Application of surface enhanced Raman scattering and competitive adaptive reweighted sampling on detecting furfural dissolved in transformer oil

Abstract

Detecting the dissolving furfural in mineral oil is an essential technical method to evaluate the ageing condition of oil-paper insulation and the degradation of mechanical properties. Compared with the traditional detection method, Raman spectroscopy is obviously convenient and timesaving in operation. This study explored the method of applying surface enhanced Raman scattering (SERS) on quantitative analysis of the furfural dissolved in oil. Oil solution with different concentration of furfural were prepared and calibrated by high-performance liquid chromatography. Confocal laser Raman spectroscopy (CLRS) and SERS technology were employed to acquire Raman spectral data. Monte Carlo cross validation (MCCV) was used to eliminate the outliers in sample set, then competitive adaptive reweighted sampling (CARS) was developed to select an optimal combination of informative variables that most reflect the chemical properties of concern. Based on selected Raman spectral features, support vector machine (SVM) combined with particle swarm algorithm (PSO) was used to set up a furfural quantitative analysis model. Finally, the generalization ability and prediction precision of the established method were verified by the samples made in lab. In summary, a new spectral method is proposed to quickly detect furfural in oil, which lays a foundation for evaluating the ageing of oil-paper insulation in oil immersed electrical equipment.