Lithium battery thermal runaway gas warning system

Abstract

The present invention belongs to the technical field of thermal runaway warning for lithium batteries in new energy vehicles, and specifically discloses a lithium battery thermal runaway gas warning system. It is used to solve the problem that existing monitoring technologies cannot quickly and accurately detect thermal runaway and provide early warning. The warning system includes a main control unit, a core processing unit, a graphical user interface, a dual-frequency drive signal generating circuit, a spectral signal dual-frequency demodulation circuit, a temperature control circuit, a constant current source and overcurrent protection circuit, a spectral data acquisition circuit, a graphical user interface, an optical system, a gas sampling device, and a warning model; the power supply of the warning system adopts a three-stage step-down design, which includes an AC-DC switching power supply, a DC-DC switching power supply, and a low-voltage difference linear regulator. The present invention is based on the off-axis integrating cavity optical path coupling technology of a dual-optical wedge group, and accurately adjusts the light beam angle through a stepper motor dual-optical wedge, without the need for a beam combiner, thereby improving light energy utilization and detection sensitivity.

Description

Technical Field

[0001] The present invention belongs to the technical field of thermal runaway warning technology for lithium batteries in new energy vehicles, and in particular relates to a lithium battery thermal runaway gas warning system. Background Art

[0002] Lithium-ion batteries, with their high energy density and excellent endurance, have become the preferred power battery for new energy vehicles. However, during the charge and discharge cycle, lithium-ion batteries may suffer internal short circuits due to mechanical damage, or the battery voltage may exceed the normal operating range due to abnormal operations such as overcharging or over-discharging, or abuse such as long-term use in high-temperature environments, all of which significantly increase the risk of thermal runaway. During thermal runaway, high temperatures can trigger side reactions such as decomposition of the negative electrode SEI film, violent decomposition of the positive electrode active material, and oxidative decomposition of the electrolyte. This not only generates a large amount of heat, but also releases toxic gases such as CO, CO2, CH4, and C2H6. The accumulation of these byproducts not only exacerbates the safety risks of the battery system, but also poses a serious threat to vehicle occupants and the surrounding environment if they leak.

[0003] It's worth noting that in practical applications, the response time of lithium battery thermal runaway detection is crucial. Thermal runaway develops extremely quickly, progressing from its initial stages to a violent reaction or even explosion in a very short period of time. If the detection system's response time is too long, it's like a delayed alarm, potentially preventing effective action from being taken in a timely manner, leading to a rapid escalation of the danger. Therefore, rapid and accurate detection of thermal runaway and early warning are crucial for ensuring the safety of new energy vehicles and their passengers, and minimizing property losses and social impact.

[0004] Current monitoring technologies have limitations in response time. While traditional temperature monitoring and voltage measurement can reflect the battery's operating status, experimental studies have shown that in the early stages of thermal runaway, changes in parameters such as temperature and voltage are often slow, making it difficult to provide timely warnings. At this stage, the electrochemical reactions within the battery have significantly intensified, and various characteristic gases (such as CO and CO2) begin to be released into the environment. This phenomenon provides a theoretical basis for the application of gas monitoring technology, which can more efficiently identify potential thermal runaway risks by detecting changes in gas composition and concentration in real time. Machine learning models, with their powerful data processing and analysis capabilities, excel at handling complex data relationships. They may provide new ideas and methods for improving the response time of lithium battery thermal runaway monitoring, helping us to more promptly detect signs of thermal runaway and achieve more efficient and accurate early warnings.

[0005] Currently, the main gas detection technologies for lithium-ion battery thermal runaway include gas chromatography, electrochemical detection, and infrared absorption spectroscopy. Compared to other technologies, infrared absorption spectroscopy not only has higher sensitivity and detection accuracy, but also can achieve real-time online monitoring of the gas being tested.

[0006] Mainstream infrared absorption spectroscopy detection technologies include non-dispersive infrared (NDIR), tunable diode laser absorption spectroscopy (TDLAS), and off-axis integrated cavity output spectroscopy (OA-ICOS). In the early stages of lithium battery thermal runaway, CO2 concentrations gradually increase, but initially may only be in the tens of ppm. NDIR may require CO2 concentrations above 100 ppm to accurately detect changes. OA-ICOS, on the other hand, can detect a significant signal at CO2 concentrations as low as 10-20 ppm. Furthermore, NDIR has a relatively long response time, typically ranging from several to more than ten seconds, while OA-ICOS can respond within one second, providing a more timely reflection of gas concentration changes. Although TDLAS also has high sensitivity, the concentration of C2H6 generated by lithium battery thermal runaway is very low, making it difficult to detect changes using TDLAS. OA-ICOS, on the other hand, has a theoretical detection limit of sub-ppb, allowing it to accurately detect changes in C2H6 concentrations.

[0007] Compared with TDLAS and NDIR technologies, OA-ICOS technology not only has higher sensitivity and faster response time, but also has a longer effective optical path, which can achieve a lower detection limit. Therefore, the development of an OA-ICOS-based lithium battery thermal runaway warning system is of great significance. Summary of the Invention

[0008] The purpose of the present invention is to provide a lithium battery thermal runaway gas warning system to effectively solve the problem that existing monitoring technologies are unable to quickly and accurately detect thermal runaway and provide early warning.

[0009] To address the above technical issues, the present invention adopts the following technical solution: a lithium battery thermal runaway gas warning system includes a main control unit (STM32F407VET6), a core processing unit (AM4378 Cortex A9), a graphical user interface, a dual-frequency drive signal generation circuit, a spectral signal dual-frequency demodulation circuit, a temperature control circuit, a constant current source and overcurrent protection circuit, a spectral data acquisition circuit, an optical system, a gas sampling device, and a warning model. The temperature control circuit comprises an ADN8834 and its peripheral circuits, the gas sampling device comprises a filter and an air pump, and the spectral data acquisition circuit comprises an ADC7606 and its peripheral circuits.

[0010] The main control unit is used to control the DAC chip to generate a low-frequency sawtooth wave signal, control the DDS chip to generate a high-frequency sine wave signal, read the second harmonic signal, laser temperature and gas pool pressure collected by the ADC chip, and transmit them to the core processing unit; the core processing unit is used to run the Linux system, take charge of running the graphical user interface, and run gas concentration inversion, calibration and early warning models; the graphical user interface is used to realize real-time display of multi-spectral information, laser parameters and concentration display.

