Small Sample Transfer Modeling of Near‐Infrared Spectra for Similar Resin‐Based Carbon Fiber Prepregs via Wavelength Selection and Pretreatment Methods

ABSTRACT Due to the different spectral response distributions of various resin matrices, the established near‐infrared spectroscopy for prepregs cannot be directly applied to prepregs with other resin matrices. Therefore, this article adopts a method combining preprocessing methods with the Competitive Adaptive Reweighted Sampling (CARS) wavelength selection algorithm to calibrate the near‐infrared spectra of carbon fiber (CF) prepreg tapes with similar resin matrix structures. The effects of preprocessing methods such as first derivative (1stDer), second derivative (2ndDer), standard normal variate transformation (SNV), multiplicative scatter correction (MSC), SNV + 1stDer, and MSC + 1stDer on the predictive performance of the Partial Least Squares (PLS) calibration model were investigated. The experimental results showed that the optimal monitoring models for the three types of prepregs had R ² values higher than 0.9 and Residual Prediction Deviations (RPD) greater than 2.5, demonstrating the models' excellent predictive capabilities. This study provides new ideas for solving redundant modeling problems.