PLS-R calibration models for fast prediction of oxidation stability of commercial biodiesel by spectroscopic techniques

Abstract

Oxidative stability is an essential quality metric for biodiesel, widely assessed using the accelerated oxidation method with Rancimat®, which requires 6 or 13 h to comply with current regulations. In this study, we propose predicting the oxidative stability of commercial biodiesel using direct and rapid analytical techniques combined with partial least squares regression (PLS-R) and ¹H NMR and FTIR-ATR spectroscopy. Statistical parameters that assess the quality of the models indicate that both approaches demonstrated good performance, with residual prediction deviation (RPD) values exceeding 1.6 and relative standard deviation (RSD) below 9.3 %. Analytical figures of merit (AFOM) also yielded favorable outcomes, with limits of quantification (LOQ) and detection (LOD) consistent with the ranges studied. For NMR data, the zgig pulse sequence (which drastically reduces ¹H–¹³C satellite signals) proved more suitable for modeling. The main compounds contributing to the NMR models’ performance were hydroperoxides and derivatives. Several significant vibrational bands contributed to the model’s performance for the compact FTIR. More predictive models could be achieved using either samples from a single feedstock or without adding antioxidants. However, this is impractical once biodiesel production can involve diverse feedstocks and antioxidant additives. Including biodiesel from different feedstocks and with antioxidants increased the dataset’s variability, leading to more realistic models with broader applicability. Compared to the standard Rancimat® method, the NMR with PLS-R considerably reduces the analysis time by about 40 times and FTIR by approximately 220 times. However, building and validating predictive models is laborious and challenging and requires knowledge of data analysis.