[0011] Furthermore, the dual-frequency driving signal generating circuit includes two DAC chips, two DDS chips, a digital potentiometer and an adder.

[0012] The two DAC chips each generate two sawtooth wave signals with equal frequencies, and one DAC chip generates a PWM wave; the two DDS chips each generate two sine wave signals with frequencies higher than the sawtooth wave signal but different from the frequency; the digital potentiometer changes the amplitude of the sine wave according to the magnitude of the DFB laser drive current and ensures that the amplitude of the sine wave meets the optimal modulation depth optimization condition; the adder uses the two sine waves with different frequencies and changed amplitudes as carriers and superimposes them on the sawtooth wave signal.

[0013] Furthermore, the spectrum signal dual-frequency demodulation circuit includes a pre-transimpedance amplifier, a dual-channel bandpass filter and a dual-channel lock-in amplifier.

[0014] The pre-transimpedance amplifier is used to convert the current signal output by the detector into a voltage signal and amplify it; the dual-channel bandpass filter is composed of two bandpass filters with different center frequencies, and is used to eliminate interference from frequency signals other than the second harmonic signal frequency; the dual-channel lock-in amplifier is constructed by two independently operating single-channel lock-in amplifiers to extract the second harmonic signals corresponding to the concentrations of two characteristic gases respectively.

[0015] Furthermore, there are two temperature control circuits, namely the first temperature control circuit and the second temperature control circuit; there are two constant current sources and overcurrent protection circuits, namely the first constant current source and overcurrent protection circuit and the second constant current source and overcurrent protection circuit, and the first constant current source and overcurrent protection circuit and the second constant current source and overcurrent protection circuit both include a high-precision operational amplifier, an N-channel field-effect transistor, a voltage comparator and an NPN transistor.

[0016] Furthermore, the optical system includes a DFB laser, a high-reflection mirror, a collimator, a converging lens, a detector, a double wedge set, a stepping motor and an off-axis integrating cavity.

[0017] The DFB laser is used to provide a stable laser light source with a wavelength that matches the absorption line of the target gas. There are two DFB lasers, namely a 1580nm DFB laser and a 1653.7nm DFB laser. The high-reflection mirror is used to make the laser reflect back and forth hundreds to thousands of times in the off-axis integrating cavity, thereby extending the effective optical path. The collimator is used to convert the divergent laser beam into a parallel beam, ensuring that the beam enters the off-axis integrating cavity in a parallel manner. There are two collimators, namely a first collimator and a second collimator. The converging lens is used to focus the transmitted light onto the detector. The detector is used to convert the optical signal into an electrical signal. The double wedge group controls the direction of the output light beam by changing the angle and position of the double wedge. There are two double wedge groups, namely a first double wedge group and a second double wedge group. The stepper motor is used to adjust the wedge angle. There are two stepper motors, namely a first stepper motor and a second stepper motor.

[0018] Furthermore, the early warning model includes a sensor module, a data acquisition module, a data preprocessing module, a random forest machine learning model and an alarm device.

[0019] The sensor module includes a temperature sensor for measuring the surface temperature of the lithium battery, an internal resistance tester for measuring the internal resistance of the lithium battery, a current sensor for measuring the current of the lithium battery, a voltage sensor for measuring the voltage of the lithium battery, a pressure sensor for measuring the internal pressure of the lithium battery, a deformation sensor for measuring the dimensional deformation of the lithium battery, and a gas sensing system for measuring the concentration of characteristic gases such as CO, CO2, CH4, and C2H6.

[0020] The data acquisition module uses ADC7609 and ADC7606 to achieve synchronous acquisition of multi-channel signals, and transmits the acquisition results to the main control unit in real time. The processed signals are uploaded to the core processing unit via SPI communication, and data such as gas concentration, voltage, current, temperature, internal resistance, deformation and pressure are obtained in the graphical user interface. The concentration data of CO, CO2, CH4 and C2H6 at different warning levels are used to form the original training samples.

[0021] The data preprocessing module uses standardization and outlier processing algorithms to clean and optimize the features of the original data; the random forest machine learning model constructs multiple decision trees based on the Bagging integrated learning framework, enhances prediction robustness through a double random mechanism, integrates the prediction results of each decision tree through a voting mechanism, and uses the warning level with the most votes as the final output; the alarm device includes a buzzer and an indicator light, and implements a three-level warning mechanism based on the prediction results of the random forest machine learning model.

[0022] Furthermore, the power supply of the early warning system adopts a three-stage step-down design, which includes an AC-DC switching power supply, a DC-DC switching power supply and a low voltage difference linear regulator.

[0023] Furthermore, the three-level warning mechanism includes level one warning, level two warning and level three warning. The warning level requirements are as follows: Temperature warning: When the surface temperature of the single battery 55℃ triggers the first level warning. 60℃ triggers the second level warning, when the surface temperature of the single battery A level 3 warning is triggered at 70°C.

[0024] Voltage and current warning: When the voltage or current fluctuation exceeds the normal range 5% or the voltage difference between single cells exceeds 0.1V, triggering a level 1 warning. 10% or the voltage difference between single cells exceeds 0.15V, triggering the second level warning. When the voltage difference between cells exceeds 20% or 0.2V, a level 3 warning is triggered.

[0025] Internal pressure warning: When the internal pressure exceeds 80% of the battery's maximum design pressure, a level one warning is triggered; when the internal pressure exceeds 90% of the battery's maximum design pressure, a level two warning is triggered; when the internal pressure exceeds 100% of the battery's maximum design pressure, a level three warning is triggered.

[0026] Deformation warning: When the battery size exceeds 5% of its design size, a level one warning is triggered; when the battery size exceeds 8% of its design size, a level two warning is triggered; when the battery size exceeds 10% of its design size, a level three warning is triggered.

[0027] Characteristic gas warning: When the carbon monoxide concentration 50ppm triggers a first-level warning. 80ppm triggers the second level warning, when the carbon monoxide concentration 100ppm triggers the third level warning.

[0028] Compared with existing technologies, the present invention offers the following advantages: It utilizes a multi-gas sensing system and a random forest machine learning model to achieve early and accurate early warning of lithium battery thermal runaway. Through feature screening and model optimization, it significantly improves detection speed and accuracy, achieving real-time early warning with millisecond-level response.

