Combining near infrared spectroscopy with predictive model and expertise to monitor herb extraction processes

Albeit extensively utilized, herb extraction process (HEP) is hard to be monitored because of its batch nature and the fluctuating quality of raw materials. Process analytical tools like near infrared spectroscopy (NIRS) can offer nondestructive examinations and collect abundant data of the process, which in principle contain the information about the quality of both the product and the process itself. However, extra effort is often required for the data mining of such process measurements, and extracting knowledge of the quality of process can be even harder. In this study, we take the extraction process of licorice as a typical HEP instance, and combine NIRS with classical partial least squared regression (PLSR) and expertise for its on-line monitoring. We show that our scheme effectively extracts information with clear physical meanings, through which we can even uncover the process fault that does not induce evident abnormalities in the product quality. Moreover, the constructed model can continuously evolve with more process data from daily operations, and the idea of the whole framework can be directly generalized to other HEP.