Residence Time Distribution characterization in a Continuous Manufacturing tableting line using PCA and PLS-DA modeling

In this work, a new approach to characterize the residence time distribution (RTD) using supervised and unsupervised learning techniques was investigated. The NIR spectral data from the feed frame was used to capture the dynamic response of a continuous tableting line. The data-driven approaches investigated in this study using MPCA and PLS-DA were able to successfully capture the main variation of the step change performed during the RTD experiments. Its good performance compared with the traditional RTD characterization methodology offers the potential to be used to predict the RTD parameters. This could contribute to making the RTD determination procedure more efficient and reduce the need to make offline measurements.