[0029] (1) The present invention establishes a gas warning model through the random forest algorithm and designs a three-level alarm mechanism (level one, level two, and level three warning). It can quickly judge the level of thermal runaway according to the change of gas concentration and issue the corresponding level of alarm signal in time, thereby improving the reliability and practicality of the warning system.

[0030] (2) The present invention is based on the off-axis integrating cavity optical path coupling technology of the double optical wedge group. The double optical wedge of the stepper motor accurately adjusts the beam angle without the need for a beam combiner, so that two laser beams are coupled into the off-axis integrating cavity in parallel with a very small distance, thereby improving the light energy utilization rate and detection sensitivity.

[0031] (3) By analyzing key parameters such as CO, CO2, CH4, and C2H6 concentrations and warning levels, this method eliminates redundant features, reduces computational complexity by over 60%, and lowers model inference complexity. Using a stratified five-fold cross-validation method, 500 decision trees were reduced to 300, and inference speed was improved to 10–15 ms while maintaining 98% accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 It is a schematic diagram of the connection structure of the lithium battery thermal runaway gas warning system of the present invention.

[0033] Description of the accompanying drawings: first double-wedge group-1; second double-wedge group-2; first stepper motor-3; second stepper motor-4; off-axis integrating cavity-5; converging lens-6; filtering device-7; air pump-8. DETAILED DESCRIPTION

[0034] Example 1: Currently, the main gases generated by lithium battery thermal runaway are CO, CO2, CH4, and C2H6. In order to meet the measurement requirements of lithium battery thermal runaway gases, this embodiment provides a lithium battery thermal runaway gas early warning system, such as Figure 1 As shown, it includes a main control unit (STM32F407VET6), a core processing unit (AM4378 Cortex A9), a graphical user interface, a dual-frequency drive signal generation circuit, a spectral signal dual-frequency demodulation circuit, a temperature control circuit, a constant current source and overcurrent protection circuit, a spectral data acquisition circuit, an optical system, a gas sampling device, and an early warning model.

[0035] In this embodiment, the main control unit (STM32F407VET6) has the following core functions: controlling the DAC chip to generate a low-frequency sawtooth signal, controlling the DDS chip to generate a high-frequency sine wave signal, and reading the second harmonic signal, laser temperature, and gas cell pressure data collected by the ADC chip. The collected data is then transmitted to the core processing unit (AM4378 CortexA9) for processing.

[0036] The core processing unit (AM4378 Cortex A9) mainly runs the Linux system, runs the graphical user interface, and performs gas concentration inversion, calibration, and graded warning model operations.

[0037] Regarding the graphical user interface, the core processing unit (AM4378 Cortex A9) can run the Linux operating system. Therefore, the applicant designed a real-time monitoring software system for multi-component gas detection based on the QT Creator development platform that supports the Linux system. It is based on a real-time control GUI system and is equipped with an LCD touch screen on the hardware to realize functions such as laser drive current parameter modification, laser temperature parameter modification, demodulation and modulation parameter control, standard concentration calibration algorithm, decoupling algorithm, concentration inversion algorithm, real-time display of multi-spectral information, laser parameter and concentration display, and host computer communication and spectral data transmission control.

[0038] The dual-frequency drive signal generation circuit includes two DAC chips, two DDS chips, a digital potentiometer, and an adder. The main control unit controls the two DAC chips and two DDS chips to generate modulation signals, and controls one DAC chip to generate PWM waves. The two DAC chips each generate two sawtooth wave signals with lower and equal frequencies, while the other generates a PWM wave. The two DDS chips each generate two sine wave signals with frequencies much higher than the sawtooth signals but with different frequencies. After the two sine wave signals pass through the digital potentiometer, their amplitudes are modified according to the drive current of the DFB laser, meeting the optimal modulation depth optimization conditions. The adder superimposes the two sine waves with different frequencies and modified amplitudes as carrier waves onto the sawtooth wave signals.

[0039] The spectral signal dual-frequency demodulation circuit includes: (1) a pre-transimpedance amplifier, which converts the current signal output by the detector into a voltage signal and amplifies it. The current signal flows through the feedback resistor, which is a sliding rheostat that can adjust the gain of the operational amplifier. The feedback capacitor acts as a negative feedback to perform primary filtering and improve the output signal quality. The adjustable bias voltage can be used to improve the adaptability of the circuit and facilitate data acquisition and processing. (2) a dual-channel bandpass filter, which divides the signal passing through the pre-transimpedance amplifier into two paths and passes through two bandpass filters with different center frequencies. Signals with frequencies greater than or less than the center frequency will be attenuated or even filtered out by the bandpass filter, which can preliminarily eliminate the interference of other frequency signals other than the second harmonic signal frequency. (3) a dual-channel phase-locked amplifier, which is constructed by two independently operating single-channel phase-locked amplifiers to extract the second harmonic signals corresponding to the concentrations of two characteristic gases. The sinusoidal excitation signal and the reference square wave signal of the same frequency (2F) are input into the multiplier for processing, and then the high-frequency components are filtered out by the low-pass filter. Finally, the signal sampling is completed by the analog-to-digital converter (ADC7606). By performing differential operation on the two second harmonic signals, accurate quantitative analysis of the change in characteristic gas concentration can be achieved.

[0040] There are two temperature control circuits: the first temperature control circuit (referred to as the first temperature control circuit) and the second temperature control circuit (referred to as the second temperature control circuit). The temperature control circuit consists of the ADN8834 and its peripheral circuits. The AND8834 compares the target temperature with the actual temperature and adjusts the TEC current using a PID control algorithm to achieve heating or cooling, thus achieving temperature control.

[0041] There are two constant current sources and overcurrent protection circuits, namely the first constant current source and overcurrent protection circuit and the second constant current source and overcurrent protection circuit. The first constant current source and overcurrent protection circuit and the second constant current source and overcurrent protection circuit both include: (1) a high-precision operational amplifier. Due to the characteristics of virtual short and virtual open, U+=U-, the input impedance is infinite. When the input voltage increases, the output will remain constant due to the existence of the feedback loop. (2) an N-channel field effect transistor. When the voltage between the gate and the source is greater than the threshold voltage, the MOS transistor is turned on. Otherwise, it is cut off. The gate voltage is controlled by the operational amplifier, which can make the constant current source work in the constant current region. (3) a voltage comparator. It generates an output signal by comparing the input voltage and the threshold voltage. If the input voltage is greater than the threshold voltage, the output is a high level, otherwise it is a low level. Its rail-to-rail output can directly drive a transistor. (4) an NPN transistor. When the voltage difference between the base and the emitter is greater than the threshold voltage, the transistor is turned on. Otherwise, it is cut off. The collector voltage controls the gate voltage of the subsequent MOS transistor.

