Lithium-ion batteries are widely used in electronic equipment, electric vehicles, and other applications owing to their advantages of high specific energy, high power, long cycle life, and little or no pollution. Using lithium-ion batteries is a good strategy to achieve the "double carbon" goal. However, abusing lithium-ion batteries may trigger a thermal runaway event, and the burning of toxic and highly flammable gases emitted from thermal runaway can cause severe fires or explosions. Analyzing the gas production behavior of lithium-ion batteries during thermal runaway is of great significance. However, the traditional technology used for detecting pyrolysis gas from lithium-ion batteries struggles to meet the needs of online detection because of many shortcomings, such as a long detection cycle, strong cross-interference, easy saturation, and the inability to detect homonuclear diatomic molecules. The basic understanding of the risk of gas release during the thermal abuse of batteries is still limited. Therefore, an effective technology that can perform a high-resolution online in-situ analysis of battery pyrolysis gas is urgently needed.

In this work, a laser Raman spectroscopy technique is used to rapidly detect the changes in the composition and concentration of the main gases released from the thermal runaway of abused lithium-ion batteries . Hence, an online analysis of the risk of pyrolysis gases of lithium-ion batteries is possible. The pretreatment methods by discrete wavelet transform (DWT) and adaptive iterative re-weighted penalized least squares (airPLS) are used to denoise and deduct the background of the spectral data, respectively. Partial least squares (PLS) is used to establish a quantitative model for the target gas, which can accurately and stably analyze the information on pyrolysis gas from lithium-ion batteries.

First, the Raman spectra of a battery with state of charge (SOC) of 100% during thermal runaway is compared with those of standard reference gases (Fig. 3), and the main gases released by battery pyrolysis are determined to be CO₂, CO, H₂, CH₄, C₂H₄, and C₃H₆. CH₄, C₂H₄, and C₃H₆ have the problems of intensive and overlapping characteristic peaks, especially in the 2502-3444 cm^-1 range. Second, to eliminate the spectral noise and baseline interference on the effective spectral information, this study uses DWT and the airPLS algorithm, respectively. The grid search strategy is used to optimize the DWT parameters, which is determined by the root mean square error of prediction (RMSEP) of the PLS model (Fig. 4). The different DWT parameters have great influence on the model, and the optimal DWT parameters from different gases varies. In addition, to quantify the composition of the unknown pyrolysis gas from the battery, the variables most relevant to the measured gas concentration are screened out in the spectral library by the classical PLS method to establish a multiple regression model (Table 4). In the prediction results on the validation set, the minimum value of predictive correlation coefficient (R²) is 0.937, and the maximum value of RMSEP is only 0.452%, indicating that the PLS model accurately extracts the concentration of the pyrolysis gas from the battery. Finally, the laser Raman spectroscopy technology is used in the thermal abuse experiment of a lithium-ion battery, and the surface temperature, voltage, internal pressure of the abused device, and gas data are recorded in real time (Fig. 6). The thermal abuse process of the battery is divided into three stages. The gas composition and concentration change significantly across the three stages. The gas composition and concentration remain unchanged after the thermal runaway ends, and the maximum error of gas concentration change is no more than 0.29% (Table 5). The gas concentration and its relationship with the SOC are consistent with the results obtained by different gas detection technologies in previous literature. These results show that laser gas Raman spectroscopy technology can effectively and stably analyze the pyrolysis gas from a battery.

The main components of the pyrolysis gas from a lithium-ion battery are determined as CO₂, CO, H₂, CH₄, C₂H₄, and C₃H₆ using Raman spectroscopy analysis. The above six gases and N₂ and O₂ from the air can accurately reflect the temperature and stage of the thermal runaway of a lithium-ion battery, and effectively evaluate the risk of thermal runaway of lithium-ion batteries. The thermal runaway gas components of a lithium-ion battery are analyzed online using a Raman spectroscopy system, and the spectral data of the gas samples are quantitatively analyzed using a spectral preprocessing algorithm and PLS model. The results show that the Raman spectroscopy can analyze the dynamic changes of the main pyrolysis gases from a lithium-ion battery and air components in seconds. The maximum root mean square error of the spectral model is no more than 0.45%, and the minimum root mean square error is only 0.04%, which effectively meets the requirements of the online analysis of typical gas products in the thermal runaway process of a lithium-ion battery. Therefore, the Raman spectroscopy is expected to become an effective technique for the online in-situ study of thermal runaway release gases from lithium-ion batteries and accurately assess the explosion risk of lithium-ion batteries.