundary Constraints: To prevent division-by-zero errors or infinite temporal outputs \((\infty)\) during nominal baseline vehicle operation, the ETTF computation engine features an automated logical gate. The algorithm remains dormant while the battery is within standard parameters \((TRRS \leq 20)\), and dynamically activates only when an active, non-zero escalation trend \((EMF > 0)\) is verified.
Operational Output Example \(\left( \mathbf{K = 10000,TRRS = 80,EMF = 6} \right)\):
\[ETTF = \frac{10000}{80 \times 6} = 20.8\text{ minutes}\]
| Risk Score |
80 |
| Confidence Factor |
0.85 |
| Mapped Probability |
85.8% |
| Estimated Time-to-Failure |
20.8 minutes |
| System Status |
High Risk |
11. Probability Mapping
A.I.T.R.E.S.™ converts the combined multi-parameter framework into an explicit, highly readable probability metric using a logistic sigmoidal mapping equation. This delivers an actionable, easy-to-interpret percentage value (e.g., Probability of Runaway: 72% / Confidence: 0.81) that can be immediately utilized by non-technical dispatchers, commanders, and emergency personnel.
12. STEMPCHEM™ Interpretation of A.I.T.R.E.S.™
The interdisciplinary nature of the A.I.T.R.E.S.™ platform is cleanly evaluated through the holistic STEMPCHEM™ framework:
Science: Grounded in understanding battery failure chemistry, heat propagation pathways, and gas generation kinematics.
Technology: Encompasses multi-sensor physical integration, telemetry data links, dashboard interfaces, and edge computing algorithms.
Engineering: Addresses physical cell geometries, structural cooling lines, structural containment, and field charging interfaces.
Mathematics: Built upon normalization arrays, linear scoring weight matrix scaling, confidence vectors, and logistic sigmoidal curves.
Physics: Tracks core thermodynamic heat transfer laws, thermal gradients, mechanical forces, and electrical circuit dynamics.
Chemistry: Evaluates SEI layer decomposition pathways, toxic off-gassing reactions, and exothermic structural cell degradation.
13. Application in Fire and Rescue Operations
For fire and rescue services, A.I.T.R.E.S.™ supports crucial tactical decision-making across all operational phases of an EV incident:
A.I.F.R.T.™ (Artificial Intelligence Fire Rescue Tactics): Integrates directly with advanced incident management models to feed real-time risk scores into tactical command modules.
P.L.A.C.E.™ (Push Isolation, Layout Blanket, Approach Safely, Cover Vehicle, Evaluate & Coordinate): Provides critical timing data for rapid deployment and positioning of specialized equipment.
L.A.C.E.™ (Locate Fire Blanket, Approach Safely, Cover Vehicle, Evaluate & Coordinate): Delivers a definitive safety window (ETTF) before teams approach a vehicle to deploy specialized containment blankets.
Post-Incident Overhaul: Guides tactical isolation decisions, perimeter distances, and monitors long-term cell cooling to mitigate thermal reignition risks.
14. Application in EV Charging Station Safety
EV charging stations represent high-stress environments due to rapid electrical and thermal energy transfer. Integrating A.I.T.R.E.S.™ into charging nodes enables early localized isolation actions before a vehicle enters an uncontrollable failure state. Warning triggers such as charging current anomalies or local gas concentration changes interface directly with facility safety loops to trigger power disconnects, deploy on-site containment assets, or automatically alert dispatch channels.
15. Research and Validation Strategy
Empirical validation requires running the predictive engine against controlled physical abuse test profiles:
Overcharging and deep external electrical short circuits.
Mechanical drop, crush, and direct nail-penetration abuse profiles.
Localized heater-induced thermal runaway and cooling loss simulations.
System optimization criteria must prioritize maximizing valid early-warning lead times while driving down false-alarm rates (FAR) and keeping missed detection rates (MDR) close to zero, ensuring absolute operational dependability for emergency crews.
16. Limitations and Future Development
Several developmental gates remain before wide-scale commercial field application:
Calibration: Fixed weight profiles must be validated across varied modern battery chemistries (e.g., LFP, NMC, Solid-State).
Hardware Durability: Physical sensor arrays must retain operational accuracy under high heat, blast forces, submersion, and impact.
Cross-Validation: The software must be tested against non-runaway stress cases to prevent system over-sensitisation.
Future targets include cross-linking predictive models with automated mitigation loops, like localized emergency cooling activation or active cell venting containment.
17. Conclusion
A.I.T.R.E.S.™ represents a structured advancement in electric vehicle battery safety, fire science, and emergency response technology. By combining thermal, electrical, gas, mechanical, contextual, and propagation indicators into a single predictive model, A.I.T.R.E.S.™ moves beyond simple threshold-based warning systems.
Its strength lies in its ability to generate a real-time Thermal Runaway Risk Score, adjust that score according to confidence, evaluate escalation momentum, estimate time-to-failure, and support operational decision-making. As electric vehicle adoption continues to grow, predictive safety systems must evolve beyond passive monitoring. A.I.T.R.E.S.™ provides a clear pathway toward proactive, intelligent, and scientifically structured thermal runaway risk prediction.
Technical Signature Statement
The A.I.T.R.E.S.™ Advanced Mathematical Predictive Model provides a structured quantitative pathway for translating battery telemetry, thermal behaviour, gas generation, electrical anomalies, mechanical damage, operating context, and propagation likelihood into real-time prediction of electric vehicle thermal runaway risk, confidence, and expected escalation time.
Author / Project Credit
Developed under the research and innovation direction of Mohd Taufiq Bin Abd Sattar, with application toward electric vehicle safety, fire and rescue emergency response, thermal runaway research, and AI-assisted predictive fire science.