[0042] The spectral data acquisition circuit consists of an ADC7606 and its peripheral circuits. The eight independent input channels collect the 2F signals (2F_SIN+, 2F_SIN-, 2F_COS+, and 2F_COS-) output by the dual-channel lock-in amplifier, the optical power signal (DC), the real-time temperature return values ​​of the two DFB lasers (T_LD1 and T_LD2), and the gas absorption cell pressure (P), and transmit the converted digital signals to the main control unit.

[0043] The optical system includes (1) DFB laser: used to provide a stable laser light source with a wavelength matching the absorption line of the target gas. There are two DFB lasers, namely 1580nm DFB laser and 1653.7nm DFB laser. (2) High reflective mirror: used to make the laser reflect back and forth hundreds to thousands of times in the off-axis integrating cavity, significantly extending the effective optical path. (3) Collimator: used to convert the divergent laser beam into a parallel beam, ensuring that the beam enters the off-axis integrating cavity in a parallel manner. There are two collimators, namely the first collimator and the second collimator. (4) Converging lens 6: used to focus the transmitted light onto the detector. (5) Detector: used to convert the optical signal into an electrical signal. (6) Double wedge group: by changing the angle and position of the double wedge, the direction of the output beam is controlled. There are two double wedge groups, namely the first double wedge group 1 and the second double wedge group 2. (7) Stepper motor: used to adjust the wedge angle. There are two stepper motors, namely the first stepper motor 3 and the second stepper motor 4. (8) Off-axis integrating cavity 5: The laser is reflected multiple times between the cavity mirrors.

[0044] The gas sampling device includes: (1) a filter device 7 for removing particulate impurities and highly active interfering components; and (2) an air pump 8 for injecting the filtered mixed gas into the off-axis integrating cavity at a controllable flow rate.

[0045] The early warning model includes: (1) sensor module: temperature sensor for measuring battery surface temperature; internal resistance tester for measuring battery internal resistance; current sensor for measuring battery current; voltage sensor for measuring battery voltage; pressure sensor for measuring battery internal pressure; deformation sensor for measuring battery dimensional deformation; gas sensor system for measuring the concentration of characteristic gases CO, CO2, CH4, and C2H6. (2) data acquisition module: high-precision ADC7609 and ADC7606 are used to achieve synchronous acquisition of multi-channel signals and transmit the acquisition results to the main control unit in real time; the processed signals are uploaded to the core processing unit via SPI communication, and gas concentration, voltage, current, temperature, internal resistance, deformation and pressure data are obtained in the graphical user interface. The concentration data sets of the four characteristic gases CO, CO2, CH4, and C2H6 under four types of early warning states (no warning, first level warning, second level warning, and third level warning) are obtained. Each state contains 2,000 sample records of each gas. (3) data preprocessing module: standardization and outlier processing algorithms are used to clean and optimize the features of the raw data. The dimension difference is eliminated by Z-score standardization, and outliers are eliminated based on the 3σ principle. Then, Min-Max normalization is used to map the eigenvalues ​​to interval, ensuring that the data distribution meets the convergence requirements of the machine learning model and improving the stability and generalization ability of model training. (4) Random Forest Machine Learning Model: Based on the Bagging ensemble learning framework, multiple decision trees are constructed. The prediction robustness is enhanced through the following double random mechanism. Finally, the prediction results of each decision tree are integrated through a voting mechanism. The warning level with the most votes is used as the final output, effectively reducing the risk of overfitting and improving classification accuracy. (5) Alarm Device: It includes a buzzer and an indicator light. Based on the prediction results of the random forest machine learning model, a three-level warning mechanism is implemented. The main control unit controls the indicator light and buzzer to perform corresponding operations.

[0046] The early warning system provided in this embodiment includes ADC and DAC chips that require high-precision reference voltages, and their drivers require a stable and reliable power supply. To meet the operational requirements of each module in the early warning system, the system's power supply utilizes a three-stage step-down design consisting of an AC-DC switching power supply, a DC-DC switching power supply, and a low-dropout linear regulator (LDO), providing both AC-DC conversion and voltage reduction.

[0047] The basic operating principle of the early warning system provided in this embodiment is as follows: the main control unit controls the DDS chip to generate sine and square wave signals, while simultaneously controlling the DAC chip to generate a sawtooth signal. A digital potentiometer adjusts the amplitude of the sine wave to meet the optimal modulation depth optimization conditions. The two sine waves are superimposed on the low-frequency sawtooth signal and synthesized by an adder to output the voltage drive signals for the two DFB lasers. Based on the principle of deep negative feedback, the constant current source circuit converts the drive voltage signal into a highly stable drive current, and the overcurrent protection circuit monitors the current status in real time. When the current exceeds the threshold, the overcurrent protection circuit quickly disconnects the circuit to prevent damage to the lasers due to overcurrent. The temperature control circuit uses a PID control algorithm and a TEC to precisely regulate the temperature of the dual DFB lasers, ensuring stable and adjustable operating temperature. The laser beam is injected into the off-axis integrating cavity 5 through a coupling system. The high-reflectivity mirror within the cavity causes the light to reflect multiple times, with a portion of the light passing through the mirror during each reflection. During the light absorption process, the detector collects changes in light intensity passing through the lens, and the spectral signal dual-frequency demodulation circuit extracts the second harmonic signals of the four gases CO, CO2, CH4, and C2H6. The gas concentration is obtained through the concentration inversion algorithm, and the random forest machine learning model determines whether to trigger a graded warning.

[0048] Example 2: This example provides an early warning method for the lithium battery thermal runaway gas early warning system described in Example 1.

[0049] (1) Main control unit (STM32F407VET6): Functions include controlling the DAC chip to generate a low-frequency sawtooth wave signal and the DDS chip to generate a high-frequency sine wave signal, thus achieving multi-frequency modulation drive of the laser. It reads the second harmonic signal, laser temperature, gas pool pressure, and other data collected by the ADC chip and transmits them to the core processing unit (AM4378 Cortex A9).

