A Online NIR Sensor for the Pilot-Scale Extraction Process in Fructus Aurantii Coupled with Single and Ensemble Methods

Abstract

Model performance of the partial least squares method (PLS) alone and bagging-PLS was investigated in online near-infrared (NIR) sensor monitoring of pilot-scale extraction process in Fructus aurantii. High-performance liquid chromatography (HPLC) was used as a reference method to identify the active pharmaceutical ingredients: naringin, hesperidin and neohesperidin. Several preprocessing methods and synergy interval partial least squares (SiPLS) and moving window partial least squares (MWPLS) variable selection methods were compared. Single quantification models (PLS) and ensemble methods combined with partial least squares (bagging-PLS) were developed for quantitative analysis of naringin, hesperidin and neohesperidin. SiPLS was compared to SiPLS combined with bagging-PLS. Final results showed the root mean square error of prediction (RMSEP) of bagging-PLS to be lower than that of PLS regression alone. For this reason, an ensemble method of online NIR sensor is here proposed as a means of monitoring the pilot-scale extraction process in Fructus aurantii, which may also constitute a suitable strategy for online NIR monitoring of CHM.

Process quality control of the manufacturing of Chinese Materia Medica by process analysis technology

“…16 Ten spectral pre-processing methods were compared in eliminating interference information. 17 Eight preprocessing methods and several variable selection methods were also optimized in monitoring the pilot-scale extraction process of Fructus aurantii, 18 respectively. Based on these optimizations of modeling parameters, robust on-line NIR models were established to monitor the quality in CMM manufacturing.…”

New sensor technologies in quality evaluation of Chinese materia medica: 2010–2015

“…Wu et al 38 , 39 also found the two systems were suitable as the process analytical technology to understand ethanol precipitation process of water extract of L. japonica . Lastly, Pan et al 40 investigated an ensemble method as the means of monitoring the pilot-scale extraction process in Aurantii Fructus, which may also constitute a suitable strategy for monitoring of CMM.…”

Detection of Adulteration in Infant Formula Based on Ensemble Convolutional Neural Network and Near-Infrared Spectroscopy

“…In the former studies, most of the ensemble learning approaches for NIR spectra processing adopted PLS, which is a classical linear regressor, as the base estimators [ 41 , 42 , 43 ]. Popular ensemble strategies, including bagging [ 44 ], boosting [ 45 ] and stacking [ 46 ], were employed to integrate and improve the predicting results of the basic PLS models. Zhou et al compared bagging-PLS with boosting-PLS in online near-infrared models for monitoring active pharmaceutical ingredients of Chinese Medicine [ 47 ].…”

“…Random selection of variable subspaces [ 39 , 51 ] is a frequently-used strategy to promote the diversity, however, may result in some poorly performing sub-models, given the selection of some inefficient wavelength combinations. On the other hand, several wavelength selection methods, including SPA [ 52 ], uninformative variable elimination (UVE) [ 41 ] and synergy interval partial least squares algorithm (SiPLS) [ 44 ] were employed to solve the redundancy and collinearity problems and improve the accuracy of sub-models. However, sharing a wavelength combination for all sub-models may harm the diversity of the sub-model pool.…”