On-line Quality Control Of A Chinese Herbal Liquid Tonic Using Near Infrared Spectroscopy

Near infrared(NIR)spectroscopy is currently the most widely used process analytical technology(PAT)in the pharmaceutical industry.Its application in the quality control of traditional Chinese medicine has also attracted much attention in recent years.Traditional Chinese medicine is often very complex in composition,and the production processes are distinctive from that for pharmaceutical chemicals.As a result,the application of NIR spectroscopy in the traditional Chinese medicine domain faces major challenges,in particular in chemometric model development.The research conducted in this paper was aimed to address the challenges.The research is summarized below:Firstly,a kind of oral liquid of Chinese herbal medicine collected from a pharmaceutical company as the research object,a system engineering method combined GA-PLS algorithm were used to establish the NIR models of measuring the content of polysaccharides and anthraquinone which were important quantity index in the oral liquid.Since these two indexes have no obvious characteristic absorption peaks in the near infrared spectral region,the performance of the model made by traditional methods is poor.Different kinds of preprocessing methods,variable selection methods combined the environment influence of production were used to make models with high accuracy and good robustness by chemometric method.The pretreatment method of polysaccharide model was detrend(linear),and the anthraquinone model's is the first order derivative with the width of window 3.The variable selection method of both is GA algorithm.The relative error between prediction values and chemical values of the calibration data,validation data and unknown data collected from production line is controlled within 10%,which has met the quality control standard of the company.At the meantime,the relative prediction error of polysaccharide Chinese herbal medicine was the lowest in current literatures.Secondly,the universality of the model in practical application was studied.1.Study of the universality of the models between parallel probes.The first model was built using the data collected from probe 1#.The result of prediction of data collected from probe 2#by first model was not good.Combined with PCA and T~2-Q statistical analysis,it is concluded that the universality of models among the parallel probes is poor.Two solutions are proposed.(1)Data collected from the single probe are used to set up the respective single-probe model,and then the single probe model will be applied to the corresponding probe.This method can accurately predict the data,with the disadvantage of a large amount of data,which will cost a huge amount of manpower and material resources.(2)Data collected from different parallel probes were combined used as the modeling data to establish a double(multi-)probe model,and then the model was applied to each probe.This solution can accurately predict data,and can save a lot of manpower and material resources compared with the solution(1).2.Study of the universality of the model when probe is replaced.Using the data collected from the original probe(the collection period is long enough and 1 year is suggested)to make model,which was used to predict the data collected from the new probe,and the result of prediction was good enough.However,using the data collected from the original probe(the collection period is 3 months only)to make model,the result of prediction of data collected from new probe was very poor.Combined with PCA and T~2-Q statistical analysis,it was concluded that,if the collection period of calibration data collected from original probe is long enough(1 year is suggested),the universality of original model will better when probe is replaced.Thirdly,Study of the model transform when the product and the quality index are constant while other conditions such as production site,equipment,the way of sampling of probe are changed.DS algorithm was used to correct the spectrum collected from the new condition,making them equivalent to the spectrum collected from the original condition.The result of prediction with the original model was good enough,with the relative error controlled within 10%.The transform of model solves the important problem of the near infrared spectrum model in the quality control of Chinese herbal medicine,avoiding the series of problems caused by the model's reestablished,and saving a lot of manpower and material resources.