[0050] (2) The core processing unit is an important component of the early warning system described in Example 1 and is primarily responsible for algorithm processing and the operation of the graphical user interface (GUI) system. The core processing unit undertakes the complex signal processing and data analysis tasks in the system, including decoupling algorithms, concentration inversion algorithms, and calibration and calibration. The core processing unit runs the Linux operating system and supports a graphical interactive interface. Its functions include real-time display of data on gas concentration, temperature, pressure, and other parameters. It also provides a dynamic adjustment interface for laser drive current, temperature, demodulation parameters, etc.

[0051] (3) Power supply: The power supply provides 5V, 12V, and 18V power for the early warning system. At the front end of the power supply, an AC-DC power supply module is used instead of an industrial frequency transformer to convert the 220V AC voltage into a 24V DC voltage. The back end of the power supply is divided into two parts. One part is a linear regulator (LDO) module with a soft start function, which steps down the 24V voltage to an 18V DC voltage with lower ripple. The other part is a switching power supply (DC-DC) module, which steps down the 24V voltage to 5V and 12V with higher efficiency and greater power, and supplies power to other parts of the entire system. At the same time, the entire power supply is installed in an aluminum shielding box to reduce interference from high-power power supplies.

[0052] (4) Dual-frequency drive signal generation circuit: The main control unit controls the two DAC chips to generate two sawtooth wave signals with lower and equal frequencies. The main control unit controls the two DDS chips to generate sine wave modulation signals with frequencies much higher than the sawtooth wave frequency, with frequencies F1 and F2 respectively. However, the frequencies of the two sine waves generated by the DDS chips are different, and the frequency difference between the two sine signals cannot be too small. After the two sine signals pass through the digital potentiometer, the amplitude of the sine wave is changed according to the size of the DFB laser drive current and meets the optimal modulation depth optimization conditions. The two sine waves with different frequencies and changed amplitudes are superimposed on the low-frequency sawtooth wave signal as carriers, and the final output is the voltage drive signal for the two DFB lasers.

[0053] (5) Spectral signal dual-frequency demodulation circuit: The current signal output by the detector is converted into a voltage signal by a pre-transimpedance amplifier and amplified. The voltage signal then passes through two bandpass filters and two lock-in amplifiers. The two bandpass filters are used to filter out signals with frequencies different from the frequency signal to be extracted; the two lock-in amplifiers are used to extract the second harmonic signal of a specific frequency. Ultimately, two second harmonic signals representing the gas concentration are successfully demodulated, thereby characterizing the concentration of the multi-component gas to be measured.

[0054] (6) Temperature control circuit: It consists of the ADN8834 temperature control chip and its peripheral circuits. The current temperature data of the laser is obtained by connecting to the thermistor inside the laser. The output channel of the DAC chip is connected to the TEMPSET pin of the temperature control chip. The main control unit converts the target temperature into the corresponding DAC chip output value through the preset temperature-voltage correspondence and writes it into the DAC chip register. The DAC chip outputs the corresponding voltage. The temperature control chip reads the voltage of the DAC chip through the TEMPSET pin and uses it as the target temperature. The temperature control chip compares the target temperature with the actual temperature and adjusts the current of the TEC through the PID control algorithm to achieve heating or cooling, thereby achieving temperature control.

[0055] (7) Constant current source and overcurrent protection circuit: ① In the constant current source circuit, when the driving voltage is input from the positive input terminal of the operational amplifier, the output of the operational amplifier controls the gate of the N-channel MOS tube. The voltage difference between the gate and source of the MOS tube is greater than the threshold voltage, and the MOS tube is turned on and works in the constant current region, forming a negative feedback loop. The driving voltage determines the driving current. When the driving current increases, the voltage at the inverting input terminal of the operational amplifier will also increase, and the difference between the non-inverting input terminal and the inverting input terminal will also decrease. The output of the operational amplifier will decrease, the gate voltage of the MOS tube will also decrease, the current flowing through the source of the MOS tube will also decrease, and the voltage at the inverting input terminal of the operational amplifier will also decrease and return to its original value, and the driving current will remain constant. ② The overcurrent protection circuit is connected to the constant current source circuit. The voltage across the sampling resistor is the input signal of the overcurrent protection circuit. After passing through the voltage follower, it enters the voltage comparator. The threshold voltage is provided by the DAC chip. When the comparator input voltage is greater than the threshold voltage, the output is high, the NPN transistor is turned on, the switch MOS tube is turned off, and the circuit is closed.

[0056] (8) Graphical User Interface: The Graphical User Interface (GUI) is the core interactive platform of the dual-component gas synchronous detection system. The system is based on the Linux operating system and uses the QT Creator development platform. The initial interface is divided into four main areas: the parameter display area displays the gas absorption cell pressure, laser temperature, drive status, and gas concentration in real time; the gas selection area provides CH4 / C2H6 and CO / CO2 selection buttons corresponding to different laser parameter settings; the function button area includes calibration, upload, curve, and data buttons, supporting system self-test, data upload, historical curve viewing, and raw data export; clicking the control button enters the laser parameter setting interface, which facilitates the user to configure detailed parameters. The first-level menu offers 11 core functional options, including system information displaying hardware version, firmware version, and network status; laser parameters entering the second-level menu, allowing adjustment of laser drive current, temperature, modulation amplitude, etc.; phase shift function setting laser phase compensation parameters; spectral processing algorithm selection signal processing algorithm; wavelength drift recovery compensation wavelength shift caused by ambient temperature; limit value setting for setting gas concentration alarm thresholds; information format configuration data output protocol; information correction manual correction of abnormal data points; concentration calibration initiation of multi-point calibration process; PLS modeling setting partial least squares modeling parameters; reference spectrum management gas absorption characteristic line database. The main parameters that can be set in the second-level menu include the start and end values ​​of the DFB laser drive current, DFB laser tuning rate, modulation amplitude, laser control temperature, spectrum averaging times, and scan null points.

[0057] (9) Optical system: Based on the absorption spectrum characteristics of CO, CO2, CH4, and C2H6 in the HITRAN database, this embodiment selects the near-infrared band (800-2500 nm) as the target detection band. In view of the strong absorption characteristics of CO2 and CO, a 1580 nm DFB laser with a central wavelength of 1580 nm is selected as the dual-component detection light source. The frequency of the sinusoidal modulation signal in its driving signal is F1, and its emission spectrum lines correspond to the 1579.57 nm absorption peak of CO2 and the 1579.74 nm absorption peak of CO, respectively. For the detection of CH4 and C2H6, a 1653.7 nm DFB laser with a central wavelength of 1653.7 nm is used. The frequency of the sinusoidal modulation signal in its driving signal is F2, covering the 1653.7 nm absorption peak of CH4 and the 1653.32 nm absorption peak of C2H6. To achieve efficient coupling and spatially separated transmission of dual-wavelength lasers, this embodiment designed and constructed an off-axis coupling system based on dual optical wedges. This eliminates the need for a beam combiner, reduces optical energy loss, and achieves a lower detection limit. The 1580nm DFB laser corresponds to the first dual optical wedge group 1, and the 1653.7nm DFB laser corresponds to the second dual optical wedge group 2.

