Research On On-line Detection Of Quality Markers Content In The Extraction Process Of Xiaoer Xiaoji Zhike Oral Liquid

The modernization of traditional Chinese medicine(TCM)industry puts forward higher requirements on the quality of Chinese medicine products.The extraction process,as a key basic link in the production of TCM,is directly related to the final quality of the products.Due to the complexity of TCM system,there is still a lack of effective online quality detection means in the TCM extraction production at current stage,resulting in serious lags in production control and poor quality consistency between batches.On-line detection of extraction process is the key to achieve process control and improve drug quality.Based on Near Infrared Spectroscopy(NIRS)and soft-sensing technologies,this paper takes the extraction process of Xiaoer Xiaoji Zhike Oral Liquid as the research object,and studies the on-line detection method of the active ingredient content in extracts to solve the problem of real-time quality monitoring during the extraction process.It is of important theoretical significance and application value for ensuring the stability of TCM product quality.Aiming at the problems of poor spectral quality and inaccurate analysis results caused by liquid impurities,bubbles,temperature and other factors,an on-line NIRS analysis platform suitable for the extraction process of TCM was designed.A secondary filtering device was set in the external-circulation bypass sampling system to filter the impurities in the medicine liquid.The temperature stability of the medicine liquid when collecting spectrums was controlled,in a way of cooling first and then fine tuning,by a cooling device and a temperature control device.Aiming at the problem of bubbles,a detection flow cell with unique structure was designed to eliminate the influence of bubbles.The optical path is 2mm,and a temperature sensor was installed to collect spectrum and temperature data at the same time,to improve the stability of NIRS analysis by a way of temperature compensation.Aiming at the problems of random noise,baseline drift and information redundancy in the raw NIRS,the spectral data were preprocessed by baseline correction and normalization,and the characteristic spectral variables were selected by ridge regression method.Finally,off-line NIRS models were established using partial least square method based on the processed spectral data.The results showed that the generalization performance of the models were good.For simephrine,forsythin A,naringin,neohesperidin four quality markers(Q-markers),the coefficient of determination(R2)reached 0.9963,0.9962,0.9983 and 0.9983 respectively.The on-line NIRS models were further constructed based on the off-line models,and the on-line model of each Q-marker was optimized by using multi pretreatments and temperature compensation methods.In the on-line detection test,the R2 of the models for synephrine,forsythin A,naringin and neohesperidin reached 0.9726,0.9719,0.9844 and 0.9848 respectively,which proved the reliability of the models for on-line detection.The content data of each Q-marker was obtained by on-line NIRS detection,and softsensing models were established with extraction process parameters data to achieve on-line detection of the liquid quality based on the process parameters.To address the problem of unknown extraction process mechanism,correlation analysis was performed between different process parameters and the Q-markers content.Further,the derived process variable was constructed by parameter combination,and finally the key extraction process variables were identified by feature selection for model training.To address the limitation of single algorithm modeling,soft-sensing models were established,fusing five regression algorithms,KNN,RF,GBDT,LightGBM,and GRNN,based on stacking ensemble strategy,and on-line detection test was conducted.The results showed that the RMSEP of the models for the four Q-markers of sinaprin,allantoin A,naringenin and neohesperidin were 0.0037 mg/mL,0.0199 mg/mL,0.0679 mg/mL and 0.0574 mg/mL,and the R2 reached 0.9773,0.9907,0.9841 and 0.9894,respectively.The detection results were good,which proved the feasibility of applying softsensing technology to on-line detect the quality change of Xiaoer Xiaoji Zhike Oral Liquid in the extraction process.It provides a new idea for the development of economical and reliable on-line detection tool for TCM pharmaceutical,removing the dependence on spectroscopic instruments.