1.Main text

In practical applications, lithium iron phosphate batteries are often exposed to risk scenarios such as overcharging, mechanical damage, and extreme temperatures. The thermal runaway caused by overcharging is particularly tricky. Its essence is the continuous accumulation of heat inside the battery, which leads to the decomposition of the electrolyte, the degradation of electrode material performance, and ultimately triggers an uncontrollable chain reaction, resulting in catastrophic consequences such as fires and explosions.

Compared to instantaneous mechanical injuries such as puncture and compression, thermal runaway induced by overcharging exhibits a unique pattern: the entire deterioration process is relatively slow and accompanied by monitored characteristic signals, including gradual temperature rise, characteristic gas escape, and battery structure deformation, providing a technical window for early warning.

However, current research has significant limitations: firstly, single modal monitoring methods are often used (such as focusing only on temperature or voltage), which have poor anti-interference ability under complex working conditions and are prone to misjudgment or missed detection; Secondly, the early warning accuracy for thermal runaway (before the safety valve is opened) is insufficient, resulting in missed opportunities for risk intervention; Thirdly, emergency response relies heavily on experience driven approaches and lacks a quantitative decision-making mechanism based on real-time data, making it difficult to cope with the severe challenges of rapid spread, high toxicity, and high temperature of battery fires.

To this end, this study conducted five sets of overcharge simulation experiments with different charging rates, comprehensively collected multimodal data, constructed a three-stage dynamic evolution model of thermal runaway, developed a dual machine learning intelligent warning algorithm, and innovatively designed a three-dimensional emergency decision-making matrix to form a full process safety management system of “data collection intelligent warning precise response”, focusing on breaking through the technical bottleneck of overcharge safety control of lithium iron phosphate batteries.

2.Experimental design and three-stage model of thermal runaway

2.1 Experimental Plan Design

2.1.1 Battery Samples and Preprocessing

Select commercial square lithium iron phosphate batteries with a rated capacity of 32Ah and a nominal voltage of 3.2V. Before testing, perform 3 charge and discharge pretreatments on the battery: first discharge at a constant current of 0.5C to 2.5V, and then charge at a constant current constant voltage (CC-CV) of 0.5C to 3.65V (current reduced to 0.5A and maintained for 30 seconds) to ensure experimental stability and data reproducibility.

2.1.2 Experimental conditions and data collection

Under a constant temperature environment of 35 ℃, set five charging rates ranging from 0.5C (16A) to 1.5C (48A), and repeat the test twice for each operating condition. Collecting data through multimodal devices:

Temperature: battery surface (average of double K-type thermocouples), flue gas layer (three thermocouple array), accuracy of 0.1 ℃

Voltage: Adjustable power supply real-time monitoring, sampling frequency 1Hz

Gas: The probe at the top 5cm detects the concentrations of four gases, including CO and H ₂

Physical signals: electronic scales record changes in quality, industrial cameras capture deformations, and voiceprint devices collect sound signals

2.2 Three stage evolution model of thermal runaway

Based on the core parameters of voltage and temperature, combined with changes in gas composition and physical form, the overcharging thermal runaway process is divided into three stages:

2.2.1 Stage I: Overcharge start to safety valve opening

Characteristics: After the voltage exceeds 3.65V, it shows a plateau period, followed by a slight decrease due to electrochemical reactions; The temperature rises to 47.6-63.65 ℃, and the shell expands by 1.48-1.78cm

Flag: Internal pressure triggers safety valve, at this state of charge (SOC) of 110.8% -115.9%

Key signal: Early deformation, presence of trace amounts of CO gas

2.2.2 Stage II: Safety valve opens until voltage rebounds

Features: Gas continuously escapes, voltage first drops and then rises (peak 4.96-5.11V), temperature rises to 70.35-92.6 ℃

Flag: The voltage has reached the maximum value of the stage, SOC 120%-125.6%

Key signals: sound of safety valve opening, fluctuation of CO concentration, detection of hydrogen gas

2.2.3 Stage III: Voltage and temperature sudden change to thermal stability

Features: Short circuit returns to zero after sudden voltage rise, temperature exceeds 300 ℃, accompanied by a large amount of thick smoke

Sign: Temperature drops to 200 ℃, thermal runaway ends

Key signals: severe fluctuations in CO concentration, double explosion sound, release of highly toxic gases

Experiments have shown that the charging rate significantly affects the thermal runaway process: under 1.5C conditions, the entire process only takes 527s, and the toxicity index (FED) reaches 0.792; At 0.5C working condition, it takes 2142s. At high magnification, gas release is severe, and the risk of explosion increases exponentially.

3.Multi modal characteristic signal analysis of thermal runaway

The dynamic changes of multimodal signals are the key basis for dividing thermal runaway stages and triggering warnings, with a focus on analyzing the core signal characteristics of gas and physics

3.1 Gas signal: CO as the core warning indicator

Experimental monitoring shows that the release patterns of four characteristic gases, CO, H ₂, CH ₄, and HF, are closely related to the process of thermal runaway:

CO: Stable detection under all operating conditions, with significant concentration fluctuations observed in Stage II after the safety valve is opened and Stage III after a short circuit. Under 1.5C charging and discharging conditions, the peak concentration of CO can reach 10000ppm, which is detected 30-60 seconds earlier than H ₂ and is the most reliable early warning signal

H ₂: Only occurs in the 1.0C-1.5C operating conditions, with concentrations consistently lower than CO and irregular fluctuations, and is not suitable for individual warning

CH ₄ and HF: The detection probability is low (only 2 working conditions) and there is a significant lag. The highest concentration of CH ₄ reaches 3.49% of the lower explosive limit, and the concentration of HF is always below the safety threshold of 30ppm, with limited warning value

The FED model evaluation based on ISO 13571 standard shows that the toxicity risk index reaches 0.792 under 1.5C condition and 0.617 under 0.5C condition. When multiple battery cells lose control simultaneously, the cumulative dose of toxic gases may exceed the lethal threshold, posing a serious threat to personnel safety.

3.2 Physical Signals: Stage Characteristics of Deformation and Sound

Deformation: In the initial stage of thermal runaway (Stage I), the battery thickness increases by 1.48-1.78cm, resulting in a mass loss of 136-137g. When the deformation exceeds 1.5cm, the probability of the safety valve opening exceeds 90%, and internal pressure changes can be monitored through visual inspection or thickness sensors

Sound: Stage II is characterized by short bursts or continuous hissing sounds produced by the opening/rupture of safety valves; Stage III experienced two significant sonic explosions due to gas production and violent reactions caused by short circuits. By analyzing the time-domain characteristics such as amplitude and frequency of sound signals, it can assist in determining the transition of thermal runaway stages

Visibility: Stage III saw a significant decrease in environmental visibility due to the release of a large amount of smoke. When the extinction coefficient exceeds 0.5m ⁻¹ and the visibility drops below 1m, remote cooling measures should be immediately activated to prohibit personnel from entering the site recklessly