[0058] Connect the output ends of the dual lasers to the first and second collimators, respectively. Use a high-sensitivity infrared photosensitive card to observe the direction of the laser beams. Use two stepper motors (first stepper motor 3 and second stepper motor 4) to adjust the angles of the corresponding dual wedges so that both beams pass through the center of the aperture. Switch to infrared lasers of their respective wavelengths and precisely calibrate the beam positions using an infrared CCD calibration plate, ensuring that the deviation between the two beams passing through the center of the aperture is less than 0.1 mm. After adjustment, lock the wedges. Place the cavity in the center of the aperture, first inserting the internal calibration disk. Adjust the position and height of the cavity front end so that visible light passes through the center, and observe the output light intensity. When the output light intensity does not change significantly before and after removing the calibration disk, the front end is in the optimal position. Adjust the back end of the cavity in the same way. At this point, the cavity can be considered approximately parallel to the beam propagation direction, and the beam passes through the cavity center.

[0059] During the optical path alignment process, coaxial adjustment is particularly critical. Initially, a reflector is installed. Using a visible red light source positioned at a distance, the beam is carefully aligned perpendicular to the reflector's surface. By adjusting two knobs on the mirror mount, the reflector's angle can be finely adjusted until the reflected light completely overlaps the incident light, ensuring the lens is perpendicular to the incident light. Based on this step, the second lens is installed, positioned closer to the laser. Once the lenses are installed, due to their incomplete reflectivity in the visible light band, some light will pass through the second lens and be reflected back by the first lens, forming two reflected light spots: one from the second lens (higher intensity) and the other from the first lens (lower intensity). By fine-tuning the knobs on the lens mount, these two reflected light spots are adjusted until they perfectly overlap with the original incident light. At this point, the two lenses are approximately parallel and perpendicular to the incident light. Finally, a plano-convex lens is installed at the end of the cavity and the lens position is adjusted to focus the beam onto the detector's photosensitive surface.

[0060] Turn on the 1580nm DFB laser and the 1653.7nm DFB laser. The two wavelength lasers propagate separately at the same off-axis angle. The wedge refraction ensures that the parallelism of the two beams is greater than 99.9%. The two laser beams form parallel and spatially separated optical paths in the cavity. An infrared calibration plate is placed 10 cm in front of the cavity. The wedge angle is adjusted to stabilize the distance between the two spot centers at 2.6 mm (the error is less than 0.05 mm), and observe the signal form output by the detector. Adjust the laser position and angle until a clear periodic sawtooth waveform is presented.

[0061] The formula for the effective optical path of OA-ICOS is:

[0062] ;

[0063] Where, It represents the equivalent value after the laser undergoes multiple reflections in the optical resonant cavity and the actual optical path is significantly extended; represents the geometric distance between the two reflectors of the resonant cavity, Indicates the reflectivity of the mirror.

[0064] When the reflectivity of the lens of the off-axis integrating cavity 5 is greater than 99.95%, the average finite optical path of CO, CO2, CH4 and C2H6 at the currently selected gas absorption wavelengths can reach more than 700m, and the theoretical detection limits of the four gas concentrations of CO, CO2, CH4 and C2H6 are in the sub-ppb level to the ppm level. The theoretical detection limits of the four gas concentrations required by the early warning system of this embodiment are several to several hundred ppm. The off-axis integrating cavity 5 used in the early warning system of this embodiment can fully meet the detection limits of CO, CO2, CH4 and C2H6 for thermal runaway of lithium batteries.

[0065] (10) Gas sampling device: When a lithium battery experiences thermal runaway, it will release a complex mixed gas. In order to ensure safety and improve the efficiency of gas treatment, the applicant designed a customized filter device 7 for pre-treatment. The filter device 7 integrates multi-stage physical and chemical adsorption units, which can not only accurately remove particulate impurities with a diameter of less than or equal to 0.3 μm, but also effectively remove highly active interfering components, thereby ensuring the stability and reliability of subsequent processes. The pre-treated gas is continuously pumped by the air pump 8 and injected into the off-axis integrating cavity at a precisely controllable flow rate (adjustable between 0.5 and 2 L / min), thereby forming a stable gas environment suitable for detection.

[0066] (11) Early warning model: Based on the group standard of thermal runaway early warning method for lithium-ion batteries, key early warning monitoring parameters such as battery temperature, internal resistance, voltage, current, internal pressure and CO concentration are clearly defined. The specific early warning level requirements are as follows.

[0067] Temperature warning: When the surface temperature of the single battery 55℃ triggers the first level warning; when the surface temperature of the single battery 60℃ triggers the second level warning; when the surface temperature of the single battery A level 3 warning is triggered at 70°C.

[0068] Voltage and current warning: When the voltage or current fluctuation exceeds the normal range 5% or the voltage difference between single cells exceeds 0.1V, triggering a level 1 warning; when the voltage or current fluctuation exceeds the normal range 10% or the voltage difference between single cells exceeds 0.15V, triggering the second level warning; when the voltage or current fluctuation exceeds the normal range When the voltage difference between cells exceeds 20% or 0.2V, a level 3 warning is triggered.

[0069] Internal pressure warning: When the internal pressure exceeds 80% of the battery's maximum design pressure, a level one warning is triggered; when the internal pressure exceeds 90% of the battery's maximum design pressure, a level two warning is triggered; when the internal pressure exceeds 100% of the battery's maximum design pressure, a level three warning is triggered.

[0070] Deformation warning: When the battery size exceeds 5% of its design size, a level one warning is triggered; when the battery size exceeds 8% of its design size, a level two warning is triggered; when the battery size exceeds 10% of its design size, a level three warning is triggered.

[0071] Characteristic gas warning: When the carbon monoxide concentration 50ppm triggers a first-level warning; when the carbon monoxide concentration 80ppm triggers the second level warning; when the carbon monoxide concentration 100ppm triggers the third level warning.

