文献类型：期刊文献

英文题名：Rapid Decomposition of Epoxy Resins via Raman Spectrometry in Combination with Machine Learning Algorithms

作者：Guan, Qiyuan[1,2] Guo, Kang[1,2] Tan, Weihong[1,2] Zhou, Yonghong[1,2]

第一作者：Guan, Qiyuan

机构：[1] Institute of Chemical Industry of Forest Products, Chinese Academy of Forestry, National Engineering Lab for Biomass Chemical Utilization, Key and Open Lab of Forest Chemical Engineering, SFA, Key Lab of Biomass Energy Sources and Materials, Nanjing, 210042, China; [2] Co-Innovation Center of Efficient Processing and Utilization of Forest Resources, Nanjing Forestry University, Nanjing, 210037, China

年份：2019

卷号：4

期号：2

起止页码：130-134

外文期刊名：Journal of Bioresources and Bioproducts

收录：EI(收录号：20220411526431);Scopus(收录号：2-s2.0-85115854798)

语种：英文

外文关键词：Adhesives - Curing - Decision trees - Least squares approximations - Learning algorithms - Mixing - Machine learning - Spectrometry - Mean square error

摘要：Epoxy resins are a group of important materials that have been used everywhere, and development of new materials of this kind with optimal mechanical properties from either bio-resources or industrial precursors has drawn great focus from scientists and engineers. By reacting different kinds of epoxy adhesives and curatives, massive kinds of epoxy resins with different characteristics are produced. Determination of original mixing ratio of epoxy adhesives and corresponding curatives of their curing products is useful in controlling and examining these materials. Here in this work, we described an efficient method based on Raman spectrometry and machine learning algorithms for rapid molar composition determination of epoxy resins. Original mixing ratio of epoxy adhesives and curatives could be calculated simply via Raman spectra of the products. Raman spectral data scanned during curing procedure was fed to random forest (RF) classification to calculate weights of Raman shift features and reduce data dimensionality, then spectral data of selected features were processed by partial least squares regression (PLSR) for model training and composition ratio determination. It turned out that ratio predictions of our model fit well to their actual values, with a coefficient of determination (R2) of 0.9926, and a root mean squared error (RMSE) of 0.0305. ? 2019 Nanjing Forestry University