基于傅里叶变换红外光谱与机器学习的防火涂料酸腐蚀周期预测研究  

Prediction of acid corrosion cycle of fire-retardant paint based on Fourier-transform infrared spectroscopy and machine learning

作　　者：岳鑫 刘培东 王今 王霁 YUE Xin;LIU Peidong;WANG Jin;WANG Ji(School of Fire Protection Engineering,China People’s Police University,Langfang 065000,China)

机构地区：[1]中国人民警察大学防火工程学院,河北廊坊065000

出　　处：《电镀与涂饰》2025年第1期37-44,共8页Electroplating & Finishing

摘　　要：[目的]准确预测在役防火涂料的耐火性能是消防安全评估工作中的重点和难点之一。[方法]借助傅里叶变换红外光谱仪,对不同酸腐蚀周期的膨胀型钢结构防火涂料试样进行测试,发现涂料中的主要成分会与盐酸溶液发生双向迁移并产生反应,酸腐蚀24 h时该反应达到峰值。利用主成分分析法对红外光谱数据进行降维处理的结果表明不同酸腐蚀周期试样存在可分性。联合使用S-G卷积平滑和标准正态变化(SNV)法对数据进行预处理,并对常用的析、LW-PLSC(局部加权偏最小二乘分类)、K-ELM(核极限学习机)、SVM(支持向量机)等4种分类方法进行评估。[结果]测试集平均分类准确度高达94.25%,最优模型分类准确率可达100%。[结论]将红外光谱技术与机器学习结合起来能快速、准确、定量地评估钢结构防火涂料的实验室酸腐蚀周期,为实现防火涂料实际耐火性能的预测提供可行的新方法。[Objective]Accurately predicting the fire resistance of in-service fireproof coatings is one of the key and difficult problems in fire safety assessment.[Method]The samples of intumescent fireproof coatings for steel structures subjected to different cycles of acid corrosion were analyzed by using Fourier-transform infrared spectrometer.It was found that the main components in the paint migrated toward and reacted with the hydrochloric acid solution,which also migrated inside the coating,and the reaction peaked when the coating was corroded by hydrochloric acid for 24 hours.The dimensionality of infrared spectral data was reduced by principal component analysis,which showed that the samples subjected to different cycles of acid corrosion were separable.The S-G convolution smoothing and SNV method were used to preprocess the data,and four commonly used classification methods i.e.PLS-DA(partial least squares-discriminant analysis),LW-PLSC(locally weighted–partial least square classification),K-ELM(kernel extreme learning machine),and SVM(support vector machine)were evaluated.[Result]The average classification accuracy of the test set was as high as 94.25%,and the classification accuracy of the optimized model was up to 100%.[Conclusion]The combination of infrared spectroscopic analysis technology with machine learning can quickly,accurately,and quantitatively evaluate the cycle number of acid corrosion in laboratory corresponding to the fire-retardant coating for steel structures,providing a new and feasible method for predicting the actual fire resistance of fire-retardant coatings.

关 键 词：防火涂料 傅里叶变换红外光谱 机器学习 酸腐蚀 周期预测 

分 类 号：TQ630.72[化学工程—精细化工]