[0072] In a laboratory environment, overcharging and needle puncture are used to simulate and trigger different levels of lithium battery thermal runaway. The sensor module measures voltage, current, internal resistance, temperature, internal pressure, deformation, characteristic gas concentration, and other data and transmits them to the data acquisition module. When multiple parameters trigger different warning levels simultaneously, the highest priority is followed, thereby constructing a large amount of lithium battery gas concentration data (samples) under no alarm, first warning, second warning, and third warning levels. , the sample mean is The Z-score method is used to filter out abnormal values ​​from the collected data. According to statistical principles, When the absolute value of is greater than 3, the corresponding data point is judged as an outlier. The specific calculation formula is:

[0073] ;

[0074] Where, represents the sample standard deviation.

[0075] Based on the normal distribution characteristics, this processing method can ensure that 99.7% of the data are at the mean. Within the range of 3 standard deviations, outliers are effectively eliminated to ensure data quality. The mean imputation method is used to fill missing values. The formula is:

[0076] ;

[0077] Where, Indicates the first Sample values, is the number of samples. This method uses the sample mean as a reasonable estimate of the missing value to maintain the basic statistical characteristics of the data. All feature variables are linearly scaled to interval, to achieve data standardization and meet the requirements of subsequent machine learning model training. The formula is as follows:

[0078] ;

[0079] Where, represents the normalized sample value, and Represents samples The minimum and maximum values ​​of .

[0080] Aiming at the application scenario of real-time monitoring of lithium battery gas in new energy vehicles with high real-time requirements, this embodiment specifically optimizes the response speed and computational efficiency of the random forest machine learning model. Through systematic parameter tuning and structural simplification, the model not only maintains excellent classification accuracy, but also significantly improves the real-time response performance. Compared with similar models, it achieves faster sample classification decisions while ensuring the accuracy of early warning. The model input features are set to four gas concentrations (CO, CO2, CH4, C2H6). According to the group standard threshold, the original data set is divided into four categories: no alarm, first-level warning, second-level warning, and third-level warning. The model is built on the basis of the Bagging idea, such as Figure 1 As shown in the figure, decision trees 1 through 1 are constructed. Each tree's subsets (subsets 1 through 2) are randomly generated from the original training samples (samples 1 through 3) using random sampling with replacement, effectively ensuring sample diversity. During training, decision trees 1 through 2 are trained on independent subsets (subsets 1 through 2) to reduce the risk of overfitting.

[0081] During the splitting process of the decision tree, a dual randomization mechanism is implemented: first, feature selection is randomized, forcing each split to be based on only one of the four features, prompting decision trees 1 through n to form differentiated discriminative paths; second, the splitting threshold is randomized, searching through all possible values ​​of the selected feature to select the optimal split point that minimizes the Gini index. As a tree-like model based on conditional judgments, the decision tree partitions the data through a series of "if-then" rules. At each step, the optimal feature and threshold are selected to gradually divide the subsets until all samples in each subset fall into the same warning level, achieving subset normalization. Decision trees 1 through n together form a random forest.

[0082] The number of decision trees is determined by using the stratified five-fold cross-validation method. This method can ensure that the distribution of samples at each warning level in each compromise is consistent with the original data. When setting the hyperparameters, the number of trees is set to , the maximum depth of the tree is set To use only one feature for each split, the minimum number of samples required for node splitting is fixed to 2. All tree values ​​are traversed using a grid search strategy, and the performance of each parameter group is evaluated using F1-score as the evaluation metric. F1-score is the core metric for evaluating the performance of the classification model. The accuracy of the model is integrated through the harmonic mean ( ) and recall ( ), which can more comprehensively evaluate the performance of the model, is calculated as follows:

[0083] ;

[0084] Where, , , represents the number of true positive samples, represents the number of false positive samples, represents the number of false negative samples, The F1-score is the evaluation metric. During training, the inference time of a single sample is dynamically monitored to ensure that it meets real-time requirements, thereby ultimately determining the optimal number of decision trees.

[0085] The random forest machine learning model runs in the core processing unit. When presented with a new prediction sample, each tree independently determines the warning level for the sample based on the input concentrations of CO, CO2, CH4, and C2H6. The random forest machine learning model uses the "majority rule" principle to determine the final prediction. The system tallies the votes of all decision trees and uses the warning level with the most votes as the model's final output.

[0086] The alarm device, which includes a buzzer and indicator light, is connected to the main control unit and implements graded warnings based on the prediction results of the random forest machine learning model: ① Level 1 Warning: When the model prediction results meet the level 1 warning trigger conditions, the buzzer outputs an intermittent pulse alarm and the indicator light remains off. ② Level 2 Warning: When the model prediction results meet the level 2 warning trigger conditions, the buzzer outputs a continuous beep and the indicator light flashes. ③ Level 3 Warning: When the model prediction results meet the level 3 warning trigger conditions, the buzzer outputs a rapid, high-frequency alarm and the indicator light remains on.

[0087] The present invention uses a multi-gas sensing system and a random forest machine learning model to achieve an early and accurate warning of lithium battery thermal runaway. Through feature screening and model optimization, the detection speed and accuracy are greatly improved, and a real-time warning with millisecond response is achieved. (1) A gas warning model is established through a random forest algorithm, and a three-level alarm mechanism (level one, level two, and level three warning) is designed. It can quickly judge the level of thermal runaway according to the change in gas concentration and issue an alarm signal of the corresponding level in a timely manner, thereby improving the reliability and practicality of the warning system. (2) Based on the off-axis integrating cavity 5 optical path coupling technology of the double wedge group, the beam angle is precisely adjusted by the stepper motor double wedge, without the need for a beam combiner, so that two laser beams are coupled into the off-axis integrating cavity 5 in parallel with a very small distance, thereby improving the light energy utilization rate and detection sensitivity. (3) By analyzing key parameters such as CO, CO2, CH4, C2H6 concentrations and warning levels, redundant features are eliminated, the amount of calculation is reduced by more than 60%, and the complexity of model reasoning is reduced. By using the stratified five-fold cross-validation method, 500 decision trees were reduced to 300, and the inference speed was increased to 10-15ms while maintaining 98% accuracy.

[0088] Of course, the above description is not a limitation of the present invention, and the present invention is not limited to the above examples. Changes, modifications, additions or substitutions made by technicians in this technical field within the essential scope of the present invention should also fall within the scope of protection of the present invention.

