Application of laser-induced breakdown spectroscopy and chemometrics for rapid identification of fire-retardant/resistant coatings from fire residues

• Published in: Construction & building materials 2022-03, Vol.325, p.126773, Article 126773
• Main Authors: Song, Weiran, Zhao, Shangyong, Zhang, Yiming, Ruan, Cheng, Huang, Ao, Hu, Xiao, Zhao, Min, Zhou, Wen, Wang, Ji, Wang, Xuebao, Wang, Hui, Hou, Zongyu, Wang, Zhe
• Format: Article
• Language: English
• Subjects: Chemometrics; Fire investigation; Fire residues; Fire-retardant/resistant coating; Laser-induced breakdown spectroscopy; Partial least squares
• ISSN: 0950-0618; 1879-0526
• DOI: 10.1016/j.conbuildmat.2022.126773

•LIBS combined with chemometrics for rapid analysis of fire-retardant/resistant coatings (FRC).•High efficiency of sample preparation, LIBS measurement and data analysis processes.•FRC were correctly identified from fire residues based on different types and heating devices.•Heating temperatures were predicted using partial least squares regression model. The use of fire-retardant/resistant coatings (FRC) is an effective and important method to delay the spread of fire or to protect the intact structure against fire. Due to the high demand for FRC in end-user industries such as construction and transportation, how to efficiently determine the identity of FRC has become a critical but difficult task in fire protection and investigation. This paper presents the application of laser-induced breakdown spectroscopy (LIBS) and chemometrics for rapid post-fire analysis of FRC. We attempt to identify four types of FRC from fire residues, identify FRC based on heating devices, and predict the heating temperatures of FRC. FRC residues were prepared as pressed pellets and then scanned by an integrated LIBS system in the ultraviolet–visible, visible and near-infrared ranges. The obtained spectra were pre-processed to select an appropriate number of variables, and the relationship between the spectral intensity and the categorical information was modelled using partial least squares (PLS). The class separability was demonstrated by visualisation in a low-dimensional space through principal component analysis (PCA) projection. Furthermore, important spectral lines that contribute to the classification and regression models were identified and explained. For the identification of FRC types and heating devices, all samples were accurately classified in repeated out-of-sample tests. For the prediction of heating temperature (°C), the R2 and RMSE were 0.96 and 20.11, respectively. These results coupled with the simplicity of the experimental processes indicate that LIBS is a promising candidate for rapid and on-site FRC analysis in fire protection and investigation.