Claims

1. Lithium battery thermal runaway gas warning system, characterized by: It includes a main control unit, a core processing unit, a graphical user interface, a dual-frequency drive signal generation circuit, a spectral signal dual-frequency demodulation circuit, a temperature control circuit, a constant current source and overcurrent protection circuit, a spectral data acquisition circuit, an optical system, a gas sampling device, and an early warning model; The main control unit is used to control the DAC chip to generate a low-frequency sawtooth wave signal, control the DDS chip to generate a high-frequency sine wave signal, read the second harmonic signal, laser temperature and gas pool pressure collected by the ADC chip, and transmit them to the core processing unit; The core processing unit is used to run the Linux system, run the graphical user interface, and perform gas concentration inversion, calibration, and early warning model operations; The graphical user interface is used to realize real-time display of multi-spectral information, laser parameters and concentration; The dual-frequency drive signal generating circuit includes two DAC chips, two DDS chips, a digital potentiometer, and an adder; the two DAC chips each generate two sawtooth wave signals with equal frequencies, and one DAC chip generates a PWM wave; the two DDS chips each generate two sine wave signals with frequencies higher than the sawtooth wave signal but different from each other; the digital potentiometer changes the amplitude of the sine wave according to the magnitude of the DFB laser drive current and ensures that the amplitude of the sine wave meets the optimal modulation depth optimization condition; The adder superimposes two sine waves with different frequencies and changed amplitudes as carriers onto the sawtooth wave signal respectively; The spectral signal dual-frequency demodulation circuit includes a pre-transimpedance amplifier, a dual-channel bandpass filter, and a dual-channel lock-in amplifier; the pre-transimpedance amplifier is used to convert the current signal output by the detector into a voltage signal and amplify it; the dual-channel bandpass filter is composed of two bandpass filters with different center frequencies, which is used to eliminate interference from frequency signals other than the second harmonic signal frequency; the dual-channel lock-in amplifier is constructed by the collaboration of two independently operating single-channel lock-in amplifiers to extract the second harmonic signals corresponding to the concentrations of two characteristic gases respectively; There are two temperature control circuits, namely a first temperature control circuit and a second temperature control circuit; There are two constant current sources and overcurrent protection circuits, namely a first constant current source and overcurrent protection circuit and a second constant current source and overcurrent protection circuit, each of which includes a high-precision operational amplifier, an N-channel field-effect transistor, a voltage comparator, and an NPN transistor; The optical system includes a DFB laser, a high-reflection mirror, a collimator, a converging lens, a detector, a double-wedge group, a stepper motor, and an off-axis integrating cavity; the DFB lasers include two, namely, a 1580nm DFB laser and a 1653.7nm DFB laser; the collimators include two, namely, a first collimator and a second collimator; the double-wedge groups include two, namely, a first double-wedge group and a second double-wedge group; and the stepper motors include two, namely, a first stepper motor and a second stepper motor; The early warning model includes a sensor module, a data acquisition module, a data preprocessing module, a random forest machine learning model and an alarm device; the sensor module includes a temperature sensor for measuring the surface temperature of the lithium battery, an internal resistance tester for measuring the internal resistance of the lithium battery, a current sensor for measuring the current of the lithium battery, a voltage sensor for measuring the voltage of the lithium battery, a pressure sensor for measuring the internal pressure of the lithium battery, a deformation sensor for measuring the dimensional deformation of the lithium battery, and a gas sensing system for measuring the concentration of characteristic gases such as CO, CO2, CH4, and C2H6; The random forest machine learning model constructs multiple decision trees based on the Bagging ensemble learning framework, enhances prediction robustness through a double random mechanism, integrates the prediction results of each decision tree through a voting mechanism, and uses the warning level with the most votes as the final output; the alarm device includes a buzzer and an indicator light, and implements a three-level warning mechanism based on the prediction results of the random forest machine learning model.

2. The lithium battery thermal runaway gas warning system according to claim 1, characterized in that: The DFB laser is used to provide a stable laser light source, and the wavelength matches the absorption line of the target gas; The high-reflection mirror is used to make the laser reflect back and forth hundreds to thousands of times in the off-axis integrating cavity, thereby extending the effective optical path; The collimator is used to convert the divergent laser beam into a parallel beam, ensuring that the beam enters the off-axis integrating cavity in a parallel manner; The converging lens is used to focus the transmitted light onto the detector; The detector is used to convert the optical signal into an electrical signal; The double wedge group controls the direction of the outgoing light beam by changing the angle and position of the double wedges; The stepping motor is used to adjust the angle of the optical wedge.

3. The lithium battery thermal runaway gas warning system according to claim 2, characterized in that: The data acquisition module uses ADC7609 and ADC7606 to achieve synchronous acquisition of multi-channel signals and transmits the acquisition results to the main control unit in real time; The data preprocessing module uses standardization and outlier processing algorithms to clean and optimize the features of the original data.

4. The lithium battery thermal runaway gas warning system according to claim 3, characterized in that: The three-level warning mechanism includes level one warning, level two warning and level three warning. The warning level requirements are as follows: Temperature warning: When the surface temperature of the single battery 55℃ triggers the first level warning. 60℃ triggers the second level warning, when the surface temperature of the single battery At 70℃, a level 3 warning is triggered; Voltage and current warning: When the voltage or current fluctuation exceeds the normal range 5% or the voltage difference between single cells exceeds 0.1V, triggering a level 1 warning. 10% or the voltage difference between single cells exceeds 0.15V, triggering the second level warning. When the voltage difference between cells exceeds 20% or 0.2V, a level 3 warning is triggered; Internal pressure warning: When the internal pressure exceeds 80% of the battery's maximum design pressure, a level one warning is triggered; when the internal pressure exceeds 90% of the battery's maximum design pressure, a level two warning is triggered; when the internal pressure exceeds 100% of the battery's maximum design pressure, a level three warning is triggered; Deformation warning: When the battery size exceeds 5% of its design size, a level 1 warning is triggered; when the battery size exceeds 8% of its design size, a level 2 warning is triggered; when the battery size exceeds 10% of its design size, a level 3 warning is triggered; Characteristic gas warning: When the carbon monoxide concentration 50ppm triggers a first-level warning. 80ppm triggers the second level warning, when the carbon monoxide concentration 100ppm triggers the third level warning.

Images